- Insérer la page de garde de l école doctorale à la place de cette page-

Transcription

1 Hybrid Emergy - Life Cycle Assessment: methodological developments and application to potable water production - Insérer la page de garde de l école doctorale à la place de cette page- 1

2

3 Remerciements Le travail développé lors d une thèse est en grande partie personnel, et autonome. Cependant il n aurait pas été possible sans le soutien de nombreuses personnes, que je tiens à remercier chaleureusement. Je remercie en particulier mes deux superviseurs, Pr Ligia Tiruta-Barna et Dr Enrico Benetto, pour m avoir proposé l opportunité de réaliser un tel projet, et pour leur aide et leur disponibilité permanente durant ces trois années passées à leurs côtés. J espère avoir été un bon élève. Je veux aussi remercier Dr Benedetto Rugani, pour son étroite collaboration et ses nombreux conseils, ainsi que pour les discussions passionnantes que nous avons pu avoir. Je te souhaite une carrière riche et épanouissante. Merci également à l ensemble de l équipe du CRTE et son directeur, Dr Paul Schosseler, pour leur accueil, leur convivialité et leur implication dans cette thèse. Je garderai d excellents souvenirs de l équipe Environmental Evaluation and Management et des moments passés ensemble. Un merci tout particulier à Daryna Pansiuk pour m avoir hébergé et pour son amitié, son soutien, si précieux lors des moments difficiles. J espère que ta thèse sera aussi enrichissante que la mienne. Je souhaite aussi remercier les membres permanents, doctorants et stagiaires du Hall GPE de l INSA de Toulouse, pour leur accueil, leurs conseils et leurs nombreux coups de mains sur les taches administratives. Merci à mes parents, ma famille, mes amis, qui ont toujours su me soutenir en m apportant la motivation nécessaire pour mener ce projet à bien. Enfin, je souhaite exprimer ma plus grande gratitude envers H.T. Odum et à l ensemble de la communauté des émergistes, pour m avoir apporté une vision éclairante et positive de ce qu on appelle développement durable. 3

4

5 Summary Environmental assessment is a scientific discipline essential for the construction of a sustainable society. The most commonly used tool is Life Cycle Assessment (LCA), in which the impact related to resource depletion is often assessed considering global available stocks, extraction rates and relative value of the resource for the user. However, this approach cannot be applied as such to renewable resources such as freshwater and ecosystem services, which are not, by definition, stocks. In contrast, Emergy evaluation (EME) determines the value of a resource (renewable or not) from the description of the natural mechanisms that produce it. Therefore, LCA and EME present an interesting hybridation potential, the former providing a detailed representation of a product's lifecycle, while the latter allows accounting for the value of both renewable and nonrenewable resources with the same rationale. In this PhD, a recently-developed hybrid framework is applied to four potable water production sites, in order to analyze the added value and current limitations of using detailed LCA databases in EME. Suggested improvements include the adaptation of emergy indicators to the hybrid framework, and freshwater resource characterization using Geographic Information Systems (GIS). Elaborated computational tools such as GIS and dynamic modeling could provide great benefits to current environmental assessment methods. The feasibility of using them is demonstrated in this study by validating the conceptual approach. Résumé L évaluation environnementale est une discipline scientifique indispensable à la construction d une société durable. L outil le plus communément utilisé est l Analyse du Cycle de Vie, dans lequel l impact lié à l épuisement des ressources y est fréquemment évalué à partir des stocks mondiaux disponibles, les taux d extraction et leur valeur relative pour l utilisateur. Cette approche n est cependant pas applicable aux ressources renouvelables, comme l eau douce et les services écosystémiques, qui, par définition, ne sont pas des stocks. L évaluation émergétique (EME), en revanche, détermine la valeur d une ressource (renouvelables ou non) à partir de la description des mécanismes naturels qui la génèrent. L ACV et l EME présentent donc un intéressant potentiel d hybridation, la première fournissant une représentation détaillée des étapes du cycle de vie d un produit, tandis que la seconde permet de comptabiliser la valeur des ressources renouvelables et fossiles avec une unité de mesure commune. Dans cette thèse, une méthodologie hybride récemment développée est appliquée à quatre usines de production d eau potable, afin d analyser la valeur ajoutée et les limites actuelles du modèle hybride. Les améliorations proposées incluent l adaptation des indicateurs émergétiques au modèle hybride et la caractérisation des ressources d eau douce grâce à l emploi de Systèmes d Information Géographique (SIG). Les outils informatiques élaborés, tels que la SIG et la modélisation dynamique, permettraient d enrichir sensiblement les méthodes actuelles d évaluation environnementale. La faisabilité de cette opportunité est démontrée dans cette étude par la validation de l approche conceptuelle. 5

6 Résumé long (français) L humanité a franchi les limites physiques de la planète (Rockström et al. 2009). Il est désormais reconnu que la Terre ne pourrait pas supporter une population de 7 milliards d êtres humains avec les conditions de vie actuelle des pays occidentaux, ce qui nous met face à de nouveaux challenges historiques. On pourrait se reprocher de «détruire» la planète, et se soucier des conséquences potentielles d une rupture de nos systèmes socio-économiques. Mais nous pouvons aussi considérer cette situation comme une source de motivation supplémentaire pour réfléchir à des systèmes de gouvernance et une gestion de notre planète plus durables (Folke et al. 2011; Steffen et al. 2011). Un développement sociétal environnementalement et économiquement durable implique de préserver l intégrité des systèmes naturels, uniques supports in fine des sociétés humaines, tout en assurant un développement économique uniquement à partir de ressources renouvelables (Moldan et al. 2012). Les deux problèmes sont fortement liés l un à l autre, car les écosystèmes fournissent à la fois les ressources renouvelables et des conditions de vie convenables à l Homme, sous la forme de ce que l on appelle communément les Services Ecosystémiques (Costanza et al. 1997; de Groot et al. 2002; MEA 2005; TEEB 2010). Ils constituent le capital naturel, dont la valeur est habituellement estimée à partir d outils économiques tels que l évaluation de marché, qui attribuent un prix à la Nature en fonction de son utilité pour l Humanité. Cependant, une approche complémentaire est nécessaire, basée sur les lois de la nature et l observation des mécanismes environnementaux à toutes les échelles d espace et de temps, afin de mieux appréhender la contribution des Services Ecosystémiques et des ressources naturelles pour le développement humain et analyser les conséquences des changements environnementaux d origine anthropique (Rees and Wackernagel 1996; Costanza et al. 1997; Odum and Odum 2000; Markandya and Pedroso-Galinato 2007). Une comptabilité environnementale formelle permettrait de compléter l information économique qui ne prend pas en compte ces «externalités», ce qui est une des origines possibles des crises économiques actuelles. Dans cette étude, deux visions de la comptabilité environnementale sont étudiées. Elles offrent des points de vue complémentaires, en particulier concernant la manière dont les systèmes humains sont décrits et comment les ressources naturelles sont caractérisées. D une part, l Analyse du Cycle de Vie (ACV) permet d évaluer les impacts environnementaux d un produit ou d un service, à partir d une description détaillée du réseau de procédés technologiques intervenant dans son cycle de vie, et des mécanismes environnementaux générant les impacts. D autre part, l Evaluation Emergétique (EEM) considère que les activités humaines sont dépendantes de leur environnement naturel global, et estime la contribution des processus naturels (en emjoules équivalent solaire, ou sej, défini comme la quantité totale d éxergie [solaire] directement et indirectement utilisée par les mécanismes qui participent à la formation d une ressource) nécessaires à une activité pour son fonctionnement. ACV et EEM partagent le même objectif de fournir des outils opérationnels de comptabilité environnementale, mais reposent sur des raisonnements différents : l ACV permet d estimer les impacts générés par les activités humaines sur trois Domaines de Protection (écosystèmes, ressources, santé humaine), tandis que l EEM cherche à étudier dans quelle mesure une activité humaine est dépendante du système Terre et des biens et services qu il lui fournit. Une telle différence entre ces deux outils induit une spécificité des objectifs, périmètres d étude, méthodes de quantification, règles de calcul, données d inventaires et interprétations pour la prise de décision, résumées dans le Tableau 1. 6

7 Tableau 1: Complémentarités entre ACV et EEM. Raisonnement Objectifs et périmètre Algèbre Impact(s) comptabilisé(s) Limites actuelles des données prises en compte (inventaire) Aide à la décision ACV Les activités humaines génèrent un impact sur des Domaines de Protection Analyser les impacts lies à l utilisation de matériaux et d énergie dans la technosphère Orienté utilisateurs (les impacts de coproduits sont répartis sur les utilisateurs) Dommages sur la santé humaine, les écosystèmes et l épuisement des ressources Travail humain Services écosystémiques Comparer des systèmes technologiques délivrant la même unité fonctionnelle EEM Le système Terre est un support pour les activités humaines Estimer le travail des processus naturels pour la formation des ressources utilisées Orienté donneur (toutes les ressources sont utilisées pour produire chaque coproduit) Energie [solaire] totale utilisée directement et indirectement Effets de la pollution Détails de la technosphère Fournir des indicateurs de performance environnementale universels Bien que l EEM et l ACV sont basés sur des points de vue, des périmètres et cadres mathématiques différents, ils peuvent aussi être considérés comme complémentaires : étudier leur intégration mutuelle peut donc s avérer être un projet de recherche précieux pour dépasser leurs limites respectives. Par exemple, afin d élargir l approche «technocentrée» de l ACV, les Valeurs Emergétiques Unitaires (VEU), dérivées de l analyse émergétique de la formation des ressources naturelles et définies comme la valeur émergétique d une ressource par unité physique correspondante (calculée en emjoules solaires (sej) par unité de ressource, par ex. sej/m 3 ), ont été proposés récemment pour caractériser les ressources dans le cadre de l ACV (Rugani 2010; Zhang et al. 2010; Ingwersen 2011; Rugani et al. 2013; Raugei et al. 2014). Cependant cette approche est rendue difficile par les différences des raisonnements soutenant chaque méthode : alors que l ACV est basée sur la notion de poids environnemental, l EEM repose sur la notion d appropriation du travail environnemental. Par conséquent l ACV quantifie des impacts que les décideurs chercheront à minimiser, tandis que la comptabilité émergétique fournit un résultat ayant seulement une signification physique qui exclut tout jugement de valeur de la performance environnementale. C est pourquoi l EEM se conclut par le calcul d indicateurs facilitant l interprétation des résultats pour les décideurs. En revanche, l EEM présente une limite importante liée au faible niveau de détail traditionnellement utilisé pour calculer la VEU des produits et services fournis par d autres activités humaines, ce qui peut entraîner d importants biais dans les résultats. Rugani et Benetto (2012) ont proposé une stratégie pour améliorer l EEM grâce à l utilisation des bases de données utilisées en ACV, permettant une vision plus précise du réseau des procédés technologiques impliqués dans la fabrication de produits humains. L inventaire des ressources élémentaires résultant de cette analyse peut être ensuite converti en termes d émergie, en utilisant un jeu de données de VEU consistant. Le principal défi est l application des règles de calcul spécifiques à l émergie aux bases de données d ACV. Dans ce but, le logiciel SCALE (Marvuglia et al. 2013) a été mis au point pour calculer la VEU des produits humains à partir d un réseau de procédés de la technosphère tout en respectant les règles de calcul de l émergie. Ce logiciel, cependant, n avait pas encore été utilisé dans le cadre d une étude de cas approfondie. Dans cette thèse de doctorat, la production d eau potable a servi d étude de cas à la première véritable application de SCALE. Une importante question résultant de ce choix concernait les outils et données disponibles pour caractériser la ressource naturelle utilisée dans ce procédé, c est-à-dire l eau douce, car celle-ci est étroitement connectée à la production à court et long termes des Services Ecosystémiques, et fortement dépendante des conditions locales. Par conséquent, trois questions de recherche ont été abordées : 7

8 Quelles sont les principaux défis pour améliorer la robustesse de l EEM? Quelle est la valeur ajoutée d un cadre hybride émergie - cycle de vie? Comment prendre en compte les hétérogénéités spatiales et temporelles des ressources renouvelables et Services Ecosystémiques dans la comptabilité environnementale (à la fois en ACV et en EEM)? Des réponses ont été proposées de manière aussi rigoureuse et claire que possible, en termes de démarches conceptuelles et d observations, d argumentations scientifiques et résultats de recherche, et de comparaisons avec l état de l art, à travers 6 publications scientifiques dans des journaux à comité de lecture ; chacun d eux correspond à un chapitre de ce manuscrit. Après avoir contextualisé les problèmes liés à la comptabilité environnementale dans la partie introductive de la thèse (chapitre 1), le chapitre 2 donne plus de précisions sur la procédure d intégration des modèles et jeux de données de l ACV dans l EEM. L émergie, grâce à son unique capacité à traduire dans une seule métrique la mémoire de l exergie (c est-à-dire «l énergie disponible» pour opérer un travail) de la géobiosphère, support de tout système technologique ou naturel, a le potentiel d offrir une nouvelle perspective à l analyse environnementale et l aider à mieux remplir son rôle d aide à la décision. La littérature scientifique a déjà montré les avantages potentiels d une approche hybride combinant EEM et ACV. En particulier, un calcul émergétique utilisant une base de données ACV de réseaux de procédés sous forme matricielle, permettrait d améliorer la fiabilité des évaluations émergétiques, et ainsi faciliter l applicabilité du concept de l émergie dans l aide à la décision. Ce chapitre identifie les principales limites des procédures actuelles de l EEM telles que : Le manque d une définition mathématique opérationnelle et universellement acceptée de l émergie, Le manque d outils automatisés assurant la fiabilité des études d EEM existantes et donc la construction d une base de données homogène, Le manque de standards pour définir précisément les frontières des systèmes étudiés, autorisant une description approximative et conduisant à des résultats subjectifs et peu rigoureux, Le manque d une définition robuste et d une analyse critique des indicateurs émergétiques. Les tentatives précédentes pour combiner EEM et ACV sont analysées, ainsi que les challenges à relever pour une intégration complète, comme illustré par la Figure 1. Ce chapitre indique les principaux obstacles à surmonter pour obtenir une intégration cohérente, mettant en relief les progrès effectués jusqu à présent dans cette direction concernant la théorie, la méthodologie, le développement de logiciel et les conséquences sur la recherche actuelle et les propositions des prochaines étapes. 8

9 H T Inventory flows and database Indicators Life Cycle Extraction S R N EGS Geobiosphere Society (technosphere) Energy Materials UEVs estimation and use Refinement Production Use End of life Emissions Donor-Value Algebra issue User-Value Figure 1: Conceptualisation des principales différences entre Emergie et Analyse du Cycle de Vie (ACV), pour évaluer une combinaison des deux approches et renforcer leur capacité à résoudre les problèmes de gestion de l environnement. La partie gauche (formes vertes) représente l approche émergétique du management environnemental. La partie droite (formes bleues) représente l approche ACV. Les éléments en rouge représentent les challenges liés à l hybridation. EGS = Services Ecosystémiques ; R = ressources locales Renouvelables ; N = ressource locales Non-renouvelables ; S = énergie de radiation Solaire ; H = chaleur issue du manteau terrestre ; T = énergie des marées. Les principaux défis méthodologiques à relever pour combiner ACV et EEM sont : i) une définition cohérente des frontières des systèmes étudiés ; ii) la précision (calcul rigoureux) des VEU ; iii) les difficultés mathématiques pour combiner des règles de calcul différentes ; iv) la pertinence des métriques et de l échelle de valeur, la transparence des indicateurs pour l aide à la décision et formulation des leviers d action. Dans nos dernières recherches ces aspects ont été étudiés avec des résultats prometteurs pour des développements futurs (voir Rugani and Benetto 2012; Rugani et al. 2012; Marvuglia et al. 2013; Tiruta-Barna and Benetto 2013). Concernant le formalisme et les méthodes de calcul, une conceptualisation et une démonstration formelles des règles de calcul de l émergie a été proposée (Tiruta-Barna and Benetto, 2013) ; un algorithme a été développé et intégré dans un logiciel pour calculer la VEU des produits technologiques à partir des bases de données ACV de procédés et de l application rigoureuse des règles de calcul émergétique (Marvuglia et al. 2013) ; une méthode alternative (définie comme «bottom-up»), basée sur une approche matricielle, a été proposée pour calculer la VEU des ressources naturelles, des Services Ecosystémiques et des produits de la technosphère (Rugani and Benetto, 2012). Bien qu encore à ses débuts, elle permettrait d améliorer significativement la précision des VEU. Cette méthode pourrait permettre notamment de profiter des avancées de la modélisation en écologie pour affiner les calculs de VEU. La définition des frontières du système étudié est sujette à évolution à la fois en ACV et en EEM : les études prospectives actuelles montrent que l inclusion du travail humain et/ou des SE en ACV aboutiront à une définition des inventaires plus transparente et complète (Rugani et al. 2012). Un cadre de travail homogène et standardisé se développe de pair avec le développement d outil mathématiques modernes et adaptés, facilitant ainsi le calcul de comptabilité émergétique (voir par ex. Arbault et al. 2014a, c.à.d. chapitre 4). Outre les efforts continus pour améliorer le calcul des VEU, il est nécessaire d améliorer la transparence du calcul des indicateurs émergétiques et faciliter leur interprétation. En effet, seule une acceptation unanime de leur définition et de leur directionalité, combinée à un calcul 9

10 rigoureux et automatisé, peut favoriser l adoption de ces indicateurs par les décideurs (voir par ex. Arbault et al. 2014b, c.à.d. chapitre 5). Le contenu du chapitre 2 a été publié dans le Journal of Environmental Accounting and Management (Arbault et al. 2013a). Le chapitre 3 est une application pratique de la méthodologie traditionnelle de l EEM. L applicabilité et la robustesse du cadre de l émergie est examinée à travers une étude de cas composée de quatre usines de production d eau potable. D après l analyse, la VEU moyenne de l eau potable en sortie d usine est de 1,06 (±0,15) E12 sej/m 3 ; cette valeur est en accord avec la littérature scientifique existante. Les intrants humains comme les produits chimiques, l électricité, le travail humain et les servies, et les matériaux utilisés pour l infrastructure sont des contributeurs importants de la VEU de l eau potable. Ce résultat est caractéristique des activités industrielles qui reposent sur une seule ressource locale renouvelable. Une conclusion générale est que l utilisation d une source d eau douce moins polluée rend son traitement moins lourd et donc diminue la VEU du produit sortant (l eau potable). De plus, les procédés de traitement basés sur l électricité semblent plus performants du point de vue environnemental que les procédés basés sur l utilisation de produits chimiques, du fait du caractère non-renouvelable des ressources naturelles utilisées pour produire ces composés chimiques. Ce chapitre illustre la relativement faible précision de l EEM traditionnelle dans la comptabilité des intrants humains, en comparaison avec l ACV. Il met également en relief la complémentarité des deux méthodes de comptabilité environnementale. Par exemple, l EEM permet une analyse plus holistique des ressources utilisées, incluant à la fois la contribution des ressources locales naturelles (ici, l eau douce), celle des produits issus de la technosphère et celle du travail humain et des services. Les indicateurs émergétiques se montrent utiles pour effectuer l analyse comparative de la performance écologique d une large gamme d activités humaines. L ACV, en revanche évalue de manière automatisée les impacts du cycle de vie de l activité sur la santé humaine, les écosystèmes et l épuisement des ressources non renouvelables. Cette étude illustre dans quelle mesure l EEM pourrait fortement bénéficier de la représentation détaillée du cycle de vie des intrants humains, dans un cadre hybride qui permettrait une procédure de comptabilité plus précise. Le contenu du chapitre 3 a été publié dans le journal Ecological Engineering (Arbault et al. 2013b). Le chapitre 4 décrit la première application consistante du logiciel SCALE (Marvuglia et al. 2013), aux études de cas présentées dans Arbault et al. (2013b, c.à.d. chapitre 3). Il apporte une première étape dans l hybridation concrète entre ACV et EEM dans ce travail de thèse. Les résultats sont comparés avec ceux proposés par la procédure traditionnelle de l EEM et avec ceux 10

11 de la méthode Solar Energy Demand (SED, Rugani et al. 2011) 1. Ils montrent que la comptabilité émergétique des intrants humains avec SCALE est plus détaillée que dans le cadre conventionnel de l EEM (grâce à l utilisation des bases de données ACV des procédés technologiques), et permet d appliquer rigoureusement les règles de calcul de l émergie, ce qui n est pas le cas avec la méthode SED. L analyse se prolonge dans la comparaison quantitative de la VEU de chaque intrant issu de la technosphère calculé avec chaque méthode, démontrant, à l aide de tests approfondis sur les études de cas, la valeur ajoutée de SCALE pour améliorer la reproductibilité, la précision et l aspect holistique d une EEM. Ce chapitre illustre également le niveau de détails élevé de l analyse fournie par l application de SCALE, puisque la valeur émergétique des produits (matériels et énergétiques) issus de la technosphère peut être encore décomposée. Cependant, l application de SCALE est limitée par la qualité et l envergure des bases de données ACV utilisées. Par exemple (voir Tableau 2), il ne peut pas prendre en compte le travail humain et les Services Ecosystémiques puisqu ils sont actuellement exclus de ces bases de données. De plus, la version actuelle de SCALE ne permet pas de considérer la VEU spécifique des ressources locales utilisées (voir par ex. Arbault et al. 2014c, c.à.d. chapitre 6), et laisse de côté les impacts de la pollution sur la santé humaine et les écosystèmes. Des recommandations sur le périmètre et la précision d une comptabilité émergétique avec SCALE sont suggérées pour mieux appréhender les prochaines étapes dans le développement du logiciel, ainsi que des méthodes prospectives de quantification des Services Ecosystémiques et du travail humain. Tableau 2: Comparaison du périmètre pris en compte en EEM conventionnelle (EEM CONV ), EEM avec SCALE (EEM SCALE ), SED et ACV (méthode ReCiPe). Scope EEM CONV EEM SCALE SED ReCiPe (ACV) Valeur des ressources (perspective orientée donneur, c.à.d. énergie solaire utilisée dans la formation des ressources naturelles primaires) Services Ecosystémiques Partiel* Ressources locales Travail humain Niveau de détail dans les intrants humains Bas Haut Haut Haut Impacts de la pollution Partiel* Impacts sur l épuisement des ressources Partiel* * Différentes approches sont actuellement en cours de développement. Le contenu du chapitre 4 a été publié dans le journal Science of the Total Environment (Arbault et al. 2014a). 1 La méthode SED est une méthode d évaluation d impacts sur les ressources naturelles. Elle utilise le cadre de calcul de l ACV (différent de celui de la comptabilité émergétique) tout en caractérisant les ressources naturelles à partir de leur VEU. C est donc une méthode différente de l EEM traditionnelle et de la méthode hybride proposée par SCALE. 11

12 Pour approfondir l étude de la robustesse du processus d hybridation entre EEM et ACV, le chapitre 5 analyse les conséquences du cadre hybride émergie cycle de vie sur les indicateurs émergétiques, revendiqués comme des résultats de l EEM utiles pour guider les décideurs vers une meilleure performance environnementale. Le calcul du Emergy Sustainability Index (ESI), en particulier, semble largement consensuel au sein de la communauté des experts en émergie. Cependant plusieurs variantes existent dans la littérature scientifique, impliquant une confusion entre l analyse du site de production avec celui de son cycle de vie, et ainsi conduisant à des interprétations erronées ou une mauvaise compréhension du résultat fourni par l ESI. Ce chapitre propose en premier lieu une étude sémantique de deux variantes de chacun des deux composants de l ESI (le Emergy Yield Ratio (EYR), qui est un ratio de rendement émergétique, et le Environmental Loading Ratio (ELR), qui représente la pression de l activité sur les ressources naturelles), afin d améliorer la standardisation et la reproductibilité du calcul des indicateurs émergétiques. Il est ensuite montré que l ESI peut être défini correctement soit pour le site de production soit avec une perspective «cycle de vie», bien que plusieurs études de cas dans la littérature scientifique utilisent une version intermédiaire présentant une inconsistance dans le choix des frontières du système étudié. Une récente définition théorique de l EYR (Brown et al. 2012), qui adopte une perspective «cycle de vie», est ici rendue opérationnelle par le développement original d un algorithme prêt à être implémenté dans le logiciel SCALE. Dans le cas de la production d eau potable, nous démontrons également que les investissements d arrièreplan (c.à.d. les intrants humains utilisés dans des procédés technologiques indirectement impliqués dans la chaîne de production étudiée) peuvent grandement influencer la VEU des produits issus de la technosphère. Par conséquent, il est conseillé de comptabiliser ces intrant avec la perspective «cycle de vie», en particulier s ils sont des contributeurs importants (en termes émergétiques) au système étudié comme c est le cas dans la plupart des activités industrielles. Tandis que SCALE permet déjà un calcul automatisé de la valeur émergétique produits issus de la technosphère, nous recommandons d implémenter dans SCALE l algorithme additionnel présenté dans ce chapitre, afin de fournir aux utilisateurs de la comptabilité émergétique un outil totalement intégré. Concernant les usines de production d eau potable évaluées, les résultats montrent que toutes les versions étudiées de l ESI proposent le même classement des usines, en accord avec les études précédentes de ces usines (Arbault et al. 2013b, 2014a; Igos et al. 2013a, 2013b). Cependant, sa valeur absolue de l ESI varie, ce qui souligne l importance d une définition, formulation et calcul robustes et consensuels des indicateurs émergétiques. L ESI est enfin décomposé grâce à l utilisation des dérivées partielles, afin d étudier l influence de chaque catégorie d intrant, et permettant de formuler des recommandations génériques. Concernant les usines d eau potable, par exemple, il a été montré que remplacer les produits chimiques issus de ressources non-renouvelables par d autres produits issus de ressources renouvelables serait moins efficace que sélectionner les fournisseurs d électricité en fonction de leur politique d achats verts. De plus le remplacement des produits chimiques nécessiterait des changements importants dans le réseau de procédés du cycle de vie de l usine étudiée, et par conséquent une nouvelle modélisation de son cycle de vie. Cette analyse de sensibilité basée sur les dérivées partielles doit être restreinte à l identification des leviers potentiels d action, mais ne peut pas anticiper leurs conséquences. Ces multiples résultats démontrent à nouveau la valeur ajoutée de l évaluation hybride émergie cycle de vie (et sa complémentarité avec le cadre conventionnel) pour identifier des actions potentielles spécifiques permettant d améliorer la performance environnementale des activités humaines. Nous avons la conviction qu adopter une perspective «cycle de vie» dans l évaluation 12

13 émergétique apporterait un meilleur niveau de reproductibilité et de précision ainsi que l utilisation d outils fournissant des indications claires pour l aide à la décision, permettant ainsi de mieux disséminer le concept de l émergie et favoriser son utilisation au sein des industries et d un plus grand nombre de parties prenantes. Le contenu du chapitre 5 a été publié dans le journal Ecological Indicators (Arbault et al. 2014b). Afin de cibler les problèmes liés à la précision et la reproductibilité du calcul des VEU des ressources naturelles, le chapitre 6 explore l utilisation de logiciels de Systèmes d Information Géographique (SIG) pour développer une base de données spatialisée de l émergie des rivières. En EEM, l eau est souvent identifiée comme le principal intrant de type ressource renouvelable, d un système humain ou naturel. Les flux d eau en EEM sont généralement caractérisés avec une perspective mondiale, c.à.d. en utilisant une VEU moyenne ignorant les différences topographiques et climatiques à l échelle régionale ou locale. Cependant la différentiation spatiale dans la caractérisation des flux d eau est essentielle pour améliorer la qualité des résultats d EEM. Ce chapitre introduit la première base de données émergétiques mondiale et spatialisée des cours d eau (illustrée par la Figure 2), développée selon le raisonnement adopté dans les quelques études émergétiques locales des rivières trouvées dans la littérature scientifique. Les calculs ont été effectués à très haute résolution (30 arcsec, c.à.d. un maillage d environ 1km²), sur les surfaces terrestres (Antarctique exclu), en utilisant des cartes disponibles gratuitement des précipitations, de l altitude et de l évapotranspiration. La valeur émergétique de l intrant «pluie», calculé pour chaque maille, pourrait s avérer utile pour affiner la base de données de comptabilité environnementale nationale (NEAD, CEP 2006). La VEU des rivières, en sej/m 3, est aussi calculée à très haute résolution, grâce à l utilisation d un algorithme d accumulation des flux. Les résultats préliminaires sont comparés avec les données disponibles sur la VEU des rivières trouvées dans les études existantes, ainsi qu avec le débit réel des principaux fleuves du monde et en France. Tandis que les débits modélisés montrent d importantes divergences par rapport aux débits réels, la comparaison de la valeur émergétique des rivières dans la nouvelle base de données avec celles disponibles dans la littérature scientifique a été rendue difficile par la présence d hétérogénéités dans les détails de calcul observés dans les études précédentes, ce qui justifie le développement d une base de données homogène. La base de données construite est disponible gratuitement pour que les utilisateurs de l EEM puissent récupérer la VEU locale des rivières qu ils étudient. Des moyennes régionales ont également été définies afin de caractériser les procédés d arrière-plan dans le cadre hybride émergie cycle de vie. A long terme, cette approche illustre la possibilité de développer un modèle de référence, spatialisé, pour évaluer en termes émergétiques les ressources naturelles. Cependant, la qualité des résultats montre que d importants affinages sont indispensables (nous recommandons fortement l intervention d experts en modélisation hydrologique). Une feuille de route a été formulée, dans laquelle les améliorations suivantes sont suggérées : 1) inclure la caractérisation des réservoirs d eau douce (glaciers, lacs, eaux souterraines, humidité du sol) ; 2) affiner la modélisation du ruissellement et du débit des cours d eau ; 3) développer une modélisation spatialisée des processus atmosphériques, afin d affiner les VEU de la pluie (potentiel chimique et géopotentiel) ; et 4) inclure dans le modèle les sédiments, les minéraux et particules en suspension, qui sont des flux émergétiques présents dans les cours d eau. 13

14 Figure 2: VEU des cours d eau du monde. Enfin, ce travail montre que l utilisation de SIG met en cause le fait de se baser, en comptabilité émergétique, sur une approche «top-down» des mécanismes naturels, ce qui empêche l utilisation d une description détaillée et spatialisée des mécanismes environnementaux, par exemple les processus atmosphériques et l ensemble du cycle de l eau. Notre approche pourrait ouvrir la voie à une évaluation émergétique intégrée, «bottom-up», des flux de ressources renouvelables fournis par les processus naturels. Comme suggéré précédemment par Rugani et Benetto (2012), cela pourrait effectivement permettre de déterminer la VEU locale des systèmes naturels (suivant Brown and Bardi 2001) et des Services Ecosystémiques et ainsi affiner l approche dominante actuelle «top-down». Le contenu du chapitre 6 a été publié dans le journal Ecological Modelling (Arbault et al. 2014c). Enfin, le chapitre 7 présente une méthode pionnière pour prendre en compte les impacts sur les Services Ecosystémiques en ACV. L état de l art dans ce domaine est de considérer des mécanismes environnementaux impactant la production des Services Ecosystémiques comme indépendants les uns des autres et dénués de dynamique temporelle ; l approche proposée ici étudie l utilisation d un modèle dynamique nommé GUMBO (Boumans et al. 2002) pour quantifier les changements induits par l homme à court et long termes sur la production des Services Ecosystémiques et sur des indicateurs de bien-être humain (voir Figure 3). Cette approche peut être vue comme un moyen d améliorer l ACV par des aspects méthodologiques qui appartiennent typiquement aux atouts de la méthode EEM, comme indiqué précédemment dans le tableau des complémentarités entre EEM et ACV (voir Tableau 1 et Arbault et al. 2013a, c.à.d. chapitre 2). 14

15 Figure 3: Approche proposée pour utiliser un modèle dynamique intégré (GUMBO) en ACV (LCA). GUMBO est un méta-modèle global simplifié, composé de 5 sous-modèles (voir Figure 3), qui connecte de manière dynamique les mécanismes des systèmes naturels (qui produisent les Services Ecosystémiques) avec ceux de la société humaine (qui consomme les Services Ecosystémiques). L anthroposphère est considérée à la fois dépendante de et affectant la géobiosphère. Les Facteurs de Caractérisation (FC) 2 calculés avec GUMBO dépendent de la durée et de la date de la perturbation, ainsi que de la présence d autres interventions environnementales et du choix de politique mondiale de développement adoptée pour le 21è siècle. L analyse des résultats obtenus montre les défis méthodologiques à résoudre pour considérer cette approche comme une alternative robuste à l approche conventionnelle des modèles d évaluation d impact en ACV. Les prochaines recherches devraient se focaliser sur l amélioration de la granularité des interventions environnementales dans les outils de modélisation afin de la faire correspondre avec les standards actuels en ACV, et sur l adaptation de l approche conceptuelle à un modèle intégré spatialisé. Le raisonnement derrière l approche proposée ici reste cependant prometteuse, car il montre que la modélisation intégrée et dynamique du système Terre peut potentiellement fournir des informations plus détaillées sur les interactions entre les interventions environnementales (principalement l extraction des ressources) et les capitaux humains et naturel, en comparaison avec les actuels jeux de données de FC constants utilisés en ACV. En effet, il peut prendre en compte les interactions potentielles entre les interventions environnementales, alors que ce n est pas le cas avec les FC actuels, et fournir des détails sur les mécanismes conduisant aux effets à court terme sans pour autant laisser de côté le tableau global et de long terme. Il permet de différencier les impacts d interventions environnementales se produisant dans le présent, dans le futur, sur une longue période, et peut être paramétré pour estimer l état futur du monde selon différents scénarios de politique de développement mondial. Ces atouts sont sans aucun doute des fonctionnalités intéressantes pour l aide à la décision. Le couplage intégré entre les compartiments de la géobiosphère et l anthroposphère permet de relier entre elles les chaînes de cause à effet, actuellement artificiellement séparées les unes des autres dans les modèles utilisés 2 En ACV, les FC permettent de convertir la valeur physique d une intervention environnementale (extraction de ressource ou émission de substance dans l environnement) en impact environnemental. 15

16 en ACV. Il peut aussi permettre de comptabiliser les changements dans les capitaux humains, qui sont considérés en ACV comme un Domaine de Protection important mais pour l instant exclu des modèles. Le contenu du chapitre 7 a été publié dans le journal Science of the Total Environment (Arbault et al. 2014d). Le chapitre 8 récapitule les conclusions de ce travail de doctorat, en résumant les principaux résultats des chapitres précédents concernant les défis et la valeur ajoutée d un cadre hybride émergie cycle de vie, ainsi que les recommandations pour les améliorations futures et une feuille de route à long terme. Il ne fait aucun doute que la popularité de l ACV dans l évaluation et le management environnementaux est dûe en partie à l existence de bases de données construites à partir de méthodologies standardisées, développées pendant des décennies d efforts continus de la part de la communauté scientifique grâce aux données fournies par les industriels, ainsi que, plus récemment, par le consensus politique à l échelle de l Union Européenne. En comparaison, l EEM est un outil principalement utilisé et étudié par une niche d écologistes et d experts en modélisation des systèmes, et est donc bien moins connu des décideurs et des industries (bien que de plus en plus), en dépit du fait que la méthode est plus ancienne que l ACV. Ce manque de popularité, qui ne doit pas être nécessairement perçu comme un inconvénient, doit probablement être relié à l absence d un cadre de travail standardisé et transparent, et de bases de données intégrées, autorisant ainsi une grande flexibilité dans la définition des frontières du système étudié et donc la sélection subjective des éléments à prendre en compte ou à laisser de côté. Cependant, l EEM a un potentiel considérable, car il permet de comptabiliser la valeur naturelle des ressources et des Services Ecosystémiques sur lesquels les activités humaines reposent, et ainsi proposer une solution pour une comptabilité intégrant économie et environnement. L EEM peut bénéficier d une adaptation de la procédure de standardisation et des bases de données disponibles en ACV, bien que cette intégration soit freinée par des différences dans les règles de calcul des deux outils, dont la combinaison nécessite une conceptualisation approfondie. L analyse critique du cadre hybride émergie cycle de vie proposé dans ce manuscrit démontre qu en dépit de certaines améliorations indispensables, le cadre hybride est plus rigoureux, transparent, utile et reproductible que le cadre conventionnel de l EEM. Les autres études proposées dans ce manuscrit sont dédiées au développement d un modèle intégré de la géobiosphère, grâce à l utilisation d outils des technologies de l information tels que la SIG, pour caractériser la valeur émergétique locale des ressources en eau douce, ou grâce à l exploitation d un modèle dynamique intégré de la géobiosphère en ACV afin de prendre en compte les interactions mutuelles entre chacun de ses composants, y compris l humanité. Il peut être anticipé que ces réalisations pourront contribuer à la consolidation du cadre hybride émergie cycle de vie, à la fois en validant les réalisations méthodologiques précédentes (tels que l algorithme intégré dans SCALE), et en proposant des idées pour l intégration formelle des développements futurs. 16

17 Références Arbault D., Rivière M., Rugani B., Benetto E., Tiruta-Barna L. Integrated earth system dynamic modeling for life cycle impact assessment of ecosystem services. Sci. Total Environ. 2014d;472: Arbault D, Rugani B, Marvuglia A, Benetto E, Tiruta-Barna L. Emergy evaluation using the calculation software SCALE: case st udy, added value and potential improvements. Sci Total Environ. 2014a;472: Arbault D, Rugani B, Marvuglia A, Tiruta-Barna L, Benetto E. Accounting for the emergy value of life cycle inventory systems: insights from recent methodological advances. JEAM. 2013a;1(2): Arbault D, Rugani B, Tiruta-Barna L, Benetto E. Emergy evaluation of water treatment processes. Ecol Eng. 2013b;60: Arbault D., Tiruta-Barna L., Rugani B., Benetto E. A semantic study of the Emergy Sustainability Index in the hybrid lifecycle-emergy framework. Ecol. Indic. 2014b; In press. DOI: /j.ecolind Arbault D., Rugani B., Tiruta-Barna L., Benetto E. A first global and spatially-explicit emergy database of rivers and streams based on high-resolution GIS-maps. Ecol Model. 2014c;281: Boumans R, Costanza R, Farley J, Wilson MA, Portela R, Rotmans J, et al. Modeling the dynamics of the integrated earth system and the value of global ecosystem services using the GUMBO model. Ecol Econ. 2002;41(3): Brown MT, Bardi E. Folio# 3: Emergy of Ecosystems. Handbook of Emergy Evaluation. Center for Environmental Policy, Environmental Engineering Sciences, University of Florida, Gainesville; Brown MT, Raugei M, Ulgiati S. On boundaries and investments in Emergy Synthesis and LCA: A case study on thermal vs. photovoltaic electricity. Ecol Indic. 2012;15(1): CEP. National Environmental Accounting Database (NEAD) Center for Environmental Policy [Internet] Available from: Costanza R, d Arge R, Groot R de, Farber S, Grasso M, Hannon B, et al. The value of the world s ecosystem services and natur al capital. Nature. 1997;387(6630): Folke C, Jansson Å, Rockström J, Olsson P, Carpenter SR, Chapin FS, et al. Reconnecting to the Biosphere. AMBIO. 2011;40(7): De Groot RS, Wilson MA, Boumans R. A typology for the classification, description and valuation of ecosystem functions, goods and services. Ecol Econ. 2002;41(3): Igos E, Benetto E, Baudin I, Tiruta-Barna L, Mery Y, Arbault D. Cost-performance indicator for comparative environmental assessment of water treatment plants. Sci Total Environ a;443: Igos E, Dalle A, Tiruta-Barna L, Benetto E, Baudin I, Mery Y. Life cycle assessment of water treatment: what is the contribution of infrastructure and operation at unit process level? J Clean Prod b;in press:dx.doi.org/ /j.jclepro Ingwersen WW. Emergy as a Life Cycle Impact Assessment Indicator. J Ind Ecol. 2011;15(4): Markandya A, Pedroso-Galinato S. How substitutable is natural capital? Environ Resour Econ. 2007;37(1): MEA. Millennium ecosystem assessment, ecosystems and human well-being: a framework for assessment. World Resources Institute, Washington, DC Moldan B, Janoušková S, Hák T. How to understand and measure environmental sustainability: Indicators and targets. Ecol Indic. 2012;17(0):4 13. Odum HT, Odum EP. The Energetic Basis for Valuation of Ecosystem Services. Ecosystem s. 2000;3(1):21 3. Raugei M, Rugani B, Benetto E, Ingwersen WW. Integrating emergy into LCA: Potential added value and lingering obstacles. Ecol Mod. 2014;271:4 9. Rees WE, Wackernagel M. Ecological footprints and appropriated carrying capacity: Measuring the natural capital requirements of the human economy. Focus. 1996;6(1): Rockström J, Steffen W, Noone K, Persson Å, Chapin FS, Lambin EF, et al. A safe operating space for humanity. Nature. 2009;461(7263): Rugani B. Advances towards a comprehensive evaluation of emergy in life cycle assessment. University of Siena: Siena, Italy; Rugani B, Benetto E. Improvements to Emergy Evaluations by Using Life Cycle Assessment. Environ Sci Technol. 2012;46(9): Rugani B, Benetto E, Arbault D, Tiruta-Barna L. Emergy-based mid-point valuation of ecosystem goods and services for life cycle impact assessment. Rev Métall. 2013;110: Rugani B, Huijbregts MAJ, Mutel C, Bastianoni S, Hellweg S. Solar Energy Demand (SED) of Commodity Life Cy cles. Environ Sci Technol. 2011;45(12): Rugani B, Panasiuk D, Benetto E. An input output based framework to evaluate human labour in life cycle assessment. Int J Life Cycle Assess. 2012;1 18. Steffen W, Persson Å, Deutsch L, Zalasiewicz J, Williams M, Richardson K, et al. The Anthropocene: From Global Change to Planetary Stewardship. AMBIO. 2011;40: TEEB. The Economics of Ecosystems and Biodiversity: ecological and economic foundations. Routledge; Tiruta-Barna L, Benetto E. A conceptual framework and interpretation of emergy algebra. Ecol Eng. 2013;53: Zhang Y, Baral A, Bakshi BR. Accounting for Ecosystem Services in Life Cycle Assessment, Part II: Toward an Ecologically Based LCA. Environ Sci Technol. 2010;44(7):

18

19 Outline 1. Introduction and research questions Global change and Environmental Accounting Two visions of Environmental Accounting Life Cycle Assessment (LCA) Emergy Evaluation (EME) Comparison and complementarities of LCA and EME Freshwater and related ES as case study for resource characterization LCA characterization of freshwater and ES EME and characterization of freshwater and related ES Research questions and methodological approach Thesis outline Accounting for the Emergy Value of Life Cycle Inventory Systems: Insights from Recent Methodological Advances 45 Abstract Introduction Methods Emergy Evaluation (EME) Life Cycle Assessment (LCA) Current status-quo on the combination between Emergy and LCA Benchmarking the methods combination Challenges to define an Emergy-LCA approach Conclusions Emergy evaluation of water treatment processes 61 Abstract Introduction Methodology and data collection Energy system diagram Emergy-based indicators Data collection from LCA studies and comparison of EmE and LCA Unit emergy values (UEVs) Results and discussion Emergy analysis of flows Comparison with life cycle assessment Uncertainty and limitations Conclusion 77 19

20 4. Emergy evaluation using the calculation software SCALE: case study, added value and potential improvements 79 Abstract Introduction Materials and methods Methodological principles Description of the case studies Results and discussion Quantitative comparison Contribution analysis Gravity analysis Differences in interpretation Limits of SCALE Conclusions and outlook A semantic study of the Emergy Sustainability Index in the hybrid lifecycle-emergy framework 101 Abstract Introduction Critical review of EYR and ELR Method Formulation of EYR, ELR and ESI variants Sensitivity of ESI to input categories Calculation of inputs Case studies Results and discussion Direct vs. Background inputs in technosphere products Ranking of the WTPs Sensitivity of ESI to input categories Conclusion A first global and spatially-explicit emergy database of rivers and streams based on high-resolution GIS-maps 119 Abstract Introduction Research question State-of-the-Art Materials and Methods General method Preparation of raster elements from data sources Water accumulation over drainage areas Modeling of rivers Characterization of freshwater use at regional scale Results and discussion

21 6.3.1 Flow direction and runoff Emergy inputs Emergy accumulation in streams Territorial averages Bottom-up emergy accounting Conclusion Integrated earth system dynamic modeling for life cycle impact assessment of ecosystem services 145 Graphical abstract 145 Abstract Introduction and goals Materials and methods Choice of an integrated earth system dynamic model: GUMBO Definition of midpoints and endpoints in GUMBO Tuning GUMBO to retrieve CFs for ES Results Example of midpoint and endpoint CFs calculation and interpretation for a specific perturbation and impact Generalization of the CFs calculation approach to all the types of perturbation and impact _ Influence of alternative modeling perspectives Perturbation occurring in the future (SI7.2.1) or over a longer period (SI7.2.2) Perturbation under different policy scenarios (SI7.2.3) Simultaneous perturbations (SI7.2.4) Discussion Need for harmonization between LCI datasets and integrated global models Added value of formal use of integrated earth system dynamic models within LCIA Is further sophistication of integrated modeling necessary for LCIA? Research roadmap Conclusions Conclusions 165 Supplementary information 171 SI3. Chapter 3: Emergy Evaluation of Water Treatment Processes 173 SI3.1. Life Cycle Inventory (LCI) and economic data 173 SI3.2. Proxy UEV for Ecoinvent elementary flows 173 SI3.3. Emergy value of land occupation 178 SI3.4. UEV of freshwater flows 178 SI3.5. Emergy indicators in literature 179 SI3.6. Emergy indicators of drinking water production facilities. 185 SI3.7. Comparative table of EmE and LCA results for the four WTPs 188 SI4. Chapter 4: Emergy evaluation using the calculation software SCALE: case study, added value and potential improvements 189 SI4.1. Influence of the minflow threshold value in EME SCALE 189 SI4.2. Comparative table of UEV CONV, UEV SCALE and SED of technospheric inputs 190 SI4.3. Decomposition of technospheric inputs UEV SCALE per resource category 192 SI4.4. Resources disregarded in SEF dataset due to double-counting

22 SI4.5. Application of the EcoLCA tool 195 SI5. Chapter 5: A semantic study of the Emergy Sustainability Index in the hybrid lifecycle-emergy framework 197 SI5.1. A brief discussion on the evolution of EYR s definition and other research on ESI 197 SI5.2. Inventory data and results of the case studies 199 SI5.3. Foreground input tracking algorithm and illustration with the case studies 200 SI6. Chapter 6: A first global and spatially-explicit emergy database of rivers and streams based on high-resolution GIS-maps 202 SI6.1. On solar radiation and wind energy inputs 202 SI6.2. Processing steps in ArcGIS 205 SI6.3. Emergy database of rivers: Shapefile 213 SI6.4. Emergy database of rivers: Territorial averages 214 SI7. Chapter 7: Integrated earth system dynamic modeling for life cycle impact assessment of ecosystem services 215 SI7.1. Method 215 SI7.2. Detailed results of simulations 216 SI7.3. Limitations 229 References

23 List of figures Figure 1.1: a) General LCA procedure (ISO 2006); b) environmental mechanisms in the LCIA method ReCiPe (Goedkoop et al. 2009) Figure 1.2: Generic diagram of a coupled natural-human system (Ridolfi and Bastianoni 2008) Figure 1.3: Cause-effect chains, from inventory of water flows to the three Areas of Protection (AoP). Kounina et al. (2013) Figure 2.1: Conceptualizing the main methodological discrepancies between Emergy and Life Cycle Assessment Figure 3.1: Energy diagram of potable water production Figure 3.2: Contribution of each type of input (feedback and raw water flows) to the Unit Emergy Value (UEV) calculated for the 4 plants and for Sites A and B including infrastructure items Figure 3.3: Comparison of Environmental Yield Ratio (EYR) scores for the production of various types of man-made products (see SI, Section 3.5) Figure 3.4: Comparison of Environmental Loading Ratio (ELR) scores for the production of various types of man-made products (see SI, Section 3.5) Figure 3.5: Comparison of Emergy (i.e. UEVs) and Life Cycle Assessment (LCA) results for the four Water Treatment Plants Figure 4.1: General framework for the application of EME CONV, EME SCALE and SED Figure 4.2: Contribution of technospheric inputs across the 4 case studies by analyzing the three emergy-based methods Figure 4.3: Ratio UEV SCALE /UEV CONV (unitless) for selected technospheric inputs (log scale) Figure 4.4: Relative influence of emergy quantification procedure on the final UEVs of technospheric inputs Figure 4.5: Decomposition per resource category of technospheric inputs in the EME SCALE and the SED of the 4 case studies Figure 4.6: Graphical comparison between traditional EME (left) and SCALE (right, including information retrieved from ecoinvent, dotted-blue box) Figure 4.7: Missing elements in SCALE-based emergy accounting Figure 4.8: Additional contribution (in sej/m 3, log scale) of Ecosystem Services to the final UEV of the functional unit (using EcoLCA, Zhang et al. 2010a) Figure 5.1: Relative contribution of direct investments (D) to the total UEV of material and energy products used as inputs in the life cycle phases of the case studies (i.e. only at the production site level, see also Table S5.1 in SI5.2) Figure 5.2: Values (dimensionless) calculated for the four variants of ESI (see Table 5.2) applied to the six case studies of WTP Figure 6.1: Flowchart of the methodological procedure to account for the UEV of streams and rivers Figure 6.2: Flow system diagram of a grid cell used in this study Figure 6.3: Relationship between altitude and UEV of rain geopotential (from Odum 2000) Figure 6.4: Schematic representation of a stream s Strahler order ( 131 Figure 6.5: Population in France and Western Europe: urban areas tend to be located along larger streams Figure 6.6: Emergy input, per grid cell (30 arcsec) and per year Figure 6.7: UEVs of stream freshwater at global scale Figure 6.8: UEV of streams (2nd order and higher), zoom on France and Western Europe Figure 6.9: Population-weighted average UEV Figure 7.1: Basic structure of GUMBO Figure 7.2: The four-step procedure to relate LCI data with GUMBO's response Figure 7.3: Example of Characterization Factors calculation steps

24 List of tables Table 1.1: Complementarities of LCA and EME Table 2.1: Summary of current advances and proposals to formalize a consistent Emergy-LCA approach Table 3.1: Calculation of the French electricity mix UEV Table 3.2: Emergy table for Site Table 3.3: Emergy table for Site Table 3.4: Emergy table for Site A Table 3.5: Emergy table for Site B Table 3.6: Comparison of emergy-based indicators for the four Water Treatment Plants Table 4.1: Comparison of the total emergy of technospheric inputs calculated for the 4 case studies of water treatment plants using EME CONV, EME SCALE and SED Table 4.2: Comparison of scope between EME CONV, EME SCALE, SED and ReCiPe method (LCA) Table 4.3: Estimated contribution of pollution impacts and societal net loss (in emergy term) to the case studies of water treatment plants Table 5.1: Foreground and background inputs to calculate EYR with a lifecycle perspective (adapted from Brown et al. 2012) Table 5.2: List of the four variants of ESI and relative characteristics, depending on the focus for EYR (site vs. lifecycle) and for ELR (local vs. global) as defined in equations ( ) Table 5.3: Sensitivity of ESI variants to input categories Table 5.4: Changes in inputs categories (in sej/m 3 and in %), per WTP s site, which would increase ESI11 by 1% Table 6.1: Data sources and characteristics Table 6.2: Differences between modeled and actual flow and catchment area of 13 world s major rivers and 5 major French Rivers Table 6.3: Stream order of several rivers according to the cutoff value in the GIS model, as compared to actual stream order Table 6.4: Comparison between modeled emergy value, flow rate and UEVs of selected sites and values retrieved from the literature (adjusted to baseline 9.26 E24 sej/yr) Table 7.1: The 4 ecosystem goods, 7 ecosystem services and 4 human capitals (i.e. production factors) included in GUMBO Table 7.2: Description of midpoint impacts and their physical unit as found in GUMBO s outputs. Midpoints are equivalent to the production factors of Table Table 7.3: Summary of CFs: impacts of perturbations on ES, human capitals, GWP and welfare

25 List of figures in supplementary information Figure S. 4.1: Relative portion of the emergy value of flows lost due to the threshold level applied in SCALE and the loop violation Figure S. 4.2: Ratios of SED value to the emergy value of SCALE inputs of technospheric products, per resource category Figure S. 5.1: Comparison between EYR as defined by Odum (1996) and the later version commonly adopted Figure S. 6.1: Total solar radiation per latitude (J/yr) in the original dataset (see Table 6.1 for source) Figure S. 6.2: Emergy input of wind force, per grid cell (30 arcsec) and per year Figure S. 6.3: Highest emergy value of available wind energy (green), rain chemical potential (blue) and rain geopotential (brown) Figure S. 7.1: Screenshot of GUMBO, showing the implementation of a perturbation (here, fossil fuel) Figure S. 7.2: Example of changes on the production factor Energy, following an additional 0.5% ore production in Figure S. 7.3: Relative change in ecosystem services according to the year of perturbation (additional fossil fuel extraction) Figure S. 7.4: Relative change in human capitals according to the year of perturbation (additional fossil fuel extraction) Figure S. 7.5: Relative change in ecosystem goods production (except energy) according to the year of perturbation (additional fossil fuel extraction) Figure S. 7.6: Relative change in ecosystem good "energy" according to the year of perturbation (additional fossil fuel extraction) Figure S. 7.7: Relative change in endpoint CFs according to the year of perturbation (additional fossil fuel extraction) Figure S. 7.8: Relative change in ecosystem goods production (except energy) according to the duration of the perturbation (additional fossil fuel extraction) Figure S. 7.9: Relative change in ecosystem good "energy" according to the duration of the perturbation (additional fossil fuel extraction) Figure S. 7.10: Relative change in ecosystem services according to the duration of the perturbation (additional fossil fuel extraction) Figure S. 7.11: Relative change in human capitals according to the duration of the perturbation (additional fossil fuel extraction) Figure S. 7.12: Relative change in endpoint CFs according to the duration of the perturbation (additional fossil fuel extraction) Figure S. 7.13: Relative change in ecosystem goods production according to the global policy scenario Figure S. 7.14: Relative change in ecosystem services according to the global policy scenario Figure S. 7.15: Relative change in human capitals according to the global policy scenario Figure S. 7.16: Relative change in endpoint CFs according to the global policy scenario Figure S. 7.17: Evolution of GW P according to selected policy scenarios (with no perturbation) Figure S. 7.18: Comparison of annual change in GWP (in Trillion US$) due to an additional fossil fuel extraction of 65 MtC in year 2000, according to selected policy scenarios Figure S. 7.19: Evolution of SSW according to selected policy scenarios (with no perturbation) Figure S. 7.20: Comparison of annual change in SSW due to an additional fossil fuel extraction of 65 MtC in year 2000, according to different policy scenarios Figure S. 7.21: Change in soil formation for several perturbations in fossil fuel extraction at year Discrete values are due to computational limits of STELLA

26 List of tables in supplementary information Table S. 3.1: LCI data (operation) for the WTPs Table S. 3.2: LCI data of infrastructure for sites A and B Table S. 3.3: Economic inputs for the 4 WTPs (in per m 3 output water) Table S. 3.4: Proxy UEVs of Ecoinvent v2.2 elementary flows Table S. 3.5: Proxy UEVs of Ecoinvent v2.2 elementary flows used in infrastructure Table S. 3.6: Example of UEV calculation for local freshwater, here of Seine River at Site 1 location Table S. 3.7: Specific data for each local freshwater source used up by the WTPs Table S. 3.8: Emergy-based indicators of selected published case studies Table S. 3.9: LCA and EmE results for the WTPs Table S. 4.1: UEVs retrieved from the application of the three methods on the man -made products used in WTPs and the production of activated carbon (x 1E11 sej/unit) Table S. 4.2: Electricity production mix for UCTE and UEVs CONV from literature (excluding human labor and services) Table S. 4.3: Decomposition of SCALE output (UEV SCALE ), SCALE input (application of rule #2 only) and SED of technospheric inputs Table S. 4.4: Data used in the EcoLCA software Table S. 5.1: Inventory of material inputs for the case studies Table S. 5.2: Labor and service inputs of the case studies. Data are in /m 3 of potable water produced Table S. 5.3: Total emergy value of inputs (foreground + background), calculated with SCALE Table S. 5.4: Emergy value of inputs: local resources (L), direct investments (D), background investments (B) and total inputs (U, equal to output UEV in these case studies), including materials and labor, for the case studies Table S. 5.5: Resulting EYR0, EYR1, ELR0, ELR1 for the case studies (dimensionless Table S. 6.1: List of calculated parameters for each stream segment Table S. 6.2: List of calculated parameters for each territorial average Table S. 7.1: Impacts on midpoints and endpoints for 1Gt fossil fuel perturbations occurring at different dates in the future Table S. 7.2: CFs for perturbations on 1Gt fossil fuel extraction, over different periods, starting in year Table S. 7.3: CFs for perturbations on fossil fuel extraction in year 2000, with selected scenarios for future global policies Table S. 7.4: Changes in midpoints due to perturbations in year 2000 on fossil fuel extraction (+65 MtC), ore production (+7.44 Gt), separately and simultaneously Table S. 7.5: Potential connections between GUMBO and LCA

27 1. Introduction and research questions

28

29 1.1 Global change and Environmental Accounting Chapter 1 Introduction and research questions Mankind is reaching the physical limits of the Earth (Rockström et al. 2009). It is even considered as a global force, shaping the surface of the planet and influencing climate as well as global cycles of carbon, nitrogen and water, to the point of creating a new geologic era called anthropocene (Steffen et al. 2007). Last century s increase of global population and technological improvements, mostly thanks to non-renewable resources (Moldan et al. 2012), comforted our vision of possible unlimited growth, while on the contrary we are becoming increasingly aware of the finiteness of our planet and fossil resources stocks. It is now widely acknowledged that the Earth cannot support 7 billion people with the living standards of occidental countries, which puts Mankind in front of historically new challenges. We may blame ourselves for destroying the planet and worry about the potential consequences on the stability of our socio-economic systems, or we could instead consider this as a motivation to search for more sustainable systems of governance and effective stewardship of our planet (Folke et al. 2011; Steffen et al. 2011). Environmental-economic sustainability involves the preservation of life-supporting systems integrity, while ensuring economic development on renewable resources only (Moldan et al. 2012). Both issues are highly intricate, since renewable resources and suitable living conditions are provided by ecosystems, under the form of commonly called Ecosystem Services, or ES (Costanza et al. 1997; de Groot et al. 2002; MEA 2005b; TEEB 2010): provisioning service supply us in food, raw material, freshwater, genetic resources; regulating services (e.g. erosion and flood control, air and freshwater quality regulation) are important contributors of the stability of our natural environment i.e. a prerequisite for human settlement; nature also delivers important cultural services such as recreational areas, spiritual and aesthetic assets, on which science, art and religion are often based. ES are necessary for human development, although they are currently considered as externalities out of the economic sphere, which hampers the acknowledgement of the natural capital s value. To deal with this discrepancy in our value system, ES may be estimated in economic terms, using market values, avoided costs or willingness-to-pay estimations (e.g. Wilson and Carpenter 1999; Guo et al. 2001; Jenerette et al. 2006; Biao et al. 2010; Ma and Swinton 2011; Vigerstol and Aukema 2011; Wise et al. 2011; Ghaley et al. 2013; Morri et al. 2013). However, estimating the costs of impaired ES with the utilitarian perspective of an economic approach may occult the complexity of socio-ecologic systems, which understanding is important to design sustainable governance systems (Cornell 2011). A complementary approach is necessary for sustainability assessment, based on the laws of nature and the observation of environmental mechanisms at all scales of space and time, to evaluate the contribution of ES and natural resources to human development and assess the consequences of human-driven environmental changes (Rees and Wackernagel 1996; Costanza et al. 1997; Odum and Odum 2000; Markandya and Pedroso-Galinato 2007). In this study, two visions of environmental accounting are investigated, which offer complementary viewpoints. On the one hand Life Cycle Assessment (LCA) evaluates the environmental impacts of a product or a service, based on a detailed description of the network of the technological processes and environmental mechanisms involved. On the other hand, Emergy Evaluation (EME) considers human activities as dependent on the natural sphere, and estimates the contribution of natural processes to maintain them. The next section introduces the two approaches in detail, as well as the potentialities of combining them and current advances in this direction. Two aspects have to be considered: how human systems are described in both methods, 29

30 and how resources are characterized. To the latter aim, a deeper focus is proposed on the characterization of freshwater and ecosystem services. 1.2 Two visions of Environmental Accounting Life Cycle Assessment (LCA) Life Cycle Assessment (LCA) is the reference framework for the estimation of the diverse impacts of human activity, and namely products and production processes, on these environmental mechanisms (ISO 2006; Huijbregts et al. 2008; European Commission 2010c). In LCA, a functional unit is defined by quantifying the function delivered by the product system. The product life cycle, which includes raw material extraction, transformation, transport, use, disposal etc., related to that functional unit is then modeled in the Life Cycle Inventory (LCI) step by collecting site and context specific data and information, complemented by the use of databases of processes such as ecoinvent (Weidema et al. 2011), linked to each other in a technological network. The latter allows quantifying all exchanges of material and energy flows between the functional unit s lifecycle and the natural environment. So-called elementary exchanges represent resources extracted, pollutants emitted, land occupied and transformed. Then, the Life Cycle Impact Assessment (LCIA) step quantifies the cause-effect chains, from these elementary flows to pollution effects and damages on the natural environment, human health and resource availability (Jolliet et al. 2003b; Margni et al. 2008; Rosenbaum et al. 2008; Goedkoop et al. 2009; Van der Voet et al. 2009; European Commission 2010a). In current LCIA methods, these cause-effect chains are described independently from each other, and quantified with constant coefficients called Characterization Factors (CFs). CFs refer to several pollution types (midpoint impacts) and their damages (endpoint impacts) on so-called Areas of Protection (AoP), which are typically the human health, the quality and diversity of ecosystems, and the depletion of resources (Figure 1.1). Concerning the AoP Resources, one possible approach is to estimate, for each resource, the future costs for further extraction in monetary or energy terms (e.g. Jolliet et al. 2003b; Goedkoop et al. 2009). Although suitable for non-renewable (fossil and mineral) resources, this rationale cannot be applied on renewable resources and ES, which are not stocks subject to depletion but are in fact continuously-regenerated flows. Impacts on these resources occur when they are consumed faster than their regeneration rate or when the environmental mechanisms that produce them are affected by human intervention. Hence, accounting for the regeneration rate of renewable resources and ES is a prerequisite for tools that aim at measuring environmental sustainability. Impacts related to biotic resources formation and availability were not paid much attention until recently in LCIA, which is considered as a caveat of current methods (e.g. Finnveden et al. 2009; Koehler 2008; Koellner et al. 2013b). Exergy (i.e. available energy) was proposed as a common unit to characterize renewable and non-renewable resources and ES, by estimating the maximum potential work that can be extracted from a substance (Gong and Wall 2001; Yi et al. 2004; Szargut 2005; Sciubba 2012). Recently-developed datasets can be directly employed in LCIA (Ukidwe and Bakshi 2004, 2007; Bösch et al. 2007; Dewulf et al. 2007; Zhang et al. 2010a; Alvarenga et al. 2013). However, characterizing resources with their exergy content holds an human centered (utilitarian) viewpoint and would therefore lead to neglecting the importance of 30

31 Chapter 1 Introduction and research questions the natural processes that deliver them (or, equivalently, the future environmental contribution to regenerate them). Figure 1.1: a) General LCA procedure (ISO 2006); b) environmental mechanisms in the LCIA method ReCiPe (Goedkoop et al. 2009) Emergy Evaluation (EME) A different approach is proposed in the emergy framework (Odum 1973, 1988, 1996). Emergy is defined as the amount of available energy of one kind (usually solar energy) previously used directly and indirectly to make a product (Odum 1996; Bastianoni et al. 2007), expressed in solar emjoules (sej). Emergy stands on the principle that natural processes sustain one another in a network of flows of available energy, generating complex systems hierarchically organized (Odum 1988, 1996). Therefore, all natural resources can be quantified with the total energy used up by natural systems to produce it. Since every human activity ultimately relies on natural processes, goods and services, this approach is also applicable to man-made products. By proposing a donor-side valuation system, emergy is complementary to the user-side viewpoint, which put values on natural resources according to their usefulness (e.g. economic value, exergy content). Hence, it seems potentially valuable to discern sustainable patterns of human societies. The emergy value of a man-made good can be considered as a measure of cumulated present and past environmental work to generate the resources used up in the product s lifecycle, or as the future environmental work necessary to replace those resources used up (Raugei et al. 2014). Transformity (Tr) is the ratio of emergy value to exergy content of a product, and calculated in sej/j. A product with a high Tr value denotes a large amont of primary energy sources required to make a unit of it. Other intensive metrics called Unit Emergy Value (UEV) can be measured accordingly, in sej per mass, area, monetary value or other appropriate extensive property of a product. Emergy is not a conservative function but a sort of cumulative function that follows a memorization logic, supported by a specific algebra based on four fundamental rules (Brown and Herendeen 1996): 31

32 1. When only one product is obtained from a process, all source-emergy is assigned to it. 2. Co-products of a process (i.e. product items showing different physico-chemical characteristics, but which can only be produced jointly) have the total emergy assigned to each pathway (no allocation). 3. When a pathway splits (originating flows showing the same physico-chemical characteristics), emergy is assigned to each leg of the split according to the percentage of total exergy flow on the pathway. 4. Emergy cannot be counted twice within a system: a. emergy in feedback cannot be double counted; b. co-products, when reunited, cannot add up to a sum greater than the source emergy from which they were derived. Rule #2 highlights the donor-side approach of emergy-based environmental accounting: it is considered that each co-product of a process could not have been produced without the contribution of all inputs. In contrast, a user-oriented accounting approach, as e.g. in LCA, would proceed differently by considering that the environmental burdens caused by the process and its inputs should be shared among the users of the diverse co-products. Rule #4 of emergy algebra is a necessary complement to Rule #2, in order to avoid double-counting. The Emergy Evaluation (EME) of a human activity or a territory provides decision-makers with performance indicators on the environmental-economic sustainability of the studied system. EME procedure (see Odum 1996 for more details) starts with a diagram of the studied system, depicting the main materials and energy storages, transformation processes and flows. Natural resources used up are sketched on the left-hand side, along with external driving forces that generate them, while man-made goods and services consumed are placed around the system, on the upper righthand corner (Figure 1.2). Flows and storages are then quantified in physical units and converted into emergy, using UEVs from previous publications or specifically calculated for the case study. To enhance the operability of the EME framework, a UEV database is currently being developed (Tilley et al. 2012). By applying the algebra rules, the system s output Yield can be calculated, enabling the calculation of the emergy-based indicators. Notably, the Emergy Yield Ratio (EYR), the Environmental Load Ratio (ELR) and the Emergy Sustainability Index (ESI) are further described and analyzed in chapters 3 and 5. The transformity or UEV of natural resources are estimated from a quantitative description of their generation processes, starting from the observation that natural systems are maintained by continuous capture of exergy flows, which ultimately originate from the three independent driving forces of the planetary system (sunlight, tidal energy and deep Earth heat), via successive transformations. These driving forces form the so-called baseline, which value, expressed in sej/yr, is still under debate (Odum 1996; Odum et al. 2000; Campbell 2001; Brown and Ulgiati 2010). The transformity of renewable resources that result from global processes, for instance wind (Odum 2000) and precipitation (Odum 2000; Buenfil 2001; Campbell 2003), is calculated by dividing the baseline by their global annual energy flow. Local renewable resources are formed by local ecosystems, which use up only a fraction of these global resources. Therefore, their UEV is calculated by accounting for the amount of incoming renewable flows (sun, wind and rain) over the ecosystem s area, per unit output (Brown and Bardi 2001; Neri et al. 2014). According to rule 4b, only the input with the highest emergy value is accounted for. Resources that form over long-time scales, which are considered non-renewable with regard to the anthropic time-horizon, are evaluated with the corresponding time scale. For instance, fossil fuel production 32

33 Chapter 1 Introduction and research questions is formed via the capture of baseline energy flows, stored into biomass (via photosynthesis) over a long period, and then concentrated into coal, oil and natural gas via the action of geological forces, land cycling and deep Earth heat (Bastianoni et al. 2005; Brown et al. 2011). Therefore, they have rather high transformities. Figure 1.2: Generic diagram of a coupled natural-human system (Ridolfi and Bastianoni 2008) EME can be considered as a holistic and operational framework to identify and quantify the natural processes that support a human activity. However, the concept of emergy faces criticism from physicists, engineers and researchers of other disciplines (Brown and Herendeen 1996; Hau and Bakshi 2004; Sciubba and Ulgiati 2005; Voora et al. 2010): the non-conservative algebra (in contrast to energy and exergy accounting) and low level of details of EME (in contrast to thermodynamic studies of industrial processes) seem to question the method s credibility. The point is that although the formal link between emergy and exergy remains discussed (Sciubba 2010), emergy should be considered as an exergy path function, with different purposes than exergy and different system boundaries (Sciubba and Ulgiati 2005; Bastianoni et al. 2007). For instance, exergy analysis of a process does not account for externalities such as labor, information (non-material) services, which may influence efficiency of a process. EME considers a process with a larger lens, to the detriment of precision. It is roughly assimilable to cumulative exergy accounting (Zhang et al. 2010a; Sciubba 2012; Rocco et al. 2014), provided the latter method is correctly extended to include both supporting natural and human economic systems. Although apparently simplistic, EME is currently the only operational method able to account for the influence of complex, global natural and human systems on local systems and human activities. To be more acknowledged as a credible environmental accounting method, EME would certainly benefit from further linkage with other thermodynamic metrics, as well as from the development of properly standardized guidelines and datasets to characterize both natural resources and manmade products (Hau and Bakshi 2004; Voora et al. 2010; Tiruta-Barna and Benetto 2013; Raugei et al. 2014). 33

34 1.2.3 Comparison and complementarities of LCA and EME Despite sharing the same purpose of providing operational tools for environmental accounting, LCA and EME rely on rather distinct rationales: LCA aims at estimating the impacts generated by human activities on the afore-mentioned Areas of Protection (AoP), whereas EME investigates to which extent a human activity is dependent on the Earth system and the goods and services it delivers. Such difference results in specific points of focus, quantification methods, algebra rules, scope limitations, and information for decision-making, as summarized in Table 1.1. Table 1.1: Complementarities of LCA and EME. Rationale Focus Algebra Impact(s) accounted for Current scope limitations Decisionmaking LCA Human activities generate impacts to the Areas of Protection Material and energy flows in technosphere User-oriented (co-products burdens are shared among users) Damages on human health, ecosystems and resource depletion Human labor Ecosystem Services Compare technological systems delivering the same functional unit EME The Earth system supports human activities Natural processes for resource formation Donor-oriented (all resources are used up for the delivery of each co-product) Total [solar] energy used up directly and indirectly Effects of pollution Details in technosphere Provide universal environmental performance indicators Since the aim of LCA is to quantify the impacts of a human activity on the three AoP (Human Health, Ecosystem Diversity and Resource Depletion), an important modeling step is to determine at which point of the studied lifecycle system the impacts are generated. Therefore, a high level of details is sought for the representation of the network of technological processes (so-called technosphere). When a technological process generates co-products (e.g. salted water electrolysis, producing caustic soda and chlorine gas), the environmental burdens are allocated to the coproducts, to be shared by the various downstream users; this is a user-side accounting scheme. Impact assessment quantifies the damages on the so-called AoP via environmental mechanisms modeled independently from each other (see e.g. section 1.2.5). The scope of technospheric flows and environmental mechanisms is gradually extended; currently limitations of operational databases concern e.g. the accounting for material and energy flows related to human labor and the impacts generated to ecosystem services. As a result, LCA does not embrace a holistic scope - though it is gradually extended - and gives a higher level of details to the description of the technosphere. Finally, LCA provides information to decision-makers about how to reduce the various impacts previously quantified. Since every human activity generates impacts, absolute values in LCA results are meaningless; LCA can only be used for the comparison of two technological systems that deliver the same functional unit. In contrast, the aim of EME is to quantify the portion of the Earth system necessary to operate the studied human activity. Therefore, more emphasis is given to the description of natural processes that deliver the local resource used up. When a natural process generates several coproducts (e.g. wind and rain from atmospheric processes), is it considered that each co-product could not have been delivered without the full contribution of all resources used up by the process; this is a donor-side accounting scheme. The aim of EME is not to identify which user the environmental burden should be assigned to, but rather to estimate how much [solar] energy was 34

35 Chapter 1 Introduction and research questions used up directly and indirectly to enable the user to run his/her activity. Therefore, the traditional accounting scope focuses on resources formation, while neither the impacts of pollution on the natural mechanisms nor the detailed tracking of material and energy flows within the technosphere are currently considered with a consistent model. Contrary to LCA, EME provides environmental performance indicators based on the classification of the various inputs to the studied system (e.g. into local, renewable resources, local, nonrenewable ones and man-made, imported ones), and thus can be applied to any kind of activity, allowing cross-sectorial benchmarking. Although EME and LCA are based on dissimilar viewpoints, scopes and mathematical frameworks, they can also be seen as complementary, bestowing their mutual integration as a valuable research project. UEVs were recently proposed as CFs for resources in the LCA framework (Rugani 2010; Zhang et al. 2010a; Ingwersen 2011; Rugani et al. 2013; Raugei et al. 2014). However, this approach is made difficult by the different meanings behind both methods: while LCA is based on the notion of environmental burdens, EME relies on the notion of appropriation of environmental work. Therefore, while impacts quantified in LCA are sought to be minimized, results of emergy accounting do not provide a clear directionality to decisionmakers, hence their further processing into indicators. In contrast, Rugani and Benetto (2012) proposed a strategy to improve EME with the use of LCI databases, allowing more precise insights of the network of technological processes involved in the production of man-made products. The resulting inventory of elementary resources could be then converted in emergy terms, using a consistent UEV dataset. The main challenge lies in the application of the specific emergy algebra to LCI datasets. To this aim, an operational software, SCALE (Marvuglia et al. 2013a), has been developed. SCALE uses the emergy algebra applied on the technospheric network of processes in order to calculate the emergy value of man-made products. However, LCI datasets do not cover the whole set of inputs to account for in EME. Therefore, although SCALE is an important step toward a comprehensive, fully operational and integrated tool for hybrid, lifecycle-based emergy evaluation, several technical challenges and recent methodological advances need to be investigated in details in order to shape a roadmap for the next steps (see chapter 2). Moreover, the hybrid, lifecycle-based EME framework needs further testing on actual case studies to identify potential improvements Freshwater and related ES as case study for resource characterization Freshwater is a vital resource for human development: 20 percent of the volume of world s rivers is withdrawn for human activities, and per-capita withdrawals are twice as important as a century ago (Shiklomanov 1997; WHO 1997). As a result, 40 percent of the world s population and 80 countries are currently concerned by water shortage (Gleick 2000). This critical situation is a threat for both human societies and ecosystems. Therefore, freshwater is an important resource to take into account in environmental accounting. It must be considered with the following important specificities. First, it is a local resource, with a highly variable availability in time and space. There is no international market able to counterbalance its heterogeneous distribution, since storage and transport on long distances are not economically reasonable options. However, its local usage can be dedicated to the production of exported goods. The recently developed Water Footprint concept (Hoekstra et al. 2011) estimates the amount of freshwater consumed and imported in a 35

36 territory (Stoeglehner et al. 2011; Zhang et al. 2011b; Feng et al. 2012) or in a product s life cycle (Chapagain and Orr 2009; Chapagain and Hoekstra 2011; Jeswani and Azapagic 2011; Berger et al. 2012; Jefferies et al. 2012; Zonderland-Thomassen and Ledgard 2012). Exchanges of virtual water (i.e. the water footprint of traded products) between territories are assessed from global to local scales (Chapagain and Orr 2009; Yang et al. 2012; Cazcarro et al. 2013), while the water impact is best modeled by considering the local conditions of water scarcity (Ridoutt and Pfister 2010; Jeswani and Azapagic 2011; Jefferies et al. 2012; Lenzen et al. 2013). A second specificity is the complex evaluation of freshwater quality. As already mentioned, exergy has been proposed as a common indicator to assess the various outcomes delivered by rivers (Chen and Ji 2007; Huang et al. 2007a; Chen et al. 2009a; Martínez et al. 2010), such as pollution dilution (chemical potential), power transmission (geopotential, kinetic and power exergy), heat regulation (thermal exergy). However, freshwater quality requirements actually depend on its usage (Boulay et al. 2011a), and are generally formulated with multiple indicators. For example, water potability is determined by measuring the concentration of multiple chemical substances and biological agents, and comparing them to local legal standards (Rickwood and Carr 2008). This aspect is not easily measured with exergy: freshwater chemical exergy measures the degree of purity, which, for potable water, does not guarantee quality since some harmful substances present even in very low quantities may prevent water from being potable. In contrast, orange juice or sparkling water include a lot of impurities, but remain drinkable beverages. Exergy was also proposed as a common unit to measure waterborne pollution, by summing up the chemical potential of dissolved and suspended substances (Huang et al. 2007a; Martínez and Uche 2010), leading to a river quality indicator. However, as described in LCIA models, the toxicity of a substance do not depend only on its concentration (Rosenbaum et al. 2008), and the various types of pollution cannot be limited to the concept of toxicity (Goedkoop et al. 2009). In addition, substances chemical potential is determined by comparing its concentration with a reference level (Szargut 2005). Hence, a global-average reference level for river composition (Chen and Ji 2007) may account for lower river quality sometimes as a negative term (e.g. a river with few dissolved oxygen), sometimes as a positive term (e.g. a river with a high load of heavy metals). The reference level for water composition should be defined on a case-by-case basis, by the local stream s natural composition of dissolved and suspended elements. A third specificity of freshwater is the large number of related Ecosystem Services (ES) (Wilson and Carpenter 1999; de Groot et al. 2002; MEA 2005a, 2005b; Bai et al. 2011; Klöve et al. 2011; Tompkins et al. 2011; Ojea et al. 2012), including freshwater-consuming services (e.g. food and fiber production, energy storage and hydropower, species habitat, spiritual and recreational areas) and freshwater-producing services (e.g. soil storage and retention of water for irrigation and drinking). The water cycle is also involved in many other ES, such as climate regulation, pollution control, erosion protection, soil formation and nutrient cycling (see e.g. MEA 2005a for an extensive list related to wetland ecosystems). The study of mechanisms related to freshwaterdependent ecosystem services is subject to flourishing research in the last decade, which highlights the complexity of natural processes involved. For instance, stream flows need to be regulated to deliver convenient habitat for fish; sediment regulation can minimize the perturbation of land cycle for terrestrial and aquatic ecosystems development and maintenance (Vigerstol and Aukema 2011). Land cover (Egoh et al. 2008; Haygarth and Ritz 2009; Posthumus et al. 2010) and terrestrial biodiversity (Bai et al. 2011) are parameters of primary importance for flow regulation and pollution control. Forest conservation is a key factor to store and supply freshwater for human consumption (e.g. Morri et al. 2013) and hydropower facilities (Guo et al. 2001) as 36

37 Chapter 1 Introduction and research questions well as to mitigate flood events (Biao et al. 2010; Bekiroğlu and Eker 2011; Ojea et al. 2012). Indirect human needs for freshwater may amount up to twice as much as direct freshwater consumption (Ferng 2007). As a result, traditional and oversimplified water management policies, focused on freshwater provisioning (and to some extent flow and quality regulation) should be broadened to include the support functions of freshwater flows (Jewitt 2002; MEA 2005b; Sullivan and Meigh 2006; Cook and Spray 2012; Lloyd et al. 2013), encompassing both waterdelivering ES and water-consuming ES. In summary, sustainable freshwater management is closely connected with the management of other renewable resources such as land; the same resource supports various users, which rely on diverse ES and different levels of quality requirements. Their actions may influence downstream beneficiaries; therefore, management policy must be envisaged over a coherent territory (typically a watershed) and overcome the local scale. Such scheme illustrates the inherent complexity of renewable resource management, which can be enhanced with sophisticated tools and models able to consider the spatial and temporal dynamics of environmental-economic systems (Bai et al. 2011; Vigerstol and Aukema 2011; Nedkov and Burkhard 2012; Ausseil et al. 2013) LCA characterization of freshwater and ES In the current LCIA framework, water-related impacts are restrained to the damages of pollution of waterborne ecosystems (Koehler 2008; Bayart et al. 2010). Impacts related to water deprivation for further use (i.e. scarcity-oriented impacts) were historically omitted because this resource was considered abundant in countries where LCA was being developed (Koehler 2008; Berger and Finkbeiner 2010). Moreover, the application of LCIA methods concerning resource depletion to freshwater is facing strong methodological challenges, which originate from the specific features of this resource illustrated in the previous section. However, recent initiatives were developed within a UNEP-SETAC workgroup dedicated to Water Use in LCA (Koehler 2008; Kounina et al. 2013). The first outcome of this workgroup is a consensual terminology of freshwater resources and use. Freshwater resources are renewed by precipitation (liquid or solid) and stored in different compartments: groundwater, surface reservoirs and streams, and soil moisture. Groundwater and surface water bodies are considered as blue water, while green water is the water stored in soil and biomass, and grey water is the additional freshwater necessary to dilute pollutants down to an acceptable level (defined by national standards). This terminology originates from the Water Footprint concept (Chapagain and Hoekstra 2011; Hoekstra et al. 2011). Bayart et al. (2010) proposed to classify freshwater use into in-stream (e.g. hydropower, transport, damming) and offstream (withdrawal from the water body for e.g. water supply, irrigation, cooling). Besides, if the amount of water abstracted from the stream is higher than the amount of water released, for instance due to evaporation loss or incorporation into a product, then the use is defined as consumptive; in contrast, when the amount of water released is equal to that of water abstracted, but the quality is altered, then the use is defined as degradative. Water users are considered in competition when appropriation by one user temporarily reduces the availability for the others. Competition may also occur between human users and ecosystems (e.g. irrigation around the Gobi desert in China, desiccating local lakes as shown in Liu et al. 2013b), whether exploited for human purposes (fishing activity) or not (habitat for aquatic ecosystem). Finally, water is considered as being depleted if the regeneration rate is longer than the overall consumption rate. This can be particularly relevant for groundwater resources. More recently, water users were 37

38 categorized according to their quality requirements (Boulay et al. 2011a), in order to identify more precisely the cases of competitive use between them. Indeed, an upstream degradative use may involve a loss of functionality of the stream for downstream users, thus involving additional treatment using energy and chemicals, which in turn may lead to additional environmental impacts elsewhere. While direct effects or freshwater use are extremely local, indirect effects can be extremely diffuse. Boulay et al. (2011a) distinguishes eleven types of users with specific quality requirements. Thresholds are defined from the acceptable levels of chemical and biological pollution (using 136 parameters in total) for each usage, thus providing 17 quality categories of blue water resources, enabling the estimation of per-category amount of available resources in a river basin. The categorization of water use enables a more precise description of mechanisms that link water uses and environmental impacts, as well as an enhanced description of water flows in future LCI databases. The variety of usages and diverse social and natural water-related contexts lead to even more various impacts on both human development and the integrity of ecosystems, as illustrated in Figure 1.2 (from Kounina et al. 2013) and further detailed below. Figure 1.3: Cause-effect chains, from inventory of water flows to the three Areas of Protection (AoP). Kounina et al. (2013). In the past few years, complementary though partly overlapping approaches were proposed to quantify these environmental mechanisms, focusing on three general environmental concerns (Bayart et al. 2010): sufficiency for human users, for ecosystems, and for future generations; they correspond to the three AoP human health, ecosystem quality and resources. Concerning human health, the environmental mechanisms covered in current developments concern the increase of 38

39 Chapter 1 Introduction and research questions infectious diseases, due to a more vulnerable hygiene when water is scarce (Boulay et al. 2011b; Motoshita et al. 2011) and the increase of diseases related to malnutrition, due to lack of water for irrigation (Pfister et al. 2009; Boulay et al. 2011b). The three methods develop a region-specific midpoint impact related to stress on (or accessibility to) blue water resources, which is then translated into damages on human health. However, the spatial scale is different: Motoshita et al. (2011) proposes country-specific results, while Boulay et al. (2011b, 2011c) considers impacts at the watershed level and Pfister et al. (2009) rely on 0.5 grid-cell maps. Besides, Boulay et al. (2011c) develop a stress index as midpoint characterization factor per watershed and water quality, and consider that human health is unlikely to be affected by water stress in high-income countries; instead, compensation mechanisms are envisaged for further treatment of lower-quality water with available backup technology, with their related impacts. Concerning impacts on ecosystems, Pfister et al. (2009) investigate the relationship between stress to surface water resources with the damages on vascular plant species, due to water-limited Net Primary Production (NPP). Van Zelm et al. (2011) focus on the effects on changes in groundwater table to soil moisture and the species richness in terrestrial vegetation in Netherlands. In contrast, Milà i Canals et al. (2008) also take into account green water as an available resource for ecosystem development, but limit their development to midpoint impacts of water deprivation due to land use change and blue water evaporative use. While Pfister et al. (2009) provide characterization factors at the grid-cell level (0.5 ), the method of Milà i Canals et al. (2008) is site-generic. Finally, impacts on resource depletion is assessed by Pfister et al. (2009) by the intervention of seawater desalination to compensate the fraction of consumptive water use that involves depletion. The impact considered is the surplus energy required by this backup technology, following the principles adopted in the LCIA method EI99 (Goedkoop et al. 1999). Milà i Canals et al. (2008) define a midpoint impact by adapting the Abiotic resource Depletion Potential (ADP) formula (Guinée et al. 2002) to water stocks, i.e. groundwater in aquifers. Exergy is also used to characterize blue water resource (Bösch et al. 2007; Dewulf et al. 2007). However, this method is common to all types of resources and does not take into account region-specific water scarcity (Kounina et al. 2013). The afore-described methods to account for impacts specific to water use in LCA were developed very recently. They are now being tested on case studies of water-intensive activities in agriculture (Ridoutt and Pfister 2010; Faist Emmenegger et al. 2011; Brandão and Milà i Canals 2013) and industry (Muñoz et al. 2010; Pfister et al. 2011; Igos et al. 2013a; Van Hoof et al. 2013). They definitely contribute to important advances in LCIA by filling a gap in covered environmental mechanisms, although time is needed for more testing and formal integration into spatially-explicit LCA frameworks under development (Weidema et al. 2011; IMPACT World+ 2012; Ecoinvent 2013). Interestingly, these methods focused their investigation on the use of water as a primary material for human health, ecosystems and future generations. But freshwater is a resource that cannot be considered only as a natural good, because of the numerous and diverse ecosystem services related to it. Heuvelmans et al. (2004) proposed a new impact category called regional water balance to account for drought and flood risk in LCA, by performing a statistical analysis of a river s daily and weekly flow rates and retrieve threshold levels of very high and very low flows. This can be considered as a pioneering work toward considering impacts on freshwater flow regulation service. However, the large amount of river-specific data necessary to build an operational dataset refrained from further development of this appealing approach. More recently, another UNEP-SETAC initiative working group (Koellner et al. 2013b) develops methods to account for impacts on ES in LCA. Their approach is to describe the environmental 39

40 mechanisms to relate intensive land occupation and land transformation (Koellner et al. 2013a) with impacts on biodiversity (Curran et al. 2011; de Baan et al. 2013; de Souza et al. 2013), biotic production (Brandão and Milà i Canals 2013) and regulating services on climate, freshwater and erosion (Müller-Wenk and Brandão 2010; Saad et al. 2011, 2013). This integrated and spatiallyexplicit framework is another important progress in LCIA, likely to be implemented soon in operational tools. However, the proposed environmental mechanisms were restricted to the midpoint level, i.e. on potential impacts on biodiversity and ecosystem services (Koellner et al. 2013b). The environmental mechanisms linking these potential impacts with damages on human health, ecosystem quality and resource depletion were not considered. As a result, there is no settled standard or agreed LCIA method for the accounting of freshwaterand ES-related impacts. Current developments need to be further enhanced and merged. Some aspects remain out of research areas, such as: the relationship between pollution and ecosystem services; the potential synergetic effects of simultaneous environmental mechanisms (e.g. influence of climate change on water availability, interdependence between soil erosion, biotic production and water availability, etc.); as well as potential cycling of environmental mechanisms (e.g. damages on human health require more health care goods and services, leading to an increased level of resource consumption and pollution; damages on ecosystems alter their ability to treat pollution, leading to increased health impacts and loss of natural resources). In other terms, in LCIA models, environmental mechanisms are assumed to operate independently from each other. This assumption may not hold much longer if impacts on ES are to be accounted for in LCIA, because they are all interdependent EME and characterization of freshwater and related ES Determining the emergy value of natural resources used up by an activity is a prerequisite to perform emergy accounting and calculate emergy indicators. By estimating the previous contribution of natural processes to deliver ES, i.e. their ecocentric value, emergy provides a viewpoint contrasting with the user-oriented, economic valuation approach (Odum and Odum 2000; Voora et al. 2010; Pulselli et al. 2011a; Coscieme et al. 2013). For example, the UEV of freshwater is built on the estimation of the exergy used up by global and local processes that drive the water cycle. Therefore, it does neither involve the notion of local scarcity (which is related to the user-side consumption rate), nor the aspect of freshwater quality (which depends on its usage), contrary to current LCIA approaches related to freshwater resource characterization. Natural resources and ES are evaluated with emergy according to their corresponding temporal and spatial scale. Supporting services provided by the hydrological, carbon and nitrogen cycles were studied at the global scale (Watanabe and Ortega 2011) and at the river basin level (Watanabe and Ortega 2014). Local ES are often the co-products of a same ecosystem: for instance, forests deliver simultaneously regulated stream flows, timber, maintenance of litter, biomass, maintenance of biodiversity, recreational areas (Tilley and Swank 2003), and an ecopark provides e.g. wastewater treatment services, as well as biodiversity protection, recreation, education support (Duan et al. 2011). The UEV of freshwater resources and related ES are determined from the global water cycle (Buenfil, 2001; Campbell, 2003; Odum, 2000). The UEV of rainfall is determined for both its chemical potential and geopotential, as co-products of global processes. While the UEV of chemical potential is a global average, the UEV of geopotential is calculated according to land 40

41 Chapter 1 Introduction and research questions elevation (Odum 2000). Therefore, a global value of river freshwater UEV could be determined (Odum 2000) and further used in emergy evaluation of territories (Odum et al. 1998; Chen and Chen 2009; Lv and Wu 2009; Giannetti et al. 2013b). However, several studies use a common method to calculate the local UEV of a stream, based on the catchment area and average elevation, precipitation and stream flow data. This method was applied to compare the emergy value of different rivers in a country in order to evaluate the importance of freshwater for the national economy (Chen et al. 2009b), to compare the monthly-varying UEV of streams between sub-basins of a watershed (Brown et al. 2010), or at selected points of interest within the basin (Odum et al. 1998; Pulselli et al. 2011b). The same conceptual framework was also used with high resolution, using GIS to visualize the spatial evolution of stream transformities within a watershed (Huang et al. 2007b). Spring water can also be included (Pulselli et al. 2011b), by considering this resource as independent from rainfall, since it is regenerated by processes operating on a different time scale. However, a deeper analysis of these applications shows disparities in data sources and accounting. A consistent, spatially-explicit dataset of freshwater UEV may be useful for further study of local activities and territories, which rely on surface water bodies (see chapter 6). Stream flow regulation and water quality regulation are less investigated in past emergy evaluation research. The flow regulation capacity of human infrastructure was evaluated in emergy terms from its storage volume (Brown and McClanahan 1996) or by the costs avoided by enhanced flood control (Kang and Park 2002). The drawbacks caused by human artificial regulation (natural costs e.g. sediment loss for downstream users; social costs e.g. population displacement; economic costs e.g. maintenance, construction, and change in human activity) could also be evaluated in emergy terms. Carey et al. (2011) looked for correlations between water quality monitored in streams (via nitrogen oxides and total phosphorous) with human disturbance estimated with emergy-based land use indexes. Other studies use dynamic modeling based on energy system diagrams to evaluate the UEV of streams and reservoirs (Tilley and Brown 2006), or the river quality and sediment composition under changing local conditions (Scariot et al. 2007). Although these models provide deeper insights of the system s functioning and deliver more precise results, they require much larger amounts of data, calibration and result analysis, along with a deeper expert knowledge on the specificities of the studied system. 1.3 Research questions and methodological approach Environmental assessment tools are meant to inform decision-makers about the long-term sustainability of their activity. To this aim, the extent to which ES support human activities and, in turn, are affected by them shall be consistently evaluated. Two complementary visions of environmental accounting were considered in this study. On the one hand Life Cycle Assessment (LCA), focusing on assessing the environmental impacts of natural resources extracted from the natural environment and pollutant emissions by product s lifecycles. On the other hand Emergy Evaluation (EME) focusing on the natural processes supporting the formation of the resources used up (directly and indirectly) by a human activity or a territory. The former approach adopts a user-oriented perspective to assign responsibility of the potential environmental impacts to the various processes involved in the lifecycle chain. As such, it relies on databases with a high level of details to represent the network of technological processes and the mechanisms relating human interventions to environmental impacts. However, current LCA methods fail short in considering the complex, interconnected natural processes that generate 41

42 renewable resources and ES. The latter approach, EME, adopt a donor-oriented and holistic approach to evaluate the contribution of natural processes to support a human activity. However, EME is criticized for its low level of standardization and low accuracy when man-made inputs are characterized, leaving room for subjective interpretation of emergy accounting framework and results. Despite their fundamental differences, recent research demonstrated the potential outcomes of a mutual integration of these two environmental accounting approaches. Regarding resource characterization, freshwater is worth being taken as a case study for such integration. As a local resource, related to many ES and supporting various usages with different quality expectations, freshwater must be managed along with other renewable resources. Current developments related to freshwater-related impacts in LCIA propose spatially-explicit models. However, these models consider freshwater only as a mere material (used in both natural and anthropic processes). Other developments focus on ES in general, but the proposed models investigate environmental mechanisms independently from each other. In contrast, EME relies on a holistic framework to characterize all kinds of resources in a similar manner, but may be improved by the use of more detailed, site-specific information. For both methods, sophisticated models and tools could help enhance the characterization of renewable resources. As a result, three research questions are addressed by this PhD project: What are the main challenges to improve the robustness of EME? What is the added value of a hybrid lifecycle-emergy framework? How to consider the spatial distribution and temporal dynamics of renewable resources and ecosystem services in environmental accounting (both in LCA and EME)? The first question is investigated in chapters 2 and 3, in which a critical analysis of strengths and weaknesses of the EME framework is proposed, to identify potential improvements, notably through the development of the hybrid lifecycle-emergy framework. Then, a case study is evaluated with the conventional EME approach, in order to illustrate the differences and complementarities with LCA. The second research question is addressed in chapters 4 and 5. The hybrid framework is applied to the same case studies, in order to shed light on its added value and suggest approaches to overcome the remaining challenges. A closer look is given to the meaning, calculation and interpretation of the emergy-based indicators, as compared to the conventional EME framework. Chapters 6 and 7 concern the third research question. GIS software is used to develop a spatiallyexplicit emergy database of freshwater resources, and a methodological approach is investigated, to develop LCIA characterization factors of ecosystem services via a global, dynamic model, which integrates the mutual influence of natural processes and human welfare. 1.4 Thesis outline Chapter 2 gives more insights on the integration process of LCA models and datasets into EME. It identifies the main limitations of current EME procedure as being a lack of an operational and universally accepted mathematical definition of emergy, a lack of automated tools to ensure the reliability of previous EMEs for further use, room for personal understanding of system boundary delineation and results interpretation, and a lack of robust definition and critical analysis of emergy-based indicators. Previous attempts to combine EME with LCA are analyzed, as well as 42

43 Chapter 1 Introduction and research questions the challenges for a full integration. The chapter presents the recent advances in theory, methodology and software development oriented toward overcoming these issues, their implication for current research and proposals for the next steps. Chapter 3 examines the applicability and robustness of the EME framework, via a case study of four water treatment plants. Man-made inputs such as chemicals, electricity, labor and services and infrastructure materials are important contributors of the UEV of potable water. This result is characteristic of industrial activities that rely on a single local, renewable resource. The chapter illustrates the relative low accuracy of EME in the accounting for man-made inputs, as compared to LCA. It also highlights the complementarity of both environmental accounting methods. For example, EME allows a more holistic assessment of resource use by better relating the contribution of local, renewable resources (here, freshwater) as well as human labor and services. The resulting emergy-based indicators are proven useful to benchmark the ecological performance of a wide range of human activities. In contrast, LCA also routinely accounts for impacts on human health and ecosystems. EME could benefit from the detailed representation of man-made inputs lifecycle, in a hybrid framework that would allow a more accurate accounting procedure. Chapter 4 relates the first consistent application of the SCALE software (Marvuglia et al. 2013a), to the case studies presented in chapter 3. Results are compared with those provided by the conventional EME accounting procedure and by the SED method (Rugani et al. 2011a). They show that emergy accounting of man-made inputs with SCALE is more detailed than in the conventional EME framework, and rigorously applies the emergy algebra, which is not possible with the SED method. This chapter also illustrates the deeper level of analysis provided by the application of SCALE, since the emergy value of man-made imported materials and energy can be further decomposed. However, the application of SCALE is limited by the quality and comprehensiveness of the LCI database used. For example, it cannot account for human labor and ecosystem services. In addition, the current version of SCALE cannot consider the specific UEV of the local resources used up, and disregards the impacts of pollution on human health and ecosystems. Complementary approaches to account for these aspects are examined in order to identify priority improvements for the software. Chapter 5 investigates the consequences of the hybrid lifecycle-emergy framework on emergybased indicators. The Emergy Yield Ratio (EYR), the Environment Loading Ratio (ELR) and the Emergy Sustainability Index (ESI) are subject to a critical review in literature, which highlights the need for a consensual, standardized delineation of system boundaries in EME. A semantic study of their definition and formulation is performed, in order to highlight their differences and complementarities in the conventional, site-oriented EME framework and in the hybrid, lifecycle oriented one. In addition, an algorithm is specifically developed to make operational the calculation of EYR in the hybrid framework, following the definition of Brown et al. (2012). Finally, the study of partial derivatives is suggested as a useful tool to identify potential actions for decision-makers and improve the ecological performance of their activity. Chapter 6 explores the use of Geographical Information System (GIS) software to develop a spatially-explicit emergy database of freshwater flows. Calculations are made at high resolution (30-arcsec, i.e. approx. ~ 1km 2 per grid cell) over terrestrial land (except Antarctica), using freelyavailable maps of precipitation, elevation and evapotranspiration. The emergy value of rain input calculated for each grid cell could be useful for example to refine the National Environmental 43

44 Accounting Database (NEAD, CEP 2006). The UEV of rivers, in sej/m 3, is also calculated at high resolution, using a flow accumulation algorithm. The resulting dataset is freely available by practitioners to retrieve local UEVs of rivers. Region-average UEVs can be used as a site-specific background dataset in the hybrid lifecycle-emergy framework. In the long term, this approach denotes the possibility of developing a spatially-explicit reference model for natural resources evaluation in emergy. However, the quality of results shows that further refinement is necessary, and next steps are formulated accordingly. Chapter 7 presents a pioneering method to take into account the impacts on Ecosystem Services (ES) in LCIA. While current developments consider environmental mechanisms impacting the production of ES as independent from each other and devoid of temporal dynamics, the proposed approach investigates the use of a dynamic model named GUMBO (Boumans et al. 2002) to quantify human-driven changes on ES production and human welfare indicators, in order to retrieve Characterization Factors (CFs). GUMBO is a simplified, global meta-model, which dynamically connects the mechanisms of natural systems (that produce ES) with those of the human society (which consumes ES). The anthroposphere is considered simultaneously relying on and affecting the geobiosphere. Calculated CFs can be adapted to the duration and/or date of the perturbation, as well as the presence of other environmental interventions and the global policy scenario adopted in the 21st century. Although this work remains preliminary, outcomes show that the conceptual approach could be beneficial to overcome the inherent limitations of current LCIA methods, composed of constant CFs. A roadmap is presented to further investigate the use of integrated dynamic models in LCA. Chapter 8 reminds the main outcomes of this study detailed in the previous chapters, concerning the promises and added value of the hybrid lifecycle-emergy framework, as well as the immediate suggestions for further improvements and the long-term roadmap. NB. This work was already published in 6 peer-reviewed scientific publications (corresponding to chapters 2 to 7). The papers content and structure correspond with the different subjects exposed here above, thus we decided to structure the thesis report by including the original papers in the related chapters. 44

45 2. Accounting for the emergy value of life cycle inventory systems: insights from recent methodological advances Published as Arbault, D., Rugani, B., Marvuglia, A., Tiruta-Barna, L., Benetto, E., Accounting for the emergy value of life cycle inventory systems: insights from recent methodological advances. JEAM 1, Abstract Emergy, due to its unique ability to translate into a single metric the memory of the geobiosphere exergy (environmental work) supporting any (technological or natural) system, has the potential to offer a new perspective of environmental assessment to support decision-making. Previous work by a number of researchers has pointed out the expected advantages of taking a hybrid approach combining Emergy Evaluation (EME) and Life Cycle Assessment (LCA). In particular, emergy calculation using Life Cycle Inventory (LCI) databases and LCA matrix-based formulation is claimed to have the potential to increase the reliability of emergy-based evaluations and thereby the applicability of the emergy concept in environmental decision-making. The paper points out the main obstacles to overcome in order to reach this consistent integration, highlighting the progresses made so far in this direction, until the most recent practical and operational advancements.

46

47 Chapter 2 Accounting for the emergy value of life cycle inventory systems: insights from recent methodological advances 2.1 Introduction Developing and testing reliable and holistic environmental accounting tools has been the objective of many scientists and public and private authorities for several years. While the literature offers many examples of tools and methods oriented to assessing and/or monitoring the effects of human activities on the land appropriation (i.e. Ecological Footprint and related methods, Rees and Wackernagel 1996) and resource consumption such as Water Footprint and related measures of water scarcity (Chapagain and Orr 2009; Pfister et al. 2009), Material Flow Accounting (see Huang et al for a historical review) and Energy and Exergy-based evaluation approaches (Bösch et al. 2007; Dewulf et al. 2007; Schneider et al. 2011), there is no universally accepted unique metric to assess comprehensively the impacts of the anthropic activities on human wellbeing, ecosystem health and biodiversity. Integrating existing and possibly complementary methods may offer mutual benefits and contribute to research advancements in the environmental accounting domain. For example, in recent years several researchers have worked on the application of hybrid models for the integration of Emergy Evaluation (EME) and Life Cycle Assessment (LCA) (Rugani 2010; Zhang et al. 2010a, 2010b; Ingwersen 2011; Rugani et al. 2011a; Rugani and Benetto 2012; Marvuglia et al. 2013a). As a nature-centered concept, Emergy provides a unique metric to account for the total contribution of the very diverse set of geobiosphere processes which are responsible for the formation of natural (but not only) resources (Odum 1996). Consequently, it potentially lends itself to be used as a suitable yardstick to assess the long-term sustainability of their extraction and use. While most methods consider mankind and the natural environment as two separate entities, emergy adopts a unique standpoint, which considers human activities as an integrated part of the global Earth system (the geobiosphere). Despite this appealing viewpoint, EME suffers from a scarce acceptance by decision-makers, most probably because of its lingering modeling uncertainties which can sometimes lead to unsubstantiated claims. There are many reasons for that, among which 1) its conceptual background lacks an operational and universally accepted mathematical formulation (Tiruta-Barna and Benetto 2013), which weakens the calculation of the emergy value of natural resources; 2) standard definitions, tools and calculation routines have been only rarely designed, which affects the reliability of previous emergy-based studies of human activities, either for further use as data sources, or for comparative purposes (benchmarking); 3) its accounting system is not fully compatible with detailed datasets of human activities, be they expressed in economic or physical terms; EMEs thus rely on coarse information, leading to inaccurate results; 4) the small size of the community of emergy experts does not foster further development and refinement: many aspects of environmental sustainability have been dealt with by emergy scholars, such as pollution and impacts on human and ecosystems (Bastianoni 1998; Ulgiati and Brown 2002; Liu et al. 2013a), recycling and waste management (Buranakarn 1998; Amponsah et al. 2011; Song et al. 2012; Giannetti et al. 2013a), ecosystem services valuation (Odum and Odum 2000; Voora et al. 2010; Zhang et al. 2010a, 2010b; Huang et al. 2011; Pulselli et al. 2011a; Watanabe and Ortega 2011), ecological footprint (Brown and Ulgiati 2001; Chen and Chen 2006; Liu et al. 2008; Siche et al. 2010; Pereira and Ortega 2012), territorial planning (Brown and McClanahan 1996; Tilley and Swank 2003; Pulselli et al. 2008a; Chen and Chen 2009), compensation mechanisms (Bastianoni et al. 2004). However, those isolated and often purely academic attempts usually miss a proper follow-up to handle the numerous interesting multi-disciplinary improvement steps necessary to progress from a speculative formulation to a practically viable application to concrete assessments. 47

48 Conversely, Life Cycle Assessment (LCA) is a user-oriented method (as opposed to nature- or donor-oriented approach), which seeks estimating the environmental impacts generated by a good or a service (ISO 2006; European Commission 2010c). This standardized approach (ISO 2006) relies on a dense network of processes built on data collected from industrial systems worldwide. The emissions and resource extraction of each process is cumulated over the product/service lifecycle and converted to environmental burden. Evaluation of impacts on resources may hinge upon the notion of scarcity or may be expressed in terms of potential thermodynamic work (Jolliet et al. 2003b; Goedkoop et al. 2009; European Commission 2010a), which are ultimately a humanoriented valuation rationale, i.e. reflecting the resource utility for humans. Although both approaches target decision making support, EME and LCA appear to be very different and disconnected from each other. While earlier studies provided a critical comparison of their advantages and drawbacks (e.g. Hau and Bakshi 2004), more recently scholars have more and more oriented their efforts towards highlighting their complementarities and the added value of both methods to their potential integration. Focusing on the use of LCA models and datasets in EME, the aim of the present paper is twofold: i) discuss the current state-of-the-art in the implementation of lifecyclelife Cycle based emergy evaluations, mostly based on our recent findings, and ii) advance methodological improvements and application proposals to overcome the critical issues that keep hampering a comprehensive and consensual integration between emergy and LCA. A brief overview and a survey on EME and LCA in section 2.2 help understand the added value of hybrid models and the extant restrictions to their development. The critical review of advances in combination and hybridization in section 2.3 points out the challenges unveiled and issues yet to be solved. Section 2.4 provides an outlook to further strengthen the proposed Emergy-LCA approach. 2.2 Methods Emergy Evaluation (EME) The concept of emergy was introduced by H.T. Odum around 40 years ago (Odum 1973). As a system ecologist, he noticed that ecosystems tend to capture sparse energy flows and concentrate them via trophic webs. The emergy value associated to a good or a service is the memory of the energy flows throughout the chain of self-organized systems that support its production, from the first capture of solar energy. It is thus defined as the direct and indirect [solar] energy used up for its production (Odum 1996). Its Unit Emergy Value (UEV) is the emergy per output unit. Therefore, EME provides a metric to estimate how much solar energy (captured via direct radiation, previously transformed into wind kinetic energy, rain chemical potential, etc. or stored in slowly-renewable or non-renewable stocks such as fossil fuels, concentrated ores, soil organic matter, etc.) has been necessary to support directly and indirectly a human activity. However, the study of energy pathways within complex systems, such as ecosystems and living organisms, geological processes and economic systems is very difficult; a comprehensive assessment of energy flows is highly challenging. To overcome this difficulty, emergy evaluation adopts a topdown approach that relies on four elementary accounting rules (Brown and Herendeen 1996; Odum 1996): 48

49 Chapter 2 Accounting for the emergy value of life cycle inventory systems: insights from recent methodological advances (i) (ii) (iii) (iv) all source emergy to a process is assigned to the process output; co-products from a process (e.g. wool and meat from sheep, or grain and straw from wheat) have the total emergy assigned to each pathway; when a pathway splits, the emergy is assigned to each leg of the split based on its percentage of the total exergy flow on the pathway; emergy cannot be counted twice within a system: (a) emergy in feedbacks cannot be double-counted, and (b) the emergy of co-products, when reunited, cannot be added to obtain an emergy value greater than the original emergy source from which they were derived. The second and fourth-(b) rules are the most critical since this allocation approach between coproducts consistently differs from most of the other environmental accounting methods. These rules, though advantageous to explain and illustrate the conceptual breakthrough of emergy, lead to several shortcomings. First, emergy scholars have not yet reached a consensus of the formal mathematical definition of emergy. The need for such formal statement can be illustrated by the research activity on this subject (Giannantoni 2006; Bastianoni et al. 2007, 2011; Kazanci et al. 2012; Tiruta-Barna and Benetto 2013). Noticeably, there exist other theories on system ecology rooted in thermodynamics, which are based on widely agreed mathematical and physical definitions but have not yet led to a concrete application to environmental accounting (Schneider 1994; Ulanowicz 1997; Fath et al. 2001; Kleidon and Lorenz 2005; Jorgensen 2006; Jorgensen and Nielsen 2007; Dewar and Porté 2008; Ulanowicz et al. 2009; Chen et al. 2010; Patten et al. 2011). Emergy rules were indeed successful in early applications of environmental accounting because they require little knowledge about the studied systems if compared to all other accounting methods (no local balances and process efficiencies needed). Second, imprecise (and sometimes contradictory) definitions of the core concepts in the emergy theory (e.g. the use of the term yield, emergy cannot be available or used, calculation of baseline; see Raugei et al. 2014) left space for personal interpretations, which did not encourage experts to standardize the accounting framework. Oppositely, so-called improvements and/or adaptation to Odum s original method flourished, without however undergoing any systematic, critical analysis. As a result, UEVs of natural resources are roughly estimated, most often without any discussion on uncertainties (Ingwersen 2011). Without a standard calculation process, sophisticated IT tools remain useless. Ultimately, the first UEV database was developed only recently (Sweeney et al. 2007). This well reflects the need for standardization tools, information sources and calculation routines. Third, the application to human systems remains highly wooly. Unsolved issues on boundary delineation of the activity to be studied relate to the accounting of Labor and Services (L&S) throughout the upstream production chain (risks of double-counting; some authors published results with and without L&S, to partially solve the issue; Sweeney et al. 2007), as well as impacts of emissions to the environment and human health via e.g. the notions of pollution and ecosystem services. In addition, this set of rules may lead to diverging results, depending on the level of detail in the decomposition of the studied system (Tiruta-Barna and Benetto 2013). Fourth, the final outcome of EME, i.e. information to support decision-making, is highly uncertain. Result interpretation remains sometimes fuzzy. Emergy-based indicators, which are 49

50 defined as aggregated indications of the system s sustainability (by comparing renewable / fossil, local / imported resources), lack robust definitions, critical analyses and transparent interpretation. The crucial question to be answered is how to have emergy-based indicators widely acknowledged as tools to pinpoint lever actions on decision making service. There is no doubt that these shortcomings largely explain the scarce adoption of EME as a tool for decision making support. Undoubtedly they make environmental accounting experts refrain from using emergy, which in turn hinders the construction of robust databases and the definition of accounting standards and ultimately the creation of a consensual vocabulary for dissemination. These criticisms of the emergy accounting framework are however meant to be constructive: despite a loose definition, the donor-side concept is truly innovative; a standardization is increasingly encouraged because emergy has an appealing added value compared to other approaches; calculated UEVs, though roughly, form a starting point that motivates a finer resolution; scoping and accounting techniques need to be refined, and existing comprehensive datasets can be useful for that purpose; emergy-based indicators, as currently stated, are a valuable specificity of EME and may benefit from a more accurate statement, without changing their general meaning. The question that motivates our research is how to improve EME and its reliability. As argued hereby, using the framework and datasets provided by LCA may favor significant enhancements to several of the afore-mentioned limitations Life Cycle Assessment (LCA) The basic principle of LCA is to estimate the overall environmental impacts of a product or a service along its life cycle, i.e. from the extraction of raw materials to the disposal of ultimate waste at end-of-use (ISO 2006; European Commission 2010c). The method starts with the definition of the functional unit to be studied. The foreground chain of processes necessary for the delivery of the functional unit is described as specifically and comprehensively as possible and covers all the stages of the life cycle; background processes are usually retrieved from comprehensive databases (e.g. Ecoinvent 2010), from which it is possible to quantify the bill of materials that are extracted from the natural environment, as well as the pollutant emissions. This physical input-output table forms the Life Cycle Inventory (LCI). LCIs are further converted into environmental impacts (Life Cycle Impact Assessment, LCIA) using static characterization factors (CFs) (Jolliet et al. 2003b; Finnveden et al. 2009; Goedkoop et al. 2009; European Commission 2010a). LCA experienced a fast development from its early stages in the 90 s to mature and transparent methodological requirements and capabilities. Present datasets (Ecoinvent 2010) gather up to 4,000 background processes into an interconnected network that describes the technosphere; extensive LCIA methods, such as ReCiPe (Goedkoop et al. 2009), are able to model the conversion of approximately 2000 so-called elementary flows (including resources, land occupation and transformation, pollutants emitted) into environmental impacts. This outstanding feature originates from the early concern for standardization, common nomenclature and procedure, which are now part of an official standard (ISO 2006). This enables the adoption of LCA by a wide community of scholars, practitioners and industries, which in turn is favored by and still favors a fast and continuous methodological progress. The mathematical background of LCA is rather simple to understand, which helps newcomers getting familiar with the method. This simple framework (based on linear systems solutions) has however been 50

51 Chapter 2 Accounting for the emergy value of life cycle inventory systems: insights from recent methodological advances continuously sophisticated and enriched by new findings (e.g. Heijungs and Suh 2002; Pfister et al. 2009; Levasseur et al. 2010; Curran et al. 2011; Saad et al. 2011; de Baan et al. 2013). LCA and EME show strongly similar procedures: the objective of the study is first defined, the boundaries of the system can be drawn accordingly, and flows across these boundaries quantified, to be finally converted into one or more impact scores, prior to results analysis and interpretation. The object of study is a man-made good or product delivered by a specific activity; the system boundaries embraces a network of processes connected by flows of materials and energy; the inventory and quantification of flows crossing the system boundaries require very similar data collection procedures; finally, the scoring into impacts is based on existing conversion coefficients, namely UEVs in EME and CFs in LCA. More precisely, LCI datasets provide robust, standard information of the connections between the technosphere and the natural environment, which are seen as valuable data sources to improve the resolution of EME results (Rugani and Benetto 2012; Raugei et al. 2014) Current status-quo on the combination between Emergy and LCA Combining EME and LCA may provide an added value for both methods. From the technical point of view, using UEVs as CFs for LCIA may bring a novel valuation system of resources, stressing their donor-side value as quantified by the amount of solar energy used up by natural systems to (re)generate them; it is complementary to existing approaches on resource depletion, which are mostly based on resource scarcity, i.e. they embed the value perceived by the user (a resource is considered more valuable if it is more scarce, because more scarcity implies higher costs to make it available to mankind). Conversely, the rich datasets of background processes used in LCI are valuable data sources for EME. From the conceptual viewpoint, LCA may benefit from the rationale specific to emergy, which is to consider human activities as an integral part of the geobiosphere. The driving question is no longer how to reduce environmental burdens but rather how to optimize the integration of human activities into its natural environment, from which they fully depend. On the other hand, LCA could serve as a reference of standardized framework for EME, thanks to its maturity and the similarities between the procedures of both methods. Figure 2.1 summarizes the main aspects of discrepancy between EME and LCA, which have fostered the research activity towards a consistent integration of the two methods. According to this goal, first attempts to combine EME and LCA (Zhang et al. 2010a; Ingwersen 2011; Rugani et al. 2011a; Brown et al. 2012; Marvuglia et al. 2013a) (further investigated in the next section) unveiled the following major issues: Issues associated with system boundary The description of system boundaries in LCA must be as large as possible, i.e. cover the most important processes that are part of the functional unit s life cycle. However, the scope of the system is reduced to industrial processes and their elementary exchanges with the natural environment, which is systematically considered outside the boundaries. On the contrary, the system boundaries of EME focus on geographically local processes, both natural and anthropic. It is up to the practitioner, however, to define what local means; though it is encouraged to consider all natural and human processes that occur within the boundaries, cutoffs are necessary to narrow down the system complexity, but difficult to set up. EME also includes human labor, which is considered a highly transformed energy source. Local human labor is evaluated inside 51

52 the system (Ortega et al. 2002) or as an imported resource. Time perspective is another point of divergence between the two methods: LCI provides information on elementary flows that occur during the life cycle of the functional unit, assuming industrial processes operating at steady-state. Conversely, EME has to consider much larger time-scales, so that the natural processes that (re)generate the local resources used up are fully considered. Long-term time scales are ultimately necessary for EME to seek a balanced integration of human activities into the natural environment they depend on. Issues of uncertainty in UEVs Thorough attempts to use UEV of natural resources as CFs revealed that they were scarcely reliable to be consistent with current LCIA standards, notably in terms of data accuracy and uncertainty estimation (European Commission 2010b; Ingwersen 2011). Most UEVs of natural resources are calculated with reference to the global scale, using an average regeneration rate of the resource. Though such approach is operational for a primary assessment of UEVs, the calculation does not consistently consider the global mechanisms at stake, such as geologic processes that concentrate minerals into ores and decompose organic matter into fossil fuels, and local water cycles that shape the landscape and form floodplains. Despite recent progresses provided a finer analysis of fossil fuel formation (Bastianoni et al. 2005; Brown et al. 2011), current UEV datasets remain coarse and without any uncertainty estimate, far from being compatible with the 130 mineral resources and fossil fuels, 42 types of land occupation and 80 types of land transformation, 5 airborne and 13 waterborne resources typically found in the Ecoinvent v2.2 database, which is the most widespread LCI database used in LCA studies worldwide. However, the latter only contains 7 types of elementary flows from biotic resources, for which UEV literature is instead richer. Using models developed for environmental economics studies (Boumans et al. 2002; for instance GUMBO and MIMES models: Boumans and Costanza 2007) may be a promising approach to describe, simultaneously and in an integrated and dynamic manner, the geobiosphere s mechanisms and the formation of natural resources, as well as interactions with mankind i.e. impacts of pollution on natural systems on one hand and influence of nature on human well-being on the other hand. The afore-mentioned models, however, suffer from not formally relying on thermodynamic processes, which prevents them from being eligible to the emergy algebra. Nevertheless, they can be inspiring to develop an integrated tool that would gather previous, static emergy-based models on resource formation and use into a single platform, and gradually incorporate knowledge from other disciplines dedicated to understanding natural mechanisms. Such approach may prove more consistent for emergy accounting than current practice, in which UEVs are retrieved from disparate studies. Discrepancy in mathematical formulation Another major difficulty of integrating both methods lays in the discrepancy of the mathematical formulation between them. Since they do adopt a fundamentally different viewpoint (donor-side approach for EME, user-side approach for LCA), they logically apply a different algebra to cumulate resource valuation along the transformation steps. Rule #2 of emergy algebra states that co-products of a process should be assigned the same emergy value, since they both require the entire upstream chain of processes to be produced. Rule #4 is thus of primary importance to avoid double-counting. In opposition, the user-side approach of LCA involves that the environmental burden should be split among all the co-products, since different users of co-products are supposed to share the impacts generated by a process. Additionally, EME is a top-down 52

53 Chapter 2 Accounting for the emergy value of life cycle inventory systems: insights from recent methodological advances approach, which aims at capturing the largest picture of a complex, coupled human-natural system. Its procedure aims at simplifying such (multi-scale) system by aggregating related processes. Though convenient, this can lead to inconsistencies in the application of emergy algebra and thus weaken the overall outcomes of the method (Tiruta-Barna and Benetto 2013). Conversely, computing the LCI of a functional unit using the detailed network of background processes is a bottom-up approach, in which a high level of detail of each process is expected (despite background processes accounted in databases are generic and do not exactly correspond to the individual background process of the studied system). Therefore, a specific algorithm was recently developed to adapt the network of processes in Ecoinvent to the specific accounting rules of emergy algebra. This algorithm, implemented in the software SCALE (Marvuglia et al. 2013a), was developed assuming that co-products of generic processes can be considered real co-products (i.e. as if they were actually produced at the same time and place), whereas actual flows that reunite in the production chain are most probably (in the real world) produced by distinct processes in time and/or space. This additional level of sophistication, which would allow reflecting the exact condition of the complex technosphere, has still to be integrated as it depends, however, from the detail of information included in LCI databases (Rugani et al. 2012a). These limitations unveiled by primary attempts on combining the methods demonstrated the added value of an EME-LCA integration, and encouraged further research in that direction. H T Inventory flows and database Indicators Life Cycle Extraction S R N EGS Geobiosphere Society (technosphere) Energy Materials UEVs estimation and use Refinement Production Use End of life Emissions Donor-Value Algebra issue User-Value Figure 2.1: Conceptualizing the main methodological discrepancies between Emergy and Life Cycle Assessment (LCA) to benchmarking their future combination and strengthening their capability to solve environmental monitoring issues. The left-hand side (green shapes) represents the emergy-based approach to environmental management. The right-hand side (blue shapes) represents the LCA-based approach. Red tags and arrow represent hybridization issues. EGS = Ecosystem Goods and services; R = Renewable local resources; N = Nonrenewable local resources; S = Solar radiation energy; H = geothermal Heat flow; T = Tidal energy. 2.3 Benchmarking the methods combination The few attempts to combine EME and LCA illustrate the gradual progress towards their mutual integration. Zhang and co-authors (2010b) first suggested the use of emergy to account for ecosystem goods and services in LCA. Since emergy provides the donor-side value of resources, it enables accounting for both renewables and non-renewables with a homogeneous rationale and 53

54 a common unit. Oppositely, the notion of scarcity does not conveniently apply to renewable resources, which partly explains why current LCIA methods only evaluate the depletion of nonrenewables. The Eco-LCA method (Zhang et al. 2010a) mostly includes provisioning and supporting services, while few regulating services are included and cultural services are left aside. UEVs of ecosystem goods and services are retrieved from the emergy literature. However, the authors adopted the conventional mathematical approach for LCA, i.e. allocation based on physical or economic flows. Therefore, this approach does not fully consider emergy algebra rules, but tries instead to reflect the real conditions of the technosphere co-production processes (as it is represented in Input-Output based LCI databases) keeping a user-side perspective. Additionally, the authors denote gaps in the inventory of flows in the database of background processes, since inputs from the natural environment typically exclude regulating services (e.g. pollination in agriculture) and supporting services (e.g. nutrient cycling, photosynthesis, soil formation). The use of UEVs in LCIA was also applied in the mining industry (Ingwersen 2011), in order to compare the contribution of man-made inputs with that of geologic processes for ore formation. That paper illustrates that using a background processes database may prove convenient to estimate the uncertainty of the results. However, such approach is limited by the low accuracy of natural resources UEV, which are computed from average regeneration rates rather than modeling of geobiosphere processes. Also, the decomposition of the system model appears to be far too simplistic and distant from reality, thus introducing significant modeling uncertainties (Tiruta-Barna and Benetto 2013). Ingwersen (2011) identified, however, the challenges of using emergy in LCIA, which are in agreement with a few of those highlighted in this paper: a different scope between methods prevented from including L&S in a consistent manner; database of background processes presents gaps that hamper the full accounting of natural inputs (e.g. renewable resources); rules for allocation cannot be easily handled in current versions of LCA software, and uncertainties in UEVs of natural resources need to be consistently estimated. Further integration of UEVs within LCIA was reached with the Solar Energy Demand (SED) method (Rugani et al. 2011a), which provided the first consistent dataset of UEV-based CFs compatible with elementary flows listed in the Ecoinvent database. Compared to other resourceoriented methods such as Cumulative Exergy Demand (CExD, Bösch et al. 2007), Cumulative Exergy Extraction from the Natural Environment (CEENE, Dewulf et al. 2007) and Cumulative Energy Demand (CED, see e.g. Klöpffer 1997; Huijbregts et al. 2006, 2010), SED assigns an intrinsic, nature-oriented value to resource consumption. Main differences in the relative impacts are found in the accounting of renewables, i.e. freshwater, biomass and land occupation. However, the afore-mentioned methods are related to the energy or exergy content of the resources, which denote their intrinsic capability to assign a value for the end user. In contrast, SED and emergy account for the work previously done by natural processes to generate the resource. According to the authors, the main limitation of the SED method is again the low accuracy of UEVs of natural resources. Notably, the approach is claimed to be an alternative adaptation of emergy to LCA, since the emergy algebra issue is explicitly disregarded. This method was then applied to all processes included in the Ecoinvent database, whereby the main outcome was to indicate the % renewability of resources used up by each process in thousands of proxies of product emergy values. The issue applying emergy algebra to LCI database has been recently addressed by Marvuglia et al. (2013a) with the development of the SCALE software. Using a graph search algorithm, 54

55 Chapter 2 Accounting for the emergy value of life cycle inventory systems: insights from recent methodological advances SCALE records the output emergy associated with each of the explored paths linking the input nodes of the network (i.e. the emergy sources) to the output nodes (i.e. the studied products). This innovative approach is a potentially useful step towards solving one of the trickiest issues in using comprehensive datasets of agro-industrial processes for EME, and demonstrates the benefits of using information communication technologies (ICT) to address sustainability problems. However, such an approach was developed for process-based LCI datasets only, for which allocation is based on physical flows. A similar rationale would not apply to macro-economic Input-Output (IO) tables and IO-based LCI datasets, which do not track the physical flows of materials and energy but aggregate economic exchanges between commodities at (most frequently) the national level. Additionally, the software needs to be tested on several case studies to both illustrate the added value of using LCI database in EME, and to shed light on potential limitations, such as gaps in the datasets, the influence of details in the process network to the results, and the limitation of low-accurate UEVs. The LCA framework was differently used by Brown et al. (2012) to refine the definition of emergy-based indicators. According to Brown et al., LCA aims at considering a local (in sense of foreground) activity embedded into a network of industrial processes that rely on each other. Therefore, it does not make much sense to apply EME to an isolated production plant. On the contrary, the ecological performance of the whole production chain is claimed to be more relevant for decision-makers. The authors suggested switching from the distinction between local (geographically) and imported to a differentiation between foreground and background activities. Brown and colleagues open a promising discussion on refining emergy-based indicators with a more formal definition compared to their original formulation. In this instance, technical aspects of LCA and EME were not combined, but the maturely standardized framework of LCA provided inspiration to strengthen EME and emergy-related concepts. The hybridization of EME and LCA remains at an early development stage, in which the preliminary proof-of-concept evolved into a first factual implementation of SED in LCIA (research projects INTERACT 3, EVALEAU 4 and HELICA 5 ). SCALE (Marvuglia et al. 2013a) can be seen as the starting point of a more advanced stage of hybridization, in which tools are developed, implemented, tested and improved. Main issues and limitations were consensually identified by the afore-mentioned studies: algebra must be adapted and made operationally applicable preferably via software development, rather than at a merely conceptual level (Rugani et al. 2012a); datasets of natural resources UEV must gain in resolution and be refined, enriched and harmonized; system boundary delineation requires a consistent, standard and consensual methodology in order to cope with boundary-related issues such as inclusion of L&S, the end-oflife stage of products, consistent accounting of ecosystem services in LCI datasets, and the impact on these services via pollution; emergy-based indicators may benefit from more robust EME framework and results, in order to consistently fulfill their role of environmental sustainability metrics for decision-makers. 3 Integrated assessment of emergy and life cycle systems, post-doc project ( ); AFR Grant by the National Research Fund (FNR) of Luxembourg; Co-funded under the Marie Curie Actions of the European Commission (FP7 -COFUND) 4 EVALEAU: French National Research Agency grant ANR-08-ECOT C Evaluation hybride émergie-acv de la production d eau potable, PhD project ( ); AFR Grant by the National Research Fund (FNR) of Luxembourg 55

56 2.4 Challenges to define an Emergy-LCA approach Table 2.1 summarizes the current advances and proposals we reckon are necessary to formalize a consistent integrated Emergy-LCA approach. In our opinion, the emergy community needs to foster the inherent meaning behind the emergy value, as this is the core basis for a consensual applicability of EME. It is well accepted that the emergy value of a resource can be mathematically defined as the integral of the exergy destroyed over its formation time (see e.g. Bastianoni et al. 2007; Tiruta-Barna and Benetto 2013). However, such theoretical definition has not been operationalized for real networks, by specifying the functional dependence of the exergy destroyed with respect to the temporal and spatial coordinates as well as to the network structure and thus assessing how this dependence affects the emergy calculation results. As a result, the terms used in EME, as well as the calculation procedure and interpretation of indicators, most often rely on somehow arbitrary interpretations. As outlined in Table 2.1, an important step in this direction was recently achieved by Tiruta-Barna and Benetto (2013), as they proposed a formal, transparent mathematical interpretation of the four rules of emergy algebra. The international scientific community may benefit from the consistent frameworks found in theoretical system ecology and network analysis (Schneider 1994; Ulanowicz 1997; Fath et al. 2001; Kleidon and Lorenz 2005; Jorgensen 2006; Jorgensen and Nielsen 2007; Dewar and Porté 2008; Ulanowicz et al. 2009; Chen et al. 2010; Patten et al. 2011), in which the mathematical formalism is a central issue. The (more practical) difficulties unveiled by preliminary hybridization proposals showed that: i) most problems find their origin in EME s low accuracy and lack of standardization in the accounting procedure, and ii) using LCA concept and dataset could help in solving these fundamental weaknesses. Process-based LCI datasets were successfully adapted to EME and integrated within the emergy calculation software SCALE; however, this hybrid model suffers the inherent gaps existing in process-based LCI datasets, i.e. missing inputs of ecosystem goods and services, missing contribution of human labor. These elements must be at best estimated separately; though such tiered approach would hardly comply with emergy algebra, it may be useful to identify relevant data sources and disclose potential double-counting issues. Rugani et al. (2012b) recently proposed a complementary approach to take into account human labor in LCA (see Table 2.1). Indeed, the human labor evaluation has been typically part of emergy evaluations, rather than of LCAs. However, the calculation of the UEV of human labor (and services) using conventional emergy approaches (e.g. via the use of the emergy-to-money ratios) may be lead to several uncertainty and double-counting issues. Therefore, being able to account for labor as a conventional input to an LCI system may support the recalculation of its UEV according to a standardized and transparent LCA framework or other extended approaches, such as macroeconomic accounting with IO tables. The impact of pollution on ecosystem services and the regeneration of renewable resources is another missing element in traditional EME. However, emergy can be also considered as an estimate of the work of the environment to replace what is used by human activities (Ingwersen 2011; Raugei et al. 2014); such rationale allows the inclusion of damages to the environment and ecosystems, as a less healthy environment is unlikely to optimize the delivery of ecosystem goods and services. Formalizing a robust framework that permits the use of high-resolution data on human activities would definitely increase the robustness and reproducibility of EME results and thus favor its adoption by decision-makers. And, conversely, attempts to use such data sources will provide new insights and formalism on the system boundary issues. 56

57 Chapter 2 Accounting for the emergy value of life cycle inventory systems: insights from recent methodological advances The development of a more robust UEV database of natural resources may also be facilitated by ecological modeling. An extrapolation of the principles exposed for LCI EME integration could be thought in case of natural processes including resources and ecological services. Ecological modeling provides networks of processes linked by mass and energy fluxes which can be used with an appropriate tool (SCALE for instance) for emergy/uevs calculation. Table 2.1: Summary of current advances and proposals to formalize a consistent Emergy-LCA approach. Issues to deal with Emergy Algebra Major recent advances Formal conceptualization and demonstration of a dynamic model to explain the principles and constraints behind the emergy algebra rules when applied to network systems (Tiruta-Barna and Benetto 2013) Set up of an algorithm and relative software interface to account for UEVs of technological products according to a rigorous application of the traditional emergy algebra (Marvuglia et al. 2013a) Consequences at the short- to medium-term The meaning of emergy algebra is mathematically explained for its non-conservative character, whereby an emergy result is found only valid for the level of detail of the network used for its calculation and shall not be reused in further calculations, for networks having different scales and levels of details For the first time, UEVs can be consistently calculated for complicated (LCI) systems Proposal for the next steps Extend rigorous emergy calculation on natural systems with consequences on the quality of calculated UEVs Possibility to properly calculate the emergy of systems at different levels (local, regional, continental) avoiding modeling errors originating from the mix of scales. Accurate calculation of emergy unit values that allows seeking EME metrics having an unambiguous directionality for the coupled natural-human systems. Unit Emergy Values Design of an alternative calculation framework to consistently estimate UEVs based on traditional LCI matrix solving techniques (Rugani and Benetto 2012) Despite that a huge amount of work would be necessary to collect a comprehensive and reliable data set that could approximate the geobiosphere complexity and the environmental work (and then be used in this matrix-based framework), the approach may highly increase the quality of natural resource-uevs to be used as LCIA characterization factors Recent findings from ecological modeling may open suitable paths to perform spatially and temporally-explicit calculations of UEVs and may be the basis for the construction of a large and transparent geobiosphere background framework to estimate emergy values of ecosystem goods and services System boundary Practical inclusion of human labor in LCA as an input to LCI systems through the use of a tiered hybrid analysis IO-LCA (Rugani et al. 2012b) Human labor can eventually be part of a LCA system boundary and thus its emergy value recalculated according to a consistent and standardized life cycle inventory framework Formalization of a guideline to frame the system boundary of an emergy calculation system and thus to: i) avoid as much as possible any double counting and ii) expand current inventories by including a number of ecosystem goods and services (which are not accounted for neither by LCA nor by EME) Decisionmaking Directionality and meaning of emergy-based indicators (Arbault 2012) Comparing the interpretations of variants of emergy-based indicators provides more insights on their actual meaning and consequences on decision-making Refine the emergy-based indicators, by: 1) mathematically translating the terms related to sustainability ; 2) adapting outcomes of EME to match these terms. Explore the combined use of ecosystem network analysis and thorough statistical correlation analysis between emergy and indicators of efficiency and resilience As demonstrated by Brown et al. (2012), the LCA framework allows reconsidering emergy-based indicators adopting a lifecycle perspective, which can be more meaningful than site-specific evaluations in a globally connected economy. For instance, the sustainability of a production chain does not rely only on the renewable flow available on the site of the last process considered, while assuming that all upstream elements of the production chain are imported inputs (F) and deemed non-renewable. The share of renewability and non-renewability of human labor and services is another important aspect, since assuming human labor is either fully renewable or fully fossil does not make sense. Though preliminary definitions of emergy-based indicators set the 57

58 grounds for the interpretation of EME results, a further overall sophistication of the framework - a prerequisite to the public acceptation of EME - will imply a refining of emergy-based indicators definition and formulas. In this respect, directionality is also very important when defining an indicator, as it provides clues on its effective monitoring of the studied system towards sustainability. The formal and consensual definition of directionality of emergy-based indicators would allow for instance to answer univocally to questions such as: is it more sustainable to promote the use of local, non-renewable resource or to import renewable ones from a remote region? From a consensual definition of sustainability, using a more standardized nomenclature of key terms (e.g. local/imported vs. foreground/background) and the intended purpose of each emergy-based indicator, the emergy community may benefit from previous experiences in the construction of indicators to refine the metrics aimed at decision-making. The advantage of using LCI databases is that it allows routine calculation of several variants of each indicator and thus may provide a wide spectrum of information to compare original indicators and their alternative formulations. 2.5 Conclusions Developing a more comprehensive and complete environmental accounting framework of anthropic systems requires the combination of complementary methods and tools. LCA is a mature and recognized method for evaluating environmental impacts on different areas of protection, including resource depletion, whose assessment is entirely built from a humancentered perspective, i.e. is rooted on the resource usefulness for human purposes. Conversely EME introduces the donor-side perspective for evaluating the total contribution of the geobiosphere processes on resource formation and on the anthropogenic systems functioning. The integration of EME and LCA is more and more investigated, but very few methodological developments have been proposed so far. The main methodological issue to be tackled for the combination of LCA and EME are: i) coherent definition of the boundaries of the studied systems; ii) accuracy (rigorous deduction) of the UEVs; iii) mathematical difficulties when combining different algebras; iv) relevance of the metrics and the value scale, transparence of indicators for decision making and action levers. In our latest research these aspects have been approached with promising outcomes for future developments. Concerning the formalism and calculation methods, a formal conceptualization and demonstration of the emergy algebra rules when applied to network systems was proposed, and a computational algorithm and related software was developed to calculate UEVs of technological products based on LCI databases (i.e. rigorous application of the emergy algebra, with the assumptions that rules #2 and #4b apply to co-production flows that may not actually occur in the reality at the same spatial and temporal production conditions). A matrix-based alternative (defined as bottom-up) method for calculating UEVs of natural resources and products of technosphere was proposed which, even though still in its infancy, would increase significantly the accuracy of UEVs. This method could allow the use of ecological modeling advances for refining the calculation of the emergy associated with natural resources and ecological goods and services. Systems boundaries are a matter of evolution in both LCA and EME. Ongoing prospective studies show that including human labor or/and ecosystem goods and 58

59 Chapter 2 Accounting for the emergy value of life cycle inventory systems: insights from recent methodological advances services in LCI/LCIA will lead to a transparent and complete definition of inventories. Using the newly developed mathematical tools, data translation will then be possible for emergy calculation. Besides the continuous efforts to improve UEV calculation, proposals for transparent and meaningful emergy indicators are at stake. It is a matter of fact that without a wide acceptance of their interpretation and directionality, the emergy accounting framework will face resistance against its wider adoption to consistently support decision making. 59

60

61 3. Emergy evaluation of water treatment processes Published as Arbault, D., Rugani, B., Tiruta-Barna, L., Benetto, E., Emergy evaluation of water treatment processes. Ecol. Eng. 60, Abstract The emergy evaluation (EmE) method is acknowledged to be a holistic approach to account for the primary (solar) energy that generates the renewable and non-renewable resource flows used up by human activities. This paper examines its application and robustness, using four water treatment plants (WTPs) as case studies. We obtained an average unit emergy value for potable water of 1.06 (±0.15) E12 sej/m 3, which is in accordance with existing literature. Chemicals and electricity were the most important man-made inputs; infrastructure, when accounted for, had a significant but lesser contribution. The application of several emergy-based indicators allowed comparing the ecological performance of water production with other types of resource extraction. These indices showed that WTPs are rather blind to economic markets and they exerted a low pressure on local non-renewable resources. A critical analysis of current EmE procedure highlighted the relative low accuracy of the method compared to Life Cycle Assessment (LCA), when man-made inputs are predominant, as well as the complementary goals and scopes of the two methods. Methodological improvements in the classification and treatment of the emergy associated with man-made inputs are necessary to make EmE indicators more straightforward and robust.

62

63 Chapter 3 Emergy evaluation of water treatment processes 3.1 Introduction Society as a whole is far from relying on natural resources in a sustainable way. Individuals and businesses must share the collective effort to reduce the pressure on resources. Appropriate tools and indicators are therefore needed to assess that pressure and provide decision-makers with an estimate of the distance-to-target between the current condition of stress and a more sustainable relationship with the natural environment (Moldan et al. 2012). Among the available environmental assessment tools, emergy evaluation (EmE) is a resourceoriented method that compares all resources on the basis of the solar-driven natural processes that contributed to their formation (Odum 1996). The EmE associated with an activity or a territory embraces a holistic picture of the studied human system embedded within a surrounding natural and economic environment and the global Earth system. It highlights the need for an activity to adjust to the local and global ecosystems that support it, instead of focusing on the local and relative efficiency of technological processes. The cumulative direct and indirect solar energy used up by natural systems to form a resource contributes to its emergy value, expressed in solar emjoules (sej; i.e. equivalents of solar energy). The Transformity of a resource is the ratio of emergy value to its available energy content (or exergy), expressed in sej/j. Specific emergy of a resource or a product is defined as its emergy value per unit mass (sej/g), while the more general term unit emergy value (UEV) is typically used when the denominator involves also other relevant physical units (e.g. volume). Average UEVs have been estimated for a wide variety of natural resources, including fossil fuels, mineral ores and renewable resources (Odum 1996, 2000; Odum et al. 2000; Brown and Bardi 2001). The emergy value associated with a natural resource accounts for the direct and indirect goods and services provided by the geobiosphere only. Concerning man-made products, each transformation step in their life cycle requires additional inputs, which are either natural resources already transformed by upstream human activities, or direct human interventions through labor and services (L&S). L&S are also fueled by extracted and imported natural (renewable and nonrenewable) resources. Accordingly, EmE enables accounting for the various forms of energy, materials and services ultimately consumed by a human activity with the sej unit. To assist decision-making, emergy-based indicators (Odum 1996; Brown and Ulgiati 1997; Ulgiati and Brown 1998; Ridolfi and Bastianoni 2008) aggregate EmE results into metrics that aim at describing the integration of the production system within its surrounding human and natural environment (section 3.2.2). EmE has been applied during the last 30 years to coupled natural-human systems of various types and sizes. The emergy evaluation of nations (e.g. Brown and McClanahan 1996; Chen and Chen 2006; Siche et al. 2008; Pereira and Ortega 2012), states, provinces (e.g. Zhao et al. 2005; Liu et al. 2008; Pulselli et al. 2008a) and regions (e.g. Campbell and Garmestani 2012) inform us on the local natural (and imported) resources used up to fuel these economies. Also, EmE has been applied to analyze the production of various commodities, e.g. in agriculture and farming (Ortega et al. 2002; Lefroy and Rydberg 2003; Castellini et al. 2006; La Rosa et al. 2008; Liu et al. 2008; Lu et al. 2009; Zhang et al. 2012), forestry (Tilley and Swank 2003), aquaculture (Lima et al. 2012), energy production (Lapp 1991; Brown and Ulgiati 2002; Paoli et al. 2008; Baral and Bakshi 2010; Yang et al. 2010; Ciotola et al. 2011; Brown et al. 2012), building materials (Buranakarn 1998; Brown and Buranakarn 2003; Meillaud et al. 2005; Pulselli et al. 2007), 63

64 recycling in industry (Mu et al. 2011, 2012; Giannetti et al. 2013a), ecological conservation or restoration (Lu et al. 2007, 2011; Dang and Liu 2012; Dong et al. 2012). EmE results for these analyses (i.e. emergy-based indicators, UEVs and transformities of the products) have been used as benchmarks to assess the ecological performance of water treatment. The focus of this paper is on the production of potable water. Few past studies refer specifically to potable water production plants (e.g. Odum et al. 1987b). The first most comprehensive survey is given by Buenfil (2001), who compared different household technologies with tap water from several municipal treatment plants in Florida. Then, Pulselli et al. (2011b) tracked the UEV of freshwater along a water course, from raw resource to water on tap, and Rugani et al. (2011b) compared ancient and modern aqueduct systems in the city of Siena, Italy. A common conclusion of those studies is that man-made inputs at the factory level make a large contribution to the final UEV of tap water. Case studies on contemporary potable water production plants (Buenfil 2001; Pulselli et al. 2011b) provide ranges of 6.9 E5 6.9 E6 sej/g, and 1.4 E5 1.4 E6 sej/j (adjusted to the 9.44 baseline, as explained in Section 3.2.4). Potable water is thus a man-made product with a high transformity relative to its specific emergy. Such particularity is due to the low exergy content of water, compared to the other types of man-made goods. Water Treatment Plants (WTPs) rely on a single local, renewable resource (freshwater), and a diverse set of man-made products and services. Local, non-renewable resources used up are apparently negligible (Rugani et al. 2011b). Such a situation can also be found in various other commodities, such as wind and solar electricity production, and organic farming (see e.g. Lu et al. 2009; Ciotola et al. 2011; Brown et al. 2012). Therefore, it seems critical to estimate the UEV of raw freshwater consistently. The water cycle (and the use of water in human activities) has been widely studied in EmE: it shapes landscapes and ecosystems, which can be used for many different activities. Freshwater-related EmEs cover a very large spectrum of situations, including dam proposals (Brown and McClanahan 1996; Kang and Park 2002), the overview of the Cache river basin (Odum et al. 1998) and water treatment via natural or artificial wetlands (Martin 2002; Cohen and Brown 2007; Carey et al. 2011; Duan et al. 2011) reflecting different aims. The most common objective of EmEs related to freshwater is to value this natural asset, i.e. its contribution to a regional or national public welfare (Tilley and Brown 2006; Chen and Chen 2009; Chen et al. 2009b; Lv and Wu 2009), its relationship with land occupation (Huang et al. 2007b) and ecosystem services (Odum and Odum 2000; Huang et al. 2011; Watanabe and Ortega 2011). EmE of the global water cycle was the subject of several studies (e.g. Buenfil 2001; Campbell 2003; Watanabe and Ortega 2011; Campbell et al. 2014). EmE was also proposed for a method to assess the full cost recovery of water management in a watershed (Brown et al. 2010). The aim of this study was to compare the outcomes of EmE associated with four WTPs located in France, in particular focusing on the UEV of the potable water produced (considering the actual quality level) and on a selection of emergy-based indicators. A particular emphasis was given to man-made inputs that are necessary to run the plant, and the computation of their emergy value. The importance of infrastructure to the overall performance of the WTPs is also investigated. Additionally, results of EmE are compared to Life Cycle Assessment (LCA) results for the same plants (Igos et al. 2013a, 2013b), in order to highlight differences and complementarities of both environmental assessment methods. The final goal of the paper was to provide new UEVs of drinking water quantified in a consistent manner along with a critical analysis of the EmE application, highlighting weak points of the method and including recommendations on how to deal with them. 64

65 Chapter 3 Emergy evaluation of water treatment processes 3.2 Methodology and data collection Energy system diagram According to the EmE methodology (Odum 1996), an energy systems diagram of the WTPs is presented in Figure 3.1. The left-hand side of the diagram shows the contribution of the surrounding environment in delivering the freshwater from a river. Geothermal heat runs geological processes that shape the landscape. Rainwater collected within the watershed is stored in soil moisture and then either evaporates or converges into streams and rivers. On the right-hand side, man-made inputs (fuels, electricity, chemicals, infrastructure materials and L&S) are used in the WTP to transform the freshwater into a product (potable water) valuable for humans. The distribution system was excluded from the system boundary, because specific data were not available, the scope of the analysis being the potable water production at the plant. Man-made inputs are the feedback (F) from the larger economy (i.e. purchased resources and human services), while raw freshwater is the only local, renewable input (R). Local, nonrenewable resources (N) were not used up in the potable water production systems investigated. Moreover, one could argue that land occupation of the site by the plant may hamper soil regeneration and could be counted as an N input. However, this was considered negligible in most of the studies presenting a similar situation (see the Supplementary Information material, hereafter SI, Table S3.8). In the present case studies, preliminary calculations showed that this emergy contribution was much smaller than any other input (SI, section SI3.3), and therefore it was disregarded. Fuel Elec. Chem. Mat. L&S Geo. Heat Rain Wind Stream Water F R 3.43 Infra. WTP Y 11.2 Sun Moist. Figure 3.1: Energy diagram of potable water production. Figures (in E18 sej/yr) are related to site A. The emergy value associated with each input was calculated by weighting its quantity (in physical units) with the corresponding UEV. When several R flows are feeding the system, only the input with the highest emergy value should be counted to avoid double-counting (Odum 1996) in the case they are all co-products of the same generating processes and are supporting local, natural processes. Only the highest contributor to R can thus be summed with all other (N and F) inputs 65

66 (which are not co-products of any local process). By definition, the emergy associated with the process outputs is Y (Odum 1996; Brown and Ulgiati 2002). When inputs are not co-products, Y is equal to the total emergy value of inputs Emergy-based indicators The aggregation of emergy inputs in the three categories R, N and F and their further combination can enable the calculation of the following indicators: The emergy yield ratio (EYR = Y/F) of a process is the emergy associated with the process output (Y) divided by the sum of the emergy inputs from the human economy (F). According to the literature (Odum 1996; Brown and Ulgiati 1997; Ulgiati and Brown 1998; Ridolfi and Bastianoni 2008; Campbell and Garmestani 2012) the EYR represents the energetic benefits gained by the human society for its investment in utilizing local, natural resources. The higher the EYR, the greater the net energetic benefit to the society. The environmental loading ratio (ELR = (F + N)/R) compares the sum of the emergy associated with local, non-renewable resources and imported resources to the emergy carried by local, renewable resources absorbed by the system. A high ELR often indicates a high intensity of nonrenewable resource use, or a high technological level accompanied by a high level of environmental stress on the local environment (Brown and Ulgiati 1997; Ulgiati and Brown 1998; Ridolfi and Bastianoni 2008). The emergy investment ratio (EIR = F/(R + N)) describes the investment made by the surrounding economy (i.e. F) into the process to exploit local resources (R and N). It indicates the matching of resources of the studied system with the inputs from the technosphere that encompasses it (Ridolfi and Bastianoni 2008). A high EIR would thus denote a system in which human investments are artificially high, and consequently likely to be affected by fluctuations in the economy. A low EIR would indicate a system beneficial for the surrounding economy and likely to receive more investments which would increase the EIR. We may conclude that in the long run, the EIR of a process tend to match the EIR value of the region in which it is embedded. %R represents the contribution of renewable input to the process output (R/Y). Processes showing a higher value of this indicator are likely to be more sustainable. The emergy sustainability index (ESI = EYR/ELR) indicates the ecological sustainability of the activity, indicated by the ratio of the net benefit to the society to the pressure on local renewable resources (Brown and Ulgiati 1997). Also, the UEV of the output of a system can be considered as an efficiency indicator, as stated by, e.g. Brown et al. (2012), UEVs are inversely related to the system efficiency on the scale of the biosphere. In other terms, a lower UEV means a more efficient overall use of resources by the coupled human-natural system. These indicators were applied in the present research to analyze the environmental sustainability of four potable water production systems. When benchmarking them to the various activities mentioned in Section 3.1, EmE results needed to be first homogenized, in order to wipe out the variability of formulations of emergy-based indicators (see SI, Section SI3.5 for all the calculation details). Since N is null in our case studies (i.e. no local, non-renewable resource like groundwater is used up), the indicators could be further simplified as follows: 66

67 Chapter 3 Emergy evaluation of water treatment processes EYR = 1 + R/F ELR = EIR = F/R %R = 1/(1 + F/R) ESI = EYR/ELR = (1 + R/F) R/F Noticeably, each indicator became a function of R/F only; consequently they would deliver the same ranking of the studied WTPs Data collection from LCA studies and comparison of EmE and LCA The four WTPs (hereafter Site 1, Site 2, Site A and Site B) are all currently operating in France. They were comprehensively studied using the LCA methodology (ISO 2006; European Commission 2010c). Sites 1 and 2 (Igos et al. 2013a) are plants located in the Paris area, which get raw water from the Seine River. The other two sites, i.e. A and B (Igos et al. 2013b), are new plants located in Brittany, taking raw water from local streams. Noteworthy, streams in Brittany are more polluted than the Seine River, and require a heavier treatment process. Detailed information on the life cycle inventory data and economic inputs for the plants are provided in the SI, Sections S3.1 and S3.2. The main difference on the life cycle inventory between the four WTP datasets is that A and B included infrastructure materials, while 1 and 2 did not. Input data (i.e. energy and material consumptions, expenditures of man-made goods and services, etc.) were calculated for the production of 1 m3 of potable water collected on the sites. Detailed LCA results were available and used for comparison with the EmE results. The scope and the accuracy of the results, as well as the divergences in interpretations, are presented in Section Unit emergy values (UEVs) A UEV is assigned to each inventory input to calculate its corresponding emergy value. Previous literature studies refer to different baselines (i.e. the sum of annual independent emergy inputs to the geobiosphere, i.e. solar radiation, tidal energy and geothermal heat, which is used as the reference to quantify the transformity of natural resources; see in Campbell 2001; Campbell et al. 2005a; Brown and Ulgiati 2010). In the present study, inputs and results are expressed with respect to the 9.44 E24 sej/yr baseline. This baseline was chosen because of its extensive use in the literature, supported by the consideration that there is currently no agreement on the choice of the reference baseline. Emergy values referred to another baseline in their original publications were converted using a simple ratio. Most of the UEVs were retrieved from the recently developed UEVs database (Tilley et al. 2012); Tables S3.4 and S3.5 in the SI report the original publication in which those UEVs are included. For some chemicals, we did not find any appropriate UEV in the available literature; thus, we used their Solar Energy Demand (SED, Rugani et al. 2011a) as a proxy. The discussion section presents the limitation of their usage for EmE. The complete data collection and elaboration procedure is disclosed in the SI, Section SI3.2. Specific UEVs for the freshwater used in the plants have been calculated (see Section ). Electricity mix was another important input whose UEV has been refined according to national specificities (Section ). The emergy values of human L&S were also specifically calculated, using national data, from emergy-money ratios available in the literature (Section ). 67

68 Local freshwater UEVs Natural energy flows shape the landscape of a catchment area and concentrate rainwater into streams and rivers (Chen and Chen 2009; Brown et al. 2010). Wind and rain are co-products of atmospheric processes driven by solar radiation. Therefore, only the highest contributor among them was counted in this study. Rain conveys two forms of available energy, namely chemical free energy (chemical exergy) and geopotential energy (physical exergy). Geothermal heat was not accounted for, since its past geologic contribution is already reflected in the geopotential energy of rainwater when it reaches the ground. Spring water from aquifers were disregarded is this study: although it can be considered as an input independent from rain at the short time scale, its contribution (in sej) is approximately 20 times smaller in the watershed studied in Pulselli et al. (2011b), which landscape rather favors the occurrence of springs. We noticed that the Seine watershed and Brittany are gently sloped, which favors infiltration and communication between deep and shallow aquifers rather than the occurrence of spring water. This lead us to assume that spring water in the studied watershed were relatively less important than for the Arno River basin studied in Pulselli et al. (2011b); consequently, the emergy value of spring water would amount for around 1% of the emergy value of rainfall. The UEV of freshwater should be calculated at the point of uptake. Since it was unknown for Sites A and B, we considered the whole watershed for the calculation, i.e. the UEV of freshwater at the estuary. Calculation details for the Seine River (near Site 1) are provided in the SI (Table S3.6). Table S3.7 displays the local characteristics of each river basin. The emergy value associated with the river was further divided by its annual flow to retrieve the freshwater s UEV. The resulting UEVs, used in Section 3.3 and ranging from 9.9 E11 to 1.4 E12 sej/m 3 (SI, Section SI3.4), are of the same order of magnitude than those estimated in Pulselli et al. (2011b) French electricity mix UEV No specific UEV for the French electricity mix was available in the literature. Hence, we used results from Brown and Ulgiati (2002) for electricity production systems in Italy and further adapted the share of production types to the French mix (Table 3.1). Nuclear power plants are the most relevant electricity production sources in France, for which we retrieved an UEV of 4.90 E4 sej/j from Campbell and Ohrt (2009). This value was calculated for nuclear electricity production from Minnesota. We assumed that this was calculated excluding L&S, i.e. inputs from the economy in the mentioned paper only consider material and energy inputs for the maintenance of the power plants and the preparation of the combustible. Table 3.1: Calculation of the French electricity mix UEV. Production type % mix [a] UEV UEV (E4 sej/j) Nuclear 78.50% Nuclear [b] 4.90 Hydropower 10.94% Hydro [c] 5.87 Hard Coal 4.47% Coal [c] 16.2 Natural Gas 3.18% Methane [c] 16.0 Oil 1.01% Oil [c] 18.7 French mix [a]: Ecoinvent v2.2 (2010), process #676; [b]: Campbell and Ohrt (2009), assumed without Labor and Services; [c]: Brown and Ulgiati (2002), excl. Labor and Services. 68

69 Chapter 3 Emergy evaluation of water treatment processes Human L&S UEVs Emergy accounts for both natural and man-made energy forms. While physical units are used to calculate the emergy value of a natural resource, the emergy associated with human labor and services is approximated using its economic price and the concept of the emergy-money ratio (EMR; Odum 1996). The latter, expressed in sej/, is the ratio between the emergy budget of a nation and its economic activity, represented by its Gross Domestic Product (GDP). It indicates the amount of emergy embodied in the monetary unit. Though 1 of different forms of L&S may have different emergy values, EMR remains the best available proxy to translate L&S costs into emergy terms. The National Environmental Accounting Database (NEAD, Sweeney et al. 2007) provided us with an EMR value of 2.8 E12 sej/$ for France in the year 2000, based on the baseline. Using a /$ conversion ratio for the year 2000 (INSEE 2012) and adjusting to the 9.44 baseline, the resulting French EMR was set to 1.81 E12 sej/ and used here to convert the L&S inventory inputs into emergy terms. 3.3 Results and discussion Emergy analysis of flows Emergy inventory calculations are provided in Tables The contribution of the local renewable resource (freshwater stream, R) to the total emergy of the plant ranges between 4.1 E18 (Site A) and 20.6 E18 sej/yr (Site 2). When compared in terms of m 3 produced (the size of the plants and the annual amounts of treated water are quite different among the four cases), the variations are due to slightly different UEVs associated with freshwater streams ( E11 sej/m 3 ) and water input/output ratios ( m 3 /m 3, see Tables ). Man-made inputs (F) range between 8.8 and 17.0 E11 sej/m 3 (see Table 3.6), and %R is between 22% and 40%. Accordingly, a variability score can be assessed in terms of standard deviation (i.e. weighted on the total production), obtaining an average UEV for drinking water production in France equal to 1.06 (± 0.15) E12 sej/m 3 (excluding. infrastructure). In potable water production, consumption of energy and chemicals is mostly determined by the quality (for the user) of the raw water. A more intensive consumption of F inputs can be observed for Sites A and B, which translates into higher output UEVs. However, we did not find a suitable emergy-based explanation to discriminate among polluted and non-polluted resources we could only notice that more polluted resources needed more emergy to be treated. However, the water resources for Sites A and B are much more polluted than the resources for Sites 1 and 2. Transformities do not provide additional information, since the specific exergy of water (in J/g) is calculated using the concentration of water in the river, which is an indicator of freshwater purity but not of its quality for drinking purposes: for instance, pure water with a small amount of highly toxic compound may be purer (i.e. present a higher concentration of water), but less potable, than bottled mineral sparkling water or orange juice. Table 3.2: Emergy table for Site 1. Items Annual amount unit UEV (sej/unit) Emergy (sej/yr) % Renewable resources ( R ) Seine River water at Site E+07 m E E+18 56% Purchased energy (F) Electricity mix, France (w/o L&S) 1.04E+07 kwh 2.13E E+18 19% Diesel 7.53E+05 MJ 6.71E E % Purchased materials (F) Activated carbon 4.18E+04 kg 1.56E E+17 6% Regenerated activated carbon 4.18E+04 kg 8.54E E+17 3% 69

70 Acrylic acid 2.02E+04 kg 3.55E E % Al 2SO E+05 kg 1.18E E+17 2% NaOCl, 15% 1.03E+04 kg 2.59E E % Labor and Services (F) Purchased inputs and Labor 4.10E E E+17 6% Coal fly ash, with services 4.17E+04 kg 1.40E E+17 5% Material Transport (truck) 2.82E+05 tkm 6.61E E+17 2% Output Potable water 1.16E+07 m E E+19 Table 3.3: Emergy table for Site 2. Items Annual amount unit UEV (sej/unit) Emergy (sej/yr) % Renewable resources ( R ) Seine River water at Site E+07 m E E+19 62% Purchased energy (F) Electricity mix, France (w/o L&S) 1.91E+07 kwh 2.13E E+18 12% Purchased materials (F) Activated carbon 1.52E+05 kg 1.56E E+18 7% Regenerated activated carbon 9.78E+04 kg 8.54E E+17 2% Acrylic acid 6.07E+03 kg 3.55E E % Al 2SO E+03 kg 1.18E E+18 3% Cl 2 gas 4.91E+04 kg 6.67E E % Lime 2.65E+05 kg 1.00E E % H 3PO 4, 85% 3.06E+03 kg 6.20E E % Caustic soda 4.09E+05 kg 1.46E E+17 2% H 2SO E+05 kg 4.15E E % Labor and Services (F) Purchased inputs and Labor 2.00E E E+18 10% Material Transport (truck) 5.37E+03 tkm 6.61E E % Output Potable water 3.53E+07 m E E+19 Table 3.4: Emergy table for Site A. % w/o infra % w/ infra Items Annual amount unit UEV (sej/unit) Emergy (sej/yr) Renewable resources ( R ) Freshwater at Site A 8.54E+06 m E E+18 35% 31% Purchased energy (F) Electricity mix, France (w/o L&S) 5.58E+06 kwh 2.13E E+18 12% 11% Purchased materials (F) Activated carbon 4.20E+04 kg 1.56E E+17 7% 6% CO 2 liquid 1.76E+05 kg 9.48E E+17 2% 1% FeCl 3, 40% 5.43E+05 kg 3.01E E+18 17% 15% Lime 3.95E+05 kg 1.00E E+17 4% 4% KMnO E+03 kg 8.24E E+17 3% 2% Caustic soda 8.36E+03 kg 1.46E E % 0.1% NaOCl, 15% 1.34E+04 kg 2.59E E % 0.3% H 2SO E+03 kg 4.15E E+14 0% 0% Infrastructure (F) Em-Building Surface 6.31E+07 mm E E+17 3% Em-Building Volume 6.34E+07 cm E E % Concrete 2.26E+08 cm E E+17 7% Copper 1.28E+08 mg 2.00E E+14 0% Glass 6.69E+06 mg 2.12E E+13 0% Plastic (PVC) 1.88E+09 mg 5.85E E % Steel 3.34E+10 mg 4.13E E % Material Transport (truck) 4.25E+05 tkm 6.61E E+17 3% Excavation 3.34E+08 cm E E+14 0% Labor and Services (F) Purchased inputs and Labor 1.05E E E+18 20% 17% Output (w/o infra) Potable water 8.36E+06 m E E+18 Output (w/ infra) Potable water 8.36E+06 m E E+19 70

71 Chapter 3 Emergy evaluation of water treatment processes Table 3.5: Emergy table for Site B. % w/o infra % w/ infra Items Annual amount unit UEV (sej/unit) Emergy (sej/yr) Renewable resources ( R ) Freshwater at Site B 9.09E+06 m E E+18 38% 35% Purchased energy (F) Electricity mix, France (w/o L&S) 4.11E+06 kwh 2.13E E+17 8% 8% Purchased materials (F) Activated carbon 4.51E+04 kg 1.56E E+17 7% 6% CO 2 liquid 2.58E+05 kg 9.48E E+17 2% 2% FeCl 3, 40% 2.32E+05 kg 3.01E E+17 7% 6% Lime 4.23E+05 kg 1.00E E+17 4% 4% KMnO E+04 kg 8.24E E+18 14% 13% Caustic soda 5.67E+03 kg 1.46E E % 0.1% NaOCl, 15% 7.53E+03 kg 2.59E E % 0.2% H 2SO E+02 kg 4.15E E+14 0% 0% Infrastructure (F) Em-Building Surface 6.66E+07 mm E E+17 3% Concrete 9.96E+07 cm E E+17 3% Copper 1.16E+08 mg 2.00E E+14 0% Glass 7.09E+06 mg 2.12E E+13 0% Plastic (PVC) 2.46E+09 mg 5.85E E % Steel 1.85E+10 mg 4.13E E % Material Transport (truck) 6.83E+04 tkm 6.61E E % Labor and Services (F) Purchased inputs and Labor 1.11E E E+18 19% 18% Output (w/o infra) Potable water 7.80E+06 m E E+19 Output (w/ infra) Potable water 7.80E+06 m E E+19 Table 3.6 shows the results from calculating the emergy indicators for all the WTPs. The highest EYR is observed for Site 2, i.e. this plant shows the highest efficiency in converting local resources into valuable goods for the larger economic system. Sites A and B need more technology-intensive processes to treat the more polluted resource, which translates into higher ELRs. EIR is times lower than the national value of (Sweeney et al. 2007), which denotes an activity that is not sensitive to economic stress. Indeed, production of potable water runs independently from the economic context as it supplies a fundamental resource to society. The size of the plant, i.e. its production capacity, does not seem influential. However, this conclusion should be counterchecked through a larger survey of WTPs. The accounting for infrastructure noticeably increases F, thereby decreasing the measured performance, as shown in the results for Sites A and B. Table 3.6: Comparison of emergy-based indicators for the four Water Treatment Plants. Site 1 Site 2 Site A w/o infra Site B w/o infra Site A w/ infra Site B w/infra R (sej/yr) 6.51E E E E E E+18 N (sej/yr) F (sej/yr) 5.13E E E E E E+18 Y (sej/yr) 1.16E E E E E E+19 EYR ELR = EIR %R 55.9% 62.2% 35.5% 37.7% 30.6% 35.0% ESI = EYR/ELR Potable water produced (m 3 /yr) 1.16E E E E E E+06 UEV (sej/m3) 1.00E E E E E E+12 In Figure 3.2, the UEVs of potable water output show a similar ranking between the treatment sites. Sites 1 and 2 are the most efficient, since the UEVs of their potable water outputs are the 71

72 lowest. These sites provide potable water with the lowest requirements of direct and indirect solar energy captured by the geobiosphere. Chemicals are the main man-made inputs (F), covering 40 55% of the total emergy value of F (except for Site 1), followed by L&S (24 31%) and electricity (12 31%, except for Site 1 where it covers 43% of F). Fossil fuels are directly used only on site 1 and their contribution is marginal (1%). Infrastructure, when accounted for, covers a significant 11 20% of the total F. Noteworthy, Sites 1 and 2 require less L&S and chemicals than Sites A and B per m 3 potable water produced. Figure 3.2: Contribution of each type of input (feedback and raw water flows) to the Unit Emergy Value (UEV) calculated for the 4 plants and for Sites A and B including infrastructure items. N.B. On-site use of fossil fuels barely visible for Site 1 and not present for the others. All the results are comparable in magnitude to the results on potable water production reported in the literature (Buenfil 2001; Pulselli et al. 2011b), with UEVs ranging between 0.69 and 6.80 E12 sej/m 3 (SI, Section 3.6). The WTPs studied in Buenfil (2001) use raw freshwater with very different UEVs, which explains the higher variability of the results. EmE studies of other types of human activities, such as agricultural systems, energy extraction and industrial manufacturing, showed a relatively high disparity of results (see SI, Section 3.5 for details on data and comparative tables). The relative closeness of the results obtained for the studied WTPs does not reveal significant differences between these case studies, in terms of economical-ecological competitiveness. Drinking water production lies amongst the studied activities with the lowest EYR (Figure 3.3), meaning that this sector provides a low net contribution to the larger economic system. Indeed, potable water is a necessity and is not expected to be a primary energy. This sector does not provide an energetic return on investment. The ELR of this sector is also relatively low (Figure 3.4) compared to other sectors, which denotes a low level of environmental stress on the environment. EIR of drinking water production shows a high variability. The same situation is observed for vegetal and animal products. The return on investment of the larger system to the local activity is thus averagely efficient. The combination of a relatively low EYR and a low ELR 72

73 Chapter 3 Emergy evaluation of water treatment processes leads to an average ranking of potable water production in terms of overall sustainability. %R and ESI in drinking water production are also average when compared to the other activities. Note that %R only relates to the use of resources that are both renewable and local. Finally, the specific emergy (sej/g) of potable water is much lower than other products, while its transformity is among the highest ones: a gram of potable water needs less indirect solar energy to be produced as compared to other products, while a joule of potable water (exergy) apparently requires more transformation of primary solar energy to be produced (see SI, Section 3.5). EYR EYR Min EYR Median EYR Max Ren Energy Drinking Water Vegetal Animal Electricity Industrial Figure 3.3: Comparison of Environmental Yield Ratio (EYR) scores for the producti on of various types of manmade products (see SI, Section 3.5). ELR (log scale) 1.00E E E E+01 ELR Min ELR Median ELR Max 1.00E E-01 Electricity Ren Energy Animal Vegetal Drinking Water Industrial Figure 3.4: Comparison of Environmental Loading Ratio (ELR) scores for the production of various types of man-made products (see SI, Section 3.5). The identification of available UEVs for chemicals in the emergy literature was critical for our case studies, due to the high number of reagents and their diversity. Their UEVs (or proxies) range between 4.15 E11 sej/kg for sulfuric acid and 8.24 E13 sej/kg for potassium permanganate. 73

74 UEVs of lime, caustic soda and gaseous chlorine were retrieved from Campbell and Ohrt (2009), not referenced in the online database (Tilley et al. 2012). The UEV of other chemicals remain not available in the existing literature to our knowledge. Indeed, this can be considered as a practical limitation of emergy-based accounting. The UEV of activated carbon and regenerated activated carbon were computed specifically for this study (see SI, Section SI3.2). For the other chemicals, we used an updated value of their SED (Rugani et al. 2011a) as a proxy (SI, Section SI3.2). Both UEVs and SEDs refer to the indirect amount of solar energy required to make a product, but the latter are computed following the rationale of LCA for allocation between co-products, which does not match the emergy algebra; however, they rely on a high level of detail in the network of industrial processes, which makes them more accurately calculated than UEVs. In the near future, the software SCALE (Marvuglia et al. 2013a), currently under development, may provide equally accurate UEVs for such products, while respecting the emergy algebra. Figure 3.2 also highlights the importance of electricity consumption in Site 1 (0.90 kwh per m 3 of produced water, vs for the other sites). Sites 1 and 2 also have lower L&S costs ( D /m 3 ) compared to Sites A and B ( D /m 3 ), which employ a more complicated treatment process. The emergy contribution of L&S is usually approximated by the economic cost of purchase. Since the highest expenditures are for energy and reagents, it may be relevant to further decompose these expenses, considering the actual labor in the supply chain, the assets, the speculation, etc. Typical UEVs of a year of human labor could not be found for the French context. In a country of similar level of industrialization, Italy, they range between 5.3 E15 and 2.8 E17 sej/yr (Brown and Ulgiati 2002; Pulselli et al. 2007, 2008a, 2011b; Rugani et al. 2011b). With an average value of 5.00 E16 sej/yr, an annual production of 7.80 E6 m 3 potable water/yr and 4.5 full-time equivalent workers to run the Site B plant (personal communication with the company), the total L&S input is worth 2.89 E10 sej/m 3. This value is rather close to the 4.58 E10 sej/m 3 found using the monetary approach ( D /m 3 of labor, Table S3.3). This result-checking somehow validates the assumption that economic inputs are mostly composed of human labor, although more in-depth analyses should definitely be carried out Comparison with life cycle assessment In Figure 3.5 (and SI, Section 3.7), the EmE results are compared to LCA results (Igos et al. 2013a, 2013b). LCA results were computed using the ReCiPe method (Goedkoop et al. 2009). Inputs from technosphere used in the infrastructure processes represent 6 7% of impacts on resource depletion, which is lower than results of EmE. Concerning the impacts of chemicals, there is no clear conclusion unanimously emerging: though both methods indicate a higher impact on sites A and B, EmE and LCA would rank Sites 1 and 2 differently. Also, EmE shows the contribution of electricity to the output emergy value is very similar between Sites 2 and B, while in LCA the impacts on resources from electricity is quite higher in Site B than in Site 2. Therefore, detailed results seem contradictory while overall results are analogous. The main reason is that impacts on resource depletion in LCIA are computed from accessible resources only, while the UEV of natural resources are computed from both accessible and inaccessible stocks. In other words, LCA takes into account the notion of scarcity of a resource, which denotes a user-side point of view, while the UEV of a resource is not calculated based on its potential utility for a user (and thus its rate of consumption), but it is rather based on a donor-side approach. The rationale of both approaches is thus complementary on this point. Another major difference between EmE and LCA lies in the inventory of inputs: while resources 74

75 Chapter 3 Emergy evaluation of water treatment processes used up for the provision of L&S are fully accounted for in EmE, they are partially included in LCA because only non-economic inputs are considered (energy and materials for transportation and sludge disposal). Human labor is at present outside the scope of LCA (Rugani et al. 2012b). The relative importance of L&S for LCA results (Figure 3.5) is due to the fossil fuels consumed by transporting chemicals and sludge. LCA also considers impacts on ecosystems and human health, while traditional emergy accounting focuses on resource use there exist however some attempts to consider pollution effects in EmE (Ulgiati and Brown 2002; Liu et al. 2013a). Finally, EmE also includes the use of the renewable resource to be transformed by the activity, and compares it with the other inputs using emergy-based indicators. LCA follows a different goal, i.e. comparing resource consumption and pollution for the production of a similar functional unit. Figure 3.5: Comparison of Emergy (i.e. UEVs) and Life Cycle Assessment (LCA) results for the four Water Treatment Plants. For LCA, the ReCiPe method is used; ecopoints are the aggregated single score of impacts on human health, ecosystems and resource depletion. The present analysis outlines several strengths of EmE, as compared to LCA. First, EmE provides a more holistic assessment of resource use by an activity, considering: (1) the natural value of a resource (defined as the solar energy necessary to regenerate it): this rationale is disregarded in current impact assessment methods (Ingwersen 2011; Raugei et al. 2014); (2) human labor 75

76 directly and indirectly required (through purchased services and products) to produce potable water, or any other man-made good. Second, this framework places the human activity within its local, natural context. For potable water production, the main local natural resource used up is freshwater, while energy and reagents are imported inputs. Third, emergy-based indicators rank different activities that produce a variety of man-made goods, as shown in Figures 3.4 and 3.5 and SI (Section 3.5). In contrast, LCA can only compare two technological solutions for an identical output (the functional unit), though with a much finer analysis of environmental impacts and a depletion-oriented approach for natural resource consumption assessment (which emphasizes scarcity). Therefore, EmE and LCA show insightful complementary features (Raugei et al. 2014). EmE also proves to be useful for territorial analysis issues. Watershed management obviously considers the provision of potable water as a priority, but a river provides other ecosystem services such as flood regulation, pollutant filtration, habitat for fisheries, communication roads, local climate regulation, etc. (Wilson and Carpenter 1999). In terms of decision-making, these ecosystem services should be handled altogether (Jewitt 2002); in emergy terms, they all are coproducts of the same ecosystem (Pulselli et al. 2011a; Rugani et al. 2013). In this connection, EmE offers a physical common denominator (rather than an economic framework) for a multiuser approach to resource management (Brown and McClanahan 1996; Tilley and Swank 2003; Cohen and Brown 2007; Huang et al. 2007b; Agostinho et al. 2010; Brown et al. 2010) Uncertainty and limitations This study revealed some current limitations of the EmE methodology as it is most often performed. Major issues are clearly related the high uncertainty of the results, which originate from two aspects. First, as illustrated by the emergy value of chemicals, available UEVs in the literature can be scarce and some useful ones are not gathered in the database(s) under construction. UEVs of industrial products require a consistent effort to be refined, if absent from published scientific and transparent work. In most cases, only generic or proxy UEVs could be identified. For example, different types of steel are aggregated into a single material, as well as plastics, transport systems, etc. (Tables S3.4 and S3.5). Critical UEVs like those for chemicals in our case studies had to be approximated by their SEDs. Using UEVs retrieved from literature requires a high but subjective level of attention, because of data may include L&S and the choice of baseline needs to be checked systematically. The phase of UEVs selection in EmE is one of the most critical issues of the methodology (Hau and Bakshi 2004; Raugei et al. 2014). Despite the 40 years of development of the method, the standardization of EmE is not as mature as for LCA (e.g. ISO 14040); to our opinion, it is certainly one of the reason why EmE is much less popular than LCA, despite the ground-breaking paradigm shift it offers. The recent publication of an online UEV database (Tilley et al. 2012) is an important step forward. Practitioners of EmE may also benefit from the recent results on how to use high-resolution LCI database (e.g. Ecoinvent 2010) to automatically calculate the UEV of man-made products (Zhang et al. 2010a; Rugani et al. 2011a; Rugani and Benetto 2012; Marvuglia et al. 2013a; Raugei et al. 2014). Similarly, economic inputs may potentially be further refined, using for instance national input-output tables. The second source of uncertainty is due to the possible omission of natural or man-made inputs. Among possible missing elements in the present study, one can mention the intervention of 76

77 Chapter 3 Emergy evaluation of water treatment processes ecosystem services to treat pollution (Ulgiati and Brown 2002; Liu et al. 2013a), the contribution of sunlight and wind on the WTP sites to provide a healthy working environment (Brown et al. 2012), the contribution of knowledge and technology (Odum 1996), the contribution of aquifers to the freshwater stream (Pulselli et al. 2011b), which participates to the longer water cycle (and thus the formation of the river) but not necessarily to the short-run river flow, and also the transformation and control of the river by human activities, which ensures WTPs a stable provision of raw freshwater. In principle, all these omitted elements could be eventually evaluated within the emergy framework, but site-specific data and detailed regional environmental surveys would be necessary to perform such an extended treatment of information. In the case of the four WTPs, the absence of a standardization procedure on how to select priority items and thus broaden the energy system diagram, as well as the lack of additional spatially explicit information related to the biophysical conditions of the territorial system (e.g. to evaluate the emergy of wind and sunlight) limited our possible investigation on that direction. Developing standardized procedures would be favored by a formal agreement on the mathematical framework, which continues to develop within the emergy community (see, e.g. Li et al. 2010; Le Corre and Truffet 2012; Patterson 2012). In particular, systemic methods of handling double-counting (Morandi et al. 2013a), though improving, remain theoretical. Besides, Tiruta-Barna and Benetto (2013) demonstrated that the calculation process is inaccurate when used in particular complicated systems because it depends on the level of details in the description of the network. Consensus among researchers may also require the decomposition of F inputs (Brown et al. 2012). For instance, man-made inputs and L&S can potentially be further decomposed into a renewable share (Fr) and a non-renewable one (Fn), as already found in literature, but with a non-standardized procedure (La Rosa et al. 2008; Paoli et al. 2008; Lu et al. 2009, 2011; Yang et al. 2010; Ciotola et al. 2011; Rugani et al. 2011b; Lima et al. 2012; Zhang et al. 2012). This will inevitably lead us to reconsider (and strengthen) the formulation of emergybased indicators, whose robustness suffers from a plethora of variants and requires a considerable work to allow comparing human-natural systems of different nature (SI, Section 3.6). Another important limitation is the inability of EmE to compare the quality (as perceived by the user) of freshwater resources, as demonstrated by the UEVs of the various freshwater streams. But this is also the case in LCA, although pioneering approaches attempt to handle this issue (Igos et al. 2013a). 3.4 Conclusion This paper applied EmE to assess and compare the resource consumption of four selected water treatment plants located in France. Results show a high stability of emergy-based indicators among these similar industrial systems. Our findings are comparable to those of other recent studies. Man-made inputs are of primary importance to run the plants, while infrastructure accounts for around 10 20% of the total emergy associated with these inputs. Regarding the operational phase, EmE highlights the need for more imported inputs to treat more polluted raw water, although the UEV of raw freshwater does not reflect its level of pollution (i.e. concentration, hazard and recalcitrance to treatment of harmful substances). Water treatment plants run on a single local, renewable resource. But like most industrial plants, they do not use local non-renewable resources (i.e. N equals zero). Therefore, emergy-based indicators become correlated (i.e. each one can be expressed as a monotone function of R/F) and thus rank the four plants identically. The predominance of man-made inputs (F) and the inherent 77

78 low accuracy of their UEV suggest that the lack of a clear and defined standardization of the method in emergy still provides users with little guidance in the choice of those UEVs. However, UEVs of man-made products could be refined by adopting life-cycle perspective and datasets, including the whole production chain within the technosphere. A formal agreement on the procedure for emergy calculation for man-made products needs to be reached, and may influence the definition and calculation of emergy-based indicators. These open questions could be partially addressed by applying hybrid emergy-lca approaches on the same case studies and compare them to the results presented in this paper. Such option may also strengthen the added value of emergy evaluation relative to other resource-oriented, thermodynamic indicators used in LCA, such as CEENE (Dewulf et al. 2007). 78

79 4. Emergy evaluation using the calculation software SCALE: case study, added value and potential improvements Published as Arbault, D., Rugani, B., Marvuglia, A., Benetto, E., Tiruta-Barna, L., Emergy evaluation using the calculation software SCALE: case study, added value and potential improvements. Sci. Tot. Environ. 472, p Abstract This paper reports the emergy-based evaluation (EME) of the ecological performance of four water treatment plants (WTPs) using three different approaches. The results obtained using the emergy calculation software SCALE (EME SCALE ) are compared with those achieved through a conventional emergy evaluation procedure (EME CONV ), as well as through the application of the Solar Energy Demand (SED) method. SCALE s results are based on a detailed representation of the chain of technological processes provided by the lifecycle inventory database ecoinvent. They benefit from a higher level of details in the description of the technological network as compared to the ones calculated with a conventional EME and, unlike the SED results, are computed according to the emergy algebra rules. The analysis delves into the quantitative comparison of Unit Emergy Values (UEVs) for individual technospheric inputs provided by each method, demonstrating the added value of SCALE to enhance reproducibility, accurateness and completeness of an EME. However, SCALE cannot presently include non-technospheric inputs in emergy accounting, like e.g. human labor and ecosystem services. Moreover, SCALE is limited by the approach used to build the dataset of UEVs for natural resources. Recommendations on the scope and accuracy of SCALE-based emergy accounting are suggested for further steps in software development, as well as preliminary quantitative methods to account for ecosystem services and human labor.

80

81 Chapter 4 Emergy evaluation using the calculation software SCALE: case study, added value and potential improvements 4.1 Introduction Emergy evaluation (EME), unlike most of the other environmental assessment tools, adopts a nature-centered viewpoint, focusing on the resources used up by a human system (an activity, a territory) and considering it as embedded within its natural environment. Emergy is defined as the direct and indirect energy of one kind to produce a product, resource or service (Odum 1996) and measured in solar emergy Joules (solar emjoule or sej). Since the concept originates from systems ecology, the EME of a human activity emphasizes more on calculating the emergy value of locally-available natural resources. Given the complexity of their production chain, the characterization in emergy terms of man-made resources produced elsewhere and used up by the activity (e.g. human labor and commodities) suffers from a low level of accuracy. Despite the valuable paradigm shift emergy brings about, further efforts are needed to make it a more popular tool for environmental sustainability (Månsson and McGlade 1993; Cleveland et al. 2000; Hau and Bakshi 2004). In contrast, other approaches such as Life Cycle Assessment (LCA) (ISO 2006; European Commission 2010c) have a user-oriented or utilitarian perspective and evaluate the environmental consequences of materials and energy flows taken from and emitted to the natural environment by human activities. For example, the impact on resources with the Life Cycle Impact Assessment (LCIA) method ReCiPe (Goedkoop et al. 2009) is based on the notion of functional utility of the resource for the user and its rate of depletion; other LCIA methods such as CExD (Bösch et al. 2007) and CEENE (Dewulf et al. 2007) assess the thermodynamic equivalent of the maximum work that is possible to extract from the resource, i.e. its exergy. These human-centered approaches, although consistent with the underlying rationale of a lifecycle perspective, could not be successfully applied to renewable resources and many Ecosystem Services (ES), which, by definition, are not stock-limited (i.e. they do not get depleted ) but are flow-limited (i.e. are regenerated at a limited pace). EME takes a donor-side approach oriented to consider the natural mechanisms that form the resources. Hence, it is complementary to LCIA methods and somehow advantageous in providing an overall characterization of renewable resources as well as non-renewable ones and ES. EME shares many similarities with the broader LCA framework: it has been demonstrated that the two approaches are complementary and can be combined (Zhang et al. 2010b; Ingwersen 2011; Brown et al. 2012; Rugani and Benetto 2012; Rugani et al. 2012a; Raugei et al. 2014). By integrating emergy within LCA (Rugani 2010; Ingwersen 2011; Raugei et al. 2014), it was shown that emergy could provide a complementary indicator for resources, as a unified measure of the provision of environmental support, and an indication of the work of the environment that would be needed to replace what is consumed (Raugei et al. 2014). EME could also benefit from the detailed description of the network of industrial and agricultural processes and the exchanges of energy and materials among them (the so-called technosphere) that LCA practitioners use to build the Life Cycle Inventory (LCI) of the functional unit under study. The LCI represents the cumulated amount of elementary flows (resources and emissions) exchanged between the technosphere and the natural environment. An important step toward the combination of the LCI database ecoinvent (Ecoinvent 2010) and EME was carried out throughout the development of the Solar Energy Demand (SED) method (Rugani et al. 2011a). In SED, the Solar Energy Factors (SEFs) of natural resources, derived from the emergy concept, are applied as characterization factors to LCI results. 81

82 The main challenge for integrating LCA into EME is, however, rooted in the specific features underlying the emergy algebra, which hamper the direct use of LCI databases (Ecoinvent 2010). While LCA adopts allocation rules among co-products of a multi-output process following a logic of conservation (of mass, energy, etc., where the environmental burden is shared among the coproducts), emergy algebra relies on a logic of memorization (Rugani et al. 2012a), as described by four algebra rules (Brown and Herendeen 1996). Two of them are particularly relevant for the discussion in this paper: rule #2 ( co-products from a multi-output process have the total emergy assigned to each pathway ) and rule #4 ( emergy cannot be counted twice within a system: emergy in feedback loops cannot be double-counted, and co-products, when reunited, cannot be added to equal a sum greater than the emergy source from which they were derived ). Marvuglia et al. (2013a, 2013b) developed an algorithm to apply the four rules of emergy algebra to large networks of interconnected processes, which was implemented in the software SCALE using the ecoinvent database and the SEFs dataset (Rugani et al. 2011a). SCALE is therefore the first tool using an LCI database for automatic emergy calculation in compliance with the emergy algebra rules. The aim of this paper is to present the first fully comparative case study performed with SCALE, in which we analyze the results obtained through: i) conventional EME (hereafter EME CONV ), meaning the application of the emergy accounting framework as usually performed in the literature (see e.g. Brown and Ulgiati 2002; Cavalett et al. 2006; Vassallo et al. 2007; Pulselli et al. 2008b; Bastianoni et al. 2009); ii) the SED method (Rugani et al. 2011a), using the ecoinvent database but not fulfilling the algebra rules and iii) SCALE (EME SCALE ), where both the ecoinvent database and the algebra rules are included. Four Water Treatment Plants (WTPs) are considered as test bed cases for the comparison. The three methods are described and compared from a conceptual point of view in section The case studies and their system boundaries are presented in section A quantitative comparison of the results (section 4.3.1) is followed by the analysis of the contribution of each technospheric input according to the three methods (sections and 4.3.3) and of the differences in interpretation (4.3.4). Some limitations remaining in using SCALE are finally discussed in section Materials and methods The three methods share a common framework (Figure 4.1): the man-made products and natural resources inputs to the studied system are inventoried and then converted into emergy per unit output (sej/m 3 potable water), using Unit Emergy Values (UEVs, defined as the emergy value of a product or resource per unit of a corresponding physical quantity e.g. mass, energy, volume) and SED or SEF of products or resources, respectively. The resulting figure is the UEV or SED of the system s output. However, the conceptual differences between the methods (section 4.2.1) involve alternative calculation of UEVs and SED for each input. 82

83 Chapter 4 Emergy evaluation using the calculation software SCALE: case study, added value and potential improvements Figure 4.1: General framework for the application of EME CONV, EME SCALE and SED Methodological principles EME CONV has recently been applied to the case studies used in this paper (Arbault et al. 2013b). When applied to a local activity, EME CONV calculates the Unit Emergy Value (UEV CONV ) of the output by accounting for all resources needed by the activity to be operational. These resources are natural, locally available (R and N, for renewables and non-renewables, respectively), or manmade (so-called feedback inputs, F). F inputs include materials and energy transformed into useful goods within the technosphere, and non-material services and labor. The underlying philosophy of the emergy approach is to assess the role of natural resources in supporting human systems (activities, economies, territories). Therefore, EME CONV gives a particular importance on depicting natural mechanisms responsible for the formation of R and N resources. Oppositely, the usual practice is to use simplified representations of the F inputs, for instance by attributing them the UEV of the corresponding natural resource (e.g. the UEV of limestone for lime product). When their contribution is assumed to be important, the transformation steps are roughly estimated (e.g. production of an electricity mix: Brown and Ulgiati 2002; soda and chlorine production: Campbell and Ohrt 2009). In contrast, labor and services are usually estimated using the national average of the emergy value of one working-hour or one monetary unit (Odum 1996). The SED method (Rugani 2010; Rugani et al. 2011a) uses the LCI s rationale underlying the ecoinvent database, which consists in calculating the cumulated amount of each resource (in the corresponding physical unit), ultimately consumed in the technosphere to provide the studied functional unit. The set of SEFs is then used to convert each amount of resource into a solar-energy equivalent. The development of this method was a remarkable attempt toward matching the set of 200+ elementary resources listed in ecoinvent and their solar-energy equivalents. However, the method cannot be considered as a rigorous application of the emergy concept: SED applies allocation rules typically used in LCA, which are not compatible with the specific emergy algebra. It can be thus considered as an effort toward integrating an emergy-like indicator into LCA, rather than embedding LCI datasets into emergy evaluation. EME SCALE also uses the detailed representation of the technosphere provided by ecoinvent (in principle, other LCI databases could also be used), and the SEFs dataset as UEVs of natural resources. However, as opposed to the SED method, it calculates the emergy of the system s output by rigorously applying the emergy algebra to the detailed network of technological 83

84 processes. The software SCALE requires a preliminary modification of allocation values originally set in the ecoinvent multi-output processes (see supporting information of Marvuglia et al. 2013a, for details). Then it applies a backtracking algorithm to trace flows of energy and materials within the modified network and to avoid double-counting. According to the rules of emergy algebra, the output emergy value is lower than (or equal to) the total emergy value of inputs, provided that the exact set of inventory data and SEFs is used. The difference depends on the number of feedback loops in the network of processes. An important consequence is that the level of details in the representation of processes influences the resulting emergy value of outputs (Tiruta-Barna and Benetto 2013). To avoid infinite calculation time, a threshold (named minflow) on the value of the flow tracked by SCALE must be set by the user (hereafter called Minflow). A high threshold value (e.g. 0.1 Msej) would lead to the omission of important feedback loops, while a low threshold value (say lower than 1E-6 Msej) would increase computation time drastically (Marvuglia et al. 2013a) with a negligible loss of information (named emergy lost in Marvuglia et al. 2013a). Supplementary Information SI4.1 discusses the influence of threshold value selection. The conceptual differences between EME CONV, EME SCALE and SED are translated into different mathematical assumptions that support the calculation steps. In EME SCALE, all resources used up by a multi-output process of the technosphere (e.g. salted water electrolysis, crop cultivation) are allocated to all co-products; in other terms, each co-product is considered to require the full use of all incoming resources. Consequently, specific algebra rules are necessary to avoid doublecounting in feedback loops included in the considered network of interconnected processes. The differences between EME CONV and EME SCALE lie in the accuracy for calculating technospheric inputs and the scope of the resources considered. In EME CONV, all forms of resources shall be taken into account: F inputs also include non-material, man-made resources such as labor and services or information. The point of the traditional accounting technique is to provide a holistic but rather simplified description of the functioning of a system and the external drivers feeding it. Therefore, it has a limited accuracy and often scarce reproducibility. In SED, the rationale for coproducts differs: the resources consumed by the multi-output process are allocated among the coproducts on the basis of their material or energy content. SED adopts a user-side viewpoint (the same as in LCA), in which the environmental burden is shared among all users of co-products Description of the case studies The characteristics of the four WTPs investigated in this paper are comprehensively described in Igos et al. (2013a) (Sites 1 and 2) and Igos et al. (2013b) (Sites A and B). Sites 1 and 2 are located in the Paris area and the raw water input comes from the Seine River. Sites A and B are new plants located in Brittany, using raw water from local streams. It is worth mentioning that water streams in Brittany are more polluted than the Seine River, and require a heavier treatment process, which explains why the UEV of output potable water from Sites A and B is higher than the UEV of potable water produced in Sites 1 and 2 (Arbault et al. 2013b). The data used for the present study focus on the UEVs of inputs from technosphere, i.e. energy, chemicals and services. Each piece of data presented below is the input required for the production of 1 m 3 of potable water and is extensively discussed in Arbault et al. (2013b). Technospheric inputs were modeled with products available in the ecoinvent v2.2 database. Since there is no feedback loop or co-production in the WTPs, they can be considered as a single process in the calculations without violating the emergy algebra. All the calculations were done 84

85 Chapter 4 Emergy evaluation using the calculation software SCALE: case study, added value and potential improvements using the 9.26E24 sej/yr baseline (Campbell 2001), which represents the sum of annual independent emergy inputs to the geobiosphere from solar radiation, tidal energy and geothermal heat (Odum 1996). 4.3 Results and discussion Quantitative comparison Table 4.1 compares the contribution of technospheric inputs in sej for EME CONV, EME SCALE and SED for the four case studies investigated. EME SCALE results were calculated with a minflow equal to 1E-6 Msej. In our case studies EME SCALE provides higher results than EME CONV because the UEVs of technospheric inputs used in both methods are different (see section 4.3.2). Table 4.1: Comparison of the total emergy of technospheric inputs calculated for the 4 case studies of water treatment plants using EME CONV, EME SCALE and SED. Calculated for a similar functional unit - 1 m3 potable water - excluding direct and indirect labor and freshwater resource (baseline: 9.26E24 sej/yr; Campbell 2001). Technospheric inputs (E12 sej/m 3 ) Site 1 Site 2 Site A Site B EME CONV EME SCALE SED The three methods provide slightly different rankings of UEVs among the case studies. For Sites 1 and 2, the UEVs are systematically lower than for the other sites, but EME CONV shows lower emergy value of technospheric inputs for Site 2 than for Site 1, while the two other methods yield the inverse ranking. Nevertheless, the order of magnitude is comparable for the three methods. In particular, the SED method provides results 20-30% lower than EME SCALE, while EME CONV delivers results 40-60% lower than EME SCALE. Site 1 is an exception, for which EME CONV is higher than SED and only 13% lower than EME SCALE. Interestingly, Site 1 has a higher consumption of electricity than the other sites and a lower consumption of chemicals (see Arbault et al. 2013b). Results of SED are useful for comparative purposes, although they do not have the same meaning as the emergy value of technospheric inputs (see section 4.2.1). As SCALE also uses SEFs, comparing results from EME SCALE and SED helps investigating the influence of emergy algebra vs. LCA-like allocation on the results. To this aim, the contribution of each individual technospheric input within the four case studies is analyzed in the following section Contribution analysis Contribution analysis investigates the relative importance, in sej, of each technospheric input (among purchased energy, chemical reagents and external services such as waste disposal and products transportation) to the system s output (1m 3 of potable water). Figure 4.2 shows that the three methods provide comparable results for electricity, which is the major input (in emergy terms) to Site 1. The main chemical products (in sej/m 3 ) used by the plants are activated carbon, lime, sodium hydroxide, iron chloride, potassium permanganate and aluminium sulfate. However, the emergy 85

86 value of their contribution varies across the methods employed. Figure 4.3 compares the ratio UEV SCALE /UEV CONV of the products for which an UEV CONV was found in literature (see SI4.2). Figure 4.2: Contribution of technospheric inputs across the 4 case studies by analyzing the three emergy-based methods. Lime and quicklime have a much higher UEV with SCALE. This has a major influence on the diverging results presented in Table 4.1 and Figure 4.2, in particular with regard to Sites A and B. This can be (only partly) explained by the UEV of the main natural resources used for each compound to calculate the UEV CONV and its SEF (used to calculate its UEV SCALE ): UEV CONV of limestone is 9.81 E11 sej/kg (Campbell and Ohrt 2009), while the SEF of calcite is 5.5 E12 sej/kg, i.e. 5.6 times higher (Rugani et al. 2011a). Both data, however, were retrieved from two different documents of the same author (respectively: Odum 1996, and Odum 2000); this highlights the need for a single and consistent dataset of natural resources UEV. The relative differences between the UEV SCALE and UEV CONV of lime and quicklime, and sodium hydroxide, also explain why EME CONV provides a lower emergy value of technospheric inputs for Site 1 than for Site 2, while EME SCALE provides opposite results. Disposal of hard coal ash (which simulates the destruction of exhausted activated carbon) and steam (used in the production and regeneration of activated carbon) have a much lower UEV with SCALE. Since the service disposal, hard coal ash is used only in Site 1 (see Arbault et al. 2013b), and its UEV CONV is much larger than its UEV SCALE and SED (see Figure 4.3 and Table S4.2), the resulting contribution of Services is more important for Site 1 evaluated with EME CONV than with the other methods, in contrast to the other sites (see Figure 4.2). To a lesser extent, SCALE provides higher UEVs for sodium hydroxide, electricity from the European grid (UCTE), natural gas, coal and activated carbon. UEV SCALE and UEV CONV present less significant 86

87 Chapter 4 Emergy evaluation using the calculation software SCALE: case study, added value and potential improvements differences for the other products. UEV CONV of coal and natural gas is rather close to their SEF: concerning coal, UEV CONV = 1.15 E12 sej/kg (Odum 1996, assuming 29.3 E6 J/kg coal) and SEF = 1.42 E12 sej/kg (Rugani et al. 2011a); for natural gas, UEV CONV = 4.35 E10 sej/mj (Bastianoni et al. 2005) and SEF = 3.67 E10 sej/mj (Rugani et al. 2011a, assuming 40 MJ/m3). The UEV CONV of mineral NaCl is equal to its SEF (9.81 E11 sej/kg, Odum 1996). Figure 4.3: Ratio UEV SCALE /UEV CONV (unitless) for selected technospheric inputs (log scale). Rather than discussing which value is most adequate to use as SEF for these resources, the focus here is on analyzing the differences in the results provided by the three methods. In the aforementioned cases, the differences concerning coal, gas and sodium hydroxide can be only explained by the additional technospheric inputs used up to transform the primary resource into a refined material, which are accounted for in the UEV SCALE, but not in the UEV CONV. The quantitative comparison of UEVs retrieved from literature and those computed using SCALE can only partly explain the differences in the results of EME SCALE and EME CONV showed in Table 4.1. The remaining difference involves technospheric inputs for which UEVs CONV were not available in literature and were thus assimilated to their SED (see Table S4.1). SEDs and UEVs SCALE are based on exactly the same network data upstream: their difference is only due to the application of respectively LCA-like allocation factors and algebra rules for UEVs 87

88 calculation. Figure 4.4 shows the relative changes when applying the two algebra approaches (of conservation in SED and of memorization in SCALE): the circle markers show the virtual UEV of each input, when rule #2 of emergy algebra is applied but not rule #4. These values have no physical meaning, they are assigned a value of 100% in order to investigate separately the influence of the additional application of rule #4 (square markers) to provide the final UEVs SCALE. Triangle markers represent the relative difference due to the application of LCA-like allocation instead of rule #2, leading to SED. The comparison between the data points marked with circles and those marked with triangles indicates the influence of the allocation rule #2, all other aspects being equal. Figure 4.4: Relative influence of emergy quantification procedure on the final UEVs of technospheric inputs due to 1) LCA-type allocation rule instead of emergy algebra rule #2 (only): triangle vs. circle markers; 2) influence of the application of emergy algebra rule #4 after rule #2 (square vs. circle markers). Figure 4.4 shows that SEDs are systematically lower than UEV SCALE : the use of LCA-like allocation criteria in SED thus provides a significant underestimation of the real emergy value of man-made products. Chemicals derived from sodium chloride (sodium hypochlorite, iron chloride, sodium hydroxide, hydrochloric acid, chlorine) present some of the highest differences between SED and UEV SCALE, which explain the variations observed in Figure 4.2. This means that the allocation rule for these processes plays a very important role: LCA-like allocation (reflected in SED results) is equivalent to accounting (in terms of environmental burdens) only for the anion chloride (Cl - ) to produce chlorine, disregarding the fact that resources in Cl - and cation sodium (Na + ) cannot be produced independently. Such reasoning makes sense only from the user-side point of view, when it is assumed that the Na + will be used by another technospheric process and thus the burden can be shared. From a donor-side perspective, both ions are necessary to form the crystal NaCl, which is then used up in its entirety by downstream users: each downstream user should take into account the whole burden even though he/she would need only the Na or Cl part 88

89 Chapter 4 Emergy evaluation using the calculation software SCALE: case study, added value and potential improvements of the crystal. This is the rationale of rule #2 in emergy algebra. Rule #4 pinpoints that doublecounting should be taken care of when both users of Na + and of Cl - combine their products. Similarly, the SED of liquid oxygen represents roughly 20% of the UEV SCALE value, because it is produced from air liquefaction, with liquid nitrogen and liquid hydrogen as co-products, with allocation factors in ecoinvent assimilated to the air composition. The differences found for fuel oil and natural gas are also explained by the fact that they are co-products of multi-output processes (i.e. considered as being extracted from the same oil reservoir). The differences for activated carbon and road transportation can be explained by those found for natural gas and fuel oil Gravity analysis Another interesting feature of SCALE and SED is the decomposition of outputs into resource categories, as illustrated in Figure 4.5. Figure 4.5: Decomposition per resource category of technospheric inputs in the EME SCALE and the SED of the 4 case studies. The classification in resource categories is not possible in EME CONV, unless a detailed decomposition of resource contribution was carried out for each technospheric input. SCALE provides higher results ( times higher) for fossil, metal, mineral and water resources, similar results for nuclear and renewable energy resources and much higher results regarding land resources (17-32 times higher). Agriculture operations are often multi-output processes (e.g. grain and straw out of wheat production), which explains the high difference. Since these items only indirectly enter the composition of chemicals and processed energy, the selected case studies do not highlight such difference. Gravity analysis of technospheric inputs, per resource category, is detailed in Figure S4.2. The ratio SED/INPUT (i.e. UEVs calculated with rule #2 only) shows the influence of allocation rules (LCA-like vs. emergy rule #2), per resource category. Relative differences are the highest for land resources, for a large majority of the studied products; however, the absolute contribution of land resource to the UEV SCALE (and SED) of technospheric inputs is marginal (the main contributors are fossil, metal and mineral resources). Excluding land resources, the highest relative differences between SED and INPUT concerns liquid oxygen (78-89

90 87% difference) and chlorine-based compounds (Cl 2 : 56-72%; HCl: 55-72%; FeCl 3 : 50-58%; NaOH: 47-66%; NaOCl: 43-54%). For each of these products, the ratio is similar among the resource types. For the other products, the difference sometimes reaches values higher than 40% (see e.g. electricity, regenerated activated carbon, ammonia via steam reforming, steam, natural gas) but only for one or two resource types. This decomposition highlights again the potentially high influence of the allocation rules, all other aspects being equivalents. In a multi-output process like salt water electrolysis to retrieve sodium hydroxide and chlorine compounds, allocation in LCA is based on mass (here: 46% chlorine, 52% sodium hydroxide, 1% hydrogen), which reflects the figures mentioned above Differences in interpretation The investigation carried out so far has highlighted two main points. Firstly, the higher level of detail achieved with SCALE as compared to traditional emergy evaluation. Processes involved in the transformation of the primary raw material into a useful product (for downstream users) are taken into account in a more detailed and homogeneous manner, thanks to the use of the ecoinvent database. Secondly, the higher resolution of the UEV of man-made products (e.g. chemicals, refined fuels and types of electricity) as compared to state-of-the-art UEV datasets (Sweeney et al. 2007). This is particularly important for the evaluation of industrial systems, in which local natural inputs are marginal compared to technospheric ones (Arbault et al. 2013b). Such enhancement may also have significant consequences on emergy-based indicators, such as the percentage of renewability and the emergy yield ratio (Odum 1996; Brown and Ulgiati 1997; Ridolfi and Bastianoni 2008), but this type of analysis is outside this paper s scope. SCALE, however, does not provide a full emergy accounting: it only gives a higher level of detail on technospheric inputs. The mathematical differences sketched up among the 3 methods and illustrated in section and can be translated in textual form, to shed light on the added values brought from SCALE to emergy accounting: 1. The UEV of potable water derived from conventional EME is the cumulated solar energy used up directly and indirectly by natural and human systems to deliver 1m 3 of potable water. 2. The UEV of potable water derived from SCALE is the cumulated solar energy used up by natural systems to form the natural resources that are used up, directly and indirectly via the supporting technosphere, to deliver 1m 3 of potable water. 3. The SED of potable water is the sum of solar-energy equivalents of resources consumed by the portion of the technosphere allocated to the delivery of 1m 3 of potable water. Figure 4.6 illustrates the conceptual difference between EME CONV and EME SCALE for the case study of water treatment plants. While SCALE includes a much more detailed description of the technospheric processes, EME CONV encompasses a larger scope of man-made contributions by including the emergy value of direct and indirect labor. The third sentence shows that SED adopts a user-side point of view: resources are allocated among users, while EME (via emergy algebra) attributes the whole value of the resources to each user. 90

91 Chapter 4 Emergy evaluation using the calculation software SCALE: case study, added value and potential improvements Figure 4.6: Graphical comparison between traditional EME (left) and SCALE (right, including information retrieved from ecoinvent, dotted-blue box). E = Energy products (fuels, electricity...); L = Labor (incl. services and non-material inputs); M = Materials (chemicals, infrastructure...); N = Non-renewable resources; R = Renewable resources; Y: Activity s output (Yield); WTP = Water treatment plant. Applied to the present case studies, each method emphasizes the large contribution of chemical reagents in the use of resources. By assuming that the lower the output s emergy value, the better (according to the conventional directionality adopted to interpret the emergy values, Arbault et al. 2013a), the final generic recommendation from all the methods is to monitor and possibly reduce the amount of chemicals consumed in the treatment plant. However, whereas the SED method implicitly suggests favoring chemical processes that provide more co-products, the emergy-based accounting methods recommend selecting only those processes that provide a similar amount of chemical reagents, but using up natural resources that are more easily delivered by natural processes (i.e. that require less direct and indirect solar energy to be produced). Despite the valuable leap forward it brings to emergy evaluation, SCALE has also some limitations in terms of scope, as listed and illustrated in section Limits of SCALE The correct application of SCALE to the technosphere depends on a good resolution and comprehensiveness of the database representing the network of industrial processes that compose the technosphere. The ecoinvent database, used in this study, remains perfectible, as illustrated in Section Moreover, it covers only a limited portion of the man-made contributions to the coupled human-natural system under study. While Table 4.2 shows the comparison of scopes between EME CONV, EME SCALE, SED and LCA (ReCiPe method, Goedkoop et al. 2009), Figure 4.7 emphasizes the identified missing elements in SCALE-based emergy accounting, which are further discussed in Sections to

92 Table 4.2: Comparison of scope between EME CONV, EME SCALE, SED and ReCiPe method (LCA). Scope EME CONV EME SCALE SED Value of resources (donor-side perspective, i.e. solar energy embodied in primary natural resources) ReCiPe (LCA) Ecosystem Services Partly* Local resources Human Labor Level of details in man-made inputs Low High High High Impacts of pollution Partly* Impacts of resource depletion Partly* * Different approaches are currently under development in the emergy community (see sections and ). These aspects are not systematically included in emergy evaluations, as commented in Arbault et al. (2013a). Figure 4.7: Missing elements in SCALE-based emergy accounting. Green shapes refer to the inventory of flows; red shapes refer to the impact assessment; dashed-blue box refers to the ecoinvent database. Atm = Atmospheric resources; E = Energy products (fuels, electricity...); EGS = Ecosystem goods and services; L = Labor (incl. services and non-material inputs); M = Materials (chemicals, infrastructure...); P = Pollutants released; R = Renewable resources; WTP = Water treatment plant. 92

93 Chapter 4 Emergy evaluation using the calculation software SCALE: case study, added value and potential improvements Algebra issues specific to ecoinvent Rule#2 of emergy algebra entails a revision of the allocation criteria, which involves a preliminary change of allocation factors in the ecoinvent database (see methods section). In the latter, most of the processes are modeled adopting a gate-to-gate perspective, i.e. each transformation step of the chain of processes is modeled separately. This is convenient, for example, when a process is involved in various production chains. It also allows a flexible, customizable allocation procedure when multi-output processes are present in the life cycle and the changes required by rule#2. However, a few processes are modeled from a cradle-to-gate perspective, i.e. in an aggregated form. An overview of the 2,000+ processes involved in the modeling of our case studies shows that around 50 of them (2.5 %) are actually modeled in an aggregated form, where only the cumulated environmental interventions throughout the whole cradle-to-gate lifecycle (i.e. from nature to the production plant output) are visible. Twenty-one of them are related to basic organic chemical reagents (e.g. benzene, toluene, esters, polyols ), which form the basis of organic chemistry and are likely to form other compounds directly used in the plant such as polymers; 16 of them are plastics (PVC, HDPE, PP ), which represent a significant part of e.g. material assets. If the concerned production chains generate co-products, emergy algebra would be inapplicable to them, because allocation rules are already embedded in the available dataset and cannot be modified manually. In order to correctly apply the emergy algebra, one possibility is to ask data providers to use SCALE by themselves and publish aggregated results, per resource category, assuming that there is no feedback loop from downstream systems that use polymers and the petroleum industry. This would be a prerequisite for an integral, correct application of emergy algebra in SCALE Human Labor, services and economic inputs In the ecoinvent database, several components necessary to run industrial and agricultural activities, such as human labor, management of economic assets e.g. banking activities, and of social welfare, e.g. education, healthcare, public administration, are omitted. These aspects may have important contributions to the overall environmental impacts of an activity. For instance, Rugani et al. (2012b) demonstrated that human labor may increase by % the overall environmental impacts of LCIA profiles. EME CONV accounts for the non-material inputs from the anthroposphere, by translating their economic cost in emergy terms. The emergy value of the economic cost of a man-made good is complementary to the emergy value of the resources used up: it is a proxy of the total direct and indirect human labor and services necessary to its production. There is thus no risk of doublecounting in using complementarily SCALE along with economic data, as long as human labor and services are not included in ecoinvent. Non-material, economic costs in our case studies are detailed in Arbault et al. (2013b), and have an emergy value of 6.4E10 (Site 1) to 2.6E11 (Site B) sej per m 3 of potable water produced, which represents 2-6% of the results found for man-made goods in the present case studies. However, this estimation should not be summed up with results provided by SCALE, since the level of details used to calculate them is not consistent with that of the description of the technosphere provided by ecoinvent. According to Tiruta-Barna and Benetto (2013), values calculated from a different level of detail of the network of processes cannot be added without creating a bias in emergy accounting. This inclusion of non-material, man-made inputs into 93

94 EME SCALE should rely on a process-based decomposition of indirect costs, whose resolution matches the level of details found in ecoinvent. A potential way forward is to consider new process categories in ecoinvent to represent several types of human labor (Rugani et al. 2012b) and public service providers, as illustrated in Figure 4.7. However, it would require updating the whole database Atmospheric resources Atmospheric resources are seldom inventoried as inputs in ecoinvent (only CO 2, Xenon, Krypton and Helium) as these resources are not directly solicited by human activities but freely available. This omission might compromise SCALE s results. For instance, atmospheric nitrogen is a major component in the production of ammonia and most fertilizers used worldwide. Assuming that 0.82 kg of atmospheric N 2 (UEV 2.28 E13 sej/kg, from Campbell et al. 2014) is consumed for the production of 1 kg ammonia, the potential additional contribution of nitrogen to ammonia (1.87 E13 sej/kg, see SI4.2) is 4.4 times higher than the UEV SCALE of ammonia from partial oxidation (4.06 E12 sej/kg, including 2.46 E10 sej/kg from renewable resources; see SI4.2). This example demonstrates that atmospheric resources should not be systematically omitted. Similarly, chemical oxidation is a very common reaction in industrial processes, and is often made with atmospheric oxygen. However, it could be claimed that the volume of airborne resources used up in industrial processes is negligible as compared to the oxygen consumption by living organisms including human bodies, nitrogen fixation by plants, and air flowing through factories useful e.g. for ventilation. Natural processes do not affect the overall balance between the production and the consumption of airborne resources, contrary to chemical reactions in industry for which there is a net consumption, potentially affecting the natural equilibrium of oxygen and nitrogen cycles and their supporting ecosystem services. Indeed, the nature-centered viewpoint of emergy can evaluate the regeneration capacities of atmospheric oxygen and nitrogen (along with carbon, water, sulfur ) by global processes, which are so-called supporting ecosystem services (Watanabe and Ortega 2011) and are important natural inputs to the technosphere. The SEF dataset (and hence SCALE) also disregards atmospheric resources and other renewable resources such as solar radiation, biomass and soil carbon, by attributing a SEF equal to zero, in order to avoid double-counting. Further discussion on such shortcomings and potential advances is provided in SI Ecosystem Services Other ES not included in ecoinvent may also provide important inputs to the technosphere (Zhang et al. 2010b). This is the case of some provisioning services (e.g. production of biogenic material), regulating services (e.g. erosion control, CO 2 sequestration) and supporting services (e.g. nitrogen and phosphorous mineralization), which are omitted in ecoinvent. To estimate their relative importance we applied to the case studies the EcoLCA method developed by Zhang et al. (2010a), available online ( EcoLCA uses economic data of the 1997 US Input- Output table to calculate the direct and indirect contribution of a selection of EGS to the generation of 1$ of output value by each industrial sector. This model converts the economic value of each input of the studied system into an emergy value using economic Input-Output tables. The application of this model to our case studies suffers from three important limitations: 1) the economic structure of France (and the price of goods) differs from that of the US; 2) Input- Output tables are suitable for macro-economic studies but not for one single plant; 3) the system described in this tool has a different scale and resolution from the technosphere as described in 94

95 Chapter 4 Emergy evaluation using the calculation software SCALE: case study, added value and potential improvements ecoinvent: their results should not be mixed (Tiruta-Barna and Benetto 2013). For these reasons, the results of EcoLCA shall not be used for the final interpretation of SCALE results but only for the purpose of obtaining a first estimation of the potential relative contribution of ecosystem services that are not yet reported in ecoinvent. The emergy value of the economic sector potable water production obtained from EcoLCA is 8.70 E11sej/m 3 (see SI4.5 for details on calculation and the underlying method). This result is in line with the results yielded by SCALE for our case studies. Detailed results (Figure 4.8) show that the main contributors are lithospheric processes to generate fossil, mineral and metal resources. These resources are also accounted for in SCALE. Although the ranking of these contributors is different, as well as the set of natural resources used up, EcoLCA and SCALE provide similar results for non-renewable resources. This implies, however, that CO 2 sequestration by forests, fish production, water provision for power plants, hydropotential energy and soil erosion control in construction sites should be investigated more in detail. Despite being inexact, the application of EcoLCA allowed us identifying ES which are potentially important inputs to the studied system. It further illustrates the need for complementing LCI databases like ecoinvent with the contributions of ecosystems and the biosphere to the functioning of the technosphere. Specific considerations related to the calculation of ES based on the application of emergy algebra in SCALE are discussed in SI Local resources Another limitation of the joint use of SCALE, the SEF dataset and ecoinvent v2.2 is the lack of geographic information. Emergy evaluation originally intended to place human activity within its direct surrounding natural environment, and therefore required the re-evaluation of the emergy value of local natural resources - while the emergy value of natural resources indirectly used up by the surrounding human environment (so-called Feedbacks) are estimated with coarser information. For example, the SEF of raw freshwater used in SCALE calculations is 3.09 E11 sej /m 3 (adapted from Buenfil 2001). Arbault et al. (2013b) calculated a specific UEV for local water resources used up in the present case studies of 5.40 (Sites 1), 5.76 (Site 2), 3.93 (Site A) and 4.33 (Site B) E11 sej/m 3, which are 29% to 86% higher than the SEF value. Consequently, UEVs and emergybased indicators are likely to change significantly. Another example is the SEF of land occupation, which is constant (i.e. equal to the empower density of the Earth; see Odum 1996) among the various types of land. This is a typical limitation of the traditional top-down approach, which must be overtaken by future research to obtain more spatially-explicit results from a rigorous application of emergy algebra. Bottom-up approaches could use Geographic Information System (GIS) tools to retrieve local UEVs, as illustrated by state-of-the-art research e.g. on freshwater and land use (Huang et al. 2007b; Agostinho et al. 2010; Mellino et al. 2014). Moreover, the ecoinvent v3 database provides a convenient structure to include geographical information in the representation of the technosphere (Weidema et al. 2011). In this case, however, it remains questionable whether natural, renewable resources captured at the same location and time should be considered as coproducts, since they are likely to have been generated at different geographical locations (e.g. wind, rain) and/or time horizons (e.g. groundwater, biomass). 95

96 Figure 4.8: Additional contribution (in sej/m 3, log scale) of Ecosystem Services to the final UEV of the functional unit (using EcoLCA, Zhang et al. 2010a). Left-hand side figure: contributions aggregated into resource and service categories; right-hand side figure: contributions detailed per ES accounted for in EcoLCA. See SI4.5 for method and calculation details. 96

97 Chapter 4 Emergy evaluation using the calculation software SCALE: case study, added value and potential improvements Pollutants and societal costs: impacts on Human Health, Ecosystems and Resources Emergy evaluation typically focuses on the consumption of natural resources. The impacts of pollution on human health and ecosystems remain unaddressed both in EME CONV and EME SCALE, though several authors suggested quantitative methods to account for them. For instance, it was suggested to quantify the costs of pollution remediation (Paoli et al. 2013), the technological costs to treat it before release (Ulgiati et al. 2007; Song et al. 2013), or the volume and mass of air necessary to dilute airborne pollutants until acceptable concentrations (Ulgiati and Brown 2002; Mu et al. 2011; Zhang et al. 2011a). This approach is close to the concept of grey water in ecological footprint (Chapagain and Hoekstra 2011) and to the ecological scarcity method in LCA (Frischknecht et al. 2006). Its weakness lays in the definition of acceptable level of pollutants, which is set by an environmental agency and is thus not determined by the analysis of the biophysical mechanisms affected by pollution. Oppositely, the latter are frequently used in LCA. For example, the ReCiPe method (Goedkoop et al. 2009) aggregates impacts into three Areas of protection : Human Health (HH), Ecosystem Diversity (ED), and (abiotic) Resource Depletion (RD). Impact indicators on HH and/or ED have already been used to calculate the emergy investment necessary to recover the impacts (Ukidwe and Bakshi 2004; Zhang et al. 2011a; Liu et al. 2013a; Reza et al. 2013). Impacts on Human Health (HH) consider the loss of human life or good health, which affects the good functioning of society. The latter needs more resources to recover such losses. This additional requirement can be attributed to the studied activity, although it does not directly use them up. The unit of impacts on HH is the DALY [disability-adjusted life years], which can be thought of as one lost year of healthy life (WHO 2013). The annual emergy budget per capita in France is 3.76 E16 sej/cap/yr (Sweeney et al. 2007, year 2000, 9.26 baselineadjusted). According to the afore-mentioned studies, this figure is supposed to estimate the emergy value of the resources used up for 1 average year of human life in France, although an average year of human life is not necessarily healthy. Nevertheless, an impact of 1 DALY can be seen, as a first proxy, as the net loss for the society equivalent to 3.76 E16 sej. Impacts on Ecosystem Diversity (ED) refer to the loss of biodiversity, which ultimately contributes to the efficient delivery of ES. An impact on ecosystem diversity of 1 species.yr means that 1 known species is lost (reversibly) by the biosphere during one year (Goedkoop et al. 2009). Assuming a biodiversity of 2 million known species on Earth (Goedkoop et al. 2009) and a global emergy budget (baseline) of 9.26 E24 sej/yr, the maintenance of 1 average species during 1 year by geobiosphere processes has an emergy value (as a first proxy) of 4.63 E18 sej/(species.yr). Impacts on Resources Depletion (RD) calculated according to the ReCiPe method (Goedkoop et al. 2009) account for the marginal (future) economic cost of resource extraction due to the presently extracted quantities. This can be seen as a (user-side) societal cost. It is thus unaccounted for in EME CONV and EME SCALE, which focus only on the regeneration of resources by natural processes. An impact of 1$ on resource depletion means that the net economic loss for society in the consumption of present resources (equivalently the societal replacement cost of presently-consumed resources) is 1$. This cost must be estimated at the societal level: the emergy value of the additional resources necessary to overcome this loss can be approximated as the local emergy-to-money ratio (1.63 E12 sej/$ for France, Sweeney et al. 2007, year 2000, 9.26 baselineadjusted). 97

98 It is worth mentioning that the rationales presented above suffer from several shortcomings for EME. In the first place, impacts on HH and ED hardly match the rationale of emergy, since they are not based on tracking energy transformation pathways but rather on statistical correlations between these impacts and the quantified amounts of various pollution types (Goedkoop et al. 2009). Secondly, assuming these results can be translated into emergy terms, they still originate from a top-down approach and are thus subjected to possible double-counting. Finally, they mix different geographical scales and rely on a very rough level of detail regarding the description of the concerned systems: French national averages for emergy budget per capita and economic production, global average for the generation of species. As previously demonstrated, results of emergy evaluation depend on the level of detail in the description of the network of processes (Tiruta-Barna and Benetto 2013). Therefore, these results can only be considered as a rough estimation of externalities that remain unaccounted for, both in EME CONV and in EME SCALE. Table 4.3 indicates the estimated contribution, in terms of emergy, of these externalities. Despite the fact that they seem to be marginal, a rigorous emergy accounting framework should not disregard them a priori. It must be highlighted, however, that these aspects are of primary importance for the society and should not be assessed only with the lens of resource accounting; EME does not deliver a single metrics on which decision-makers could rely blindly. Table 4.3: Estimated contribution of pollution impacts and societal net loss (in emergy term) to the case studies of water treatment plants. Impacts per m 3 of potable water Site 1 Site 2 Site A Site B Total sej/m E E E E12 HH (DALY) 5.25 E E E E-7 HH (sej/m 3 ) 1.97 E E E E10 % sej/m % 1.71% 2.31% 1.65% ED (species.yr) 1.88 E E E E-9 ED (sej/m 3 ) 8.70 E E E E10 % sej/m % 0.84% 1.15% 0.85% RD ($) 1.33 E E E E-2 RD (sej/m 3 ) 2.17 E E E E10 % sej/m % 1.93% 2.54% 2.43% HH: human health; ED: Ecosystem Diversity; RD: Resource Depletion. Total sej/m 3 is retrieved from the EME CONV method (Arbault et al. 2013b). It includes technospheric inputs, labor and services, and the local UEV of freshwater. Impacts on HH, ED and RD were retrieved from Igos et al. (2013a, 2013b). 4.4 Conclusions and outlook For emergy practitioners, using SCALE provides a pragmatic advantage: technospheric inputs are calculated automatically in compliance with the emergy algebra rules and with a level of details never achieved before. This enhancement is useful for nearly all the case studies related to human activities. The precision of SCALE, however, relies on the comprehensiveness and level of detail with which the network of processes is described in the selected representation of the technosphere (i.e. in the used LCI database). Similarly to LCA, the selection of the LCI database depends on the location and sector of the studied activity. As mentioned in Marvuglia et al. (2013a), all commercial, process-based LCI datasets can in principle be used with SCALE, even when further enriched with process-specific datasets (such as activated carbon in the present 98

99 Chapter 4 Emergy evaluation using the calculation software SCALE: case study, added value and potential improvements study). Therefore the extent to which certain elements (such as e.g. human labor) are taken into account in SCALE mostly depends on the level of information contained in the used database about these elements. There is no technical limitation in SCALE itself to handle these elements and include them in the assessment. However, it was found that the LCI database applied in this case study relies on aggregate information for the plastic industry, and omits some potentially important elementary flows like airborne resources and ecosystem services. The current version of SCALE does not provide definitive results for emergy evaluation, since inputs such as labor and services and local resources must be calculated separately. Summing up these results computed independently raises issues on the heterogeneity of their levels of detail and spatial-temporal scales used for the calculation. A possible way forward is considering human labor and services as additional technospheric processes. Finally, the limitations identified in the use of SEFs in SCALE have very different origins: 1) the traditional top-down approach employed in their calculation leads to disregarding some resources in the dataset (like atmospheric resources) to avoid double-counting; 2) the current representation of the technosphere (i.e. the ecoinvent database) is site-generic: the SEF dataset cannot accurately represent local resources with a high spatial variability (e.g. freshwater and land); 3) since EME focuses on material and energy resources, it does not yet conveniently account for pollution impacts and societal costs. Further research is needed to replace a top-down, site-generic, material-resources oriented SEF dataset of natural resources by a bottom-up, spatially-explicit, ecosystem-services oriented dataset. Such improvements are necessary to apply rigorously the emergy algebra. A first step, as proposed by Rugani and Benetto (2012), could be a fully-fledged, bottom-up approach made of a matrix representation of geological and biophysical processes, in which SCALE could be jointly applied. Acknowledgment This project was supported by the National Research Fund, Luxembourg (Ref ) and the French National Research Fund (project EVALEAU ANR-08-ECOT C0238). Supporting Information SI includes all details of UEVs CONV, UEVs SCALE and SED of technospheric inputs used in this paper, as well as the dataset used in the EcoLCA software. 99

100

101 5. A semantic study of the Emergy Sustainability Index in the hybrid lifecycle-emergy framework Published as Arbault, D., Tiruta-Barna, L., Rugani, B., Benetto, E., A semantic study of the Emergy Sustainability Index in the hybrid lifecycle-emergy framework. Ecol. Indic. In press. DOI: /j.ecolind Abstract Emergy-based indicators are claimed to be useful outcomes of the emergy evaluation framework, which aims at guiding decision-makers toward environmental sustainability. The calculation of the Emergy Sustainability Index (ESI), in particular, seems widely consensual among emergy scholars, but several variants actually exist in the scientific literature, which may lead to different interpretations or misunderstanding of the ESI result. This paper proposes a semantic study of two variants in both components of the ESI (the Emergy Yield Ratio and the Environmental Loading Ratio, respectively EYR and ELR), to enhance standardization and reproducibility in the calculation of emergy indicators. It is shown that ESI can be consistently defined at the level of the production site as well as from a lifecycle perspective, although several case studies in the literature use an intermediary version with inconsistent system boundaries. A recent definition of lifecycle-oriented EYR is made operational by the development of an algorithm to be implemented in the emergy accounting software SCALE. However, the classification of foreground inputs needs further precision. ESI is also decomposed using partial derivatives, in order to analyze the influence of each input category and retrieve generic recommendations. These multiple outcomes demonstrate the added value of hybrid lifecycle-emergy evaluation to identify specific potential actions toward enhancing ESI of human activities.

102

103 Chapter 5 A semantic study of the Emergy Sustainability Index in the hybrid lifecycle-emergy framework 5.1 Introduction Decision-makers need operational environmental sustainability indicators in order to monitor territorial policies and assess businesses performance. On the one hand, national or regional environmental policies are often evaluated using composite indicators, i.e. synthetic aggregations of independent parameters, reflecting stakeholder values and expert weighting (e.g. Spangenberg and Bonniot 1998; Sébastien and Bauler 2013). On the other hand, business- and product-oriented indicators are usually derived by adopting a Life Cycle Assessment (LCA) perspective of the production chain, in order to identify the best improvement opportunities and to avoid pollution transfer from one step to another of the lifecycle (European Commission 2010c). With regard to the assessment of resource depletion, one set of indicators is based on the notion of scarcity (e.g. Frischknecht et al. 2006; Goedkoop et al. 2009), using data on consumption rates and remaining stocks. Alternative indicators are based on exergy, i.e. the maximum potential useful energy that can be retrieved from a resource (Szargut 2005; Bösch et al. 2007; Dewulf et al. 2007). Exergy accounting assesses the thermodynamic efficiency of an activity in converting the useful work embodied within resources. Moreover, the concept can be extended beyond the traditional boundaries of LCA to include the resource costs of labor, capital and environmental remediation (Sciubba 2013). Despite their pertinence, these approaches do not necessarily take into consideration that the physical limits of human exploitation of the planet may have been reached already (Rockström et al. 2009), due to the increasing global population and technological improvements (Moldan et al. 2012). The comparison of the relative importance among resources (i.e. the potential impact of their shortage) essentially reflects their utility values, which potentially deviate from the planet s physical limits. However, prominent studies demonstrate the effectiveness of exergy accounting in modeling what is potentially lost at Earth scale, suggesting novel approaches to study the physical limits of the globe and the efficiency of production processes, especially those that consume exhaustible resources (Wall and Gong 2001; Szargut 2005; Chen 2006; Chen et al. 2006; Hermann 2006; Liao et al. 2012). In any case, the effects on the ability of the geobiosphere processes to (re)generate scarce and non-renewable resources after human intervention cannot be assessed, i.e. the contribution of natural processes and ecosystems in the formation of renewable resources is systematically omitted, which makes these indicators more adapted to account for the depletion of non-renewable resources from a user-oriented perspective. The study of energy and exergy flows in systems ecology (Lotka 1922; Schneider 1994; Fath et al. 2004; Kleidon and Lorenz 2005; Jorgensen and Nielsen 2007; Puzachenko et al. 2011; Skene 2013) aims at relating the energy used up by natural systems to the amount of renewable resources they deliver, providing scientific indications on the maximum consumption rate of renewable resources that human systems can afford in the long run. Odum (Odum 1988, 1996) proposed the concept of emergy for comprehensive environmental accounting. Emergy was defined as the total direct and indirect (solar) energy used up to deliver a product. Therefore, emergy encompasses in its definition the contribution of both geological and biological processes, as well as transformation steps by human activities. The novelty in the emergy concept is the nature-centered perspective to the evaluation of human activities, which are considered as embedded within and dependent on the surrounding natural environment. Although the mathematical framework of emergy and its relationship with thermodynamics remain debated (e.g. Brown and Herendeen 1996; Sciubba and Ulgiati 2005; Bastianoni et al. 2007, 2011; Lazzaretto 2009; Li et al. 2010; Amponsah et al. 2011; Le Corre and Truffet 2012; Patterson 103

104 2012; Tiruta-Barna and Benetto 2013; Morandi et al. 2013a), hybrid emergy-lca models were proposed, either using emergy as an indicator for resources in LCA (Zhang et al. 2010a; Ingwersen 2011; Rugani et al. 2011a, 2013; Raugei et al. 2014), or using detailed datasets from LCA to enhance the resolution of emergy evaluations (Rugani and Benetto 2012; Arbault et al. 2013a, 2014; Marvuglia et al. 2013a) and to calculate Unit Emergy Values (UEVs), i.e. emergy per product unit. Commonly adopted but less discussed achievements of the emergy evaluation framework are emergy-based indicators. In his emergy masterpiece (Odum 1996), Odum first considered the calculation of investment ratios as the ultimate step of the emergy evaluation of a human system. To this end, he classified the system s inputs into four categories: free, renewable (R) and non-renewable (N) resources, which are those retrieved directly from the natural environment by the activity under evaluation, and imported materials (M) and services (S), i.e. those purchased from the larger economy. As argued in the literature (Odum 1988; Ulgiati and Brown 1998; Odum and Odum 2001; Campbell and Garmestani 2012), natural systems do not operate at steady-state conditions but rather follow cycling and oscillating patterns. Therefore, according to the emergy rationale, an activity cannot be defined as sustainable by referring to a particular value to minimize in order to reach a steady sustainable level of resource consumption, because it depends on the oscillating patterns of resource production by natural systems. Instead, an activity is estimated environmentally sustainable when it anticipates and adapts to the changes in the surrounding environment (Ulgiati and Brown 1998). Hence, this assumption would be better reflected by ratios that consider a human system within its economic and environmental context. Among the most widely used emergy-based indicators, the Emergy Yield Ratio (EYR) evaluates the level of integration of the activity within its surrounding human context, while the Environmental Loading Ratio (ELR) reflects the intensity of human development around the exploitation of environmental resources (see e.g. Brown and Ulgiati 1997, 2004; Raugei et al. 2005; Ridolfi and Bastianoni 2008). The Emergy Sustainability Index (ESI) was introduced by Brown and Ugliati (1997) as the ratio of EYR by ELR. Accordingly, the authors defined sustainability as a function of yield, renewability, and load on the environment. A sustainable process should be both environmentally and economically sound, i.e. operate with a low dependence on non-renewables and provide a suitable yield to society. ESI is thus a ratio of two ratios, which evaluate both environmental and economic compatibility of a system according to changes in its driving forces (Ulgiati and Brown 1998), i.e. the human and natural context. Therefore, it seems that ESI provides emergy evaluation scores with a clear directionality, i.e. a higher ESI refers to a more sustainable system (Brown and Ulgiati 1997). However, the sustainability assessment through ESI is subjected to different interpretations and potential misunderstandings, because several variants of EYR and ELR have been recently proposed (Ortega et al. 2002; Brown and Ulgiati 2004; Raugei et al. 2005; Ridolfi and Bastianoni 2008; Lu et al. 2009; Agostinho et al. 2010; Yang et al. 2010; Duan et al. 2011; Zhang et al. 2012; Brown et al. 2012; Campbell and Garmestani 2012; Lima et al. 2012; Tao et al. 2013; Wilfart et al. 2013), while other authors have suggested new indicators to enhance the emergy sustainability evaluation framework (Ortega et al. 2002; Lu et al. 2003, 2007, 2009; Mu et al. 2011; Zhang et al. 2011a; Song et al. 2012, 2013; Li et al. 2013; Reza et al. 2013). The aim of this study is to further analyze the implications of the diversity of definitions and formulations of EYR and ELR proposed in the emergy methodology, with regard to the 104

105 Chapter 5 A semantic study of the Emergy Sustainability Index in the hybrid lifecycle-emergy framework mathematical decomposition of ESI. Six case studies of water treatment plants are used to illustrate the results. Two variants of ELR and EYR are selected, covering both the production site and the whole chain of production (lifecycle perspective) as system boundaries. In addition, the operational definition of the lifecycle-based EYR proposed by Brown et al. (2012) is tested by developing and using a specific algorithm. Consequently, four variants of ESI are described, with explicit meanings and covering different system boundaries. A sensitivity analysis of the 4 ESIs with respect to each type of input (local, foreground, background, renewable or non-renewable) is further performed, from which generic trends are derived to enhance the characterization of ESI. 5.2 Critical review of EYR and ELR The Emergy Yield Ratio (EYR) is defined as the total emergy of inputs to a system divided by the imports from the larger economy: EYR = (N+R+M+S)/(M+S) (Brown and Ulgiati 1997, 2004; Raugei et al. 2005; Ridolfi and Bastianoni 2008). This index is interpreted as the ability of the local system to exploit local resources in order to deliver real wealth to the larger economy. More details on the evolution of its definition in the literature and interpretation are provided in Supplementary Information material, section SI5.1. When applied to an activity, EYR reflects its efficiency in processing local resources: the smaller the emergy of imports (M+S), the higher the EYR, the more efficient is the activity. Recently, Brown et al. (2012) argued that this definition is misleading for the evaluation of technological systems (i.e. chains of processes), because typically they do not have a specific location in the global economy. The authors rather proposed switching from local vs. imported to foreground vs. background inputs by adopting a lifecycle perspective when calculating EYR for an industrial process. Foreground input flows were defined as flows that are directly input to the process expressed in the emergy of the raw resources from which they are derived and background inputs as the investments required previously to extract, refine, and deliver foreground input flows (Brown et al. 2012). For example, in the case of diesel oil input to the modeled process, the emergy of crude oil is considered as foreground input, while background investments cover the additional emergy inputs throughout the production chain used up to extract crude oil and refine it into diesel oil. Table 5.1 illustrates these definitions with typical inputs to a technological system. This variant of EYR is intended to indicate the efficiency of the production chain in processing natural resources necessary to run the production activities. The operability of such new definition requires however the study of the full production system s lifecycle, as opposed to conventional practice in emergy evaluation that only focuses on a single site. Interestingly, the recently developed emergy calculation software SCALE (Marvuglia et al. 2013a) can generate lifecyclebased accounting of man-made material and energy inputs to a system. However, SCALE cannot discern yet between foreground and background inputs, and therefore cannot provide a direct calculation of lifecycle-based EYR. Table 5.1: Foreground and background inputs to calculate EYR with a lifecycle perspective (adapted from Brown et al. 2012). Foreground inputs Background inputs Locally available renewable and non-renewable resources used up in the production site Non-renewable feedstock (fossil fuels, metal and mineral ores, nuclear resources) in ground (e.g. at mine or well), and renewable resources that constitute the fuels, electricity and materials used up on the production site, i.e. excluding those used up in the various steps of processing, transport, treatment, etc. Materials for infrastructures (excluding extraction, processing, etc.) Labor at operation phase Renewable and non-renewable inputs such as fuels, electricity and materials used for extraction, processing, packaging, storage, transport and delivery of foreground inputs, for both operation and infrastructures. Services used in the supply chain and the treatment of waste. 105

106 The Environmental Loading Ratio (ELR) is defined as the total emergy value of non-renewable and invested resources divided by the emergy value of renewable ones (ELR = (N+M+S)/R). A high ELR indicates a high intensity of non-renewable resource use (N) and/or a high technological level (M+S) of the process. In addition, it is often claimed that a high ELR highlights a high level of environmental stress on the local environment (Brown and Ulgiati 1997; Ulgiati and Brown 1998; Ridolfi and Bastianoni 2008), although the notion of environmental stress is arguably adapted for emergy accounting, whereby it focuses on the use of resources but usually excludes pollution-related impacts (for further discussion about accounting for pollution impacts in emergy evaluation, see Arbault et al. 2014). Therefore, in this paper, we consider the original meaning of ELR as a nonrenewable-to-renewable ratio, as labeled by Odum (1996). We also acknowledge that the formulation of this indicator should be revised to account for pollution impacts, as soon as a commonly accepted accounting method is made available. A global re-formulation of ELR has been also increasingly used in the recent years (see e.g. Ortega et al. 2002; Lu et al. 2009; Agostinho et al. 2010; Yang et al. 2010; Duan et al. 2011; Lima et al. 2012; Zhang et al. 2012; Tao et al. 2013; Wilfart et al. 2013). In contrast to the original definition, man-made inputs were split in a renewable part (Mr+Sr) and a non-renewable part (Mn+Sn); thus, global ELR was calculated as the ratio of both local and invested non-renewable resources by local and invested renewables (ELR = (N+Mn+Sn)/(R+Mr+Sr)), mixing up technological and natural inputs. As a consequence, such definition ignores the evaluation of the technological level of the studied system, focusing just on the dependence of non-renewable vs. renewable resources with a lifecycle perspective. Following that rationale, the renewable fraction of purchased inputs can be generally estimated using prior emergy evaluation studies, or assigning either 0% or 100% value depending on the type of products (for example, seeds, food and log would be considered as 100% renewable, while peat and minerals would be considered 0% renewable). Such approximations neglect the complex lifecycle chain to produce and deliver these materials and services. In that respect, SCALE can decompose the lifecycle-based emergy value of inputs into several resource categories (described in Rugani et al. 2011a), providing a clear distinction of the renewable and non-renewable parts of the output emergy, thus eventually of the global ELR too. Other alternatives concerning both ELR and EYR integrated waste management and the accounting for pollution impacts through the additional technological requirements or landscape restoration necessary to deal with them (Mu et al. 2011; Song et al. 2012, 2013; Li et al. 2013). However, a lifecycle approach would already consider such technological requirements as manmade inputs of the studied system. Therefore, no additional consideration of these inputs is necessary here. Bakshi (2000) proposed a theoretical framework to account for ecosystem services of pollution remediation in emergy evaluation. But this rationale would require a comprehensive modeling of environmental mechanisms (Bastianoni 1998; Xu et al. 1999; Campbell 2000) which, to the best of our knowledge, has not been convincingly simplified by emergy practitioners so far. As a result, diverging approaches are currently adopted (e.g. on the ecosystem service of waste assimilation: Mu et al. 2011; Zhang et al. 2011a; Reza et al. 2013; accounting of ecosystem services actual value: Campbell and Brown 2012). Concerning labor and service inputs, different strategies were adopted: local labor was either considered as 100% renewable or split as 90% non-renewable and 10% renewable (e.g. La Rosa et al. 2008; Lu et al. 2009), while purchased services were considered as 100% non-renewable (e.g. Agostinho et al. 2010; Yang et al. 2010; Ciotola et al. 2011). Other studies proposed new 106

107 Chapter 5 A semantic study of the Emergy Sustainability Index in the hybrid lifecycle-emergy framework emergy indices focusing on labor empowerment (see e.g. Ortega et al. 2002). We consider that in view of environmental sustainability, local labor and purchased services should be accounted for like any other man-made input in emergy accounting, i.e. using a convenient unit (work-hours or monetary value) and the corresponding UEV. Moreover, the share of renewability in labor should be that of the regional economy that delivers this workforce (if additional details to define labor production are not available), reflecting the impact due to goods and services consumption in the respective country. This emergy approach could also match the lifecycle approach on human labor proposed by Rugani et al. (2012b), in which labor is considered as a technosphere process in the same way as any other unit process included in lifecycle inventory databases. All these examples show that, in contrast with the LCA practice and guidelines (European Commission 2010c), the selection of system boundaries and accounting method for inputs classification and calculation are not yet standardized procedures in the emergy framework. This paper focuses on the influence of system boundaries on emergy indicators. The comparison of accounting methods is out of the scope of the present work. 5.3 Method Formulation of EYR, ELR and ESI variants The conventional variant of EYR (hereafter EYR0) discerns local inputs, available on-site, from imported ones (Odum 1996), whereas the version of EYR (EYR1) introduced by Brown et al. (2012) differentiates foreground inputs from background ones. These terms originate from the LCA framework (European Commission 2010c), and were adapted to the emergy evaluation approach to distinguish the natural resources that are directly used up in the studied production site from those that are used up elsewhere in the life cycle (see section 5.2 and Table 5.1 for Brown et al. s quoted definition and examples). By matching the two versions, one can classify the inputs into three categories: foreground inputs are split between local resources (L), i.e. recovered from the natural environment at the production site and directly used in the process without previous transformation (e.g. the wind energy captured by a wind turbine), and direct investments (D), which gathers all other foreground inputs listed in Table 5.1. The third category of input is the background investments (B). In the present study, these three categories are further split between renewable and nonrenewable resources by adding, respectively, a lowercase r and n, resulting in 6 items: Lr, Ln, Dr, Dn, Br, Bn. D and B are usually aggregated into F (sometimes split into Fr and Fn), while Lr and Ln are typically named R and N, respectively. This traditional nomenclature was adapted in our paper for the sake of consistency with the following equations. By noting U the sum of the emergy of all independent inputs, the two versions of EYR can be reformulated as follows: (5.1) The two definitions use different system boundaries: EYR0 focuses on the production site, while EYR1 adopts a lifecycle perspective that also includes inputs upstream to the production site. (5.2) 107

108 Similarly, the two variants of ELR, i.e. local (ELR0) and global (ELR1), can be formulated as follows: (5.3) Hence, the four variants of ESI are defined accordingly in Table 5.2. (5.4) Table 5.2: List of the four variants of ESI and relative characteristics, depending on the focus for EYR (site vs. lifecycle) and for ELR (local vs. global) as defined in equations ( ). Investment on local resources (EYR0) Investments on lifecycle local and direct resources (EYR1) Load on local environment (ELR0) Load on global environment (ELR1) ESI00 corresponds to the conventional version defined by Brown and Ulgiati (1997). It sets the boundaries of the studied system at the site level. ESI01, which is often found in the literature (e.g. Lu et al. 2003, 2009; Yang et al. 2010; Duan et al. 2011; Zhang et al. 2012; Wilfart et al. 2013), assesses the ability of the production site in extracting natural resources while evaluating the use of non-renewable vs. renewable resources at the global scale. ESI10 investigates the efficiency of the production chain in processing the natural resources necessary to run the production site, i.e. where the output of the whole production chain is delivered, while focusing on the exploitation of the environmental resources located at the production site only. Finally, ESI11 encompasses the global-scale technological system necessary to run the production site. We do not recommend the use of ESI01 (although often used in literature) and ESI10 because these versions mix the system boundaries of the local production site and its lifecycle. However, they can be useful to decompose and investigate the differences between ESI00 and ESI11, which, in contrast, present coherent boundaries and complementary meanings Sensitivity of ESI to input categories The partial derivatives of a function f(x,y,z), usually noted as f/ x, f/ y and f/ z, are calculated by considering all variables as fixed parameters - except the one over which the function is derived. For instance, f(x,y,z)/ x is calculated by considering y and z as constant parameters. The method helps understanding the influence of each variable in a multivariable function. Therefore, it is particularly useful in the present study to depict the possible diverging behavior of the ESI versions with respect to the variation of the different input categories. A positive (respectively negative) partial derivative indicates that the function increases (respectively decreases) when the variable increases. If the sign is mathematically conditioned by the relative value of all variables, as it happens in this study, a Monte-Carlo analysis can be performed to evaluate the probability of the condition to be verified: each variable is assigned a random value in a defined range (in this study: between 1.00 E-6 and 9.99 E+9), and the condition is checked. The operation is iterated a large number of times (100,000 times in this study) to 108

109 Chapter 5 A semantic study of the Emergy Sustainability Index in the hybrid lifecycle-emergy framework retrieve probability results. The absolute value of the range is not important here because the partial derivative does not include constants: only the relative value of all variables needs to be controlled. Monte-Carlo analysis is increasingly used in emergy evaluation, for example to calculate confidence intervals of UEVs (Brown et al. 2011; Hudson and Tilley 2013) or of the baseline (Brown and Ulgiati 2010). Brown and Ulgiati (2011) also used this technique to verify the sustainability inequality proposed by Harizaj (2011), in a very similar fashion as it is used here Calculation of inputs The calculation of Lr and Ln can be performed using traditional methods of emergy evaluation applied to the generation of natural resources (Odum 1996). However, the separate calculation of the other inputs needs to be further detailed. For example, the emergy value of labor and services throughout the chain of processes is usually approximated with their monetary value and the regional emergy-money ratio. Some studies also consider the extended energy and material requirements to produce human labor (Pulselli et al. 2007; Bastianoni et al. 2009). Labor needed on the production site can be categorized as direct investments (D), while labor and services required outside the production site, and reflected in the market price of purchased materials and services, are seen as background investments (B). Their emergy value per monetary unit and renewable portion was assumed equivalent to that of the regional economy (France in the present study, see section 5.3.4), i.e E12 sej/ and 0.5% renewable (Sweeney et al. 2007) considering 9.26 E24 sej/yr as baseline for all results of this study. The emergy value of material investments are calculated using SCALE (Marvuglia et al. 2013a), by modeling their lifecycle using the Life Cycle Inventory (LCI) database ecoinvent v2.2 (Ecoinvent 2010), which involves thousands of unit processes in the calculation. SCALE provides, among others, the emergy value of natural resources entering the life cycle aggregated in 8 categories: renewable (airborne, energy, land occupation, waterborne) and non-renewable (fossil fuels, metals, minerals, nuclear) resources. These items correspond to (Dr+Br) and (Dn+Bn), respectively. However, SCALE cannot distinguish direct investments from background ones. Therefore, a specific algorithm was implemented in order to track the flows of direct investments within the chain of processes up to the corresponding natural resource. The algorithm was developed as an Excel macro and directly relies on ecoinvent datasets. It delivers the list of D flows, which can then be summed up into Dr and Dn using the corresponding UEVs. Br and Bn are calculated by subtracting respectively Dr and Dn from SCALE results. A usable copy of the algorithm is available in SI5.3. Some direct investments are not natural resources but by-products from other processes, namely: hydrogen (by-product of sodium chloride electrolysis); sulfur (by-product of crude oil and natural gas fractioning); blast furnace slag cement (waste from iron production); glass from public recycling; aluminum and iron scrap; secondary copper; disposed coal ash; MOX (recycled fuel for nuclear power plants); carbon dioxide (retrieved from waste gas). In such cases, the emergy value of the recycled material is assimilated to that of the cumulated work used up for its concentration and purification, following the concept of emformation (Brown 2005). Therefore, those listed byproducts are assigned an emergy value of 0 (see e.g. Amponsah et al. 2011; Tilley 2011; 109

110 Agostinho et al for emergy accounting of recycling processes), while energy and material inputs of the recycling processes are clearly background investments. Discerning direct from background investments presents however ambiguous cases: for instance, sodium chloride electrolysis can be seen as the transformation of mineral salt and water into caustic soda and chlorine, or as the transformation of electricity exergy into chemical exergy stored in a solution of sodium cations and hydroxide anions and the chemical bonds of gaseous chlorine. It depends if the primary source to be considered is only the material support or if the energy stored in chemical bounds should also be included. This aspect is generic for the material transformation in the chain of processes, i.e. both material and energy inputs contribute to form the physico-chemical structure of a product. The following processes were found particularly ambiguous in our case studies: electrolysis of mineral sodium chloride to prepare sodium hydroxide, gaseous chlorine and sodium hypochlorite; electrolysis of bauxite to prepare aluminium hydroxide; preparation of acetic anhydride via electric oxidation of acetaldehyde or acetic acid; preparation and regeneration of activated carbon using heat from natural gas; conversion of water into of steam using heat from natural gas. The list of direct investments to each process of the production chain is detailed in SI5.3. The influence of these options (i.e. including material only or material & energy as direct investments) is illustrated in the results section Case studies A numerical application of the proposed revised procedure of emergy index calculation is performed for four water treatment plants (WTPs). Two of them (Site 1 and Site 2) are WTPs located in Paris area and use freshwater from the Seine River, while the other two (Sites A and B) are new plants located in smaller and more polluted streams in Brittany. Infrastructure data are available for Sites A and B only. Therefore, in the present work a total of six models are analyzed: Site 1; Site 2; Site A (no infra); Site B (no infra); Site A (with infra); Site B (with infra). More information about the treatment processes, material and energy inventories and lifecycles of the four WTPs is available in Igos et al. (2013a, 2013b). We have shown in a previous study that the operation phase of Sites A and B requires more emergy (per m 3 of produced potable water) than Sites 1 and 2, which indicates that more polluted waters imply a larger bestowment of natural resources (estimated in emergy) to be made potable (Arbault et al. 2013b). For the present research, results on local resources and labor and service inputs were retrieved also from Arbault et al. (2013b), while results on material inputs obtained with SCALE (Dr+Br and Dn+Bn) from Arbault et al. (2014). Numerical datasets previously elaborated on the case studies are available in SI Results and discussion Direct vs. Background inputs in technosphere products Figure 5.1 shows the relative contribution of direct (D) and background (B) investments to the total UEV of the various materials and energy used up in the WTPs as lifecycle inputs at the level of the production site (see also Table S5.1 in SI5.2). The production of lime and quicklime, which are the main products consumed to treat water, has few transformation steps upstream in their lifecycle. Therefore, the contribution of the direct investments is very high (96-97%). In other 110

111 Chapter 5 A semantic study of the Emergy Sustainability Index in the hybrid lifecycle-emergy framework terms, the emergy value associated with the human investments to e.g. extract the calcite ore, refine it into lime and transport the resulting chemical, is marginal compared to the emergy value associated with the natural resource itself. In contrast, the similar ratio obtained for potassium permanganate is due to the high UEV of manganese, which counterbalances the relatively low demand (i.e. low input amount in kg/m 3 drinking water) for this input in the process model. Figure 5.1: Relative contribution of direct investments (D) to the total UEV of material and energy products used as inputs in the life cycle phases of the case studies (i.e. only at the production site level, see also Table S 5.1 in SI5.2). The high ratios obtained for plastics, nylon and acrylic acid originate from a limitation of the ecoinvent models regarding the chains of production of petroleum-derived compounds, which are only available in a lifecycle aggregated form. Therefore, it is not possible to distinguish the amount of fossil resources incorporated in the final product from those products that are used up in the production chain as background investments. The other products have a ratio between 10% and 50% of direct investments, which means that the background investments to process the 111

112 natural resources are the main contributors to their UEV. The emergy value of liquid carbon dioxide is only due to background investments, because the only direct investment is CO 2 from waste gas, which was assigned an emergy value of 0. Finally, Figure 5.1 shows that the difference between the two options to define direct investments (i.e. materials only or materials & energy ) has a considerable influence on products, for which the choice is ambiguous, e.g. regarding chemicals derived from sodium chloride and activated carbon. It is therefore necessary to define more precisely the notion of direct investments. A too restrictive definition (e.g. only processes under direct control of the operator) would lead to considering most of the inputs as background investments. Conversely, a too broad definition (e.g. considering that all energy transformation steps should count as direct investments) would result in including every input as direct investments. In this study, these options yielded significant differences (as shown in Figure 5.1) with products derived from electrolytic and thermic processes (listed in section 5.3.3), through which the exergy embedded in the chemical bonds of the output (hence its usefulness) originates from resources different from the supporting material. In addition, other processing steps (e.g. centrifugation, filtration, drying), involving a change in the product s exergy, may require further attention Ranking of the WTPs The decomposition and automatic tracking of direct and background flows allowed us to calculate the emergy indicators. Figure 5.2 compares the four variants of ESI computed for the 6 case studies (see Table S for details; the material only option as direct investments was selected). The four ESI variants show similar rankings (ordered from the highest to the lowest value: Site 1, Site 2, Site B no infra, Site B with infra, Site A no infra, Site A with infra), but provide different absolute values. In addition, the highest value is obtained in each case study for ESI11, followed by ESI01, ESI10 and ESI00. Figure 5.2: Values (dimensionless) calculated for the four variants of ESI (see Table 5.2) applied to the six case studies of WTP. The ratio of ELR0 / ELR1 is between 2.22 and 2.70, while the ratio EYR1 / EYR0 is between 1.13 and Therefore, the definition of ELR has more influence than that of EYR in these case 112

113 Chapter 5 A semantic study of the Emergy Sustainability Index in the hybrid lifecycle-emergy framework studies, mainly because of the relative importance of Br (see Table S5.4) and its different position in equations (5.3) and (5.4). The influence of EYR s definition is more important when direct investments (Dr+Dn) are comparable (in magnitude) with background investments (Br+Bn), which is the case for Site B (with and without infrastructure). The EYR definition of Brown et al. (2012) is an important step toward a lifecycle-oriented emergy evaluation. The present study demonstrates that the definition is operational provided that the inputs are properly calculated. In our work, the inputs have been calculated according to the rules of emergy algebra applied to the whole lifecycle network of processes (Marvuglia et al. 2013a; Tiruta-Barna and Benetto 2013). Accordingly, it is suggested to incorporate the algorithm developed here in a revised version of SCALE, in order to provide an automatic calculation of lifecycle-based EYR, ELR and ESI. The global version of ELR, already employed by several authors (e.g. Ortega et al. 2002; Lu et al. 2009; Agostinho et al. 2010; Yang et al. 2010; Duan et al. 2011; Lima et al. 2012; Zhang et al. 2012; Tao et al. 2013; Wilfart et al. 2013) is more accurately calculated in this work thanks to the use of the SCALE software. However, the definition of direct investments needs further discussion: the material & energy option (which can be recalculated using SI5.3) reduces EYR1 by % in these case studies, involving the same change on ESI10 and ESI11 (i.e. it has no influence on ELR). The output UEV (equal to U in these case studies) provides a slightly different ranking (see Table S5.4), i.e. from lowest to highest: Site 1, Site 2, Site A no infra, Site B no infra, Site B infra, Site A infra. In addition, LCA results on resource depletion rank as follows, from lowest to highest impact (Arbault et al. 2013b): Site2, Site 1, Site A no infra, Site A infra, Site B no infra, Site B infra. For all ranking systems, the relatively small difference between Sites A and B should be substantiated by consistent uncertainty analysis. The ranking provided by ESI is thus in line with the previous results. All these indicators show that it is more environmentally sustainable to produce potable water from a less-polluted resource (than from a more-polluted one), such as in Sites 1 and 2 (Arbault et al. 2013b), as well as using electricity rather than chemicals in the treatment process, such as in Site 1 (Arbault et al. 2014). However, the rationale is specific to each assessment tool, which leads to the following statements: ESI is an indicator based on both the activity and its natural and economic context, i.e. it considers the origin of resources (renewable or non-renewable, through ELR) used up throughout the system s lifecycle and the outcomes for the larger economic system (through EYR); UEV provides the total (solar-equivalent) amount of resources used up by the activity s lifecycle, regardless of their origin and the activity s outcomes; LCA indicators used for resource depletion account for the cost of future requirements to extract the same amount of non-renewable resources used up by the activity s lifecycle Sensitivity of ESI to input categories The four variants of ESI were analyzed using partial derivatives, to investigate the influence of marginal changes of the six variables (Lr, Ln, Dr, Dn, Br, Bn), i.e. the sensitivity of the indicator with respect to each input category. The six variables are assumed to be independent from each other, i.e. each one can vary without changing the value of the others. Indeed, changing the value of a variable would alter the configuration of the supporting technological network and therefore affect all variables. However, the assumption can be considered valid as far as the study of 113

114 derivatives is limited to marginal changes, by indicating whether the result would increase or decrease following a small change in one variable. Table 5.3a shows the sign of partial derivatives. A positive (respectively negative) sign means that ESI would increase (resp. decrease) if the corresponding variable (slightly) increases. The partial derivatives of ESI01 and ESI11 by variables Dr and Br can be either positive or negative, depending on the value of all the six variables, as shown in Table 5.3a (see footnotes (1) and (2)). Table 5.3b shows that (1) and (2) are given the same sign in 67 % of simulations from the Monte- Carlo analysis (see section 5.3.2). Moreover, it was not possible to satisfy simultaneously the conditions (1) < 0 and (2) > 0, while in 33% of cases it resulted that (1) > 0 and (2) < 0. The increase of Br may lead to an increase of ESI01 but not necessarily of ESI11, whereas, as shown in Table 5.3a, by increasing Dr, ESI11 always increases but it is not necessarily the case for ESI01. Table 5.3: Sensitivity of ESI variants to input categories. a) Influence of the six components (Lr, Ln, Dr, Dn, Br, Bn) on the four variants of ESI. A positive (resp. negative) sign of the partial derivative means that ESI would increase (resp. decrease) if the corresponding variable increases. b) Monte-Carlo analysis on the correlations between the signs of (1) and (2). a) by Partial derivative of ESI00 ESI10 ESI01 ESI11 Lr > 0 > 0 > 0 > 0 Ln < 0 < 0 < 0 < 0 Dr < 0 < 0 (1) > 0 Dn < 0 < 0 < 0 < 0 Br < 0 < 0 (1) (2) Bn < 0 < 0 < 0 < 0 (1) > 0 ELR1 EYR0 + 2 > 0; (2) > 0 ELR1 EYR1 + 2 > 0 (these relations are deduced from the formulas of the respective partial derivatives). b) (1) > 0 (1) < 0 (2) > 0 48 % 0 % (2) < 0 33 % 19 % Using R = (Lr + Dr + Br), the condition (2) > 0 is equivalent to each following condition: (5.6) (5.7) In other terms, the increase of renewable background investments (Br) has a positive effect on ESI11 only if the relative contribution of renewable inputs R/U is smaller than B/(U-B), i.e. B/(D+L), the ratio of background by foreground inputs. This condition is more precise and more detailed than previous conclusions e.g. mentioned by Ulgiati and Brown (1998): Minimization of (5.8) 114

115 Chapter 5 A semantic study of the Emergy Sustainability Index in the hybrid lifecycle-emergy framework imports are important measures of sustainability. However, their statement remains valid for ESI00, i.e. for site-based emergy accounting, as shown in Table 5.3a. The four variants behave similarly regarding the influence of the other variables. A purely mathematical analysis would lead to the conclusion that in order to increase the ESI11 of an activity or a product s lifecycle, Ln, Dn and Bn should be minimized whereas Lr and Dr should be maximized. It seems logical to increase Dr and decrease Dn simultaneously, for example by purchasing renewable-based electricity and chemicals; or to decrease Bn while increasing Br (provided that ESI11/ Br > 0, i.e. ELR1 EYR1 + 2 > 0, which is the case for all the present case studies see Table S5.5 in SI5.2), e.g. by selecting suppliers committed to restrict the use of non-renewable resources in their supply chain. Table 5.4 shows the results of the numerical application of these marginal changes to the six case studies, with a special focus on the sensitivity of ESI11 (the highest score showed in Figure 5.2) with respect to the input categories. The level of change of each input (the others being kept constant) required to increase ESI11 by 1% is depicted. The results show that the indicator is more sensitive to Bn, followed by Lr, Br, Dn and Dr. On average, a 1% increase of ESI11 requires an increase of Br by 4.3% (+1.5 E10 sej/m 3 ) or a decrease of Bn by 1.06% (-7.6 E9 sej/m 3 ). Alternatively, it requires increasing Dr by 75% (+5.93 E9 sej/m 3 ) or decreasing Dn by 7.1% (-3.3 E10 sej/m 3 ). In other terms, requiring suppliers to rely less on non-renewable resources for the operation of their processes seems to be the less demanding effort (in emergy terms), while the same increase of ESI11 would require replacing the large majority of mineral-based chemicals with chemicals made out of renewable resources, which would in turn involve re-considering the whole technological system. For instance, producing sodium-chloride based chemicals from sea water rather than from mineral ore (i.e. increasing Dr and decreasing Dn, respectively) is an interesting option to increase ESI11. Other examples concern limiting the transport operations and the use of renewable energy sources (i.e. decreasing Bn and increasing Br, respectively). Of course, the whole production chain of this alternative would need to be modeled again for a novel evaluation, which is out of the scope of this paper. Table 5.4: Changes in inputs categories (in sej/m 3 and in %), per WTP s site, which would increase ESI11 by 1%. Lr Dr Dn Br Bn sej/m 3 % sej/m 3 % sej/m 3 % sej/m 3 % sej/m 3 % Site E E E E E Site E E E E E Site A 5.59 E E E E E Site B 6.27 E E E E E Site A w. infra 5.83 E E E E E Site B w. infra 6.46 E E E E E Average 5.92 E E E E E Ln is null in the case studies and thus disregarded in this table. In the present case study, ESI11 exhibits significant sensitivity to Lr, which is the freshwater used for providing 1 m3 of drinking water. However, increasing Lr per output unit (i.e. in the present case study using more raw water to produce 1 m3 of potable water) hardly makes sense and does not provide any demonstrated benefit to the environment. This apparent contradiction could be solved by extending the scope of the study to the ecosystem services provided by the river. Accounting for the loss of freshwater-related ecosystem services due to an increased appropriation of freshwater by the treatment plant (i.e. extraction rate greater than replacement 115

116 rate) would be considered as an Ln input, which would decrease the value of ESI11. Such option could help accounting for the environmental impacts (i.e. environmental loading ) in the ELR. However, as mentioned in section 2, emergy-based accounting methods for the loss of ecosystem services are not yet operational: this input was therefore disregarded in this study (further discussion available in Arbault et al. 2014). 5.5 Conclusion The bibliographic survey and the formal analysis of the Emergy Sustainability Index (ESI) presented in this study showed that its use and formulation in the literature often rely on an unclear definition of the boundaries of the studied system, leading to mixing the assessment of the production site with its supply chain. Besides, this situation does not prevent from misinterpretation of emergy indicators, especially ELR. The aim of this paper was not to discuss the definition of these indicators, but rather to discuss the consequences of using different system boundaries. Two versions of the ESI were defined: a site-oriented version, which corresponds to the original formulation (Brown and Ulgiati 1997; Ulgiati and Brown 1998), and a lifecycleoriented version. They have different meanings, which are usefully complementary. But they must be used carefully, in accordance with coherently selected boundaries of the studied system. We demonstrated that background investments may highly influence the UEV of purchased materials and energy. Therefore, these inputs should be accounted for with a lifecycle perspective, especially if they are important contributors (in emergy terms) to the studied system as it is the case in most industrial activities. The global version of ELR can be consistently calculated using results from the software tool SCALE, while the lifecycle-oriented version of EYR proposed by Brown et al. (2012) was made operational through the formulation of a specific algorithm, delivered in the supporting material (SI5.3). However, the definition of foreground inputs needs further consideration. A consensual agreement is necessary to provide a consistent and reproducible framework for hybrid lifecycle-emergy evaluation. Finally, the study of partial derivatives showed that decision-makers may efficiently increase the environmental sustainability of their activities by e.g. selecting suppliers who are already committed to replace their use of non-renewable resources by renewable ones. Besides, sensitivity analysis is useful to identify specific potential actions and alternative scenarios to increase the environmental sustainability of the studied activity. However, it is important to stress that the present analysis does not allow forecasting the outcomes of such actions. It rather helps understanding the significance of the different formulations of ESI and the qualitative influence of each input category. Indeed, replacing a material by another (e.g. a non-renewable resource by a renewable one) in the supply chain would require significant changes in the network of processes of the system s lifecycle (see Srinivasan et al for a discussion of this concept applied to building construction). As a consequence, the value of several inputs in the calculation of ESI is expected to change and the ESI of the alternative scenario would need to be compared with the current situation. Therefore, the aim of the sensitivity analysis based on partial derivatives and marginal changes shall be restricted to the identification of potential actions but cannot anticipate their outcome. While SCALE allows an automated calculation of the emergy value of inputs supplied by the global network of technological processes, we recommended implementing the additional algorithm presented in this paper (SI5.3) in the software, to automatically retrieve the emergy- 116

117 Chapter 5 A semantic study of the Emergy Sustainability Index in the hybrid lifecycle-emergy framework based indicators and provide a fully integrated tool to emergy practitioners. However, the calculation procedure used in this paper to compute background investments does not fully comply with the emergy algebra rules for all possible application cases. Recommendations for its improvement are also provided in SI5.3. Concerning the evaluated Water Treatment Plants, results show that all studied versions of ESI provide identical rankings, in line with the conclusions of previous assessment of these case studies (Arbault et al. 2013b, 2014; Igos et al. 2013a, 2013b). However, they exhibit different absolute values; this underlines the importance of robust and consensual definition, formulation and calculation of the emergy-based indicators. As a general conclusion, using less polluted freshwater resources makes the treatment process lighter, thus enhances ESI and reduces output UEV, while electricity-based treatment processes seem more environmentally sustainable than chemical-based ones, due to the non-renewability of natural resources used up to produce the chemicals. The first concrete studies attempting to enumerate the emergy shortcomings and provide a roadmap to improve its evaluation method date back to 2004 (Hau and Bakshi 2004). Despite much has been made afterwards to formalize the emergy concept and expand its acceptance (e.g. Brown and Ulgiati 2010, 2011; Bastianoni et al. 2011; Ju and Chen 2011; Morandi et al. 2013b; Tiruta-Barna and Benetto 2013), emergy is not yet broadly employed in environmental sustainability assessment practices, likely because of the (perceived) lack of transparency in calculation procedures and the difficulty in understanding its meaning (Herva et al. 2011). We are convinced that adopting a lifecycle perspective in emergy evaluation, with a consequent adoption of higher level of reproducibility and accurateness, and using operational tools that deliver clear indications to decision-makers, would help disseminating the concept and use of emergy among industries and a broader number of stakeholders. Acknowledgments This project is supported by the National Research Fund Luxembourg (Ref ) and the French National Research Fund (project EVALEAU ANR-08-ECOT C0238). The authors thank the two anonymous reviewers for their contribution to improve the quality of this paper. Supplementary information The supplementary materials include a discussion on the evolution of EYR s definition and other research on ESI (SI5.1), inventory data and results of the cases studies (SI5.2), and a macroenabled spreadsheet file operational to run the algorithm presented in this paper (SI5.3). 117

118

119 6. A first global and spatially-explicit emergy database of rivers and streams based on high-resolution GISmaps Published as Arbault, D., Rugani, B., Benetto, E., Tiruta-Barna, L., A first global and spatially-explicit emergy database of rivers and streams based on high-resolution GIS-maps. Ecol. Model. In press. DOI: /j.ecolmodel Abstract In emergy evaluation (EME), water is often identified as the main renewable resource input of a natural or human system. Water flows in EME have been generally examined with a global perspective, i.e. without considering topographical and climatic differences at regional or local scales. Hence, spatial differentiation in water flows characterization is essential to improve the quality of EME results. This paper introduces the first global, spatially-explicit emergy dataset of freshwater flows, developed following the rationale found in prior EMEs of rivers. The Unit Emergy Value (UEV) of a stream was calculated as the highest value between rain, chemical potential emergy and rain geopotential emergy over the stream s catchment area, divided by the stream flow rate. This approach was applied with a high resolution and a global coverage, using Geographic Information System (GIS) software and, notably, world maps of precipitation, evapotranspiration and elevation, to estimate accumulation patterns of rainfall emergy value and flow rates. Preliminary results are compared with available data on river s UEVs retrieved from previous studies and with the actual stream flow of major rivers in the world and in France. While flow rates modeled in the database show important differences as compared to actual data, the comparison of the modeled emergy value of rivers with prior studies was made difficult by the heterogeneity in calculation details observed previously. Therefore, it is highly recommended for the emergy community to foster the use and improvement of such high-resolution, spatiallyexplicit dataset instead of using regional or global UEV averages, which should only be used when reliable local values are not available. Hence, territorial averages were computed in order to characterize background processes in the hybrid lifecycle-emergy accounting framework, as this approach can complement and enrich the conventional EME with the inclusion of detailed information on supply-chain processes. To this aim, data were aggregated over major watersheds and administrative regions, and weighted with a proxy for urban surface water consumption. The next steps identified to enhance our prospective work include: 1) the characterization of water reservoirs (glaciers, lakes, groundwater, soil moisture), 2) the improvement of runoff modeling and stream flows, 3) the spatial assessment of atmospheric processes to refine transformities of rain (chemical potential and geopotential), and 4) the inclusion of additional elements such as sediments, minerals and particulate matter as a flow of emergy in rivers.

120

121 Chapter 6 A first global and spatially-explicit emergy database of rivers and streams based on high-resolution GIS-maps 6.1 Introduction Research question According to the emergy concept of the ecologist Howard T. Odum (1988, 1996), Emergy Evaluation (EME) is a method of environmental accounting for human activities and territories (Odum 1996). The Unit Emergy Value (UEV) (Brown and Cohen 2008) of a natural resource, which is expressed in solar emjoules (sej) per unit of resource (mass, energy, volume, etc.), is defined as the total (solar) energy directly and indirectly used up by natural systems to produce one physical unit of the resource (while the resource s Transformity, Tr, is expressed as its emergy value divided by its available energy, i.e. exergy, in sej/j). EME relies on the emergy value of natural resources required by the studied anthropic system in order to calculate the output(s) UEV and a number of environmental performance indicators. In addition, natural resources are indirectly used up through the consumption of imported, man-made goods and services. For instance, freshwater can be used up directly in a water treatment plant (Odum et al. 1987b; Buenfil 2001; Pulselli et al. 2011b; Rugani et al. 2011b; Arbault et al. 2013b) or in various activities over a territory (e.g. Odum et al. 1987a, 1987b, 1998; Brown and McClanahan 1996; Kang and Park 2002; Campbell et al. 2005b; Pulselli et al. 2008a; Campbell and Ohrt 2009; Chen and Chen 2009; Chen et al. 2009b; Giannetti et al. 2013b), but activities and territories may also import electricity and food, which indirectly require freshwater in their production lifecycle. As a consequence, almost all natural processes and human activities need, directly and indirectly, freshwater. But this resource is not uniformly shared on terrestrial land, leading to diverse climatic conditions and landscapes, and influencing the human settlement, which typically tends to develop nearby rivers and water-abundant areas (even though this is obviously not the only determining factor). Therefore, though the emergy literature already offers several examples of water resource assessment, it is important to correctly assess the local emergy value of water with a commonly-acknowledged and consensual method. A short critical review about EME studies in the field of water analysis is provided in Section Site-specific evaluations of resources UEV are increasingly determined with Geographic Information System (GIS), as this enhances the accuracy and reproducibility of local assessments in EME (see also Section 6.1.2). Applications are diverse: evaluation of crops and land (Agostinho et al. 2010), territorial assessment of renewable emergy budget (Mellino et al. 2014), spatial patterns of empower density, i.e. emergy flow per unit area (Pulselli 2010). Huang et al. (2007b) used an accumulation function over freshwater streams to determine the relationship between land transformity and the convergence observed in the hierarchical patterns in a watershed, already noticed in Odum (1996). A global evaluation of atmospheric resources was performed with GIS by Brandt-Williams and Brown (2011). To our knowledge, however, such exercise has never been attempted on rivers worldwide, although GIS-based characterization of freshwater resources is being developed in other environmental assessment fields such as Life Cycle Assessment (see e.g. Pfister et al. 2009; Boulay et al. 2011c), water footprint and Input- Output Accounting (Ridoutt and Pfister 2010; Lenzen et al. 2013), as well as hydrological modeling for risk assessment or resource management (e.g. Döll et al. 2003; Chormanski et al. 2011; Assefa 2013; Flörke et al. 2013). The aim of the present paper is to present the first attempt to develop a worldwide, high resolution spatially-explicit database of rivers UEV. To make this contribution a useful and consensual tool, transparency and reproducibility are considered as key factors in the methodological 121

122 development. In addition, results are compared with actual river data and published UEVs for cross-checking, with the purpose of carrying out a prospective work in the domain of river UEVs assessment. Other freshwater resources, such as snow cover (Campbell and Ohrt 2009), lakes and aquifers (Buenfil 2001; Brown et al. 2010; Pulselli et al. 2011b; Díaz-Delgado et al. 2014), sediments and minerals in rivers, and glaciers are excluded from this study, as well as stream transformities (Huang et al. 2007b; Chen et al. 2009b), due to a lack of available data. The computed database is finally complemented with territorial averages, in order to derive an operational UEVs dataset suitable for hybrid lifecycle-eme analysis State-of-the-art In EME, rainfall and rivers are characterized with two forms of exergy (Odum 1996, 2000; Odum et al. 2000): chemical exergy (or chemical potential), which is used up to dissolve or suspend matter, and geopotential energy, which is used up for mechanical work (Odum 2000). More details can be found in Romitelli (1997) and Odum (2000). The global average UEV of rivers (Odum et al. 2000) was estimated to as 4.0 E11 sej/m 3 from the ratio between the baseline 6 (15.83) and the total annual runoff. Therefore, this represents an average of river UEV at estuary, which was used to determine a global average transformity of river chemical potential (8.1 E4 sej/j, baseline 15.83) from average water Gibbs free energy (4.94 J/g in Odum et al. 2000), and a global average transformity of river geopotential (4.7 E4 sej/j, baseline 15.83), from average river elevation (estimated at 875m). With the same approach, Campbell (2003) updated rain UEV and transformity of chemical potential with uncertainties of global flows, determined with estimations from several authors. These global transformities of rivers were used in several EMEs, such as for example in territorial and watershed analyses (e.g. Odum et al. 1986, 1998; Campbell et al. 2005b; Campbell and Ohrt 2009; Chen and Chen 2009; Lv and Wu 2009; Campbell and Garmestani 2012), in ecosystem services evaluation (Huang et al. 2011; Watanabe and Ortega 2014), and in river engineering projects (e.g. Brown and McClanahan 1996; Kang and Park 2002; Martin 2002; Tilley and Brown 2006; Wu et al. 2013). Buenfil (2001) refined Odum s work to determine a global-average UEV for each type of freshwater reservoir and rainfall pattern, by assuming that they are all co-products of baseline-driven, global processes. However, his definition of reservoir categories is debatable: for example, groundwater was considered as a single, homogeneous category, while different mechanisms are involved in the formation of shallow and deep aquifers, which should lead to different transformities; moreover, rainfall was split into tropical and temperate categories (Buenfil 2001), whereas there are more than two climate patterns around the world. Should the author have accounted for three climate patterns, resulting UEVs would have been different (and higher). As a result, while that top-down calculation appeared useful to determine global averages, it did not depict the underlying environmental mechanisms that actually form the resources and their regional specificities. Therefore, Buenfil s approach cannot be used to calculate site-specific UEVs. 6 The baseline is defined as the total annual emergy value of the energy received by the Earth s crust system (or geobiosphere) from three independent driving forces (solar exergy, tidal energy and deep Earth heat). It represents the starting point to accoun t for the downstream emergy value of natural resources and ecosystem services. Baseline values have been differently determined by Odum (1996), Odum et al. (2000), Campbell (2001) and Brown and Ulgiati (2010) to as 9.44, 15.83, 9.26 and 15.2 E24 sej/yr, respectively. 122

123 Chapter 6 A first global and spatially-explicit emergy database of rivers and streams based on high-resolution GIS-maps A more local-oriented approach for rivers UEV calculation, based on a more explicit though simple description of environmental mechanisms, can be applied to any point in a stream or river network. It considers that the UEV of a river at a given location is the total emergy value of renewable atmospheric resources feeding the upstream catchment area, divided by the river flow. Because atmospheric resources are co-products of the same global mechanism, and due to the specific rules of emergy algebra (Brown and Herendeen 1996), only the atmospheric input with the highest emergy value is accounted for. The latter is typically rain, chemical potential or rain geopotential. This method has been widely adopted (Tilley 1999; Huang et al. 2007b; Chen et al. 2009b; Brown et al. 2010; Pulselli et al. 2011b; Arbault et al. 2013b), but some divergences in its application are worth being noticed. For example, Pulselli et al. (2011b) summed up the emergy value of both rainfall and spring water, considering that these sources are delivered by environmental mechanisms operating at different time scales. Therefore, they should be considered as independent resources instead of co-products. In contrast, Arbault et al. (2013b) noticed that in their studied watersheds, exchanges between surface water and shallow aquifers are more likely to happen, because of the rather gentle slope over the drainage basin. Hence spring water was considered as a co-product of runoff generated by precipitation, thus left unaccounted for. Brown et al. (2010) used elevation-specific transformities of rain geopotential adapted from Odum (2000). However, while the original study calculated transformities as the ratio of UEV (in sej/g) and chemical exergy (in J/g), Brown et al. (2010) applied these transformities to the geopotential energy of rainfall (instead of its chemical exergy), leading to distorted results. Similarly, Pulselli et al. (2011b) combined rain geopotential exergy with the transformity of rain chemical potential calculated by Campbell (2003). Huang et al. (2007b) also used the geopotential energy value of rain, but combining it with a global average transformity of physical energy of rain retrieved from Odum (1996). However, as they calculated rain emergy with GIS software, Huang and co-authors could have relied on the available set of elevation-specific transformities instead of using a global constant average. In contrast, Chen et al. (2009b) considered rain chemical potential as the renewable resource with the highest emergy value and, therefore, they eventually disregarded the calculation of rain geopotential emergy. Remarkably, a global, spatially-explicit database could be useful for the emergy community to benefit from the important methodological advances provided in published works and to avoid prior discrepancies in manual recalculations. Such database may also be useful for other purposes, e.g. to update the National Environmental Accounting Database (CEP 2006), to provide regionspecific background datasets in hybrid lifecycle-emergy evaluation (Rugani and Benetto 2012; Arbault et al. 2013a, 2014; Marvuglia et al. 2013a), or to develop a bottom-up baseline (Rugani and Benetto 2012; Arbault et al. 2013a). 6.2 Materials and methods General method The complete methodological procedure is outlined in Figure 6.1. It starts with the collection of physical data to elaborate emergy inputs and flow accumulation maps, which finally enabled the calculation of streams and rivers UEVs. 123

124 Physical data Solar radiation Wind speed Precipitation Temperature Elevation Emergy input maps* Sun Wind Rainfall, chemical pot. Rainfall, geopotential Flow direction Flow accumulation Chemical pot. emergy Geopotential emergy Stream flow Population Solar and wind energies do not accumulate along the drainage basin Database Evapotranspiration Highresolution UEV Regional averages Figure 6.1: Flowchart of the methodological procedure to account for the UEV of streams and rivers. * solar and wind energies are disregarded from the accumulation process along the drainage basin. The emergy value of a stream at any location was calculated as the total emergy value of renewable atmospheric resources incoming upon the upstream drainage area (Huang et al. 2007b; Chen et al. 2009b; Brown et al. 2010; Pulselli et al. 2011b; Arbault et al. 2013b), as shown in equation (6.1). ( ) ( ( )) (6.1) where Em(stream) P is the year-average emergy value of the stream at location P (sej/yr), D P is the drainage area (in km 2 ) upstream location P, and Em j (res i ) is the year-average emergy value of atmospheric input i (sun, wind, rain chemical potential and rain geopotential) on grid cell j within the drainage area D P (sej/yr/grid cell). In the present work, however, solar radiation and wind energy inputs were not accounted for. Two reasons motivated this choice: 1) lack of consistent, high-resolution and freely-available datasets of solar radiation and wind energy used up by the terrestrial mechanisms contributing to the formation of streams and drainage basins (we only found data related to the availability of these resources); 2) by first mapping available solar radiation and wind energy in each grid cell and assuming these available inputs are larger than the fraction actually used up, we showed that the emergy value of solar and wind inputs are lower than the emergy value of rain inputs for the vast majority of lands - except areas experiencing very low annual precipitations (e.g. Sahara, Northern latitudes, Santa Cruz province in Patagonia, Argentina). Therefore, apart from those regions, disregarding solar and wind inputs had virtually no influence on the stream s emergy value. Details on this methodological step are depicted in the Supporting Information SI6.1. The emergy value of a stream was thus estimated using rainfall volume, elevation (for the calculation of geopotential emergy value of rain) and temperature (for the calculation of chemical exergy value of rain as shown in equation (6.8)). Stream flow (Q P, m 3 /yr) and catchment area (D P ) were also necessary data to calculate stream UEV (sej/m 3, equation (6.2)) and the yearly average 124

125 Chapter 6 A first global and spatially-explicit emergy database of rivers and streams based on high-resolution GIS-maps emergy input per unit area of the drainage basin 7, noted Em(D P ) in equation (6.3) and expressed in sej/km 2 /yr. ( ) ( ) (6.2) ( ) ( ) (6.3) Em(stream) P and D P have been calculated using the GIS software ArcGIS 10.0 function flow accumulation. This function uses a flow direction raster, derived from an elevation map, to calculate the accumulated flow into each grid cell. A weight raster can be applied. Therefore, all Em j (res i ) raster elements were used to retrieve a Em(stream) P raster, and a raster representing the area of each grid cell was used to calculate D P. A necessary condition is the good accuracy of the flow direction raster. However, there was no worldwide, spatially-explicit dataset available for Q P. Accordingly, a raster was created to estimate this value, for each location P, by applying the flow accumulation function weighted on the estimated runoff (assumed to be equal, in each grid cell j within the drainage area, to the difference between precipitation (PP j, m 3 /yr/grid cell) and actual evapotranspiration (ET j, m 3 /yr/grid cell)), as shown in equation (6.4) and Figure 6.2. ( ) (6.4) Figure 6.2: Flow system diagram of a grid cell used in this study. For each grid cell, water flow is accumulated over the drainage basin. Emergy value of e ach resource is accounted for similarly. This procedure assumed a steady-state for freshwater reservoirs such as soil moisture, aquifers, lakes and glaciers. Incorporating the short-term variation of freshwater reservoirs would have required a much more detailed approach, supported by additional spatially-explicit datasets extremely difficult to retrieve or to model. Therefore, this aspect was left out from the scope of this study. 7 This term, quantified as sej/km 2 /yr, is similar to empower density, although in our opinion the latter terminology is more appropriate for dynamic system modeling, in which time is a variable. 125

126 These results were then compiled into a shapefile for further exploitation and analysis. This shapefile included more than 4 million vectors (each one represents a portion of stream between two confluences) with the following values: Strahler stream order (see Figure 6.4), catchment area, flow rate, emergy value of accumulated rain chemical potential, emergy value of accumulated rain geopotential, UEV, emergy value of input per unit area of drainage basin (see SI6.3). For further use in hybrid lifecycle-eme, we calculated average UEVs of river water used by human activities, per country, province (defined as the largest sub-country administrative level) and major watershed in the world (SI6.4). To this aim, population density around rivers was used as a proxy for surface water consumption. Sections specify the procedures used for each step, while technical details are enumerated in SI Preparation of raster elements from data sources Worldwide, spatially-explicit information on rainfall, elevation, evapotranspiration (ET), average temperature and population were retrieved from freely-available sources indicated in Table 6.1. However, each dataset was provided with its own specific format, and required further manipulation to obtain maps with identical geodetic reference system (WGS84), grid cell size (30 arcsec i.e. approx. 1km 2 at the equator), extent and coverage. These parameters were those of maps retrieved from worldclim.org (see Table 6.1). Noticeably, these datasets do not cover Antarctica, which was thus excluded from the present accounting. Table 6.1: Data sources and characteristics. Data Solar radiation* Wind force* Precipitation, Mean temperature, Elevation Elevation (bis) Evapotranspiration Population Watersheds shapefile Countries shapefile Provinces shapefile Spatial resolution 1 10 arcmin (~15 km) 30 arcsec (~1km) URL to dataset c/ ds Comments Monthly averaged diffuse radiation incident on a horizontal surface (kwh/month/km 2 ). Wind speed 10 m elevation, monthly averages (m/s). For precipitation and temperature, interpolation from measurement stations and available database; monthly averages (mm/month and C). 250 m Used for the flow direction map. 30 arcsec arcmin #NA #NA #NA gpw-v3-population-count/data-download tadata.show?id= ads/10m-cultural-vectors/10m-admin-1- states-provinces/ Includes only data on terrestrial vegetated land, i.e. excludes urban areas, deserts, and surface waters. Distribution adjusted with global UN data (file glp00ag). 254 major basins only. Provinces or equivalent highest subcountry administrative level. *Data related to solar radiation and wind force were not used in the calculation of streams UEV (see SI6.1). 126

127 Chapter 6 A first global and spatially-explicit emergy database of rivers and streams based on high-resolution GIS-maps Grid cell area Grid cell area is typically necessary to convert values available per km 2 (or m 2 ) into values calculated per grid cell and, then, enable flow accumulation steps. Indeed, the flow accumulation function in ArcGIS sums the value of grid cells, which have a variable area depending on the latitude (see equation (6.6)). However, such function does not exist in ArcGIS. Therefore, a specific procedure was developed: a constant raster was created, with the extent similar to the reference map (i.e. 60 S, 89 N, 180 W, 180 E). Each grid cell was assigned a value of 64, which indicates the upwards direction in a flow direction raster. Flow accumulation was then applied to this raster map, so that grid cells at the bottom row (latitude 60 S) of the resulting raster got a value of 0, whereas those at the top row (89 N) got a value of Latitude was then calculated as a linear function of grid cell value: the latitude of the center of a grid cell at the bottom row was arcsec, and that of a grid cell at the top row was arcsec, leading to equation (6.5): ( ) ( ) (6.5) where lat k is the latitude at the center of grid cell k in, Acc k is the flow accumulation value of grid cell k (between 0 and 17999), S is the grid cell size ( in a 30-arcsec raster), and lat(acc 0 ) (= ) is the latitude of the center of a bottom grid cell (which Acc value is 0). Then, the cell area A k in km 2 was calculated via equation (6.6): ( ) (6.6) where A eq is the grid cell size at the equator (0.857 km 2 ). Finally, the elevation map from worldclim.org was used as a mask to limit the geographical coverage of the resulting grid cell area raster, e.g. to filter out oceanic areas. The sum of all grid cell areas was 13.4 E7 km 2. This value was very close to the actual global terrestrial land (13.5 E7 km 2 without Antarctica), which substantiated our approach to calculate the area of a grid cell. This was an important validation step, as the procedure employed was developed manually and the resulting map was used in all following calculation procedures Rainfall, chemical potential emergy The emergy of rain chemical potential denotes the energy used up to evaporate water from sea surface, and stored in water vapor as latent heat. Transformity of rain chemical potential is typically calculated to as a global average (17,481 sej/j; Odum et al. 2000, baseline-converted), i.e. ratio of the baseline (9.26 E24 sej/yr in this study) with the annual global precipitation (1.05 E20 g/yr; Odum 1996) and Gibbs free energy of pure water (estimated at approx. 5 J/g in Odum 2000). Chemical potential energy (or chemical exergy) of rain was thus obtained by multiplying the rainfall volume (in m 3 /grid cell) with a density factor (1E6 g/m 3 ) and specific Gibbs free energy (ex rain, in J/g), which depended on temperature as shown in equation (6.7) (after Odum 1996): ( ) (6.7) 127

128 where R is the universal gas constant (8.314 J/mol/K), T is the temperature (K) of the substance (retrieved from the temperature map), w is the molar mass of water (18 g/mol), c 0 is the water concentration in rain (999,990 ppm) (Odum 1996) and c 1 is the water concentration in reference sea water (935,000 ppm) (Odum 1996). Precipitation and temperature maps were retrieved from the website worldclim.org as 30 arcsec, monthly average rasters (in mm/month and 0.1 C, respectively) of a 50-year period ( ). Precipitation maps were summed up to obtain a raster of yearly average rainfall, in mm/yr, further converted into m 3 /grid cell/yr using the grid cell area raster. Total rainfall in this work was calculated to be E14 m 3 /yr. With an additional precipitation over Antarctica estimated to 2.16 E12 m 3 /yr (166 mm/yr according to Vaughan et al. 1999, over 1.3 E7 km 2 ), our result was thus very close to the one calculated by Odum, i.e E14 m 3 /yr (Odum 1996, p42). Monthly average precipitation was multiplied by the grid cell area and ex rain, calculated for each month with the specific mean temperature of the month. Results were summed up to obtain the yearly-average rain chemical potential in each grid cell (in MJ/grid cell/yr) and finally converted in emergy value using the transformity of rain chemical potential Rainfall, geopotential emergy The UEV of rain geopotential energy was determined at given altitudes by considering the atmosphere as a hierarchically-organized system of layers (Odum 2000): the higher the water vapor, the more energy used up to elevate it, hence the higher its UEV. The use of geopotential transformity of rain was found quite misleading in the scientific literature: for example, Odum (2000) computes altitude-dependent transformity using altitude-dependent UEV of rain divided by an average value of chemical potential of water (see Odum 2000, Table 4). To avoid possible propagation of errors, we preferred using the UEV of rain geopotential instead of its transformity. Accordingly, we interpolated a continuous function of rain geopotential UEV with altitude, from Table 4 in Odum (2000), as shown in Figure 6.3. Two important remarks need to be underlined here: 1) in the original table, rainfall at sea level is assumed to be formed at 110m altitude (Romitelli 1997; Odum 2000; Brown et al. 2010); we also kept this assumption; 2) in order to keep a crescent function, Figure 6.3 s equation was applied also in the present study to all grid cells with an altitude higher than 120m, while grid cells with an elevation lower than 120m were assigned a constant UEV of 78,479 sej/m 3, which corresponds to the UEV at 120m elevation. Rainfall and elevation maps were retrieved from the website world clim.org (see Table 6.1) and, along with Figure 6.3 s equation, this enabled to directly produce the map of rain geopotential emergy. 128

129 Chapter 6 A first global and spatially-explicit emergy database of rivers and streams based on high-resolution GIS-maps Figure 6.3: Relationship between altitude and UEV of rain geopotential (from Odum 2000) Evapotranspiration (ET) ET was used for the calculation of stream flow accumulation. The retrieved ET data were made available by the Numerical Terradynamic Simulation Group, University of Montana (Table 6.1), in tiff format. Those were yearly averages given in 1 E-4 m/yr (i.e. 0.1 mm/yr) over the period with a 30 arcsec resolution. Values were provided only for terrestrial vegetated areas; other land types (urban areas, deserts, surface water bodies) were not covered. First, tiff files were converted into ArcGIS raster elements. Then, the mean value of the 11 covered years was calculated for each grid cell. The gaps in the resulting raster (due to urban areas, deserts and surface water bodies) were filled using an interpolation algorithm (details in SI6.2). It is anticipated that these areas have an evaporation rate similar to that of the surrounding vegetated areas. This is an arguable but necessary assumption, because no alternative dataset was available. Before proceeding with that assumption, two other approaches were tested to model runoff and stream flows: the first one was the use of the non-interpolated ET raster directly in flow accumulation, and the second one was the use of Turc equation (Turc 1954; Mellino et al. 2014), which is an empirical relationship that estimates ET from annual precipitation and mean temperature. However, the resulting modeled runoff and stream flow values were found farther from real data (Table 6.2) than the results obtained with the interpolated ET raster. 129

130 6.2.3 Water accumulation over drainage areas Flow direction Flow accumulation was performed using a flow direction raster weighted with the raster that contained the value to be accumulated. The flow direction map is always computed from an elevation map. However, the elevation map provided by world_clim.org provided biased flow direction, because the value of grid cells is an average elevation. For instance, water flowing through the Leman Lake, in Switzerland, was shown to flow northwards to the Rhine watershed (instead of flowing eastwards to the Rhône watershed). Therefore, a more precise map with a 250m resolution, available at the CGIAR-CSI website (Table 6.1), was aggregated into a 30 arcsec map by selecting the lowest corresponding elevation in the 250m raster. As a result, the retrieved raster was an elevation map with the lowest altitude in each grid cell. The resulting flow direction map presented fewer discrepancies Stream flow and catchment area It was assumed that water reservoirs such as glaciers, lakes, soil moisture and aquifers are at equilibrium in the long run. Therefore, the stream flow through a grid cell was calculated as the difference between accumulated rainfall over the grid cell s drainage area and the accumulated ET (see Figure 6.2). Noticeably, the accumulation function in ArcGIS calculates values upstream of a grid cell, i.e. the value of a flow entering the cell. It does not include the cell value itself. For instance, if the grid cell had a higher average altitude than the surrounding ones, then its accumulation value was 0. Therefore, to get the value of the flow leaving the grid cell, the original map should be added to the flow accumulation map. The catchment area was computed using the grid cell area as the value to be accumulated. Results were compared to the data found for 13 main rivers in the world, plus 5 in France (section 6.3.1) Emergy accumulation in streams The emergy values of rain chemical potential and geopotential were accumulated accordingly over the catchment area of each grid cell. Then, the rain input with the highest emergy value was divided by stream flow and catchment area to obtain, respectively, the stream flow UEV of each grid cell and the upstream basin emergy density Modeling of rivers In the accumulated stream flow raster, almost all grid cells had a non-zero value, meaning that all grid cells would contain a river. In order to discard cells without any stream, a cutoff value needed to be determined. To this end, a series of cutoff values were tested (1 E4, 1 E5, 1 E6 and 1 E7 m 3 /yr; see section 6.3.1). For each cutoff, a mask was generated to indicate which grid cells could contain a stream. This mask was then combined with the flow direction map, in order to calculate the Strahler stream order corresponding to each grid cell. The Strahler stream order (Strahler 1957) is a widely-used indicator on the importance of a stream, calculated as follows (Figure 6.4): when a stream has no affluent, its order is 1. When two streams of the same order n merge, the resulting stream is of order n+1. Moreover, when two streams with a different order merge, the order of the resulting stream is that of the affluent with the highest order. For instance, the Amazon River has the largest Strahler order, with a value of 130

131 Chapter 6 A first global and spatially-explicit emergy database of rivers and streams based on high-resolution GIS-maps 12 at its estuary. Nile and Mississippi Rivers have a Strahler order of 10. In Europe, Rhône and Danube Rivers have a Strahler order of 8, while Garonne, Po, Rhine, Ebro and Douro Rivers are of 7th order. Comparing modeled stream orders with the actual values of these rivers was useful to select the convenient cutoff value: comparative results (Table 6.3) showed that the most convenient cutoff value was 1 E6 m 3 /yr. Figure 6.4: Schematic representation of a stream s Strahler order ( The resulting raster of streams network was then converted to a shapefile containing vectors, each one representing the course of a river segment between two confluences. However, due to computational constraints, 1st-order streams were discarded in the resulting shapefile. The middle of each segment was then identified, to extract the corresponding value of each accumulation raster (stream order, stream flow, catchment area, accumulated emergy value of rain chemical potential and geopotential, UEV and drainage basin empower density). Finally, these values were assigned to the corresponding vector in the shapefile. This outcome (available in SI6.3) can be already consulted at this stage by practitioners seeking a specific emergy value for case studies, or willing to further analyze the global dataset (see the Conclusions section) Characterization of freshwater use at regional scale The following computational approach of region-specific average UEVs was specifically developed for use in the hybrid lifecycle-eme framework (Rugani and Benetto 2012; Arbault et al. 2013a, 2014; Marvuglia et al. 2013a), which typically underlies a user-side evaluation perspective. Indeed, the final goal is to account for the local specificities of freshwater resources used up by background processes, which location is imprecisely known. Accordingly, average UEVs were calculated for each delineated region by using weighting factors that could be representative of surface water consumption activities by the target processes ( user-side perspective). Indeed, the number of small river streams was much higher than the one of large streams. As a consequence, a plain average UEV would be more representative of the former and, as human and industrial activities tend to settle near larger streams, the use of average UEVs would not work. This rationale can be justified as follows: suppose a background process uses river water (e.g. for cooling or cleaning purposes); its location is known to be in the Seine River watershed, but is not known more precisely. It can only be reasonably assumed that this industrial process is likely to be located in the largest industrial area of the Seine River watershed, i.e. nearby Paris, France. More generally, in order to estimate the UEV of the freshwater used up by 131

132 this process, the best alternative would be to use a raster map of industrial water consumption within the Seine River watershed as a weighting factor. Because worldwide, high-resolution maps of industrial water consumption were not available, we chose population as a proxy weighting factor, as this can be reasonably considered compliant with the user-side perspective of the hybrid lifecycle-emergy framework (Arbault et al. 2013a, 2014). A 2.5 arcmin resolution (i.e. ~ 25 km 2 at the equator) map of population at year 2000 was freely-available (see Table 6.1) and used here. Figure 6.5 shows that urban areas tend to be located nearby larger rivers (screenshot on Western Europe only). In order to weight river UEV with the size of surrounding population, the UEV raster was aggregated to a 2.5 arcmin resolution, by keeping the UEV that corresponds to the highest flow rate and emergy value. This procedure was equivalent to assuming that urban activities consume surface freshwater from the largest stream present within a 5 km radius. The population raster was further processed by filtering out grid cells with no stream, which corresponds to the (debatable) assumption that populations farther than 5 km of a stream do not use surface water but groundwater. Then, the aggregated UEV raster was multiplied with the filtered population raster, summed up over the selected territory, and divided by the sum of filtered population over the territory (see SI6.2 for more details on the technical procedure). Three levels of territorial averages were ultimately computed following the same procedure (see Figure 6.10: country, province, i.e. the highest subnational administrative level, and world s major river basins). Figure 6.5: Population in France and Western Europe: urban areas tend to be located along larger streams. 6.3 Results and discussion Flow direction and runoff The flow accumulation function in ArcGIS highly depends on the quality of the flow direction raster, which was subject to a specific map processing, described in section In order to 132

133 Chapter 6 A first global and spatially-explicit emergy database of rivers and streams based on high-resolution GIS-maps verify the correctness of flow direction, the catchment area of 13 major rivers worldwide and 5 rivers in France are surveyed (Table 6.2). Out of the 18 sample rivers, 13 of them present a ratio GIS estimations vs. actual data (i.e. annual flow estimated by the GIS dataset / actual annual flow) comprised between 0.94 and 1.01: Rio Grande, Mississippi, Yangtze, Mekong, Congo, Nile, Murray, Danube, Volga, Loire, Garonne (+ Dordogne), Seine, Rhine (+ Meuse) and Rhône Rivers. The other 5 rivers show a larger difference (Amazon 0.86, St Lawrence 0.68, Amur 1.72, Yellow River 1.29, Nile 1.20), notably due to the discrepancies in the elevation model of the Himalaya and the Mongolian plateau and to the accounting of Sahara desert land as part of the Nile River watershed. It is also important to underline that the catchment area of rivers were often estimated, and different values may be found in the literature, with sometimes remarkable differences. For example, in the Major Basins shapefile introduced in Table 6.1, the catchment area (calculated using our grid cell area map) of Amazon, St Lawrence, Amur, Yellow River and Nile is respectively 6.12, 1.05, 2.02, 0.88 and 3.06 E6 km 2. Therefore, the method used for catchment area modeling is considered quite reliable. In other terms, results from flow accumulation of runoff and emergy inputs are not highly affected by errors in the flow direction raster, as rainfall is found very low in these areas. Table 6.2: Differences between modeled and actual flow and catchment area of 13 world s major rivers and 5 major French Rivers. River Catchment area, km 2 Flow, m 3 /yr Actual Modeled Ratio Actual Modeled Ratio Amazon 6.92E E E E Rio Grande 8.96E E E E Mississippi 3.23E E E E St Lawrence 1.60E E E E Amur 1.86E E E E Yellow River 7.52E E E E Yangtze 1.94E E E E Mekong 8.11E E E E Congo 3.82E E E E Nile 3.35E E E E Murray 1.07E E E E Danube 8.16E E E E Volga a 1.40E E E E Loire 1.15E E E E Garonne + Dordogne b 7.89E E E E Seine 7.87E E E E Rhine + Meuse c 1.99E E E E Rhône 9.56E E E E Notes. [a]: In the GIS model, Volga River is prolonged after its estuary, due to poor modeling of the Caspian Sea. Data were checked around the city of Astrakhan. [b]: Garonne and Dordogne rivers share the same estuary. [c]: In the GIS model, Meuse River flowed into Rhine River, and both flowed into the wrong estuary. Indeed, their common delta is highly channeled in the Netherlands. [d]: References for actual flow and catchment areas. 1: Gupta (2008); 2: Benke and Cushing (2011); 3: Wikipedia (2013); 4: Litvinov et al. (2009); 5: Giret (2012). Ref. d 133

134 In contrast, stream flows are overestimated for Rio Grande, Nile and Murray rivers, with modeled values 27, 11 and 22 times higher, respectively. Mississippi, Yellow river and Congo present other cases of overestimated flows. The 12 other sampled rivers present a ratio of modeled vs. actual flows between 0.54 and Runoff (i.e. rainfall minus ET) seems, for instance, underestimated in hot regions and overestimated in temperate regions. However, Congo, Amazon and Mekong rivers are located in regions of similar climates, whereas their ratio of modeled vs. actual flow is quite different. Therefore, it is not possible to formulate general conclusions and recommendations to refine rainfall and ET data at the present stage. It must be also considered that the available data on ET and rainfall cover different periods (precipitation: years ~ ; ET: years ). Besides, ET in urban areas and on surface waters is estimated from surrounding vegetation, which can be questionable. A flow cutoff value is implemented to assess the presence of a river in a given grid cell. Stream order is a robust characteristic of a river (Strahler 1957); in the present model, its calculation was found to be highly influenced by the choice of flow cutoff value. Table 6.3 shows the resulting stream order of a selection of rivers, as a function of cutoff value. Out of the several cutoff values tested on six rivers (Amazon, Mississippi, Ohio, Nile, Garonne and Rhône Rivers), 1 E6 m 3 /yr (i.e l/s) is observed to be a convenient cutoff value. This value is further tested on five major rivers in Europe (Danube, Douro, Ebro, Po, Rhine), confirming the choice of such cutoff. Table 6.3: Stream order of several rivers according to the cutoff value in the GIS model, as compared to actual stream order. River Modeled order using cutoff value Actual 1 E4 m 3 /yr 1 E5 m 3 /yr 1 E6 m 3 /yr 1 E7 m 3 /yr order Amazon Garonne Mississippi Nile Ohio Rhône Danube Douro Ebro Po Rhine Although Table 6.3 shows that cutoff value is a determining parameter to correctly model stream order, its influence seems not homogeneous worldwide. Indeed, the presence of a stream in a grid cell may not be only dependent on precipitation and ET. Rainfall variability and seepage from/to aquifers may also be important factors Emergy inputs Figure 6.6 shows that the emergy values of chemical potential and geopotential of rain present similar patterns. Notably, the larger differences are observed for elevated areas, such as the Himalaya Plateau, Central Andes, Rocky Mountains, Alps, Armenian Highlands, and Persian Plateau, where geopotential UEV of rain is higher than chemical potential UEV. The chemical potential emergy map is proportional to the average annual rainfall (the influence of temperature is marginal in this case). Therefore, it presents similarities with climate patterns, showing (hot and cold) desert areas, steppe and bush, and the tropical belt. Figure 6.6d indicates which emergy input is the highest, per grid cell. 134

135 Chapter 6 A first global and spatially-explicit emergy database of rivers and streams based on high-resolution GIS-maps Emergy accumulation in streams Figure 6.7 shows the UEV of modeled streams worldwide. Streams of 1st order are discarded from the display. Among the 4.16 million segments modeled, around one half of them are of 2nd order, one fourth of 3rd order, one eighth of 4th order, etc. As a result, large streams cannot be visible: Figure 6.7 map is mostly representative of 2nd order streams. It shows that the UEV of small streams is not dependent on climate (in contrast to rainfall) and is mostly comprised between 1 and 45 E11 sej/m 3 (with the exception of some regions such as Central Andes, Himalaya, African eastern coast, Brasilian Nordeste, Siberia, northern latitudes, and Western Sahara). Areas of high elevation and/or high ET tend to experience higher UEVs. Figure 6.8 is a zoom on the UEVs map of Western Europe, with focus on France. It shows that UEVs are heterogeneous within a watershed (boundaries defined by black lines in Figure 6.8) and that the flow direction map used for the accumulation of runoff does not coincide exactly with published maps of watershed boundaries, because some streams cross these boundaries. Using a more precise elevation model (e.g. with a 90m resolution) could result in less mismatch, but this was, however, beyond the scope of the present work. Besides, the inconsistencies may also originate from the river basin shapefile, these differences remaining relatively marginal. Figures 6.7 and 6.8 illustrate that the variation of UEV is higher within a watershed than between watersheds. In other words, we observe that the scale of emergy values across watersheds typically varies of, at least, two orders of magnitude (see e.g. Mississippi, Amazon, Mekong, Garonne: these rivers appear to have the same variability of UEVs within their watersheds, but rather similar UEVs at their estuary; Figure 6.7). As a result, it is not recommended to define freshwater UEV only as a watershed average, whereas the dataset developed in this study (sections and 6.3.4) should be used only to characterize background industrial processes (when its location remains approximate) in the hybrid lifecycle-eme framework. UEVs along the river course do not follow a clear trend: as shown in Figure 6.8, large rivers do not exhibit particularly high or low UEVs when compared to their tributary streams. Accordingly, one would assume that the transformity of a river gradually increases along its course, as illustrated in Odum s book for the Mississippi river basin (Odum 1996, p29). This apparent contradiction is driven by the fact that the present model does not track exergy flows and eroded materials, but only freshwater flows, which thus result not transformed or concentrated: freshwater remains the same all along its terrestrial cycle, from rainfall to estuary, and its chemical exergy does not typically change along the river course. Water is just accumulated over a drainage area. In the energy system theory, the transformity of a flow does not change when it is accumulated in a reservoir, but rather when this reservoir also concentrates the substance (Odum 1996). Therefore, to observe an energy hierarchy pattern in our stream s models, transformities should be calculated including those elements that streams typically carry, such as several forms of exergy e.g. chemical, mechanical, geopotential, thermal (Chen et al. 2009a; Martínez and Uche 2010), and transfers of sediments with potentially high emergy from highlands to lowlands. Numerous additional datasets become then necessary to calculate the total exergy of rivers. Ultimately, these different exergy forms may not have the same (donor-side) energy quality, i.e. it is important to understand the conversion processes (and related losses) between all these forms of exergy. 135

136 136 Figure 6.6: Emergy input, per grid cell (30 arcsec) and per year. a) rain, chemical potential; b) rain, geopotential; c) color scale for emergy input maps; d) highest rain emergy input, per grid cell.

137 Figure 6.7: UEVs of stream freshwater at global scale. The displayed values are representative of 2nd order streams UEV, due to their large number in comparison with larger streams and display aggregation. Chapter 6 A first global and spatially-explicit emergy database of rivers and streams based on high-resolution GIS-maps 137

138 138 Figure 6.8: UEV of streams (2nd order and higher), zoom on France and Western Europe. Black lines indicate the border of main watersheds. UEV is not homogeneous in a watershed.

139 Chapter 6 A first global and spatially-explicit emergy database of rivers and streams based on high-resolution GIS-maps Emergy value, flow rate and UEV of case studies found in literature (Chen et al. 2009b; Pulselli et al. 2011b; Arbault et al. 2013b) are compared to the modeled values (Table 6.4). Although the orders of magnitude are the same, differences are remarkable and concern both emergy accumulation and flow rates. They have several potential origins. First, although global estimations of rain are considered of good quality (see section ), local estimations may be biased by the interpolation process used by the producers of the rainfall map and the different periods covered in each case study and model. Second, precipitation maps are provided with integer values, thus presenting some truncation of monthly precipitation, which may have propagated in the calculation of annual values and then over the drainage area in the flow accumulation process, resulting in potentially remarkable errors in the calculation of stream flow and accumulated emergy. Third, the model uses a variable UEV for rain geopotential, which may have influenced the emergy value of streams that presented an elevated drainage basin. Geopotential emergy of rain is typically determined using the average elevation of the drainage basin, while in this study it is determined from the elevation of each grid cell. Therefore, geopotential emergy is calculated differently between the GIS model and the published case studies. Besides, prior calculations of stream UEV presented in Table 6.4 are also subject to discrepancies, as detailed in the introduction section. Fourth, although the modeling of drainage basins seems coherent (see section 6.3.1), the estimation of runoff as the difference between precipitation and ET seems to induce an important bias. The model should be improved with a more precise calculation of flow rate, e.g. by incorporating an infiltration factor. However, it would require additional data, such as land cover and underground rock composition, as well as a validation process from experts in hydrological modeling. Table 6.4: Comparison between modeled emergy value, flow rate and UEVs of selected sites and values retrieved from the literature (adjusted to baseline 9.26 E24 sej/yr). Data point Emergy (E20 sej/yr) Flow (E9 m 3 /yr) UEV (E11 sej/m 3 ) ref model Δ (% ) ref model Δ (% ) ref model Δ (% ) Seine (Site 1) [a] % % % Seine (Site 2) [a] % % % Vilaine [a] % % % Couesnon [a] % % % Heilongjiang (Amur) [b] % % % Liao [b] % % % Haihe-luanhe [b] % % % Yellow [b] % % % Huaihe [b] % % % Yangtze [b] % % % Pearl [b] % % % Anconella [c] % % % Pontassieve [c] % % % upper Sieve [c] % % % Sutra [c] % % % [a]: Arbault et al. (2013b); [b]: Chen et al. (2009b); [c]: Pulselli et al. (2011b)3.4.. Data from [c] were recalculated without the contribution of spring water Territorial averages Figure 6.9 shows population-weighted average UEV for provinces, countries and the major watersheds. The importance of the weighting factor (i.e. population) is best observed for countries and watersheds. In addition, Figure 6.5 displays that in some countries (e.g. France) population is 139

140 determined for each grid cell, while in some other cases (e.g. Germany) population is extrapolated from regional averages, thus hampering the quality of this weighting raster. Therefore, the average UEVs obtained shall be considered as preliminary values for research purposes, to be further refined. For instance, these records may be used for implementation in the framework of hybrid lifecycle-eme (Rugani and Benetto 2012; Arbault et al. 2013a), applying regionalized UEVs to the characterization of surface water inventories at the scale of background processes of product, or for the estimation of emergy imports and exports in the NEAD database (CEP 2006). Tables corresponding to Figure 6.9a-c are available in SI Bottom-up emergy accounting It is worth mentioning that another important issue, which has been found unsolved in prior studies, is the adaptation of emergy accounting to a bottom-up framework. Prior research on determining rivers UEV adopted a top-down approach, by roughly considering atmospheric resources (solar radiation, wind, rain chemical potential and rain geopotential) as co-products of the same natural processes. Moreover, it is presumable that a large portion of solar radiation and wind energy are used up by atmospheric processes, whose contribution is already accounted for (theoretically) in the UEV of rain. In contrast, only a tiny fraction of those resources is used up by terrestrial processes that contribute to the formation of streams and drainage areas, which should be added up to the emergy value of rainfall feeding the stream. In the longer term, such methodological approach could be greatly enhanced by adopting a bottom-up rationale. To this aim, we suggest incorporating a spatially-explicit description of atmospheric processes and of the whole water cycle, including water balance between surface and Earth crust (Williams 2007). Finally, the accounting rationale adopted in our study, illustrated by equation (6.1), is to consider that the emergy value of a stream corresponds to the highest emergy value of rainfall inputs (chemical potential or geopotential) over the whole catchment area. An alternative rationale, proposed by Romitelli (1997), is to consider that the emergy value of a stream in a given river section is equal to the emergy value of the river upstream this section, plus the highest emergy input over the section. Such rationale corresponds to equation (6.8), which clearly differs from equation (6.1): ( ) ( ( ( ))) (6.8) Equation (6.1) is preferred in this study because it better corresponds to the conventional approach found in literature to calculate rivers UEV, enabling a cross-check with our results (Table 6.4). Hopefully, the instances provided with the present work will bring useful elements to feed the current discussions within the community of emergy scholars, and contribute to addressing weak points of the emergy calculation procedures. 140

141 Figure 6.9: Population-weighted average UEV. a) per province; b) per country; c) in major watersheds. d) per 2.5 arcmin grid-cell population. Chapter 6 A first global and spatially-explicit emergy database of rivers and streams based on high-resolution GIS-maps 141

142 6.4 Conclusion This work attempted to provide a worldwide high-resolution (30 arcsec) database for the emergybased characterization of rivers, which is made available in SI6.3. It resulted in a shapefile composed of more than 4 million stream segments (i.e. portions of rivers between two confluences), each of them being characterized with the following modeled values: stream order, annual flow rate, catchment area, emergy values of rain (chemical potential and geopotential) accumulated over the drainage area, UEV and the emergy density of the drainage basin. For computational reasons, streams of 1st Strahler order were discarded from the outputs. UEV were then averaged over countries, provinces (i.e. sub-country highest administration level), and major watersheds, in order to produce datasets to be used in the hybrid lifecycle-eme framework, which typically adopts a user-side perspective (Rugani and Benetto 2012; Arbault et al. 2013a). In effect, resulting UEVs shall only be used to account for surface freshwater use in activities located in urban areas, e.g. industries and potable water production, and not for other purposes, e.g. to characterize surface freshwater use for irrigation. This approach, however, could be applied to natural systems and agricultural activities by selecting another weighting factor, such as for example river capacity and repartition of irrigated cropland, respectively. In addition, the proposed method allows generating new datasets with regional averages for any territory. The comparison with existing case studies of rivers local UEV calculation showed that a refinement of the present, prospective work is recommendable. The main limitations are worth being recalled: i. Spring water sources were not accounted for, because groundwater bodies were considered as co-products of global processes composing the emergy baseline, in a longterm perspective (e.g. precipitation data are 50-year averages). Further investigation and modeling is thus necessary to determine to which extent they are actually co-products or splits. ii. Solar and wind emergy were discarded from this study. Although this methodological choice deviates from the conventional emergy algebra rationale (Brown and Herendeen 1996; Odum 1996), it was motivated by 1) the lack of useable, high-resolution datasets, and 2) the relative low emergy value of wind and solar energy, as compared to rainfall (see SI6.1). iii. The temporal dynamics of freshwater reservoirs was excluded from the scope of this study. Although this aspect may considerably influence the stream flow (and hence the time-specific) UEV of rivers, such calculation approach was extremely time-consuming for this prospective work. iv. Modeled stream flow rates from precipitation and evapotranspiration showed important discrepancies only when compared to actual data. Therefore, other mechanisms need to be considered, including the modeling of sediments, minerals and particulate matter transportation by the streams, calling for more expertise and knowledge in hydrology. The rationale used in this work is a compromise between data availability, model complexity and time constraints. v. The use of GIS questions the reliance on a top-down approach for emergy accounting, which prevents from using a detailed, spatially-explicit description of environmental mechanisms, e.g. atmospheric processes and the whole water cycle. Such an approach could open the path for an integrated, bottom-up emergy evaluation of renewable flows from natural processes, in order to refine current top-down approaches (Rugani and 142

143 Chapter 6 A first global and spatially-explicit emergy database of rivers and streams based on high-resolution GIS-maps Benetto 2012, Neri et al. 2014). A detailed description of mechanisms driving the water cycle would more logically deliver a single, spatially-explicit UEV for rainfall. Such a bottom-up, regionalized baseline could also help determine the local transformities of natural systems (following Brown and Bardi 2001) and therefore of ecosystem goods and services. Despite these limitations, the proposed methodological approach presents the advantage of being homogeneously applied worldwide, as well as fully reproducible. Although the method and underlying assumptions can be improved, this work should be seen as a starting point to develop a reference emergy map of freshwater and atmospheric resources for the emergy analysts. The priority for the next steps is to identify good-quality data for solar energy available at the surface, for freshwater reservoirs such as snow cover (Campbell and Ohrt 2009), lakes and aquifers (Buenfil 2001; Brown et al. 2010; Pulselli et al. 2011b) and glaciers, and for high-resolution freshwater consumption. The calculation of the transformity of stream chemical potential would require a high-resolution dataset (or modeling) of dissolved and suspended solids. Priority improvements of the model include the calculation of rain geopotential emergy and the modeling of soil storage and infiltration to groundwater. The intervention of experts in hydrological modeling is thus highly recommended. Finally, the emergy results normalized for population could be compared in the future to water stress values (Pfister et al. 2009; Boulay et al. 2011c), to see whether a numerical correlation exists between UEV and water stress. For example, if a strong correlation is demonstrated, then the UEV of rivers could be used as a proxy for water quality assessment. Acknowledgments This project was supported by the National Research Fund, Luxembourg (Ref ) and the French National Research Fund (project EVALEAU ANR-08-ECOT C0238). The authors thank the two anonymous reviewers, who greatly contributed to improve the quality of this paper. Supporting information SI6.1 provides information on solar radiation and wind energy maps; SI6.2 contains all detailed procedures used for maps processing in ArcGIS 10.0; SI6.3 includes a freely-available shapefile of the river emergy database; and SI6.4 includes the shapefile and tables related to the computed territorial average emergy values. 143

144

145 7. Integrated earth system dynamic modeling for life cycle impact assessment of ecosystem services Published as Arbault, D., Rivière, M., Rugani, B., Benetto, E., Tiruta-Barna, L., Integrated earth system dynamic modeling for life cycle impact assessment of ecosystem services. Sci. Tot. Environ. 472, Graphical abstract Abstract Despite the increasing awareness of our dependence on Ecosystem Services (ES), Life Cycle Impact Assessment (LCIA) does not explicitly and fully assess the damages caused by human activities on ES generation. Recent improvements in LCIA focus on specific cause effect chains, mainly related to land use changes, leading to Characterization Factors (CFs) at the midpoint assessment level. However, despite the complexity and temporal dynamics of ES, current LCIA approaches consider the environmental mechanisms underneath ES to be independent from each other and devoid of dynamic character, leading to constant CFs whose representativeness is debatable. This paper takes a step forward and is aimed at demonstrating the feasibility of using an integrated earth system dynamic modeling perspective to retrieve time- and scenariodependent CFs that consider the complex interlinkages between natural processes delivering ES. The GUMBO (Global Unified Metamodel of the Biosphere) model is used to quantify changes in ES production in physical terms leading to midpoint CFs and changes in human welfare indicators, which are considered here as endpoint CFs. The interpretation of the obtained results highlights the key methodological challenges to be solved to consider this approach as a robust alternative to the mainstream rationale currently adopted in LCIA. Further research should focus on increasing the granularity of environmental interventions in the modeling tools to match current standards in LCA and on adapting the conceptual approach to a spatially-explicit integrated model.

146

147 Chapter 7 Integrated earth system dynamic modeling for life cycle impact assessment of ecosystem services 7.1 Introduction and goals Ecosystem Services (ES) result from ecosystem functions (Costanza et al. 1997; Daily 1997; Burkhard et al. 2012; de Groot et al. 2012), which are the capacity of natural processes and components to provide goods and services that satisfy human needs, directly or indirectly (de Groot et al. 2002). Over the last 15 years, scientific studies flourished in the economic and biophysical valuation of ES (e.g. Gómez-Baggethun et al. 2010; TEEB 2010). In particular, the Millennium Ecosystem Assessment (MEA) classified four categories of ES (MEA 2005b): provisioning, regulating, cultural and supporting services. The MEA has represented the consensual umbrella for all the ES valuation approaches developed afterwards. For example, The Economics of Ecosystems and Biodiversity (TEEB) approach, which is one of the most recommended frameworks to target ES and pursuing their benefits, especially at the country scale, incorporates many of the concepts, classification schemes and criteria developed by MEA (TEEB 2010). However, valuing the contribution of ES to human welfare demands robust methods to define and quantify ES (Crossman et al. 2013), especially if the accounting perspective aims to address both economic, environmental and social (triple-bottom line) aspects (de Groot et al. 2010; Cardinale et al. 2012; Haines-Young et al. 2012; Crossman et al. 2013; Maes et al. 2013). Focusing on the environmental dimension, the growing interest for ES valuation has permeated the larger environmental assessment field, and namely Life Cycle Assessment (LCA), which is a widely accepted methodology to evaluate the environmental impacts of a product or service throughout its life cycle (ISO 2006). Within LCA, the Life Cycle Impact Assessment (LCIA) step translates the elementary flows (resources consumed and pollutants emitted) into environmental impacts, which are either problem-oriented (midpoint approach) or damage-oriented (endpoint approach) (European Commission 2010c). To this aim, so-called characterization factors (CFs) are developed using impact assessment models, reflecting the values associated with three main Areas of Protection (AoP): Human Health (HH), Natural Resources (NR), and Natural Environment (NE). Whereas there is scientific consensus on the scope of HH, the evaluation of NR and NE remains debatable, because of the intrinsic cross-linkages between the two areas (European Commission 2010b; de Baan et al. 2013). The AoP of NR should cover the damage associated with the exploitation of natural resources, which can affect the delivering of ES. However, LCIA indicators (and related CFs) have essentially been developed with regard to the usefulness of natural resources for human purposes (see e.g. European Commission 2010b, for a comprehensive list and analysis). Most of the indicators focus on the assessment of mineral and fossil resource scarcity, by evaluating the future marginal cost of extraction/use of these resources. With regard to the AoP of NE, the aim is to quantify the negative effects on the function and structure of natural ecosystems as a consequence of the exposure to chemicals or other physical interventions (European Commission 2010b). Recent researches have addressed the link among elementary flows of Life Cycle Inventory (LCI) (mainly land occupation and land transformation) and novel LCIA midpoint impact categories called potential damage on Ecosystem Services (Koellner et al. 2013a). Accordingly, spatially differentiated CFs were developed to assess potentials of e.g. biodiversity damage (de Baan et al. 2013), climate regulation (Müller-Wenk and Brandão 2010), biotic production (Brandão and Milà i Canals 2013), erosion and freshwater regulation and water purification (Saad et al. 2013), and water supply from groundwater (Van Zelm et al. 2011). The assessment of functional diversity within the different taxonomic groups of mammals, birds and plants was recently proposed as a complement to the assessment of species richness (de Souza et al. 2013). While highlighting the 147

148 lack of completeness of existing LCI databases, one of the main objectives reached is the definition, harmonization and ranking of a large set of land use and land use change elementary flows created as a common inventory database for both global and local (regionalized) assessments of ES, which can be directly used in the LCIA practice. Other research streams have tried to integrate LCA with the emergy concept and method (Odum 1996), providing an explicit LCIA of ES that underlies a pure ecological orientation (Marvuglia et al. 2013a; Rugani et al. 2013). However, this approach is not fully operational yet because of some computational and system boundary constraints associated with the combination of the two methods, further hampered by a lack of consensus on emergy in the LCA community (Arbault et al. 2013a; Raugei et al. 2014). The insufficient coverage of ES in the current LCIA practice (Zhang et al. 2010b; Curran et al. 2011; de Baan et al. 2013) is hampering the consistent application of LCA to a number of sectors which are very concerned by ES, e.g. agriculture. Despite the significant breakthrough of the recently developed methods, four aspects were identified deserving further attention. First, cause effect chains originating from land occupation and transformation are modeled independently (Koellner et al. 2013a), i.e. without considering the interconnections between the mechanisms of natural processes. It is widely recognized that natural processes influence each other in many complex and indirect ways and that indirect effects can be delayed in time and widespread over the globe (Folke et al. 2011). This is therefore a significant simplification, as for instance an increased terrestrial acidification is likely to alter the biological properties of soil, thus changing local biodiversity, which in turn may change the sensitivity of ecosystems to toxic substances. CFs should thus not be constant, but time dependent as a function of all emitted inventoried substances over time. Second, environmental mechanisms are investigated up to the midpoint level, whose impacts are expressed in physical units (Koellner et al. 2013a). However, ES is a user-oriented concept, which has been developed to assess the benefits that human societies yield from nature. Although the health of ecosystems can be measured in physical terms, benefits are commonly expressed in terms closer to human values, such as economic development or contribution to welfare (Costanza et al. 1997; de Groot et al. 2002; MEA 2005a; TEEB 2010). As a result, the investigation should include also endpoint targets. Third, the potential damages in LCIA are usually assessed by adopting a marginal and short time perspective (Goedkoop et al. 2009). The focus is therefore on how the current state of ecosystems would be altered, in the short term, by a perturbation, usually defined at local scale and in a simplistic manner, because of the granularity of LCI databases. An example could be the assessment of the effects of occupying a small piece of land on the local species diversity during a short period. As a result, only marginal effects occurring right after the perturbation are included. Such an approach misses the holistic perspective, in which environmental mechanisms are influencing each other at various spatial scales and with both short-term and long-term effects. A more comprehensive analysis should therefore include, for instance, the effects on global climate change to local water regulation and soil erosion during the next decades. Fourth, nature and mankind interact in many complex ways. Human-driven systems use natural resources and generate waste and emissions that can affect the ecosystems. The production of ES occurs at a limited pace. Over-exploitation of renewable resources may lead to a complete 148

149 Chapter 7 Integrated earth system dynamic modeling for life cycle impact assessment of ecosystem services collapse of the local ecosystem. In turn, a degradation of the natural environment may challenge human welfare, so that in order to sustain our living standards we may need to extract more renewable resources, leading to even higher degradation of the environment. Such vicious circle already occurred in the past (Diamond 2006) and still happens at present time (Steffen et al. 2007; Folke 2010). On the contrary, when the benefits of preserving this natural capital are evaluated against the long-term costs of destroying it, the trend may change. While nowadays there is not enough empirical evidence that comparing benefits of preserving natural capital (in monetary terms) against the long-term costs of destroying it (also in monetary terms) may necessarily lead to a virtuous circle, the precautionary principle of preservation underlies the ES concept and the rationale behind their valuation in economic terms (TEEB 2010). Therefore, understanding the complex feedback between the natural capital and human development is crucial for decisionmaking support. Considering these four aspects altogether could possibly lead to introduce a dynamic perspective to better understand and model the interrelationships (i.e., feedback flows) between or among subsystems over time, allowing to evaluate how specific overall system behaviors may be generated (Halog and Manik 2011). CFs for ES could therefore be time-dependent, because of the time dependency of the inventoried substances and of feedback flows. In contrast, the approaches adopted by current LCIA models (Curran et al. 2011; Saad et al. 2011, 2013; Van Zelm et al. 2011; de Baan et al. 2013; de Souza et al. 2013; Koellner et al. 2013b) essentially lead to constant CFs, which barely take these dynamic interactions into account. The use of purely ecological models can provide a way forward to tackle some of these aspects, but only a complex and dynamic model integrating the anthroposphere within the natural sphere, from local to global scales, would be able to encompass all of the four mentioned aspects. In this paper, the feasibility of an alternative approach to account for damages on ES using integrated earth system dynamic modeling is studied in order to comprehensively tackle the aforementioned limitations. GUMBO (Global Unified Metamodel of the Biosphere, Boumans et al. 2002) model was selected, which simulates context-dependent, cause effect chains between the natural sphere and the human sphere (anthroposphere). In Section the rationale for the choice of GUMBO is presented. Then the approach to define midpoints and endpoints is developed in Section 7.2.2, and the CFs calculation procedure is detailed in Section Because the paper focuses on the feasibility of the conceptual approach, the results presented in Section 7.3 mainly aim at illustrating the interconnectedness between all mechanisms included in GUMBO and how the model allows interpreting the associated consequences on CFs. The case of an additional fossil fuel extraction in the year 2000 and the associated impacts on water use and Gross World Product (GWP, as defined in Boumans et al. 2002) over the whole century is taken as example in Section 7.3.1, because of the global interest for these resources which are conventionally considered in the assessments as completely independent and separated. Section introduces all CFs that could have been retrieved from the simulations, in order to highlight the extent of these mechanisms and the practical differences and similarities with current LCA datasets. Alternative simulations (Section 7.3.3) show the higher flexibility of using dynamic and integrated modeling for LCIA than using constant CFs. Opportunities and threats posed by such integrated approach to ES valuation are finally discussed in Section 7.4, along with the priorities towards making the approach more practical (Section 7.4.1) and the added value of using integrated models within LCIA (Section 7.4.2). Section discusses additional requirements for integrated models to match LCIA standards. A roadmap is proposed accordingly (Section 7.4.4). 149

150 7.2 Materials and methods Choice of an integrated earth system dynamic model: GUMBO The choice of a pertinent integrated dynamic model is not trivial since most of the existing models are ecological oriented, i.e. focus mainly on natural mechanisms (Halfon et al. 1996; Jiang et al. 1999; Ito and Oikawa 2002). Indeed, these models have also been used as decision-making or forecasting tools to provide insights on the consequences of human interventions such as management practices or climate change (Portela and Rademacher 2001; Ainsworth et al. 2008; Ooba et al. 2010; Aitkenhead et al. 2011; Johnston et al. 2011; Shanin et al. 2011; Taner et al. 2011; Shang et al. 2012). Rather independently from the development of these (mainly biophysical) models, the evaluation of ES has been undertaken in monetary terms (TEEB 2010) through the use of different concepts, among which avoided, replacement or travel costs, hedonic pricing, factor income, contingent or group valuation, and marginal product estimation (Waage et al. 2008). A general overview of techniques and methods is provided by Farber et al. (2006). These approaches have been diversely combined to physical datasets and implemented within numerous ecological models and multi-ecosystem services assessment tools, such as ARIES (Bagstad et al. 2011), or InVEST (Tallis et al. 2013), whose relative strengths and weaknesses have been deeply investigated by Waage et al. (2008). A common aspect underlying those models is the implementation of land use and/or land use change patterns, which remains a prevailing challenge to valuate ES. Land use information is also widely implemented in several equilibrium models through the link between sectorial (partial equilibrium models) or national/multi-regional (computable general equilibrium models) economic systems and biophysical systems at more or less aggregated spatial scales 8. The usefulness of these tools resides in their ability to respond to specific regional or global policies (e.g. for agriculture, energy sector, etc.) and evaluate the direct and indirect effects on the socio-economic context, including the economic or environmental impacts to or from marginal sectors of the economy. Nevertheless, none of these models fully consider the complex interdependencies and feedback flows among ecosystems and socioeconomic systems at large scale. Based on this review, a model designed at the global scale was researched, which could balance the level of details between natural and human capitals and integrate macroeconomics within the global ecology. The selected model, GUMBO, includes dynamic feedback between the geobiosphere and the anthroposphere at the global scale in an integrated earth system (Figure 7.1) and presents numerous advantages for the approach investigated in this paper. Following the rationale used in Costanza et al. (1997) for the valuation of ES, GUMBO monitors the evolution of ES in relevant physical units and relates them to the evolution of macroindicators of GWP and human welfare (Sustainable Social Welfare, SSW), using Cobb Douglas functions: 8 With reference to models like: CAPRI: Common Agricultural Policy Regionalised Impact modeling system; last access October Available at: org; GLOBIOM: A global model to assess competition for land use between agriculture, bioenergy, and forestry; last access October Available at: web/home/research/modelsdata/globiom/globiom.en.html; GTAP: Global Trade Analysis Project; last access October Available at: LUMOCAP: Dynamic land use change modeling for CAP impact assessment on the rural landscape; last access October Available at: 150

151 Chapter 7 Integrated earth system dynamic modeling for life cycle impact assessment of ecosystem services Figure 7.1: Basic structure of GUMBO. The hydrosphere, lithosphere, and biosphere are reproduced for each global biome (Boumans et al. 2002). (7.1) (7.2) In Eqs. (7.1) and (7.2), {x i } and {x j } are the production factors that influence GWP and SSW, respectively. {α i } and {β j } are the constant outputs elasticities; c and c are constants which include global parameters, such as the total factor productivity and the production reduction factor. {α i }, {β j }, c and c are determined by calibrating GUMBO with data from year 1900 to Table 7.1 displays the 15 factors used in the calculation of GWP and their respective elasticity in the corresponding Cobb Douglas function. Concerning SSW, the number of production factors and their elasticity coefficients are different and more elaborated. As compared to GWP, the calculation of SSW does not include e.g. provisioning services, whereas it includes human factors such as mortality and consumption. Besides, {β j } is dependent on the policy scenario selected (see Section 7.3.3). A more complete description is outside the scope of this paper. The reader is invited to refer to Boumans et al. (2002) for more details. The Cobb Douglas function is based on the empirical evidence observed in industrialized countries (Douglas 1976), that GDP growth is proportional to the growth of two production factors, labor and capital, with coefficients of proportionality constant over time. The Cobb Douglas function was extended to include other production factors such as natural capital (Solow 1974; Stiglitz 1974) or more precisely net primary production (Richmond et al. 2007). It is also used in microeconomic case studies of natural resource extraction (Barbier 2007). For example, Markandya and Pedroso-Galinato (2007) list the studies in which different forms of natural capital or resources are production factors of the Cobb Douglas function. Ecological models usually translate the value of ES by using microeconomic functions that balance local supply and demand, or the willingness-to-pay approach, in opposition to the use of Cobb Douglas function. Willingness-to-pay is, however, not suited to highlight the two following facts: 1) long-term, global interactions between nature and human activities are influenced by present policy decisions, and 2) the anthroposphere relies on ES for its functioning. On the contrary, the Cobb Douglas function can take these aspects into account: when one of the x i 151

152 factors increases by 1%, then GWP increases by α i % (< 1%); its marginal productivity decreases and the marginal productivity of other factors increases. This function considers that all inputs are substitutable to each other, provided that all inputs remain strictly positive. Table 7.1: The 4 ecosystem goods, 7 ecosystem services and 4 human capitals (i.e. production factors) included in GUMBO. They are displayed with their elasticity coefficients in the Cobb-Douglas function for Gross World Product (GWP) (after Boumans et al. 2002). In this function, ES (including both ecosystem goods and services) are considered necessary inputs for the functioning of the anthroposphere. Ecosystem Goods Ecosystem Services Human Capitals Production Factors x i Elasticity β i Energy 0.40 Organic Matter 0.07 Ore production 0.05 Water Use 0.03 Gas Regulation 0.05 Climate Regulation 0.05 Disturbance Regulation 0.05 Soil Formation 0.10 Nutrient Cycling 0.01 Waste Treatment 0.10 Recreational & cultural* 0.02 Built Capital 0.10 Knowledge 0.30 Social Network 0.03 Labor Force 0.30 * These services hinge on an ecosystem s ability to provide for recreational activities such as e co-tourism and sport fishing as well as cultural activities like worship and aesthetic appreciation (Costanza et al., 1997; in: Boumans et al., 2002). In the case of GUMBO, ES are substitutable by each other for the production of GWP, but they also can be substituted by human capitals. The model seems to adopt the viewpoint of weak sustainability. Unfortunately, the Cobb Douglas function does not assume limits on substitutability between production factors (Georgescu-Roegen 1971) and this may be ultimately a limitation of the model. Nevertheless, GUMBO still respects the notion of strong sustainability: while more built capital, social capital or human capital can substitute for less natural capital in the production of GWP or SSW at the margin in the production function, these capitals themselves cannot be produced without natural capital (Boumans et al. 2002). This means that the natural capital is modeled in GUMBO as being substitutable by the other capitals up to a certain threshold over which it starts losing the ability to regenerate, showing a spontaneous decline (up to falling to 0) regardless of the level of other (substitution) inputs. Thus, this approach indeed fulfills the principles of strong sustainability (Boumans et al. 2002). The GUMBO model and its documentation are freely available ( and the model runs on conventional PC using the STELLA platform. The graphical user interface allows effortless retrieval of intermediary calculations, as illustrated in the results section, and easy tuning by non-expert users. Another advantage of GUMBO is that it is a simplified metamodel of the earth system, including parameters and variables, allowing a transparent interpretation of test results. The components of the model could, however, be replaced by more detailed ecological models for the natural sphere and socioeconomic models for the anthroposphere. It is worth mentioning, for the sake of clarity, that the nomenclature used during GUMBO's development ( Ecosystem Goods and Services ) was replaced in this paper by 152

153 Chapter 7 Integrated earth system dynamic modeling for life cycle impact assessment of ecosystem services ES, to be aligned with the most consensual nomenclature as defined in MEA (2005b) and TEEB (2010). Also, it must be stressed again that the aim of this study is not to retrieve ready-to-use CFs to be implemented in existing LCIA methods but, as a first step, to test the feasibility of the conceptual approach of using integrated earth system dynamic modeling for LCIA Definition of midpoints and endpoints in GUMBO In the proposed approach, LCI results are seen as marginal changes at the interface between the geobiosphere and the anthroposphere. The response of GUMBO to these marginal changes provides midpoint impacts called change in [ES production], expressed in relevant physical units. Changes in GWP and human welfare are considered as endpoint impacts. Therefore, CFs are retrieved by relating the magnitude of output changes with that of the initial perturbation (at different time horizons) that mimics LCI results. The midpoint impacts as defined above are in compliance with the conventional definition adopted in LCA, i.e. a point positioned half-way along the environmental mechanism linking (man-made) interventions to the AoP (Goedkoop et al. 2009). In contrast, the endpoint impacts represent a departure from the conventional definition in LCIA. In GUMBO, the components that model human health, natural resources and the natural environment are not perceived as endpoints to be preserved per se. Instead, they are intermediate contributors of welfare in the Cobb Douglas functions, presenting dynamic feedback between each other. GWP and human welfare are at the end of the environmental mechanism in the model's valuation system. Despite this departure, the term endpoint was maintained to help the reader figuring out the parallel between valuation systems of conventional LCIA models and GUMBO. In addition to midpoints related to environmental impacts, GUMBO similarly considers impacts of human interventions on four types of human capitals, as depicted in Table 1. They were also considered as midpoints in the present approach, in accordance with previous LCIA frameworks that defined the human-made environment as an important impact category (Jolliet et al. 2003a; Margni et al. 2008) Tuning GUMBO to retrieve CFs for ES In order to tune GUMBO to relate the LCI result with changes in ES production, a four-step procedure was developed, as summarized in Figure 7.2. As a first step, a perturbation function on the resource extraction flow, represented by an artificial Dirac impulse occurring at year 2000, was considered. This perturbation represents an additional amount of resource extracted (see Supplementary Information, hereafter SI, Figure. S7.1). To be consistent with the conventional approach adopted in LCIA, the additional amount is marginal as compared to actual flows, i.e. in the specific case the global extraction rates. Typically, a perturbation ranging from 0.1% to 0.9% of the value of the corresponding global flow or stock in year 2000 provides a response to all midpoints and endpoints that is proportional to the perturbation (exceptions are due to the computational limits of the platform STELLA ; see SI, section SI7.3). This assures that the changes observed in the model's responses are not due to external noise, i.e. that the model reacts to the perturbation only. As a result, simulations provide a proportional relationship between inputs (the marginal change of resource extraction in year 2000) and outputs (the so-called midpoints and endpoints, as defined in Section 7.2.3). 153

154 Figure 7.2: The four-step procedure to relate LCI data with GUMBO's response. GWP = Gross World Product; SSW = Sustainable Social Welfare. The outputs from the simulation, i.e. changes of the production factors (ES and human capitals, hereafter named midpoints), were compared to the baseline scenario, without perturbation (see Section SI7.1). In GUMBO, these changes have repercussions over the whole observation period (Figure 7.2, step 3 and Fig. S7.2 in SI). For each midpoint, the yearly net changes were cumulated over the period and then divided by the magnitude of the perturbation as follows: ( ) (7.3) CF i,p is the characterization factor on midpoint impact i due to the perturbation p. C * i,t is the physical value of midpoint i at year t (between 2000 and 2100) in the simulation where perturbation p was implemented, and C 0 i,t is the physical value of midpoint i at year t in the baseline simulation. C * i,t and C 0 i,t are expressed in the physical unit corresponding to midpoint i (e.g. teraliters, TL, for the ES water use). p is expressed in the physical unit of the perturbation (e.g. kg of carbon-equivalents, kgc, for the perturbation on fossil fuel extraction). Therefore, CF i,p is expressed as an intensity factor between two physical units (e.g. TL/kgC for the cumulated change of global water use due to a marginal change in fossil fuel extraction at year 2000). Table 7.2 indicates the units of each midpoint impact and its definition. They are formulated with GUMBO's original terminology, for the sake of transparency in our approach, although these terms should be refined in accordance with the latest terminology of MEA. 154

155 Chapter 7 Integrated earth system dynamic modeling for life cycle impact assessment of ecosystem services Table 7.2: Description of midpoint impacts and their physical unit as found in GUMBO s outputs. Midpoints are equivalent to the production factors of Table 7.1. Ecosystem Goods Ecosystem Services Human Capitals Midpoint Unit Description of impact indicator*** Water Use TL water Water use Ore production Gt ore Ore production Organic Matter Gt biomass Net harvest of autotrophs and consumers Energy Gt (fossil fuel)* Fossil fuel extraction and renewable energy use Gas Regulation kg/kg Bioavailable C per kg Gross Primary production (GPP) Climate Magnitude of temperature regulation per Gt C/Gt biomass Regulation biomass Disturbance Biomass for disturbance regulation (water Gt biomass Regulation regulation and erosion control) Soil Formation Gt DOM Dead organic matter (DOM) for soil formation Nutrient Cycling Gt nutrients in OM Nutrient cycling Waste Treatment Gt Waste assimilation capacity Combined marginal change of biomass Recreational & Unit* production minus marginal change of Social Cultural Capital Index Built Capital Gt Built material Knowledge Trillion US$ Knowledge capital Social Network Social Capital Index** Social Capital Index Labor Force Bln workers Number of workers * Energy is the sum of fossil fuel extraction (defined as the minimum between extractible fossil fuel and nonrenewable energy demand) and of renewable energy use (defined as the minimum between available renewable energy and alternative energy demand). Renewable energy use is expressed in Gt of fossil fuel, with is somewhat arbitrary and opaque unit conversion (c.f. formulation in GUMBO); ** unit (Recreation Cultural) = ln (Gt biomass / SCI); *** SCI is a normalized ratio between the value creation from social capital formation (Trillion US$ ) by level of conformation induced by the creation of rules and norms (c.f. formulation in GUMBO); **** Cumulated change unless otherwise stated. For perturbation ranges higher than 1%, not all responses follow a linear relationship, which indicates that internal feedback flows of the model become important and discredit constant CFs. A perturbation higher than 1% could then be considered non marginal. For example, fossil fuel extraction worldwide in year 2000 was 7.25 GtC and the marginal additional extraction ranged between 7.25 and MtC: linearity was still respected at that range. Endpoints CFs were computed following the same procedure, exception given for GWP in which annual change was weighted by a depreciation factor (Eqs. (7.4) and (7.5)), i.e. the Net Present Value (NPV) of a monetary unit of year t at present time (year 2000). NPV has already been used in dynamic LCIA (Levasseur et al. 2010). ( ) (7.4) with ( ) ( ) (7.5) The term d is the discount rate, which is constant over time: a high value of d means that shortterm economic consequences are considered more important than long-term ones. Oppositely, if d is set to 0, economic consequences for each year between 2000 and 2100 have equal importance: in this case more emphasis is put on future generations. The value of d has therefore a critical influence on the endpoint assessment. It reflects the priorities set by decision makers as 155

156 beneficiaries of the LCIA. The endpoint CF related to the human welfare (SSW) was computed without discount rate, assuming that all human beings have equal importance, regardless in which year they live. 7.3 Results Example of midpoint and endpoint CFs calculation and interpretation for a specific perturbation and impact The case of an additional fossil fuel extraction in the year 2000 and the associated impacts on water use and GWP over the whole century is taken as example. As shown in Figure 7.3a, additional fossil fuel extraction in year 2000 results in additional water use the year after. Because of the indirect effects on other ES, water use is shown to drop dramatically later in the next century. The total cumulated change, expressed in TL (teraliters), is observed to be negative. The simulations show a linear relationship between the magnitude of the input perturbation and the changes in GUMBO's outputs (Figure 7.3b). Therefore, a CF can be retrieved for the midpoint impact, worth 0.07 L water per kg of fossil fuel. In other terms, each additional kg of fossil fuel extracted in year 2000 is expected to decrease the global, 21st-century water use by 0.07 L. It is, however, not straightforward to understand a priori whether this is a beneficial or detrimental consequence from a LCIA perspective, i.e. relative to the AoP (Areas of Protection). Water use depends on both supply (freshwater made available by the natural system) and demand (human needs for freshwater). A reduction in water use could reflect a lower performance of ecosystems in delivering freshwater, but it could also express a lower demand by human societies, due to e.g. a more efficient resource management or a reduction of global population. However, a deeper analysis of the data generated by GUMBO showed that the simulated change in water demand (generated by an additional marginal extraction of fossil fuel in year 2000) follows the same pattern as the change in water use, while the change in available water for each biome follows completely different curves. This indicates a demand-limited use, while the actual amount of available water has merely no influence on water use. In addition, in GUMBO, the modeling of water demand does not include an efficiency parameter: it is proportional to the sum of each capital's value (the four human capitals displayed in Table 7.1 plus the natural capital, based on the value of all ES) weighted by constant water need rates defined for each couple capital-biome. Therefore, it can be concluded that in GUMBO the reduction of water use due to a marginal additional fossil fuel extraction in year 2000 is directly related to the forecasted marginal regression of economic activity during the mid-21st century (as compared to the baseline scenario, where no additional fossil fuel extraction is simulated). Such interpretation has a far larger scope than the one of the AoP of Natural Resource. It is thus important to consider GUMBO's response as a whole, i.e. to track the modeled effects of the perturbation. The holistic analysis of the model's results provides a more comprehensive picture than a single CF and highlights the importance of interconnected mechanisms, dynamic feedback, and long-term effects, which are presently not considered in LCIA methods. The consequences at the endpoint level are illustrated by calculating the marginal cumulated change in GWP over the century, which is negative if no discount rate is applied (Figure 7.3c). The rationale in GUMBO is that marginal fossil fuel extraction generates more future losses than current gains in GWP: present-time additional fossil fuel extraction has negative effects if future generations are given a high priority (discount rate d lower than 0.5%). The result is the opposite in the case of higher discount rate. For instance, if a 3% rate is applied to each year, then short- 156

157 Chapter 7 Integrated earth system dynamic modeling for life cycle impact assessment of ecosystem services term benefits are considered more important than long-term costs and today's fossil fuel extraction is beneficial. The assumptions behind the calculation of GWP in GUMBO have to be carefully considered here. Noticeably, the elasticity coefficient (see Table 7.1) for the ES energy (composed of fossil fuel extraction and renewable energy use and expressed in Gt fossil fuel, as shown in Table 7.2) is the highest (Boumans et al. 2002), meaning that GWP mostly follows the evolution of fossil fuel extraction. The influence of these elasticity coefficients on the results is a significant limitation inherent to GUMBO, as these were determined manually during the calibration process by matching the evolution of 20th century GWP with the Cobb Douglas equation (Boumans 2012) and are therefore affected by uncertainty. Considering GWP as an endpoint may sound odd to LCA practitioners, because economic indicators are generally disconnected from environmental impacts. This consideration holds if GWP is calculated traditionally, e.g. as the total market value of world production. This is not the case with the Cobb Douglas function, which remains empirical and is only valid at the macroeconomic level to assess the influence of ES on economic production, as mentioned in Section Therefore, GWP should be considered here as an alternative proxy for normalization and weighting traditionally used in LCA to calculate a single-score indicator. These techniques remain well suited for current LCIA methods based on constant CFs. a) b) c) Figure 7.3: Example of Characterization Factors calculation steps. a) Marginal changes of water use (TL/yr) over the 21st century (compared to water use in the baseline scenario) induced by a marginal additional extraction of fossil fuel (FF) in yr 2000; b) Linearity between marginal perturbation (in fossil fuel extraction, at yr 2000) and marginal impacts (in water use, over the period ), see equation 7.3; c) Endpoint CF GWP from fossil fuel use as a function of the discount rate (see eqs. 7.4 and 7.5). 157

158 7.3.2 Generalization of the CFs calculation approach to all the types of perturbation and impact Effects on other midpoint and endpoint indicators were further simulated for several types of perturbations. Table 7.3 summarizes the retrieved CFs for the following perturbations: fossil fuel extraction, ore production, surface water and groundwater use (for each one of the 11 biomes modeled in GUMBO 9 ), autotroph net harvest, consumer net harvest, and change in land use (from forest to urban). For example, one can notice that the extraction of ore increases the ES ore use (CF = 1.04), as well as the extraction of surface water or groundwater increases the ES water use (CFs = 1.05 and 1.10). As a result of the interconnection of the environmental mechanisms and feedback, the relationship between the perturbation and the ES is higher than one and more than one ES is affected by the same perturbation. Concerning ES, results like e.g. loss of climate regulation due to additional fossil fuel extraction, biomass harvest or conversion of land from forest to urban are straightforward. Other CFs are however more difficult to explain. For instance, the model does not show any influence of water use on soil formation. Recreational & cultural service is increased with the extraction of abiotic materials (fossil fuels, ore, water) and harvest of consumer biomass (e.g. animals) but decreased with regard to harvest of autotroph biomass (e.g. photosynthetic plants). These results would deserve a systemic in-depth analysis, as exposed in Section 7.3.1, which full development is outside the scope of this paper. An overall analysis of endpoint CFs shows that, in contrast to GWP, SSW is negatively impacted by all marginal changes in the simulated human interventions. This shows that these endpoints have different meanings: welfare cannot be assimilated to economic production Influence of alternative modeling perspectives In order to further explore the opportunities unveiled by the coupling of an integrated earth system dynamic model with LCIA, alternative simulations, considering different modeling perspectives, were run and results were compared with Table 7.3. For the case of a marginal additional extraction of fossil fuels, the following cases were considered: a) a future extraction (time horizons: 2001, 2003, 2005, 2010, 2020, 2030, 2050, 2070 and 2090) to represent an elementary flow that would occur e.g. at the end-of-life phase of a product system; b) extraction occurring over a longer period e.g. in the use phase of a product system (5, 20, 33, 50, 75, 91 and 100 years); c) different policy scenarios, already defined in GUMBO ( Star Trek, Big Government, Mad Max and Ecotopia ), reflecting decision-makers' perspectives different than the business-as-usual ( Base Case ); see Boumans et al. (2002) for details on the scenarios; d) simultaneous perturbations (with extraction of ore), to assess the robustness of CFs retrieved independently from each other. The sets of retrieved CFs help discuss and validate the approach described in this paper, but shall not be directly used in LCIA applications. Full details of these simulations and results are given in SI7.2. The most insightful conclusions are discussed here. 9 Despite they may arguably be better named as land cover types, these 11 systems have been labelled as biomes in the pr esent paper according to the nomenclature used by Boumans et al. (2002) for encompassing the ecosystem types in GUMBO. 158

159 Table 7.3: Summary of CFs: impacts of perturbations on ES, human capitals, GWP and welfare. OM: Organic Matter; En: Energy; GR: Gas Regulation; CR: Climate Regulation; DR: Disturbance Regulation; SF: Soil Formation; NC: Nutrient Cycling; WT: Waste Treatment; RC: Recreation Cultural; BC: Built Capital; K: Knowledge; SN: Social Network; LF: Labor Force; GWP0: Gross World Product, 0% discount rate; GWP3: Gross World Product, 3% discount rate; SSW: Sustainable Social Welfare. Units for each CF is the one displayed in Table 7.1, per unit of perturbation (shown in parenthesis in this table: Gt: Gigatons; TL: Teraliters; Mha: Million hectares). F: Forest, G: Grassland, W: Wetland, L: Lake, U: Urban, O: Ocean, CO: Coastal Ocean, Cr: Cropland, D: Desert. Chapter 7 Integrated earth system dynamic modeling for life cycle impact assessment of ecosystem services 159

160 7.3.4 Perturbation occurring in the future (SI7.2.1) or over a longer period (SI7.2.2) The temporality of human perturbations on the geobiosphere leads to very different impacts, on natural systems and human capitals (midpoint CFs) as well as on human welfare (endpoint CFs). Perturbations occurring in the far future do not really make sense for decision-makers, since the business-as-usual scenario is unlikely to last until But one can also observe significant differences between present-time perturbations, perturbations occurring within 10 years, and those occurring within a range of years. Interestingly, regulating services are less affected by perturbations spread over longer periods. This observation highlights the robustness of the GUMBO model. As a general remark, it would be extremely difficult to obtain similar detailed results using a conventional LCIA approach with constant CFs Perturbation under different policy scenarios (SI7.2.3) In GUMBO, the scenarios proposed are based on two opposed viewpoints on the overall future state of the world and types of policies implemented (optimistic vs. skeptic) (see Boumans et al for a comprehensive description). The choice of global policy options influences a significant number of parameters in the model and therefore the calculation of all CFs. For example, when an additional marginal extraction is simulated at year 2000, midpoint impacts on water use, ore extraction and human capitals seem most influenced by assumptions made on the future state of the world. In contrast, the midpoint impact on energy is more dependent on the type of policies implemented. Interestingly, endpoint impact GWP0 is less affected when skeptical policies are implemented, although the cumulated change remains negative after an additional extraction in year Similarly, in the LCIA method ReCiPe (Goedkoop et al. 2009), various cultural perspectives are proposed (Hierarchist, Individualist, Egalitarian). Decisionmakers can thus rely on different approaches and LCA practitioners must provide the keys to interpret the retrieved results and to understand environmental challenges Simultaneous perturbations (SI7.2.4) Table S7.4 shows the changes of midpoints and endpoints CFs generated by a perturbation on fossil fuel only, on ore production only, and by both perturbations considered simultaneously. The relative difference is less than 1% for most of the midpoints. As a result, one can consider that these perturbations do not influence each other significantly and hence the CFs can be used independently. The simulation should however be extended to the whole set of retrieved CFs throughout a pairwise analysis for all of the perturbations displayed in Table 7.3 (see discussion section). 7.4 Discussion The results allowed validating the general conceptual approach of using integrated earth system dynamic modeling approach to retrieve time-dependent CFs accounting for the dynamic environmental mechanisms and feedback flows underlying their calculation. However, limitations to this approach were also identified. Both advantages and drawbacks are discussed in this section, along with a roadmap for future research. 160

161 Chapter 7 Integrated earth system dynamic modeling for life cycle impact assessment of ecosystem services Need for harmonization between LCI datasets and integrated global models GUMBO considers only 4 resource categories: mineral ore, freshwater, organic matter and fossil fuel. Freshwater is split into surface water and groundwater. Organic matter is divided in autotrophs and consumers. Only two land occupation types (urban and rural) are included. This resolution is much coarser than current LCI datasets. For example, ecoinvent 2.2 (Ecoinvent 2010) has 7 elementary flows of biotic resources, 130 mineral resources, 42 types of land occupation (and 80 transformation types), 5 airborne resources and 13 waterborne resources. The LCIA method ReCiPe (Goedkoop et al. 2009) translates 55 mineral resources, 5 fossil fuel resources, 15 types of urban land occupation, 25 types of agricultural / sylvicultural land occupation and 16 types of natural land transformation (to/from artificial land) into endpoint impacts. As shown in Table S7.5 with a tentative matching between the type and categorization of flows found in LCI with those found in GUMBO, it seems unfeasible to reach such resolution in GUMBO without radical changes in the model that may jeopardize its stability. Despite the intrinsic limitations in terms of ecosystem data resolution, process aggregation and certain basic calculation assumptions (e.g. the use of the Cobb Douglas function, as discussed in Section 7.2.1), GUMBO is able to take into account the inter-linkages among natural and humandriven systems, their feedback and the resulting connections among multiple environmental mechanisms in a holistic way. These features are poorly included in current LCIA models. For example, while modeling through land use does not take into account the major complexity of connections between natural processes, it implicitly includes some of this complexity e.g. through species-area curves, although in a simplistic manner (de Schryver et al. 2010; de Baan et al. 2013). This seems to be justified in LCIA by the criteria established in the ISO 14044:2006, where impact categories, respective indicators and related CFs are proposed to avoid double counting. Therefore, this apparent demand for simplification in LCIA may prevent from considering the complex role of ecosystems in the cause effect chains of life cycle models. The use of GUMBO and, in general, integrated ecological socioeconomic modeling would overcome this limitation (see Section 7.4.2) Added value of formal use of integrated earth system dynamic models within LCIA Former attempts to integrate ES into LCIA delivered midpoint CFs expressed in physical units, which are yet to be related to endpoint impacts. However, the complex interconnections between natural and human systems involve many (interrelated) indirect effects. Trying to decompose natural mechanisms into a set of independent cause effect relationships impacting the three conventional AoP (HH, NE, NR), as well as the human-made environment (see Section 7.2.2) separately, seems a dead-end. For instance, a change in the natural environment may affect human health; deterioration of human health may increase the need for resources; the generation of renewable resources is again dependent on a healthy natural environment. According to the overall rationale of GUMBO, the endpoints (GWP and SSW) are considered as the ultimate indicators of global human development. The advantage of using the Cobb Douglas function is that it enables relating these indicators with important factors that are excluded in neo-classical economics accounting, such as health factors for labor efficiency and, more important within the goal of this paper, ES. The simulation results show that integrated earth system dynamic models like GUMBO can provide a convenient platform to better internalize these externalities in the assessment. Another advantage of using indicators computed in an integrated model is to shrink 161

162 the level of hidden subjectivity somehow existing in the calculation of endpoints in current LCIA methods. Instead of relying on an average weighing of three AoP considered as independent from each other, decision-makers and practitioners could more easily be asked to select policy scenarios (see SI7.2.3) that correspond to their vision of the 21st century. In order to compute a dataset of CFs for LCIA, it is necessary to investigate the degree of mutual influence between perturbations and (midpoint and endpoint) impacts. In the specific case of GUMBO, perturbations on fossil fuel extraction and ore production have a negligible effect on each other. Such investigation shall be performed for each couple of perturbations. If all perturbations that mimic LCI results do not interfere with each other, the computed CFs could be considered as constants and thus be used directly for LCIA, independently from GUMBO. In contrast, if they are found to be influencing each other, it is no more possible to retrieve constant CF. A possible solution would be to tune the model to automatically upload LCI results and directly retrieve midpoint and endpoint impacts. Moreover, the information that could be retrieved from the model would be significantly richer than the one usually gathered from constant CFs. The model outputs may provide sound explanations on the evolution of variables and help identifying the main cause-effect chains, which constitute an added value for stakeholders and decision markers Is further sophistication of integrated modeling necessary for LCIA? Despite the distinction between rural and urban land occupation and the differentiation across ecosystem types (11 biomes) are thoroughly implemented in GUMBO, the model is not designed to address local or regional issues. For example, Table 7.3 shows that changes in surface water use are the same for several biomes (forest, grassland, wetlands, lake, urban). Such similarities are explained by the identical tuning of each hydrologic parameter for these biomes, although different settings could be expected. Concerning groundwater use, GUMBO is tuned to give priority to the use of surface water. Therefore, all the CFs related to a change in groundwater use are null, except for the biome desert, where surface water is not present. Recent developments in LCIA clearly head towards spatially-explicit impact assessment, based on geo-referenced datasets, in order to provide finer clues for decision-making (Saad et al. 2011; IMPACT World+ 2012; de Baan et al. 2013; Koellner et al. 2013a). There is therefore the need to use more sophisticated ecological and economic sub-models, integrated in a meta-model following GUMBO's architecture. GUMBO's developers recently released MIMES (Boumans and Costanza 2007), which is designed to embed also local assessment of ES within a global system. MIMES can model multiregional systems, which are all inter-connected, and includes powerful GIS mapping tools for multi-scale characterization and dissemination purposes. Hence, MIMES appears as an interesting candidate to further explore the approach unveiled in this paper following a spatially-explicit concern. Numerous conceptual challenges and technical issues may arise, however, while designing a generic meta-model. First, the list of relevant ES is location-specific, as well as their contribution to the economy. A generic model should take into account such specificities in nesting local systems in the global one (Andrade et al. 2010). Second, their importance to the various economic sectors changes from place to place. The metamodel may take advantage of the mature structure of input output tables for the design of a generic model, possibly extended with environmental intervention datasets (e.g. Lenzen et al. 2012; Tukker et al. 2013). Third, the Cobb Douglas 162

163 Chapter 7 Integrated earth system dynamic modeling for life cycle impact assessment of ecosystem services functions may no longer be applicable in a multi-scale model and the welfare endpoints described here may thus need to be defined differently. The relationships between human capitals and economic sectors may also need further refinements. Other issues not identified in this discussion are likely to arise, should such multi-scale model be developed Research roadmap A preliminary roadmap for the development of such a meta-model shall thus give priority to implement multi-regional input output tables (Lenzen et al. 2012) to first downscale the anthroposphere at national scale. A second step should be to develop the description of ES, which is currently aggregated into four ecosystem goods and seven ecosystem services in GUMBO. Local ES could be then modeled with GIS tools as in MIMES, and connected to the nation-wide economic model. This would lead to a multi-scale meta-model: global, national and local. Spatial information should be consistently harmonized or, at least, not have too high resolution, otherwise the link between local ES and national input output tables may become incongruous. Ultimately, dynamic models would potentially enhance the modeling of the human activity over a local territory, but the user of such model should be informed on its limited resolution. Undoubtedly such project would require a large amount of data, heavy processes for quality-check, and multiple scientific skills to be coordinated. 7.5 Conclusions Preliminary results from the GUMBO simulations showed that it is feasible to retrieve CFs for ES and to connect midpoint changes to endpoints indicators such as GWP and SSW. The CFs, however, cannot be yet straight coupled to current LCIA methods because GUMBO does not match the granularity of LCI datasets, and endpoints differ significantly from the common LCIA practice, nomenclature and definition of AoP. The rationale behind the approach proposed here is however promising, as it shows that integrated earth system dynamic modeling can potentially provide more detailed information on the interactions between human interventions (mostly resource extraction) and the natural and human capitals than plain datasets of constant CFs. It can take into account the potential interactions between human interventions, whereas it is not the case with current CFs, and provide insights on the mechanisms that lead to short-term, midpoint impacts, without disregarding the global, long-term picture. It can distinguish between human interventions occurring at present time, in the future, and over a long period and can be tuned up to estimate future world state conditions according to different policy scenarios, which is undoubtedly an interesting feature for decision-makers. The integrated coupling between the geobiosphere compartments and the anthroposphere enables linking cause effect chains that are currently artificially separated in LCIA models. It may also account for changes in human capitals, which are considered in LCA as an important AoP but not yet formally included in current models. Moreover, the human capital midpoint impact indicators considered in this study (Table 7.1) could be notably adapted for use in Social LCA, Life Cycle Costing or, more broadly, Life Cycle Sustainability Assessment (jointly with ecosystem service indicators). Despite these attractive potentialities, the integration with LCIA is far from being readily operational. A research roadmap has been envisaged accordingly, including the use of more sophisticated and spatially explicit models such as MIMES. 163

164 Acknowledgment This project was supported by the National Research Fund, Luxembourg (Ref ) and the French National Research Fund (project EVALEAU ANR-08-ECOT C0238). Appendix A. Supplementary data SI includes a detailed presentation of GUMBO, methodological details, as well as in-depth results and interpretation. Supplementary data associated with this article can be found, in the online version, at 164

165 8. Conclusions

166

167 Chapter 8 Conclusions In the context of global change, environmental accounting increasingly becomes as a prerequisite for planetary stewardship. It appears as a necessity for the development of a socio-economic system that allows taking into account the natural capital. It is necessary to reach a consensus enabling a fair share of the responsibilities of environment burdens among the various actors of the society. It can also prove useful to compare possible alternatives of development policies and industrial projects compatible with a healthy environment. However, developing a proper accounting framework for the environmental impacts of human activities is a complex challenge, which requires settling a uniform assessment scope while handling numerous data from diverse information sources and different levels of details. It has to be based on strong consensuses reached in the scientific community, and reflect our understanding of the functioning of ecosystems and our interactions with them. It must be designed and applied in a rigorous, transparent and reproducible way, and packaged with useful modeling tools and interpretation rules in order to motivate decision-makers to using it. There is no doubt that the popularity of Life Cycle Assessment (LCA) in environmental evaluation and management is favored by the existence of datasets built on standardized methodologies, developed by decades of continuous efforts of the scientific community and more recently by consensus at the European policy level. In contrast, Emergy Evaluation (EME) is much less - though increasingly - known to decision-makers and analysts, despite the fact that the method is indeed older than LCA. This lack of popularity is likely to be related to the absence of such standardized and transparent accounting framework and integrated datasets, allowing a large flexibility in defining system boundaries and thus the subjective selection of elements to consider or discard. However, EME also offers considerable potentialities, by accounting for the natural value of the natural resources and Ecosystem Services (ES), on which mankind relies, and address the challenges of an integrated ecological-economic accounting. EME can benefit from adapting the standardization procedure and datasets available in LCA, although this integration is hindered by the differences of accounting rules in both frameworks. A strategic roadmap towards more standardization of EME, by hybridation with LCA, was proposed in chapter 2. The first step consists in using of LCI datasets with respect to the emergy algebra, and was made operational by the release of the software SCALE. The following step includes the development of an integrated model of the geobiosphere, in order to account for the natural value of resources in a uniform and robust manner. Other issues related to e.g. the standardization of core definitions of the EME framework, the identification of system boundaries in case studies, and the design of emergy indicators, need to be addressed by reaching consensual agreement among the emergy community. A test bed case composed of 4 Water Treatment Plants, previously studied with LCA, was performed with the traditional EME framework (chapter 3), in order to illustrate its current limitations, such as the fact that pollution impacts are left unaccounted for. In addition, the current availability of Unit Emergy Values (UEVs) of man-made products is quite limited, which is a severe limitation for the evaluation of industrial systems. This chapter also points out the complementarities with LCA, in terms of system boundaries and result interpretation through the emergy indicators, as well as the potentialities of using LCA datasets to strengthen emergy accounting. This reference study was then compared to results from SCALE and from the Solar Energy Demand (SED) method, in order to perform a critical analysis of the current state of the hybrid lifecycle-emergy framework. Results demonstrated the added value of SCALE to enhance reproducibility, accurateness and completeness of an EME, although the software is limited by the 167

168 scope of the datasets used; typically, the emergy value of the contribution of e.g. local freshwater resources, human labor, atmospheric gaseous resources, ecosystem services must be accounted for manually and added to the results provided by SCALE. However, such limitations could be addressed by enriching the scope of the datasets used, i.e. without update of the software. Another limitation discussed in this chapter is related to the exclusion of pollution impacts and the emergy value of the efforts spent by natural and human systems to absorb, treat or compensate them. This aspect was left out of the scope of this PhD. Finally, although the computation algorithm embedded in SCALE provides a rigorous emergy accounting of man-made products, results still need to be further processed to retrieve the final emergy indicators. Therefore, a study aiming at adapting the definition of the emergy-indicators to the hybrid framework was proposed (chapter 5), and their calculation was made operational with the development of a specific algorithm to be implemented in a future version of SCALE. The proposed changes in the formal definition of the indicators may hopefully contribute to the standardization of EME. These achievements demonstrate that despite some necessary improvements, the hybrid lifecycleemergy framework is more rigorous, transparent and reproducible than the traditional application of EME. The following investigations proposed in this PhD are dedicated to the development of an integrated model of the geobiosphere. A specific focus on freshwater resource characterization showed the usefulness of using Information Technology (IT) tools, such as Geographic Information Systems (GIS). The study produced the first global, spatially-explicit emergy database of rivers, developed from physical maps and high resolution, and based on the conventional approach to calculate the UEV of freshwater (chapter 6), along with regional averages in order to feed the datasets of resources UEV in the hybrid framework. Results were poorly comparable with previous case studies, in which the local UEV of freshwater were computed manually, because of the slight variations adopted by the practitioners. Similarly, this prospective work shows that IT tools can foster the standardization of calculation procedures and data sources in emergy accounting. Finally, it was also observed that LCA could benefit from further exploitation of an integrated and dynamic model of the geobiosphere, in which the mutual interactions between its components, including mankind, are taken into account (chapter 7). This work remained prospective and limited to validating the conceptual aspects, as well as the added value, of such scenario for future developments in LCA. It is advanced that these achievements will contribute to the consolidation of the hybrid lifecycleemergy framework, both by validating previous methodological achievements (such as the algorithm embedded in SCALE), and proposing hints for the formal integration of future developments e.g. the inclusion of human labor in LCA datasets, the uniform calculation of resources UEV, the definition of emergy indicators. It is recommended to enlarge the scope of the case studies used in this work to include e.g. pollution impacts of the lifecycle system, changes in ecosystem services functioning due to water abstraction by the WTPs, in order to further address the current limitations of the hybrid framework and of SCALE. Assuming that a large diversity of case studies is naturally useful to test and improve an existing methodology, the case study potable water production could be extended to compare household production systems of potable water and water-saving strategies. It may also be useful to identify other case studies that enable comparing traditional industrial processes with ecologically-engineered processes, e.g. wastewater treatment. Finally, my personal feeling is that environmental accounting methods can be greatly enhanced by a smart integration of IT tools, such as dynamic modeling and GIS software. LCA actually 168

169 Chapter 8 Conclusions already proposes commercial software to process the large datasets and increasingly relies on modeling and GIS to develop impact assessment methods. However, the formal integration of such tools could be even further explored, in order to consistently consider the dynamics of the interactions between human activities and ecosystems, their long-term consequences, and the influence of policy-making. As demonstrated in the present study, EME may greatly benefit from following such path. There are countless models, in both fields of ecology and economy, built-up by specialists, which may offer promising resources to further develop the hybrid lifecycle framework, provided IT tools are fruitfully exploited. Such integration, further strengthen by the necessary standardization of core concepts and system boundaries delineation, could result in a consensual and integrated environmental accounting model. This approach could even be proposed as a general platform for developers to insert their latest advances into a global metamodel. A common platform would undoubtedly increase the adoption of emergy by practitioners, are provide useful feedbacks for developers and theoricians. For example, the integration of detailed ecological models may help describe the energetics of ecosystems and thus refine the emergy theory, e.g. by comparing its results with those retrieved from other concepts in non-equilibrium thermodynamics and information theories (e.g. Ulanowicz 1980, 1997; Schneider 1994; Ulanowicz and Abarca-Arenas 1997; Dewar 2003; Kleidon and Lorenz 2005; Dewar and Porté 2008; Skene 2013; Fath et al. 2001, 2004; Herrmann-Pillath 2011). 169

170

171 Supplementary information

172

173 Supplementary information SI3. Chapter 3: Emergy Evaluation of Water Treatment Processes SI3.1. Life Cycle Inventory (LCI) and economic data Data in Tables S3.1-S3.3 were retrieved from Igos et al. (2013a, 2013b). They detail operational inputs for the four WTPs and infrastructure inputs for Sites A and B. Reported data refer to the delivery of 1 m 3 of potable water at the plant outlet. The reader may find all details on each input (purity of chemicals, abbreviation employed in the label ) in the Ecoinvent database. Excavation (Table S3.2) was not further categorized and considered, due to missing corresponding UEV in the literature. SI3.2. Proxy UEV for Ecoinvent elementary flows Most of the UEVs presented below were retrieved from the International Society for the Advancement of Emergy Research s database (Tilley et al. 2012); their original reference is indicated in column Ref. Some UEVs were however absent from this database: UEVs from Bastianoni et al. (2005, 2009), Campbell and Ohrt (2009), Pulselli et al. (2007). UEVs computed in this study for local freshwater and French electricity (details in main text). UEVs of several chemicals, which were not found in the literature. As a proxy, we recalculated the Solar Energy Demand (SED, Rugani et al. 2011a), with a modified Solar Energy Factor (SEF) for sodium chloride, since this value is highly discussed in this paper and strongly influences the results of some chlorine-based reagents. The SEF for sodium chloride was set to 9.81 E05 MJse/kg (from Odum 1996). Then, we re-computed the SEF of selected chemicals using ecoinvent 2.2 database and SimaPro. Note that SEDs are computed in Rugani et al. (2011a) on the 9.26 baseline. UEVs for activated carbon and regenerated activated carbon were calculated similarly, using data listed below Table S3.4. The reader may find all details on each LCI item in the Ecoinvent database, by searching the exact label reported. UEV of activated carbon Data retrieved from Igos et al. (2013a): producing 1 kg of activated carbon requires 1 kg of hard coal for the main material, 0.04 kg of hydrochloric acid, 3 kg of steam, 30.4 MJ of hard coal (burned), 196 MJ of natural gas, kwh of electricity, medium voltage UCTE, and 0.6 tkm of road transportation (lorry, 16-32t). Using UEVs listed in Table S3.4, we estimated a UEV for activated carbon of 1.56E+13 sej/kg. UEV of regenerated activated carbon Data updated from Igos et al. (2013a): regenerating 1 kg of activated carbon requires 0.1 kg of activated carbon, 0.3 kg of steam, 3 MJ of hard coal (burned), 108 MJ of natural gas, kwh of electricity, production mix, UCTE, and 2.54 tkm of road transportation (lorry, 16-32t). Using 173

174 UEVs listed in Table S3.4, we estimated a UEV for regenerated activated carbon of 8.54 E12 sej/kg. The original activated carbon to be reactivated was not accounted in its UEV, to avoid double-counting. Table S. 3.1: LCI data (operation) for the WTPs. Input Type Energy Chemicals Services UEV Name Ecoinvent input Unit Site 1 Site 2 Site A Site B Electricity mix, France Diesel Activated carbon Regenerate d activated carbon Acrylic acid Al 2 SO 4 CO 2 liquid Cl 2 gas FeCl 3, 40% Lime Lime H 3 PO 4, 85% KMnO 4 Caustic soda NaOCl, 15% H 2 SO 4 Coal fly ash, with services Material Transport (truck) Electricity, low voltage, at grid [FR] Electricity, medium voltage, at grid [FR] Heavy fuel oil, burned in power plant [FR] kwh 8.96E E E E-01 kwh E MJ 6.49E kg 3.60E E-03 - kg 3.60E E E E-03 Acrylic acid, at plant [RER] kg 1.74E E Aluminium sulphate, powder, at plant [RER] Carbon dioxide liquid, at plant [RER] Chlorine, gaseous, diaphragm cell, at plant [RER] Iron (III) chloride, 40% in H 2 O, at plant [CH] Lime, hydrated, packed, at plant [CH] Quicklime, milled, packed, at plant [CH] Phosphoric acid, industrial grade, 85% in H 2 O, at plant [RER] Potassium permanganate, at plant [RER] Sodium hydroxide, 50% in H 2 O, production mix, at plant [RER] Sodium hypochlorite, 15% in H 2 O, at plant [RER] Sulphuric acid, liquid, at plant [RER] Disposal, hard coal ash from stove, 0% water, to municipal incineration [CH] Transport, lorry >16t, fleet average [RER] Transport, lorry >32t, euro3 [RER] Transport, lorry t, fleet average [RER] Transport, lorry t, fleet average [CH] kg 1.79E E kg E E-02 kg E kg E E-02 kg E E-02 kg E- 02 kg E kg - - kg E-02 kg 8.89E-04 - kg E E E E E E E E E-04 kg 3.60E tkm 2.08E tkm 3.10E tkm E tkm 4.07E

175 Supplementary information Table S. 3.2: LCI data of infrastructure for sites A and B. UEV Name Ecoinvent input Unit Site A Site B Building, hall, steel construction/ch/i mm E E+00 Em-Building, Building, hall, steel construction/ch/i and mm E+00 Surface building, hall/ch/i Building, hall/ch/i mm E+00 - Em-building, Building, multi-storey/rer/i cm E+00 - Volume Concrete Concrete, normal, at plant/ch cm E E+01 Copper Copper, at regional storage/rer mg 1.54E E+01 Glass Foam glass, at plant/rer mg 8.00E E-01 Epoxy resin, liquid, at plant/rer mg 3.52E E+02 Glass fibre reinforced plastic, polyester resin, mg 5.79E E+01 hand lay-up, at plant/rer Nylon 6, at plant/rer mg 5.52E E+01 Plastic (PVC) Polyethylene, HDPE, granulate, at plant/rer mg 5.21E E+02 Polyethylene, LLDPE, granulate, at plant/rer mg 3.29E E+00 Polyvinylchloride, at regional storage/rer mg 1.23E E+01 Styrene-acrylonitrile copolymer, SAN, at mg 8.36E E+00 plant/rer Synthetic rubber, at plant/rer mg 8.30E-01 - Chromium steel 18/8, at plant/rer mg 1.01E E+01 Steel Reinforcing steel, at plant/rer g 2.38E+00 - Reinforcing steel, at plant/rer mg 1.30E E+03 Steel, low-alloyed, at plant/rer mg 3.09E E+02 Material Transport, lorry 16-32t, euro3/rer tkm 3.00E E-03 Transport Transport, lorry 20-28t, fleet average/ch tkm 4.67E E-03 (truck) Transport, lorry t, fleet average/ch tkm 1.11E E-04 Excavation Excavation, hydraulic digger/rer cm E+01 - Table S. 3.3: Economic inputs for the 4 WTPs (in per m 3 output water) Economic input Site 1 Site 2 Site A Site B Labor 8.62E E E E-02 Energy 1.28E E E E-02 Reagents 7.46E E E E-02 Exploitation expenditures 1.29E E E E-05 Maintenance 2.58E E E-03 Sludge management 2.58E E E-03 Annual allocation for renewal E E

176 Table S. 3.4: Proxy UEVs of Ecoinvent v2.2 elementary flows. Flow type Energy Chemicals Services proxy UEV value (sej/unit) UEV unit Ref LCI Item LCI Unit UEV (sej/lci unit) electricity, low voltage, at grid [FR] kwh Electricity mix, electricity, medium voltage, at grid kwh France (without 5.91E+04 J [a] [FR] 2.13E+11 L&S) electricity, medium voltage, production UCTE, at grid [UCTE] kwh hard coal mix, at regional storage Coal (kg) kg 1.17E+12 [UCTE] 4.00E+04 J [b] hard coal, burned in industrial Coal (MJ) MJ 4.00E+10 furnace 1-10MW [RER] Diesel 6.71E+04 J [c] heavy fuel oil, burned in power plant [FR] MJ 6.71E+10 Natural gas 4.43E+04 J [d] natural gas, burned in industrial furnace >100kW [RER] MJ 4.43E+10 Activated carbon 1.56E+13 kg [a] - kg 1.56E+13 Regenerated activated carbon 8.54E+12 kg [a] - kg 8.54E+12 Acrylic acid 3.55E+12 kg [e] acrylic acid, at plant [RER] kg 3.55E+12 Al 2 SO E+12 kg [e] aluminium sulphate, powder, at plant [RER] kg 1.18E+12 CO 2 liquid 9.48E+11 kg [e] carbon dioxide liquid, at plant [RER] kg 9.48E+11 Cl 2 gas 6.67E+09 g [f] chlorine, gaseous, diaphragm cell, at plant [RER] kg 6.67E+12 HCl, 30% 3.00E+12 kg [e] hydrochloric acid, 30% in H 2 O, at plant [RER] kg 3.00E+12 FeCl 3, 40% 3.01E+12 kg [e] iron (III) chloride, 40% in H 2 O, at plant [CH] kg 3.01E+12 Lime g [f] lime, hydrated, packed, at plant [CH] kg [f] quicklime, milled, packed, at plant [CH] kg 1.00E+12 H 3 PO 4, 85% 6.20E+12 kg [e] phosphoric acid, industrial grade, 85% in H2O, at plant [RER] kg 6.20E+12 KMnO E+13 kg [e] potassium permanganate, at plant [RER] kg 8.24E+13 Caustic soda 1.46E+09 g [f] sodium hydroxide, 50% in H 2 O, production mix, at plant [RER] kg 1.46E+12 NaOCl, 15% 2.59E+12 kg [e] sodium hypochlorite, 15% in H 2 O, at plant [RER] kg 2.59E+12 H 2 SO E+11 kg [e] sulphuric acid, liquid, at plant [RER] kg 4.15E+11 NGCC Steam steam, for chemical processes, at 1.33E+09 g [c] turbine plant [RER] kg 1.33E+12 disposal, hard coal ash from stove, Coal fly ash, with 1.40E+10 g [b] 0% water, to municipal incineration services [CH] kg 1.40E+13 Material Transport (truck) 9.65E+11 tonmile [g] transport, lorry >16t, fleet average [RER] transport, lorry >32t, EURO3 [RER] transport, lorry 16-32t, EURO3 [RER] transport, lorry t, fleet average [RER] transport, lorry t, fleet average [CH] tkm tkm tkm tkm tkm 6.61E

177 Supplementary information References: [a]: this study. [b]: Odum (1996); with E6 J/kg coal. [c]: Bastianoni et al. (2009). [d]: Bastianoni et al. (2005). [e]: adapted from Rugani et al. (2011a), see text above table. [f]: Campbell and Ohrt (2009). [g]: Buranakarn (1998); with 1 ton.mile = 907 kg x m = tkm. Table S. 3.5: Proxy UEVs of Ecoinvent v2.2 elementary flows used in infrastructure. UEV value UEV Re LCI proxy UEV LCI Item (sej/lci (sej/unit) unit f Unit unit) Concrete 1.54E+09 g [a] Concrete, normal, at plant/ch cm E+09 Glass 2.12E+09 g [a] Foam glass, at plant/rer mg 2.12E+06 Material Transport (truck) 9.65E+11 tonmile [a] Plastic (PVC) 5.85E+09 g [a] Steel 4.13E+09 g [a] Transport, lorry 16-32t, EURO3/RER Transport, lorry 20-28t, fleet average/ch Transport, lorry t, fleet average/ch Epoxy resin, liquid, at plant/rer Glass fibre reinforced plastic, polyester resin, hand lay-up, at plant/rer Nylon 6, at plant/rer Polyethylene, HDPE, granulate, at plant/rer Polyethylene, LLDPE, granulate, at plant/rer Polyvinylchloride, at regional storage/rer Styrene-acrylonitrile copolymer, SAN, at plant/rer Synthetic rubber, at plant/rer tkm tkm tkm mg mg mg mg mg mg mg mg 6.61E E+06 Chromium steel 18/8, at plant/rer mg 4.13E+06 Reinforcing steel, at plant/rer g 4.13E+09 Reinforcing steel, at plant/rer mg 4.13E+06 Steel, low-alloyed, at plant/rer mg 4.13E+06 Copper 2.00E+09 g [b] Copper, at regional storage/rer mg 2.00E+06 Em-Building Volume 1.07E+15 m 3 [c] Building, multi-storey/rer/i cm E+08 Em-Building 7.49E+15 m 2 Building, hall, steel construction/ch/i [d] Surface and Building, hall/ch/i mm E+09 [d] Building, hall, steel construction/ch/i 2 mm [d] Building, hall/ch/i mm 2 Excavation 7.30E+11 m 3 [e] Excavation, hydraulic digger/rer cm E+05 [a]: Buranakarn (1998); we assumed 2.3 g/cm3 concrete; we used UEV of concrete ready-mixed with fly ash (by product), without services for concrete, and UEV of post-consumer glass (without services) for foam glass; we used the average UEV of PVC for plastics, because no other value was found in literature. [b]: Lapp (1991); includes labor. [c]: Pulselli et al. (2007); includes construction materials and labor; baseline [d]: recalculated from Pulselli et al. (2007); calculated for a 30x50x7m3 building (Ecoinvent settings); em-building Surface = em-building Volume (1.07 E15 sej/m3) * 10,500 m3 / 1,500 m2 = 7.49e15 sej /m2 (baseline 15.83). [e]: proxy from Rugani et al. (2011a). 177

178 SI3.3. Emergy value of land occupation The typical WTP area is 37,000 m 2 (personal communication with the operating companies). Emergy value of land occupation is approximated to 6.29 E10 sej/(m 2 yr) (Odum 1996, empower density of the Earth, baseline 9.44). The emergy value of land occupation is thus 2.33 E15 sej/yr, which is negligible compared to the total value of R and F for each WTP (see Tables in Chapter 3). SI3.4. UEV of freshwater flows Table S3.6 illustrates the calculation of freshwater UEV, using local data. Table S3.7 summarizes the local information available for the calculation of al freshwater UEVs used in this paper. Table S. 3.6: Example of UEV calculation for local freshwater, here of Seine River at Site 1 location. UEV is computed using baseline 15.83; results shown in Tables 3-6 are tuned for the baseline Exergy Note Item Unit (Unit/yr) RENEWABLE RESOURCES UEV (sej/unit) Emergy (sej/yr) 1 Sun J 1.78E E20 2 Wind J 2. 32E E E19 3 Rain, chemical J 1.45E E E21 4 Rain, geopotential J 1.21E E E21 EMERGY OF SEINE WATER at Site 1 5 Freshwater flow m E E E21 Inputs # 1-4 are co-products of regional natural processes. Therefore, to avoid double-counting, only the input with the highest emergy value is taken into account. 1. Sun exergy = Area (m 2 )*(1-albedo)*annual avg. insulation (J/(m 2.yr)) Drainage Seine Area at site 1 location: see table S3.7 Albedo = 0.17 (estimation, for deciduous forests and bare soils; en.wikipedia.org/wiki/albedo) Annual average insolation = see Table S3.7 (average sum of global irradiation per square meter received by a flat plane PV collector, at optimal angle; JRC 2007) UEV = 1 sej/j, by definition (Odum 1996) 2. Wind exergy = Area (m 2 ) *air density (1.3 kg/m 3 )*drag coefficient (0.002 for land)* wind velocity (m/s) * 3.15E7 (s/yr) Drainage Seine Area at site 1 location: see table S3.7 Wind velocity: see table below UEV = sej/j (Odum et al. 2000) 3. Rain, chemical exergy = Area (m 2 )* Rain (m/yr) *water density (1 E6 g/m 3 )*Gibbs free energy (4.69 E-3 J/g) Drainage Seine Area: see table S3.7 Rain: see table S3.7 UEV = sej/j (Odum et al. 2000) Gibbs free energy of rainwater, assuming 999,990 ppm concentration in rainwater, 965,000 ppm concentration in seawater and at 12 C (estimated average air temperature for the considered catchment areas (JRC, 2007) = 4.69 J/g. 178

179 Supplementary information 4. Rain, geopotential energy = Area (m 2 )* Rain (m/yr) *water density (1 E3 kg/m 3 )* average height (m) *gravity (9.8 m/s 2 ) Drainage Seine Area: see table S3.7 Rain: see table S3.7 Average height: see table S3.7 UEV = sej/j (Odum et al. 2000) 5. UEV of freshwater = Total emergy input, without double-counting (sej/yr) / ( River flow (m 3 /s) * 3.15 E7 (s/yr) ) Average river flow: see Table S3.7 Table S. 3.7: Specific data for each local freshwater source used up by the WTPs. Seine River Seine River Watershed Watershed Data at Site 1 at Site 2 of Site A of Site B Annual average insolation (kwh/m 2.yr) 1,370 [a] 1,390 [b] 1,470 [c] 1,430 [d] Area of effective drainage (m 2 ) 4.35E10 [e] 3.08E10 [f] 1.05E10 [g] 1.12E9 [h] Wind velocity (m/s) 6.5 [i] 6.5 [i] 6.5 [j] 6.5 [j] Annual rainfall on drainage area (mm/yr) 710 [k] 710 [k] 600 [g] 600 [l] Average height of catchment area (m) 400 [m] 400 [m] 150 [n] 150 [n] Average river flow (m 3 /s) 328 [e] 218 [f] 71.3 [g] 7.1 [h] Resulting UEV (sej/m 3, this study) 5.51E E E E11 References: [a] JRC (2007), data at N, E; [b] JRC (2007), data at N, E; [c] JRC (2007), data at N, W; [d] JRC (2007), data at N, W; [e] ; [f] ; [g] [h] [i] National Environmental Accounting Database (CEP 2006); [j] assumed a similar value as for the Seine River watershed; [k] Curie et al. (2007), rainfall average of the whole Seine watershed; [l] assumed same as for the Vilaine watershed; [m] estimated; Curie et al. (2007) provide an average elevation of 300m for the Seine watershed; [n] assumed; the actual value is probably lower. SI3.5. Emergy indicators in literature Data retrieved from literature needed to be adapted to the definition of emergy-based indicators chosen for this study. For example, imports (F) are sometimes split into a renewable part and a non-renewable one, using %R of each input (e.g. in Paoli et al. 2008; Lu et al. 2009; Zhang et al. 2011a). La Rosa et al. (2008) included local labor in R, which is a usual practice for farmlands. Some authors also summed all R inputs instead of selecting only the highest one. Original data were adapted to our calculation framework of emergy indictors, if necessary. Modifications and relevant information can be found in Table S

180 Table S. 3.8: Emergy-based indicators of selected published case studies. 180

181 (References: [a] Zhang et al. 2012; [b] La Rosa et al. 2008; [c] Lima et al. 2012; [d] Giannetti et al. 2013a; [e] Brown and Buranakarn 2003; [f] Bargigli and Ulgiati 2003; [g] Brown and Ulgiati 2002; [h] Castellini et al. 2006; [i] Agostinho et al. 2010; [j] Lefroy and Rydberg 2003; [k] Ortega et al. 2002; [l] Pulselli et al. 2011b; [m] this study; [n] Paoli et al. 2008; [o] Lu et al. 2009; [p] Mu et al. 2012; [q] Ciotola et al. 2011; [r] Zhang et al. 2011a; [s] Yang et al. 2010; [t] Buenfil 2001) Supplementary information 181

182 182 The following tables and charts rank emergy-based indicators according to the median of values found in Table S3.8:

183 Supplementary information 183

184 184

185 Supplementary information SI3.6. Emergy indicators of drinking water production facilities. The following figures compare EmE of water production systems found in Table S3.8 (same referencing code, in brackets). Full bars correspond to WTPs, while hatched bars correspond to other alternatives such as household technologies and bottled water. Emergy-based indicators were recalculated according to the definition detailed in the main text to avoid bias in comparing different studies. The purposes of these figures are: 1) to check compatibility of the results presented in this paper with previously published works; 2) to compare the ecological performance of WTPs (full bars) with household technologies (hatched bars); and 3) to compare the relative performance of the WTPs studied in this paper with previous work. These figures clearly show that (1) our results give values very close to previous results, (2) that some WTPs reported in past studies have better ecological performance than the WTPs studied in this paper, which are of average rank. EYR may vary by a factor 3. In this study, ELR = EIR and these UEVs vary by more than an order of magnitude, whereas, ESI varies by 2-3 orders of magnitude, and %R ranges from 0 to 60 %. These ranges may seem to be high, but they must be compared to values of the emergy indices calculated for other activities (see section 5 of the Appendix). As explained in the main paper, such variability is due to the different technologies used (because of the different qualities of the local freshwater resource). potable water - [l] potable water - [l] potable water - [l] potable water 1 - [m] potable water 2 - [m] potable water A (without infra) - [m] potable water B (without infra) - [m] potable water A (with infra) - [m] potable water B (with infra) - [m] filtered water - [t] boiled water - [t] solar distilled water - [t] solar distilled water - [t] bottled water - [t] WTP WPB - [t] WTP Tampa - [t] WTP Gainesville - [t] WTP Tampa Bay - [t] WTP Dunedin - [t] WTP Tampa Bay 2 - [t] WTP FL Keys - [t] WTP Stock Island - [t] EYR 0 0,5 1 1,5 2 2,5 3 3,

186 potable water - [l] potable water - [l] potable water - [l] potable water 1 - [m] potable water 2 - [m] potable water A (without infra) - [m] potable water B (without infra) - [m] potable water A (with infra) - [m] potable water B (with infra) - [m] filtered water - [t] boiled water - [t] solar distilled water - [t] solar distilled water - [t] bottled water - [t] WTP WPB - [t] WTP Tampa - [t] WTP Gainesville - [t] WTP Tampa Bay - [t] WTP Dunedin - [t] WTP Tampa Bay 2 - [t] WTP FL Keys - [t] WTP Stock Island - [t] ELR = EIR 0, ESI potable water - [l] potable water - [l] potable water - [l] potable water 1 - [m] potable water 2 - [m] potable water A (without infra) - [m] potable water B (without infra) - [m] potable water A (with infra) - [m] potable water B (with infra) - [m] filtered water - [t] boiled water - [t] solar distilled water - [t] solar distilled water - [t] bottled water - [t] WTP WPB - [t] WTP Tampa - [t] WTP Gainesville - [t] WTP Tampa Bay - [t] WTP Dunedin - [t] WTP Tampa Bay 2 - [t] WTP FL Keys - [t] WTP Stock Island - [t] 1E-04 1E-02 1E

187 Supplementary information %R potable water - [l] potable water - [l] potable water - [l] potable water 1 - [m] potable water 2 - [m] potable water A (without infra) - [m] potable water B (without infra) - [m] potable water A (with infra) - [m] potable water B (with infra) - [m] filtered water - [t] boiled water - [t] solar distilled water - [t] solar distilled water - [t] bottled water - [t] WTP WPB - [t] WTP Tampa - [t] WTP Gainesville - [t] WTP Tampa Bay - [t] WTP Dunedin - [t] WTP Tampa Bay 2 - [t] WTP FL Keys - [t] WTP Stock Island - [t] 9.44 UEV /g potable water - [l] potable water - [l] potable water - [l] potable water 1 - [m] potable water 2 - [m] potable water A (without infra) - [m] potable water B (without infra) - [m] potable water A (with infra) - [m] potable water B (with infra) - [m] filtered water - [t] boiled water - [t] solar distilled water - [t] solar distilled water - [t] bottled water - [t] WTP WPB - [t] WTP Tampa - [t] WTP Gainesville - [t] WTP Tampa Bay - [t] WTP Dunedin - [t] WTP Tampa Bay 2 - [t] WTP FL Keys - [t] WTP Stock Island - [t] 0% 20% 40% 60% 80% 1E+05 1E+06 1E+07 1E

188 SI3.7. Comparative table of EmE and LCA results for the four WTPs Table S3.9 shows the results of EmE (Chapter 3) and LCA (Igos et al. 2013a, 2013b) for the four WTPs. See main text for comparison between results provided by both methods. Table S. 3.9: LCA and EmE results for the WTPs. Data Igos et al., ReCiPe (ecopoints /m3) Site 1 Site 2 Site A Site B Total HH 1.04E E E E-02 Total ED 4.16E E E E-03 Total Res Elec 3.76E E E E-03 Total Res Fossil fuels 2.07E Total Res Infra E E-04 Total Res Chemicals 4.36E E E E-03 Total Res other (Services) 3.20E E E E-06 EmE results (sej/m3) Site 1 Site 2 Site A Site B Electricity 1.91E E E E+11 Fossil fuels 4.35E Chemicals 1.17E E E E+11 Infrastructure E E+11 Labor and Services 1.30E E E E+11 Freshwater 5.61E E E E+11 HH: Impacts on Human Health; ED: Impacts on Ecosystems Diversity; Res: Impacts on resources depletion; LCA Results are calculated using the ReCiPe method and provided in ecopoints (Goedkoop et al. 2009) per m 3 of potable water produced. EmE results are expressed in sej per m 3 of potable water produced. 188

189 Supplementary information SI4. Chapter 4: Emergy evaluation using the calculation software SCALE: case study, added value and potential improvements SI4.1. Influence of the minflow threshold value in EME SCALE The backtracking algorithm employed in SCALE considers the studied system as a network of interlinked processes and performs a graph search; the emergy content of each node of the graph is tracked along the different paths from the inputs (resources) to the output (studied product). When a path (i.e., in emergy terms, an emergy flow) splits, and the emergy value assigned to a branch is lower than the threshold value set beforehand, then the algorithm stops propagating the emergy flow downstream and starts exploring a different path, visiting a new node (the order in which the nodes are visited is established by the search algorithm used, which is a depth-first search in SCALE). We refer to the flow that is not accounted for due to this circumstance (that we call minflow violation) as flow lost due to minflow violation. This threshold level must be optimized by the user to balance calculation time with the emergy value of flow lost due to minflow violation. The other halt condition for the flow propagation (i.e. the loss of a part of the emergy flow accounted for) in SCALE is the case of a feedback loop occurring: we refer to this portion of the flow as flow lost due to loop violation. Figure S4.1 shows the portion of flows lost due to minflow violation and due to loop violation. An optimal threshold is sought to balance result s precision with calculation time. The numbers in brackets (next to the WTP s name) indicate the negative log of the threshold value; e.g. Site A (5) refers to the application of SCALE on Site A, with threshold = 1E-5 Msej. The calculation time increases exponentially with this value (Marvuglia et al. 2013a): in our case studies, typical calculation times for a threshold of 1E-3, 1E-4, 1E-5 and 1E-6 Msej (using a 2.67GHz Intel Core i7 laptop, running with MS Windows 7) were respectively 1, 3, 24 and 240 minutes. Figure S4.1 shows that a rigorous application of emergy algebra affects the results by as little as 3% (flow lost due to loop violation), meaning that the ratio between the emergy value of the inputs and the emergy value of the output(s) is 97%. The choice of the threshold influences the amount of flow lost due to sheer algorithmic constraints (flow lost due to minflow), while increasing (to a lesser extent) the amount of flow lost due to loop violation. From Figure S4.1, one can conclude that threshold values of 1E-4 or 1E-5 Msej are the best tradeoffs. 189

190 Flow Lost to Minflow Violations Flow Lost to Loop Violations 0% 5% 10% 15% 20% 25% Site 1 (2) Site 1 (3) Site 1 (4) Site 1 (5) Site 1 (6) Site 2 (2) Site 2 (3) Site 2 (4) Site 2 (5) Site 2 (6) Site A (2) Site A (3) Site A (4) Site A (5) Site A (6) Site B (2) Site B (3) Site B (4) Site B (5) Site B (6) Figure S. 4.1: Relative portion of the emergy value of flows lost due to the threshold level applied in SCALE and the loop violation. SI4.2. Comparative table of UEV CO NV, UEV SCALE and SED of technospheric inputs Table S4.1 lists the Unit Emergy Values (UEVs) calculated for the materials used by the WTPs and the production and regeneration of activated carbon (Arbault et al. 2013b). UEVs SCALE are calculated with a minflow of 1E-4 Msej (considering the high number of products, lower threshold values e.g. 1E-5 Msej would have drastically increased the overall calculation time). An important point, extensively discussed in the supplementary materials of Rugani et al. (2011a), concerns the SEF of mineral sodium chloride (NaCl): the value in the original publication is 9.89 E13 sej/kg (baseline 9.26), which is two orders of magnitude higher than the reference UEV found in literature (9.81 E11 sej/kg, Odum 1996, converted to baseline 9.26). Considering the high importance of NaCl in the technosphere, results highly depend on the choice of this value. The results presented below, both for SED and EME SCALE, are calculated with the UEV of mineral sodium chloride 9.81 E11 sej/kg. The table also includes ammonia (produced with two distinct technologies) and liquefied oxygen. Data of Table S4.1 were built under the following considerations: the minflow value of the retrieved UEVs SCALE is 1E-4 Msej. The SEF (and UEV SCALE ) of mineral NaCl was modified from 9.89 E12 to 9.81 E10 sej/kg, according to the rationale provided in Rugani et al. (2011a, supplementary material). 190

191 Supplementary information UEV CONV of electricity for UCTE (Union for the Co-ordination of Transmission of Electricity) was calculated following the procedure of Arbault et al. (2013b) for electricity in France, but with the UCTE production mix (Table S4.2). When UEV CONV were approximated with SED in Arbault et al. (2013b), no data were reported in Table S4.1 s EME CONV column. Table S. 4.1: UEVs retrieved from the application of the three methods on the man-made products used in WTPs and the production of activated carbon (x 1E11 sej/unit). Category Input Unit EME CONV Ref (EME CONV ) EME SCALE SED Electricity low voltage (FR) kwh 2.09 [a] Energy Chemicals Services Electricity medium voltage (FR) kwh 2.09 [a] Electricity, medium voltage (UCTE) kwh 3.88 [b] Electricity production mix (UCTE) kwh 3.88 [b] Hard coal mix kg [c] Hard coal, burned MJ 0.39 [c] Heavy fuel oil MJ 0.66 [d] Natural gas burned MJ 0.43 [e] Acrylic acid kg Activated carbon kg [a] Aluminium sulfate powder kg Ammonia, partial oxidation kg Ammonia, steam reforming kg Carbon dioxide, liquid kg Chlorine, gaseous kg [f] Hydrochloric acid (30%) kg Iron (III) chloride (40%) kg Lime, hydrated, packed kg 9.81 [f] Oxygen, liquid kg Phosphoric acid (85%) kg Potassium permanganate kg Quicklime, milled, packed kg 9.81 [f] Regenerated activated carbon kg [a] Sodium hydroxide (50%) kg [f] Sodium hypochlorite (15%) kg Steam kg [d] Sulfuric acid liquid kg Disposal, hard coal ash kg [c] Transport, lorry > 32t tkm 6.48 [g] Transport, lorry 16-32t EURO3 tkm 6.48 [g] Transport, lorry 3,5-16t tkm 6.48 [g] Transport, lorry 3,5-20t (CH) tkm 6.48 [g] References: [a]: Arbault et al. (2013b). [b]: this study (Table S 4.2). [c]: Odum (1996); with E6 J/kg coal. [d]: Bastianoni et al. (2009). [e]: Bastianoni et al. (2005). [f]: Campbell and Ohrt (2009). [g]: Buranakarn (1998); with 1 ton.mile = 907 kg x m = tkm. 191

192 Table S. 4.2: Electricity production mix for UCTE and UEVs CONV from literature (excluding human labor and services). % mix [h] UEV ( 1E4 sej/j) ref Cogeneration 0,95% - - Wind 2,01% 5,78 [i] Coal and lignite 31,15% 15,89 [i] Hydropower 13,56% 5,76 [i] Natural gas 16,60% 15,69 [i] Nuclear 31,28% 4,81 [f] Oil 4,43% 18,34 [i] Photovoltaic 0,03% - - UCTE mix 100,00% 10,77 References: [f]: Campbell and Ohrt (2009). [h]: ecoinvent v2.2 (2010), processes # , 7191, [i]: Brown and Ulgiati (2002). SI4.3. Decomposition of technospheric inputs UEV SCALE per resource category Figure S4.2 shows the ratio of SED to the emergy value of SCALE inputs, for each technospheric product of Table S4.1 and Figure 4.4, per resource category. It can be observed that chloridebased co-products and liquid oxygen show similar ratios for the most of the resource categories, and that land resources have a very low ratio for most products, contrary to nuclear, renewable energy and water resources. Fossil, metal and mineral resources show more heterogeneous behaviors. A more detailed, per-resource decomposition is necessary to deepen the analysis. 192

193 Supplementary information Figure S. 4.2: Ratios of SED value to the emergy value of SCALE inputs of technospheric products, per resource category Table S4.3 provides the detailed results of SCALE and the SED method applied to the technospheric inputs. 193

194 194 Table S. 4.3: Decomposition of SCALE output (UEV SCALE ), SCALE input (application of rule #2 only) and SED of technospheric inputs. Nfo:Non-renewable fossil resources. Nme: Non-renewable metal resources. Nmi: Non-renewable mineral resources. Nnu: Non-renewable nuclear resources. Ren: Renewable energy resources. Rla: Renewable land resources. Rwa: Renewable water resources. Tot: Total. Results in Msej/kg, except for electricity products (in Msej/kWh). Product labels refer to the nomenclature in the ecoinvent database.

195 Supplementary information SI4.4. Resources disregarded in SEF dataset due to double-counting SCALE relies on the SEF dataset to convert the results of graph-search algorithm into emergy values. Some resources are assigned a characterization factor equal to zero: the atmospheric resources listed in ecoinvent (CO 2, Krypton, Xenon) are considered as ground-state resources i.e. with no solar energy requirement, while biomass-related resources (airborne CO 2, soil organic carbon, biomass energy and wood resources) as well as direct solar energy, land transformation and land volume occupation are considered already accounted in land occupation SEFs. Emergy accounting was initially developed with a top-down approach: the annual baseline (the Earth energy budget expressed in emergy values) supports all geobiosphere processes, whose outputs are considered as co-products of the global system. This is consistent with the specific rationale for allocation (rule #2) in emergy algebra. Rule #4 was thus introduced to deal with double-counting issues e.g. of rain and wind, since both resources are driven by the same (atmospheric) processes. Most often, freshwater has the highest emergy value of the renewable resources used up in a coupled natural-human system, so that there is no need to account for other renewables. Though often justified, this assumption is by no means a rigorous application of emergy algebra, and should not be implicitly incorporated in an algorithm that claims coping with approximation in conventional emergy evaluation. In addition, UEVs of raw materials suffer from scarce variability and low representativeness: typically, only one value is consistently available from the emergy literature. These UEVs are not homogeneously calculated: for instance, petroleum-based resources are evaluated from a description of natural mechanisms involved in their formation (Bastianoni et al. 2005, 2009; Brown et al. 2011), while mineral resources are mostly based on a top-down approach, in which the annual emergy budget (baseline) is divided by the regeneration rate of the resource (Rugani et al. 2013). A bottom-up approach at the level of the geobiosphere could be implemented to retrieve adequate characterization factors of natural resources in compliance with the emergy algebra within SCALE; such approach would focus on modeling geobiosphere processes and their co-products, using material and energy flows, so that the application of emergy algebra would prevent double-counting of co-produced resources. As a first step, a model development could rely on the framework proposed by Rugani and Benetto (2012), in which the geobiosphere is modeled with an array of static geological and biological processes and driven by the three independent external energy sources (sunlight, tidal potential energy and crustal heat); the graph-search algorithm used in SCALE could be once again applied to this matrix framework. For the time being, this is however far from real applications due to difficulties in developing such a comprehensive model and retrieving the necessary data about the static geological and biological processes (Rugani and Benetto 2012). SI4.5. Application of the EcoLCA tool The EcoLCA tool was applied in order to get indications on potentially important emergy values of some ES that are not accounted for in the SCALE-based methodology. The following data were used to simulate the water industry. They are based on Site A information (Igos et al. 2013b): 195

196 Table S. 4.4: Data used in the EcoLCA software. Sector ID Sector name Expenses ( /yr) Oil and gas extraction 4, Power generation and supply 350, Manufacturing and industrial buildings 193, Other maintenance and repair construction 15, Industrial gas manufacturing 28, Other basic inorganic chemical manufacturing 153, Plastics material and resin manufacturing 13, Other miscellaneous chemical product manufacturing 62, Lime manufacturing 35, Offices supplies, except paper, manufacturing 5, Office Administrative services 15,000 Total production (m 3 /yr) 8,390,000 The quantitative results are not expected to be implemented in EME, but are just presented for interpretation purposes and for discussion on ES accounting in EME. Indeed, there are many discrepancies to remind in both the data used and the application of the method, among which: The EcoLCA method applies to the US economy, not the EU or France (location of the case study of WTPs in the present research). Data are retrieved for a single plant, while IO-based environmental assessment methods are meaningful for macroeconomic activities only. The correspondence between the exact expenses item in Site A and US commodity sector (Table S4.4) is not always straightforward. Besides this, it is shown (Tables S4.1 and S4.3) that chemicals can have very different UEVs SCALE, while most of them are aggregated into other basic inorganic chemical manufacturing and other miscellaneous chemical product manufacturing (except lime and plastics/polymers). It is important to remind that the EcoLCA method does not comply with emergy algebra: it uses matrix algebra operations which are common in the manipulation of Input-Output Tables (the Gosh inverse, see Zhang et al. 2010a), which rely on a linear relationship between inputs and outputs and a monetary allocation of multi-output commodities. It can be concluded that the EcoLCA method is conceptually analogous to the SED method, though the former is applied to IO-based LCI and the latter to process-based LCI. 196

197 Supplementary information SI5. Chapter 5: A semantic study of the Emergy Sustainability Index in the hybrid lifecycle-emergy framework SI5.1. A brief discussion on the evolution of EYR s definition and other research on ESI Odum (1996, p84) defined the Emergy Yield Ratio (EYR) as the ratio of the yield output emergy flow (Y) divided by the sum of the feedback emergy from the economy (M+S). It estimates the real (natural) wealth an activity is able to provide to the surrounding economy, by comparing the value of the system s output and that of external investments. Quoting Odum: if the emergy for the transformations is being supplied by the economy, the net emergy yield decreases (Odum 1996, p146). According to the maximum empower principle, if those systems prevail that produce more emergy and utilize it more efficiently, then systems with greater empower (useful emergy flow) may be better and more likely to continue. [ ] The displacement of primitive systems by economic ones is consistent with the maximum empower principle as long as there are available sources of cheap emergy to purchase (Odum 1996, p166). Consequently, a high EYR indicates an activity that yields more to its next larger system. The order and organization of the larger system may be more sustainable because it can use the higher emergy yield for its maintenance. However, a slight but important change occurred in EYR s definition: most, if not all, subsequent studies defined EYR as the total input emergy (instead of output emergy, as stated by Odum) divided by the emergy of imports (see e.g. Brown and Ulgiati 1997, 2004; Raugei et al. 2005; Ridolfi and Bastianoni 2008; Campbell and Garmestani 2012). Such change has no consequence in most studies on human activities that do not build feedbacks to the surrounding environment, as mentioned by Raugei et al. (2005) quoting Odum and Odum (2001, chapter 6): free harvests from the environment [ ] are not sustainable in the long run because the humans exploiting them do not reinforce the environmental production system. Therefore, the output emergy is equal to the total input emergy (Y = R+N+M+S). Should the activity develop such reinforcing feedbacks, e.g. through ecological engineering, Y becomes lower than R+N+M+S (see Figure S5.1). Odum s EYR compares the investments with the output yield, whereas the more recent one directly compares the investments with the natural resources exploited by the activity. Raugei et al. (2005) suggested renaming EYR as e.g. Emergy Appropriation Ratio to better match its meaning. Figure S5.1 shows an example that illustrates these two different EYRs. Raugei (2011) demonstrated that the later definition of EYR means that more emergy associated with a process' output could not be interpreted as a larger contribution to the economy, but rather as a larger responsibility in the use of resources (i.e. more appropriation of the past environmental work), thus questioning the directionality of the EYR. To this aim, he calculated that the EYR of a coal-based electricity power plant was higher than photovoltaic, illustrating that the benefit for the economy was not well described by the EYR. In their response, Ulgiati and Brown (2012) addressed the apparent contradiction, by stating that the apparent low EYR of processes relying on renewables indicates a larger investment to extract energy and produce electricity, which deprives other processes from available sources, thus causing the global economy to shrink. The three authors overcame the divergent positions by adequately revising the definition of EYR of a process. In doing so, they adopted a lifecycle-based boundary definition and distinguished 'foreground investments' from 'background investments' in replacement from local vs. imported 197

198 inputs (Brown et al. 2012). The presence of loops in the network of technological processes (Marvuglia et al. 2013a, 2013b), which can be assimilated to feedback controls, may re-open the discussion on which definition of EYR should be adopted. Figure S. 5.1: Comparison between EYR as defined by Odum (1996) and the later version commonly adopted. The presence of reinforcing feedbacks from the activity to the natural environment leads to different values. Modifications of ESI were also recently proposed. In particular, Harizaj (2011) derived a sustainability inequality from the mathematical definition of the ESI, in order to formulate a simplified criterion for sustainability. The author concluded that a system is sustainable if R >> (M+S) >> N. Despite the usefulness of this formulation for quick decision-making, this statement was found inexact (Brown and Ulgiati 2011), because other combinations are possible to maximize ESI. Lu et al. (2003, 2007) suggested a modified version of EYR and ESI by multiplying them with the Emergy Exchange Ratio, EER, defined as the ratio of the emergy value of the yield product by the solar emergy of the money paid by the buyer (Odum 1996, p84). Because EER is influenced by market, culture and ethics, it is a social indicator (Lu et al. 2003), complementary to the EYR and ELR. The approach is appealing because it incorporates the perceived value by the larger system (through the money paid by the user) in the measurement of sustainability. Indeed, an activity that provides no perceived value for the larger system is unlikely to prevail in the long term. Despite it may appear as an important contribution toward revisiting ESI, we decided to exclude it from the present study because EER does not reflect the physical limits of the Earth but rather the distance between perceived (user-side) and natural (donor-side) values. We nevertheless recommend using EER and ESI complementarily and separately. 198

199 Supplementary information SI5.2. Inventory data and results of the case studies Table S. 5.1: Inventory of material inputs for the case studies. (operation phase for Sites 1, 2, A, B and infrastructure for Sites A and B). Data inventoried per m 3 of potable water produced (Arbault et al. 2013b; from Igos et al. 2013a, 2013b). Process (Operation phase) Unit Site 1 Site 2 Site A Site B Electricity, low voltage, at grid [FR]* kwh 8.96 E E E E-01 Electricity, medium voltage, at grid [FR] kwh 3.79 E-01 Heavy fuel oil, burned in power plant [FR] MJ 6.49 E-02 Activated carbon kg 3.60 E E E E-03 Regenerated activated carbon kg 3.60 E E-03 Acrylic acid, at plant [RER]** kg 1.74 E E-04 Aluminium sulphate, powder, at plant [RER] kg 1.79 E E-02 Carbon dioxide liquid, at plant [RER] kg 2.10 E E-02 Chlorine, gaseous, diaphragm cell, at plant [RER] kg 1.39 E-03 Iron (III) chloride, 40% in H 2O, at plant [CH]*** kg 6.50 E E-02 Lime, hydrated, packed, at plant [CH] kg 7.46 E E-02 Quicklime, milled, packed, at plant [CH] kg 4.73 E-02 Phosphoric acid, industrial grade, 85% in H 2O, at plant [RER] kg 8.66 E-05 Potassium permanganate, at plant [RER] kg 3.59 E E-03 Sodium hydroxide, 50% in H 2O, production mix, at plant [RER] kg 1.16 E E E-04 Sodium hypochlorite, 15% in H 2O, at plant [RER] kg 8.89 E E E-04 Sulphuric acid, liquid, at plant [RER] kg 6.88 E E E-04 Process (Infrastructure) Unit Site 1 Site 2 Site A Site B Building, hall, steel construction/ch/i**** m E E-06 Building, hall/ch/i m E E-06 Building, multi-storey/rer/i m E-06 Concrete, normal, at plant/ch m E E-05 Copper, at regional storage/rer kg 1.54 E E-05 Foam glass, at plant/rer kg 8.00 E E-07 Epoxy resin, liquid, at plant/rer kg 3.52 E E-04 Glass fibre reinforced plastic, polyester resin, hand lay-up, at plant/rer kg 5.79 E E-05 Nylon 6, at plant/rer kg 5.52 E E-05 Polyethylene, HDPE, granulate, at plant/rer kg 5.21 E E-04 Polyethylene, LLDPE, granulate, at plant/rer kg 3.29 E E-06 Polyvinylchloride, at regional storage/rer kg 1.23 E E-05 Styrene-acrylonitrile copolymer, SAN, at plant/rer kg 8.36 E E-06 Synthetic rubber, at plant/rer kg 8.30 E-07 Chromium steel 18/8, at plant/rer kg 1.01 E E-05 Reinforcing steel, at plant/rer kg 2.38 E-03 Reinforcing steel, at plant/rer kg 1.30 E E-03 Steel, low-alloyed, at plant/rer kg 3.09 E E-04 *[FR]: France; **[RER]: Rest of Europe; ***[CH]: Switzerland; ****I: Infrastructure process. 199

200 Table S. 5.2: Labor and service inputs of the case studies. Data are in /m 3 of potable water produced. Economic input Direct Labor Energy Reagents Exploitation expenditures Background Maintenance Sludge management Annual allocation for renewal (infrastructure) Site E E E E E E-03 0 Site E E E E E E-03 0 Site A 2.07 E E E E E-02 Site B 2.53 E E E E E E E-02 Emergy-money ratio for France in 2004: 1.78E+12 sej/ ; percent renewable in French economy: 0.5% (Sweeney et al, 2007) Table S. 5.3: Total emergy value of inputs (foreground + background), calculated with SCALE. Data are in sej/m3 of potable water produced (baseline 9.26 E24 sej/yr). NFo NMe NMi NNu REn RLa RWa Site E E E E E E E11 Site E E E E E E E11 Site A 3.22 E E E E E E E11 Site B 2.78 E E E E E E E11 Site A w. infra 3.46 E E E E E E E11 Site B w. infra 2.89 E E E E E E E11 NFo: Non-renewable, fossil resources. NMe: Non-renewable, metal ores. NMi: Non-renewable, mineral resources. NNu: Non-renewable, nuclear energy resources. REn: Renewable, energy resources. RLa: Renewable, land resources. RWa: Renewable, water resources. Table S. 5.4: Emergy value of inputs: local resources (L), direct investments (D), background investments (B) and total inputs (U, equal to output UEV in these case studies), including mate rials and labor, for the case studies. The renewable and non-renewable parts are marked with r and n, respectively. Data are in sej/m 3 of potable water produced. Lr Ln Dr Dn Br Bn U Site E E E E E E12 Site E E E E E E12 Site A 4.02 E E E E E E12 Site B 4.93 E E E E E E12 Site A w. infra 4.02 E E E E E E12 Site B w. infra 4.93 E E E E E E12 Table S. 5.5: Resulting EYR0, EYR1, ELR0, ELR1 for the case studies (dimensionless EYR0 EYR1 ELR0 ELR1 Site Site Site A Site B Site A w. infra Site B w. infra SI5.3. Foreground input tracking algorithm and illustration with the case studies Supplementary information shapefile is available as Appendix of the original publication, on the editor s website. The SI file contains the algorithm used to track foreground inputs in the network of technological processes (listed in the sheet Raw data from ecoinvent ) used for the case studies presented in 200

201 Supplementary information Chapter 5, up to the natural resources and by-products used up (also listed in the sheet Raw data from Ecoinvent ), for a given production system (described in the sheet DemandVector ). Macros can be freely edited. The production system is called the Demand Vector. The user fills the inputs of the production system: ID in ecoinvent in col. A, personal label in col. B, amount (in the unit defined in ecoinvent) in col. C. The sheet Raw data from ecoinvent lists all processes and resources involved in the production chain. Information from cols. A-K is directly downloaded from the website database version2.2 (restricted user access). Processes included in the Demand Vector are listed (ID, label, unit and region respectively in cols. A, B, C and D according to the ecoinvent features). For the sake of readability, these elements are mentioned on a specific line. The following lines represent all inputs of this process (ID, category, subcategory, label, location, unit and amount respectively in cols. E, F, G, H, I, J and K); emissions can be disregarded to clean the table. The user must indicate in col. L if the input is a direct investment (mark 'F' for foreground ). Inputs marked with an 'E' (for energy foreground ) in col. L are direct investments only when the 'materials & energy' option is considered (cf main paper). They are not accounted for by the algorithm. To include them, simply replace all E in col. L by F. The letter 'R' in col. L refers to ecoinvent elements that are resources (some of them are absent from ecoinvent; their ID is custom). The letter 'C' refers to recycling processes. When the algorithm meets the letter 'R' or 'C', it stops and writes a new line in the OutputVector sheet. The Command Sheet proposes 3 buttons. The first one is to ensure that the sheets DemandVector and Raw data from ecoinvent are correctly filled out, i.e. no process of the production chain or resource is missing or is listed twice. The second button runs the algorithm and fills out the sheet OutputVector with each pathway from an element of the Demand Vector to the natural resource or the recycling process of an industrial co-product. The third button aggregates this Output Vector to show the total amount (in physical units) of each resource per element of the Demand Vector. The user may note that the algorithm parses data using their ID, which must not be changed. The process and resources name can be changed, and it will be reflected in the Output Vector. The user has to organize his (her) data according to the template provided: processes should be ordered by their ID number, with the process information line preceding the list of all inputs. The algorithm is in agreement with SCALE calculation method, except a particular case which is the possible occurrence of feedback loops in the production chain of direct investments. In this case the Excel algorithm could lead to overestimate D. In order to compute more correctly D and B flows, SCALE could be run with only D flows present in the network of processes, then run again with all flows, except those that characterize the natural resources considered as direct investment (for instance, modeling mineral sodium chloride extraction with all inputs from technological processes but disregarding the flow of mineral resource from nature). D and B would thus be computed independently from each other with respect to the emergy algebra, but could not be summed up because of double-counting. However, this limitation does not hamper the methodological approach proposed in this paper. It remains an important point to keep in mind if the algorithm is to be implemented into SCALE. 201

202 SI6. Chapter 6: A first global and spatially-explicit emergy database of rivers and streams based on high-resolution GIS-maps SI6.1. On solar radiation and wind energy inputs SI Solar energy Original data were retrieved from NASA (National Aeronautics and Space Administration) as a text file (Table 6.1), with an original resolution of 1 (zonal averages), in kwh/m 2 /day. They represent monthly averages (over the 22-year period from 1983 to 2005) of the solar diffuse radiation incident on a horizontal surface of the Earth, under all-sky conditions. For each point of the downloaded file, the yearly average radiation (total of each monthly average) was calculated in kwh/m 2 /day, and coordinates were adjusted. Then, it was converted to a point shapefile, further rasterized into a 1 -resolution map. The resulting map was converted into a 30 arcsec raster, masked with the elevation map and assigned the same reference coordinate system (WGS84). Finally, the kwh/m 2 /day map was converted into MJ/grid cell/yr, using the grid cell area raster using the following equation: Output map (MJ/grid cell/yr) = Input map (kwh/m 2 /day) grid cell area map (km 2 per grid cell) 1 E6 (m 2 /km 2 ) (days/yr in the data period) 3.6 (MJ/kWh) (S6.1) Total solar irradiation on terrestrial surface (without Antarctica) was 2.85 E23 J/yr. Odum (1996) used a net (i.e. total minus albedo) solar input value of 3.93 E24 J/yr. Considering that terrestrial land (without Antarctica) composes 26.4 % of the globe (51.0 E7 km 2 ), solar irradiation over the area covered in this study following Odum s top-down calculation was 1.04 E24 J/yr. The large difference (approx. 70%) could be partly explained by the different types of data used: Odum considered the normal radiation on top of the atmosphere (minus surface albedo), while measures of the diffuse radiation that reaches the ground were used in this study. Such choice was motivated by our willingness to account only for the solar energy available on the ground for terrestrial natural mechanisms, which interfere for the formation of streams and drainage areas. The large difference between both datasets suggests that 70% of incident sunlight can be assumed to be directly caught up by the atmospheric processes, and thus never reaches the ground. However, despite the fact that the data used in the present study were arguably more accurate than the global average value proposed by Odum (1996), we decided to discard our results on solar radiation for two reasons: first, it was found that the original dataset of solar radiation presented an important discontinuity at latitudes 45N and 45S, as shown in Figure S6.1. Unfortunately, the reason for this discontinuity remained unexplained by the data provider, and no suitable dataset of better quality could be found. Second, we found out that the emergy value of available solar radiation reaching the ground, i.e. assumed higher than the fraction used up my terrestrial natural processes, is negligible in comparison with rain emergy, except in a few hot desert areas such as Western Sahara. Noticeably, this study was performed according to the 9.26 E24 sej/yr baseline (Campbell 2001), i.e. the lowest available, so that the relative contribution of solar radiation emergy was the highest possible. Therefore, the estimation of solar radiation energy flows (and 202

203 Supplementary information the repartition of its usage among atmospheric and terrestrial natural processes), needs further investigations, but this caveat does not affect our resulting database. Figure S. 6.1: Total solar radiation per latitude (J/yr) in the original dataset (see Table 6.1 for source). SI Wind energy Data were made available by CRU (Climatic Research Unit) as a text file dataset of monthly average wind speed (in m/s, 10 m above ground), averaged over a 30-year period ( ) and with an original resolution of 10 arcmin. The yearly average wind speed was computed for each data point, then converted into a point shapefile, further rasterized into a 10 arcmin map (cell size: 1/6 = ). The resulting map was converted into a 30 arcsec raster (each 30 arcsec grid cell within a 10 arcmin cell gets the same value), masked with the elevation map and assigned the same reference coordinate system (WGS84). Wind speed was converted to wind energy using equation (S6.2) (Emanuel 1999; Mellino et al. 2014). (S6.2), where E wind,k is the annual available wind energy in grid cell k (J/grid cell/yr), A k is the area of grid cell k (in m 2 /grid cell), ρ air is the air density (set at kg/m 3 ), C D is the drag coefficient (assumed 0.002) 10 ; v k is the annual average wind speed over k (m/s), t is the duration to consider (3.15 E7 s/yr). Note that the grid cell area map was in km 2 /grid cell: the formula used (SI6.2) incorporated a 1 E6 (m 2 /km 2 ) multiplication factor. Finally, the E wind map was converted into MJ/grid cell/yr, then converted into sej/grid cell/yr using a transformity of 1,433 sej/j (Odum et al. 2000, baseline-converted). Figure S6.2 shows the resulting map: 10 The drag coefficient is not constant over the Earth, but depends on the land configuration (soil, altitude, steepness, etc.). It was estimated ranging between and (Garratt 1977). For the sake of simplicity, we assumed a constant, mean value equal to

204 Figure S. 6.2: Emergy input of wind force, per grid cell (30 arcsec) and per year. Total terrestrial wind energy was 3.93 E20 J/yr, which gave an average wind speed on land of 3.22 m/s from equation (S6.2). Odum (2000) used a global wind power of 0.4 W/m 2, equivalent to a global average wind speed of 5.47 m/s and global wind energy of 6.54 E21 J/yr. If wind force was globally invariant over lands and seas, the total wind energy over the area considered in this study would have been estimated to E20 J/yr if the rationale of Odum (2000) were followed. This value is 4.4 times higher than the total calculated in this work. However, wind speed is lower on land. The cubic power of wind speed ratio global vs. terrestrial is 4.9, which approximately corresponds to the ratio between terrestrial wind energy calculated using Odum s global average vs. that calculated in this work. Therefore, this comparison validated the present result. The consequence is that the large majority of wind power (93%) is found over the oceans and Antarctica, as illustrated by the difference between terrestrial wind speed and the global average (see e.g. Brandt-Williams and Brown 2011). However, the use of equation S6.2 was here motivated by its prior use by emergy scholars (Mellino et al. 2014), but should be questioned, because the cubic power of an average value is not equal to the average value of cubic powers, though it may be suitable for estimations at the order of magnitude. Figure S6.2 displays the emergy value of the wind energy available in each 30 arcsec grid cell, which can be assumed higher than the wind energy actually used up by terrestrial natural processes that contribute to the generation of streams and drainage basins. In addition, available wind force emergy was largest in 6.7% of grid cells (located in hot desert areas as well as North Pole and South-Argentina, where precipitation was apparently very low), as shown in Figure S6.3. Considering i) the arguable approach to calculate total available wind energy, ii) its relative low emergy value as compared to rain inputs, and iii) that only a fraction of wind energy is actually used up by terrestrial mechanisms, we decided to discard the wind input from the calculation of emergy accumulation along streams. It is important to notice that this does not affect the resulting database, except in areas of low precipitation, where streams scarcely form. 204

205 Supplementary information Figure S. 6.3: Highest emergy value of available wind energy (green), rain chemical potential (blue) and rain geopotential (brown). SI6.2. Processing steps in ArcGIS SI Grid cell area Gives the orthonormal area (km 2 ) of each grid cell. Input map: elv_1km (elevation map from world_clim.org) Spatial Analsyt Tools > Raster Creation > Constant Raster o constant value: 64 (= flow direction bottom to top) o output cell size: elv_1km o output extent: same as elv_1kmv o output file: const1km Spatial analyst Tools > Hydrology > Flow accumulation o input flow direction raster: const1km o output file: flowacc Spatial Analyst Tools> Map Algebra > Raster Calculator o flowacc=0 <-> ymap = arcsec; flowacc=17999 <-> ymap =89-15arcsec o formula: "flowacc"* o output file: ymap Spatial Analyst Tools> Map Algebra > Raster Calculator o (Cell size in km)^2 multiplied by the cos of latitude (in rad) o Cell area in equator = [40000 km / (360 * 60 * 2)]^2 = km2 o formula: *Cos("ymap" * / 180) o output file: area_cell Spatial Analyst Tools> Map Algebra > Raster Calculator o to produce a grid cell area map with the coverage of elv_1km o elv_1km - elv_1km + area_cell Output file: area1km2 SI Solar energy Data from NASA/SS, original resolution 1degree, units kwh/m 2 /day Data are regional average, not point data Point value is the annual average insolation at the lower left corner of the 1x1 degree region 205

206 o Data were corrected (+0.5 lat, +0.5 lon) Output file: Diffuse_radiation_centered.txt Conversion to shp using tools > add xy data Conversion into raster using ArcMap - o conversion tool > to Raster > point to raster o Cell alignment MEAN o Cellsize 1 Output file: sun_1deg Conversion from 1, kwh/m 2 /day map into a 1km MJ/yr/grid cell Open first elv_1km for georeferencing Input maps : sun_1deg; area1km2 Spatial Analyst Tools> Map Algebra > Raster Calculator o formula: "elv_1km" - "elv_1km" + "sun_1deg" o Environment option: Output Coordinates: same as Layer elv_1km Processing Extent: same as layer elv_1km Cartography: Cartographic Coordinate System: same as elv_1km Raster Analysis: cell size same as layer elv_1km; Mask: none o output file: tempsun Spatial Analyst Tools> Map Algebra > Raster Calculator o Output map = tempsun (kwh/m2/day)* grid cell area (km2 per grid cell)* (m2/km2) * (days/yr in the data period) * 3.6 (MJ/kWh) o formula: "temp_sun" * "area1km2" * o Environment option: Output Coordinates: same as Layer elv_1km Processing Extent: same as layer elv_1km Cartography: Cartographic Coordinate System: same as elv_1km Raster Analysis: cell size same as layer elv_1km; Mask: none Output File: SunMJ_yr_px SI Wind energy Data from CRU, monthly averaged aggregated into annual averages, original resolution 10 min, unit in m/sec (average wind speed), values 10m above ground. Monthly averages, converted to annual average Output file: wind_avg.txt Conversion to shp using tools > add xy data Conversion into raster using ArcMap - o conversion tool > to Raster > point to raster o Cell alignment MEAN (1 value per cell so the cell alignment type does not matter) o Cellsize Output file: wind_10min Conversion from 10 min, m/s cell into 1km MJ/yr/grid cell Open first elv_1km for georeferencing o Input maps : wind_10min; area1km2 Spatial Analyst Tools> Map Algebra > Raster Calculator o formula: "elv_1km" - "elv_1km" + "10min" o Environment option: 206

207 Supplementary information Output Coordinates: same as Layer elv_1km Processing Extent: same as layer elv_1km Cartography: Cartographic Coordinate System: same as elv_1km Raster Analysis: cell size same as layer elv_1km; Mask: none o output file: tempwind Spatial Analyst Tools> Map Algebra > Raster Calculator o Output map = grid cell area (km2 per grid cell)* (m2/km2) *1.225 (kg/m3 air density)*0.002 (surface drag area assumed)* tempwind (m/s)^3*3.15e7 (sec/yr)/1e6 (MJ/J) o formula: "tempwind" * "tempwind" * "tempwind" * "area1km2" * o Environment option: Output Coordinates: same as Layer elv_1km Processing Extent: same as layer elv_1km Cartography: Cartographic Coordinate System: same as elv_1km Raster Analysis: cell size same as layer elv_1km; Mask: none Output File: wind_mj_yr_px SI Precipitation Data from world_clim.org, 1km resolution for each month. Values in mm/month Raster calculator (in Toolbar Spatial Analyst), sum of all monthly values mm/yr Output file: rain_mm-yr Conversion from 30 arc sec, mm/yr/grid cell into 1 km, m3/yr/grid cell Open first elv_1km for georeferencing Input maps : rain_mm-yr; area1km2 Spatial Analyst Tools> Map Algebra > Raster Calculator o formula: "elv_1km" - "elv_1km" + " rain_mm-yr " o Environment option: Output Coordinates: same as Layer elv_1km Processing Extent: same as layer elv_1km Cartography: Cartographic Coordinate System: same as elv_1km Raster Analysis: cell size same as layer elv_1km; Mask: none o output file: temprain Spatial Analyst Tools> Map Algebra > Raster Calculator o Output map = temprain (mm/yr)*0.001 (m/mm)* grid cell area (km2 per grid cell)* (m2/km2) o formula: "temprain" * "area1km2" * 1000 o Environment option: Output Coordinates: same as Layer elv_1km Processing Extent: same as layer elv_1km Cartography: Cartographic Coordinate System: same as elv_1km Raster Analysis: cell size same as layer elv_1km; Mask: none Output File: rain_m3_yr_px SI Rainfall, chemical exergy Chemical exergy of rain fall is the rainfall mass (in g) multiplied by the Gibbs free energy (in J/g), which depends on the temperature. Spatial Analyst Tools > Map Algebra > Raster Calculator 207

208 o formula: Rain (MJ/month/grid cell) = Volume (m3/month/grid cell) *1E6 (density: g/m3) *R (8.314 J/mol/K) * (273+T (C)) * ln (c0/c1) / 18 (g/mol)*1e-6 (MJ/J) with c0 = 999,990 ppm and c1 = 965,000 ppm) Rain (MJ/month/grid cell) = Volume (m3/month/grid cell) * (273 + T(month, C) * Volume (m3/month/grid cell) = pp (mm/month/grid cell) * cell area (km2) * 1E- 3 (m/mm) * 1E6 (m2/km2) Rain (MJ/month/grid cell) = pp (mm/month/grid cell) * (273 + T(month, C) * Temperature maps in C *10 Rain (MJ/month/grid cell) = pp (mm/month/grid cell) * ( T(month, C *10) * Rain (MJ/yr/grid cell) = sum of (Rain (MJ/month/grid cell) Rain (MJ/yr/grid cell) = sum of (Rain (pp/month/grid cell) * ( T(month, C*10) ) * cell area (km2) * ("prec_1" * ( "tmean_1") + "prec_2" * ( "tmean_2") + "prec_3" * ( "tmean_3") + "prec_4" * ( "tmean_4") + "prec_5" * ( "tmean_5") + "prec_6" * ( "tmean_6") + "prec_7" * ( "tmean_7") + "prec_8" * ( "tmean_8") + "prec_9" * ( "tmean_9") + "prec_10" * ( "tmean_10") + "prec_11" * ( "tmean_11") + "prec_12" * ( "tmean_12")) * "area1km2" * o Environment option: Output Coordinates: same as Layer area1km2 Processing Extent: same as layer area1km2 Cartography: Cartographic Coordinate System: same as area1km2 Raster Analysis: cell size same as layer area1km2; Mask: none Output File: chem_mj_yr_px SI Evapotranspiration Data from Numerical Terradynamic Simulation Group, University of Montana. Values in1e-4m/yr (multiply by 0.1 to get data in mm/yr). Data only for terrestrial vegetated areas; other land (urban, deserts, surface waters) get a value of raster file per year. Make raster: o In ArcCatalog: right_click on the tiff file and export > raster to different format output name with no extension to make an esri raster options: No Data value = o Output files: et_[year] Clean raster values higher than 32,000 ( =3,200 mm/yr) o Spatial Analyst Tools > Conditional > SetNull input conditional raster: et_[year] expression: "VALUE" >32000 input false raster: et_[year] o Output file: [year]_clean Average ET value o Spatial Analyst Tools > Local > Call Statistics input rasters: the 11 [year]_clean statistics type: Mean ignore NoData: checked Output file: ET_terr 208

209 Supplementary information ET_terr gaps on urban, areas, deserts and surface waters. Aggregate data with a ~10min resolution then convert to a point shapefile to interpolate with IDW. Then use map algebra to select only grid cells that represent terrestrial land, keeping original ET in priority and interpolated ET otherwise. Spatial Analyst Tools > Generalization > Aggregate o Input Raster: ET_terr o Cell factor: 20 o Aggregation technique: MEAN o expand extend: checked o ignore NoData: checked o output file: aggreg10min Conversion Tools > From Raster > Raster To Point o input raster: aggreg10min o output point feature: point10min.shp Spatial Analyst Tools > Interpolation > IDW o input point features: point10min o Z value field: Grid_Code o Output cell size: elv_1km o Power: 2 o Search radius: Variable o Number of points: 30 o Maximum distance: none o Environment option: Output Coordinates: same as Layer elv_1km Processing Extent: same as layer elv_1km Cartography: Cartographic Coordinate System: same as elv_1km Raster Analysis: cell size same as layer elv_1km; Mask: none o output file: IDW10min Spatial Analyst Tools> Map Algebra > Raster Calculator o Output map = if (ET_terr <>0, then et_terr; otherwise IDW10min) *0.1 (mm/yr)*0.001 (m/mm)* grid cell area (km2 per grid cell)* (m2/km2) o formula: "area1km2" * 100 * Con(IsNull("et_terr"), "idw10min", "et_terr") o Environment option: Output Coordinates: same as Layer elv_1km Processing Extent: same as layer elv_1km Cartography: Cartographic Coordinate System: same as elv_1km Raster Analysis: cell size same as layer elv_1km; Mask: none Output File: ET_m3_yr_px SI Emergy input maps Physical maps are converted into emergy values, using transformities from emergy folios 1 &2 (Odum, 2000; Odum et al, 2000), converted to baseline 9.26 E24 sej/yr Spatial Analyst Tools > Map Algebra > Raster Calculator o "wind_mj_yr_px" * 1433 o Output file: wind_msej Spatial Analyst Tools > Map Algebra > Raster Calculator o "chem_mj_yr_px" * o Output file: chem_msej Spatial Analyst Tools > Map Algebra > Raster Calculator o "rain_m3_yr_px" * Con("elv_1km" < 120, 78479, * "elv_1km" * " elv_1km" * "elv_1km" ) 209

210 o Output file: Gpot_Msej Which is max: Spatial Analyst Tools > Map Algebra > Raster Calculator o Con("wind_msej">"chem_msej",Con("wind_msej">"gpot_msej",2,4),Con("chem_ msej" >"gpot_msej",3,4)) o Output file: highest SI Flow direction Direction map from elv_1km gave wrong results due to a too high aggregation. A more precise map (STRM_250m) was used, and aggregated into a 1 km map to keep calculation time of flow accumulation reasonable. The 1-km grid cell value retained is the lowest of the elevation in the 250m STRM map. 3 maps (W, NE, SE). Spatial Analyst Tool > Generalization > Aggregate o Cell factor 4 o Aggregation technique: MINIMUM o Expand if needed: checked o Ignore NoData: checked o output extent = elv_1km o output maps : temp\[w/ne/se]aggr Spatial Analyst Tools> Map Algebra > Raster Calculator o formula: Con(IsNull("se_aggr2"),Con(IsNull("ne_aggr2"),Con(IsNull("w_aggr2"),"elv_1km","w_ aggr2"),"ne_aggr2"),"se_aggr2") o Environment option: Output Coordinates: same as Layer elv_1km Processing Extent: same as layer elv_1km Cartography: Cartographic Coordinate System: same as elv_1km Raster Analysis: cell size same as layer elv_1km; Mask: none o outputfile: world Spatial Analyst Tools> Map Algebra > Raster Calculator o formula: Con(IsNull("elv_1km"),"elv_1km","world") o Output File: elv_low Spatial Analyst Tools > Hydrology > Fill o Input file: elv_low o No option o Processing time: 55min o Output map : fill1km Spatial Analyst Tools > Hydrology > Flow Direction o Input file: fill1km o No option o Processing time: 24min Output map: dir_low SI Stream flow It is assumed that on the long term, land water reservoirs such as glaciers, lakes and groundwater are at equilibrium. Therefore, runoff could be estimated as the difference between precipitation and evapotranspiration, cumulated over the catchment area. 210

211 Supplementary information Runoff is estimated as the cumulated rain minus evapotranspiration (which can be negative for some grid cells) over the catchment area of a grid cell. Flow accumulation tool gives the accumulated values entering the cell: need to add the cell value to get total value of flow leaving the cell. Input maps: dir_low; rain_m3_yr_px ; ET_m3_px_yr Spatial analyst > Hydrology > Flow Accumulation o Input flow direction: dir_low o Input weight raster: rain_m3_yr_px o Output data type: float o Output File: rain_in Spatial analyst > Hydrology > Flow Accumulation o Input flow direction: dir_low o Input weight raster: ET_m3_yr_px o Output data type: float o Output File: ET_in o Processing time: >6h Spatial Analyst Tools> Map Algebra > Raster Calculator o formula: rain_in + rain_m3_yr_px - ET_in - ET_m3_yr_px Output file : runoff_out SI Catchment area Spatial analyst > Hydrology > Flow Accumulation o Input flow direction: dir_low o Input weight raster: area1km2 o Output data type: float o Output File: area_in o Processing time: 6h45min Spatial Analyst Tools> Map Algebra > Raster Calculator o formula: ( area_in + area_1km2 ) Output File: area_out SI Flow accumulation or rain emergy Chemical emergy and geopotential emergy are cumulated using flow acc, then only data corresponding to streams are kept. Flow accumulation function of these two files o chem_msej chem_in o gpot_msej gpot_in then o (chem_in + chem_msej) * stream_1e6m3 chem_out; o (gpot_in + gpot_msej) * stream1e6m3 gpot_out SI Modeling rivers Streams order Keep only grid cells that are part of a stream with flow higher than 1E6m3/yr (31.7 l/s) Con(runoff1_out > ,1) o output file: stream_1e6m3 Spatial Analyst > Hydrology > Stream Order o input files: stream_1e6m3; dir_low 211

212 o o Strahler output file: stream_order_6 Filter the accumulation rasters to keep only values in streams Input files: stream_1e6m3; runoff_out ; area_out ; chem_out; gpot_out Spatial Analyst Tools> Map Algebra > Raster Calculator o formula: stream_1e6m3 - stream_1e6m3 +[accumulation raster] o Output File: [accumulation raster]f Calculate UEV and empower density: Max of the two emergy-accumulated rasters Spatial Analyst Tools> Map Algebra > Raster Calculator o con(chem_outf > gpot_outf, chem_outf, gpot_outf) o output file: max_em_outf UEV o Spatial Analyst Tools> Map Algebra > Raster Calculator max_em_outf / runoff_outf output file: UEV Empower density: o Spatial Analyst Tools> Map Algebra > Raster Calculator max_em_outf / area_outf output file: emp_density Build shapefile Spatial Analyst > Conditional > SetNull o stream_order6 o value =1 o output file: str_ord_1_out Spatial Analyst > Hydrology > Stream_to_Feature () o input: str_ord_1_out o processing time: 2h47 o output file: str_to_feat Data Management Tools > Features > Feature_Vertices_to_Points o input: str_to_feat o processing time: 23min o MID o output file: feat_to_pts Assign data to points: o Spatial Analyst Tools > Extraction > Extract Values to Points this will create a new table for each value extracted Input rasters: stream order (already in the point shapefile); runoff_outf; area_outf ; chem_outf; gpot_outf; UEV; emp_density Create new fields in the shapefile o Join files o Copy values in the new fields o Remove join SI Aggregation over territories For each territory shapefile (with correct projection): 212

213 Supplementary information Spatial Statistics Tools > Zonal Statistics as Table o shp as zone input (country, region, watershed) o in_value raster : UEV, emp_density, area1km2 o out_table: [Shp]_[in_value] o Ignore no data: checked o statistics type: all (includes mean, majority, max, median, min, range, std, sum, variety) Assign data to polygons: o Open shapefile attribute table o Create new fields (min_uev, max_uev, avg_uev, stdev_uev, Avg_EmPden, Std_EmPden, area) o Join output tables o Copy relevant values in the new fields from UEV: min, max, mean, std from emp_density: mean, std from area1km2: sum o Remove joins Major basins: some polygons were errors. Menu Customize > Toolbars > Editor; Editor > Start Editing; then in the attribute table, the erroneous polygons can be deleted. Aggregation weighted with population cell resolution 2.5 Spatial Analyst > Generalization > Aggregate o input: uev; population o by factor 5 o take value MAX o Environment option: Output Coordinates: same as Layer population Processing Extent: same as layer population Cartography: Cartographic Coordinate System: same as population Raster Analysis: cell size same as layer population; Mask: none o output file: uev_2-5min Map algebra > Raster calculator o population map with mask uev_2-5min o uev_2-5min - uev_2-5min + population o output file: pop_mask Map algebra > Raster calculator o uev_2-5min * pop_mask o output file: uev_2-5bypop Zonal statistics: sums of population, of pop_mask and of uev_2-5bypop Assign data to polygons: o Open shapefile attribute table o Create again 4 new fields (tot_pop, pop_mask, uevbypop, UEV_pop) o Join output tables o Copy relevant values in the fields tot_pop, pop_mask and uevbypop o Remove joins o Calculate: UEV_pop = uevbypop / pop_mask o Delete fields uevbypop and pop_mask SI6.3. Emergy database of rivers: Shapefile Supplementary information shapefile is available as Appendix of the original publication, on the editor s website. 213

214 Shapefiles can be read with most GIS software. If no GIS software is installed, it is suggested to use ArcGIS Explorer Desktop ( It is free. To read the file on ArcGIS Explorer, uncompress then copy files in C:\Streams\ Caution: another location (such as C:\Streams\Streams\) would not work! Then open ArcGIS Explorer, and select the layer file to display (stream order or UEV) It takes a few minutes to load... Zoom on the region of interest, click on a segment to display the modeled characteristics. Table S. 6.1: List of calculated parameters for each stream segment. Field Description Unit Stream_ord Stream order - Flow Flow rate m 3 /yr Catch_area Upstream basin catchment area km 2 Chemical Accumulated emergy of rain chemical potential Msej/yr Geopot Accumulated emergy of rain geopotential Msej/yr UEV Unit Emergy Value Msej/m 3 EmDensity yearly average emergy input per unit area of the drainage basin Msej/yr/km 2 SI6.4. Emergy database of rivers: Territorial averages Supplementary information shapefile is available as Appendix of the original publication, on the editor s website. Shapefiles can be read with most GIS software. If no GIS software is installed, it is suggested to use ArcGIS Explorer Desktop ( It is free. To read the file on ArcGIS Explorer, uncompress then copy files in C:\Territories\ Caution: another location (such as C:\Streams\Streams\) would not work! Then open ArcGIS Explorer, and select the layer file to display (Countries, Provinces or Major Basins). Click on a polygon to display the population-weighted average UEV (UEV_pop, Msej/m 3 ). Table S. 6.2: List of calculated parameters for each territorial average. Field Description Unit min_uev minimum UEV recorded Msej/m 3 max_uev maximum UEV recorded Msej/m 3 avg_uev non-weighted average Unit Emergy Value Msej/m 3 stdev_uev standard deviation of UEVs % Avg_EmPden non-weighted yearly average emergy input per unit area of the drainage basin Msej/km 2 /yr Std_EmPden standard deviation of yearly average emergy input per unit area of the drainage basin % area Area m 2 tot_pop Total population at year 2000 hab UEV_pop population-weighted average UEV Msej/m 3 214

215 Supplementary information SI7. Chapter 7: Integrated earth system dynamic modeling for life cycle impact assessment of ecosystem services SI7.1. Method A perturbation function on resource extraction flows, represented by a Dirac impulse occurring at year 2000, was implemented in GUMBO (Figure S7.1). Figure S. 7.1: Screenshot of GUMBO, showing the implementation of a perturbation (here, fossil fuel). This perturbation represents an additional amount of resource extracted. To be consistent with the conventional (marginal) approach adopted in LCIA, the additional amount is marginal as compared to actual flows, i.e. in the specific case the global extraction rates: the size of the perturbations implemented ranged between 0.1 % and 0.9 % of the value of the corresponding flow or stock in year Figure S7.2 illustrates this step by relating change in energy extraction over the period , due to a marginal change in ore production in year Figure S. 7.2: Example of changes on the production factor Energy, following an additional 0.5% ore production in Left part: energy extraction (Gt/yr) in the baseline scenario; right part: relative difference in energy extraction, between a perturbation (here a +0.5% change of ore production at year 2000) and baseline scenario. Noteworthy, STELLA proposes several temporal integration methods (Euler, Runge-Kutta 2, Runge-Kutta 4), which can significantly alter the results. We selected the Euler method because it makes it possible to recalculate of the cumulated impacts via spreadsheet extraction with a simple procedure, in order to plot the relationships described in equations ( , see Chapter 7). 215

216 SI7.2. Detailed results of simulations SI Dirac perturbation on fossil fuel extraction in the future Perturbations can be simulated not only at present time, but also at different dates in the future. This feature is convenient for instance to better evaluate end-of-life impacts of lasting products (e.g. infrastructure deconstruction). For this experiment we applied the same Dirac perturbation (in absolute value) at different dates. Results are summarized in Table S7.1. Figure S7.3 shows that the ecosystem goods (as called in Boumans et al. 2002) water use, ore production and organic matter extraction behave similarly: while an additional fossil fuel extraction occurring in the near future further reduces the delivery of these ES, perturbations occurring in the far future actually increase their production drastically. This can be explained by the temporal windows set up in GUMBO, which ends in year When the perturbation occurs a few years before this date, the global system remains perturbed in year It is likely that a larger temporal window would make these results smoother concerning far-future perturbations. The same remark holds for energy extraction (Figure S7.4): the drastic increase for a perturbation in year 2090 results in a 40-fold increase of this ecosystem good, while perturbations occurring between 2001 and 2050 actually decrease its cumulated delivery over the whole century. Results concerning ecosystem services (Figure S7.5) show smaller changes when the year of perturbation is delayed, but different trends: Gas regulation seems more active for perturbations occurring in the next decade than for perturbations that occur later. It remains however more active than in the baseline scenario (i.e. with no perturbation), since CFs are always positive. Climate regulation service is more affected with perturbations occurring in the first half of the 21st century. Whatever the year of the perturbation, it remains hindered compared to the baseline scenario. Disturbance Regulation and Soil Formation follow an exponential decrease, with a slight difference in far-future perturbations: while soil formation shows no difference with the baseline (CF = 0), disturbance regulation is less efficient than in the baseline (CF < 0). The later the perturbation, the lower the service Nutrient Cycling, down to a null CF for a perturbation occurring late in the century (i.e. there is no difference with the baseline scenario). Finally, the Recreation Cultural service is lower and lower for a perturbation occurring in the first decade, but then stable for perturbations between 2010 and Later perturbations have smaller impacts (the later the perturbation, the smaller the impact), while remaining higher than in the baseline. It ultimately becomes lower than in the baseline scenario, with a negative CF, for perturbations occurring after Human Capitals (Figure S7.6) also show diverse trends. One can however observe that each human capital experience a local extremum for perturbations occurring between 2010 and Finally, Figure S7.7 shows how endpoint indicators depend on the date of the perturbation. GWP cumulated with no discount rate is highly influenced: while its CF decreases with perturbations between 2001 and 2010 and remains lower with perturbations up to 2050, it shows an increase for later perturbations with positive values. It means that late perturbations have lower negative impacts. This is mainly due to Labor Force capital, which follows a similar curve. GWP with a 3% discount rate logically follows an exponential, decreasing curve, since this endpoint give priority to short terms impacts. SSW is mainly influenced, in GUMBO, by Social Network, Knowledge and Recreation Cultural, as well as mortality and consumption levels. Impacts seem worst for a perturbation occurring in the far future (CF < 0), and in the very near future, where CFs are close to the scenario where perturbation occurs in

217 Table S. 7.1: Impacts on midpoints and endpoints for 1Gt fossil fuel perturbations occurring at different dates in the future. The distinction between Ecosystem Goods and Services belongs to Boumans et al. (2002). Supplementary information 217

218 Figure S. 7.5: Relative change in ecosystem goods production (except energy) according to the year of perturbation (additional fossil fuel extraction). Figure S. 7.6: Relative change in ecosystem good "energy" according to the year of perturbation (additional fossil fuel extraction). Figure S. 7.3: Relative change in ecosystem services according to the year of perturbation (additional fossil fuel extraction). Figure S. 7.4: Relative change in human capitals according to the year of perturbation (additional fossil fuel extraction). 218

219 Supplementary information GWP 0 GWP 3 SSW -4 Figure S. 7.7: Relative change in endpoint CFs according to the year of perturbation (additional fossil fuel extraction). SI Perturbation on fossil fuel extraction occurring over several years In LCA, the use phase may last rather long periods. This is the case for equipment goods such as household electric devices, transportation vehicles, infrastructures (buildings, roads) But current tools prevent the user from differentiating instant emissions like in production chains and long-term emissions that occur in the use phase. Recent advances in this research question suggest using time-dependent CFs (Levasseur et al. 2010). Dynamic ecological models allow the specifying human interventions as instantaneous or long-termed. We used GUMBO to model a fossil fuel extraction of same absolute value but spread across several periods. Results are displayed in Table S7.2. All perturbations start in year Retrieved CFs on ecosystem goods production show similar patterns (except energy extraction), as displayed in Figure S7.8: additional extraction of fossil fuel over 33 years generate the highest impact - up to twice the value of the 1-year perturbation, for organic matter extraction. Energy extraction (Figure S7.9) shows decreasing impacts with longer perturbation, down to a minimum for 50-year perturbation. The CF then increases drastically. A 100-year perturbation generates a CF twelve times higher than a 1-year perturbation. Out of the 7 ecosystem services modeled in GUMBO (see Figure S7.10), 5 are less affected by fossil fuel perturbations occurring over longer periods; Gas Regulation is slightly more affected by fossil fuel extractions that last up to 40 years, but it is less affected by longer perturbations. Climate regulation is more affected by longer perturbations, with a maximum between 33 and 50 years. Oppositely, human capitals show very different patterns (Figure S7.11). Built Capital is most affected by a 33-year long perturbation, as well as Labor Force. However, impact on Built Capital drastically decreases with longer perturbations, and even becomes positive with those longer than 60 years. Labor Force remains negative, but impacts are smaller for long-period perturbation (over 90 years) than for the 1-year perturbation scenario. Impact on Knowledge capital is lower for longer perturbations, with relatively stable CFs for perturbations longer than 20 years. Finally, Social Network capital is less and less affected by longer perturbations. 219

220 Figure S7.12 shows the changes on endpoint CFs with the duration of the perturbation. Impact on GWP with no discount is higher with perturbations that last less than 75 years, with a maximum impact for a 33-year perturbation. It then remains negative but smaller than the impact of a 1-year perturbation. Such behavior can be explained by the evolution of its most influential factors - knowledge, labor force and energy extraction. GWP with a discount rate of 3% shows an exponential decrease with longer perturbations, due to the emphasis on short term impacts of perturbations. SSW shows a minimum impact for perturbations that last around 50 years, and a maximum impact for shortest perturbations. As important as the date of the human intervention, its duration matters. Ecological modeling provides valuable information to better integrate the temporality of a human intervention in LCA. SI Use of different global policy scenarios for the 21st century GUMBO provides different options for taking into account the choices formulated by global leaders in 21st century s world governance. All results presented above consider a business-asusual scenario (hereafter called Base Case). Authors provide a transparent set of assumptions that define other global policies. Default scenarios can be schematically summed-up as follows (Boumans et al. 2002): Star Trek (ST): an optimistic viewpoint on the future state of the world, with optimist policies implemented Mad Max (MM): a skeptic viewpoint on the future state of the world, with optimist policies implemented Big Government (BG): an optimistic viewpoint on the future state of the world, with skeptic policies implemented Ecotopia (ET): a skeptic viewpoint on the future state of the world, with skeptic policies implemented Table S7.3 provides alternative CFs for the afore-mentioned policy scenarios. Note that the baseline scenario (i.e. where no perturbation is applied) differs for each policy option. More details are available in Boumans et al. (2002). Figure S7.13 shows the relative differences of ecosystem goods CFs for each alternative policy option. Water use and ore production are mainly driven by the assumptions made on the future real state of the world, since the ST and BG options have similar CFs, as well as MM and ET options. For organic matter extraction, both parameters (assumptions and policies) seem of equal influence, while for energy extraction the choice of policy implemented. For most ecosystem services (Figure S7.14), implementing positive policy options slightly increase the magnitude of all ecosystem services modeled, except for Recreation Cultural service, for which assumptions of the real state of the world are most important. Skeptical assumptions enhance its delivery over the century. Human capitals (Figure S7.15) seem most influenced by assumptions on the real state of the world. 220

221 Table S. 7.2: CFs for perturbations on 1Gt fossil fuel extraction, over different periods, starting in year The distinction between Ecosystem Goods and Services belongs to Boumans et al. (2002). Supplementary information 221

222 Figure S. 7.10: Relative change in ecosystem services according to the duration of the perturbation (additional fossil fuel extraction). Figure S. 7.11: Relative change in human capitals according to the duration of the perturbation (additional fossil fuel extraction). Figure S. 7.8: Relative change in ecosystem goods production (except energy) according to the duration of the perturbation (additional fossil fuel extraction). Figure S. 7.9: Relative change in ecosystem good "energy" according to the duration of the perturbation (additional fossil fuel extraction). 222

223 Supplementary information 2 1,5 1 0, GWP 0 GWP 3 SSW Figure S. 7.12: Relative change in endpoint CFs according to the duration of the perturbation (additional fossil fuel extraction). Endpoint CFs experience diverse results (Figure S7.16): GWP with a 3% discount rate is rather independent from the policy option. On the contrary, GWP with no discount rate is highly influenced by assumptions formulated on the future state of the world. Positive assumptions imply that a marginal extraction of fossil fuel at present time involve a loss of cumulated GWP higher than in the Base Case policy option. On the contrary, Skeptical assumptions imply that the cumulated GWP remains positive over the century. Figure S7.18 provides more insights on the variations of GWP. It can be seen that under skeptical assumptions, changes in GWP induced by an additional extraction of fossil fuel at present time leads to a lower loss of future GWP. The main reason is the stagnation of GWP under these scenarios, as shown in Figure S7.17. Impacts on SSW (Figure S7.20) are much more swinging. MM, ET and Base Case options undergo similar evolution of SSW (Figure S7.19) and an additional fossil fuel extraction has rather similar impacts. Changes in BG and ST options are more complex to interpret at a glance; they lead to quite different CFs for the endpoint SSW. 223

224 224 Table S. 7.3: CFs for perturbations on fossil fuel extraction in year 2000, with selected scenarios for future global policies. The distinction between Ecosystem Goods and Services belongs to Boumans et al. (2002).

225 Figure S. 7.15: Relative change in human capitals according to the global policy scenario. Figure S. 7.16: Relative change in endpoint CFs according to the global policy scenario. Figure S. 7.13: Relative change in ecosystem goods production according to the global policy scenario. Figure S. 7.14: Relative change in ecosystem services according to the global policy scenario. Supplementary information 225