Identifying optimum strategies for agricultural management considering multiple ecosystem services and climate change

Transcription

1 International Congress on Environmental Modelling and Software Brigham Yong University BYU ScholarsArchive 5th International Congress on Environmental Modelling and Software - Ottawa, Ontario, Canada - Jly 2010 Jl 1st, 12:00 AM Identifying optimm strategies for agricltral management considering mltiple ecosystem services and climate change A. Holzkämper P. Calanca J. Fhrer Follow this and additional works at: Holzkämper, A.; Calanca, P.; and Fhrer, J., "Identifying optimm strategies for agricltral management considering mltiple ecosystem services and climate change" (2010). International Congress on Environmental Modelling and Software This Event is broght to yo for free and open access by the Civil and Environmental Engineering at BYU ScholarsArchive. It has been accepted for inclsion in International Congress on Environmental Modelling and Software by an athorized administrator of BYU ScholarsArchive. For more information, please contact scholarsarchive@by.ed, ellen_amatangelo@by.ed.

2 International Environmental Modelling and Software Society (iemss) 2010 International Congress on Environmental Modelling and Software Modelling for Environment s Sake, Fifth Biennial Meeting, Ottawa, Canada David A. Swayne, Wanhong Yang, A. A. Voinov, A. Rizzoli, T. Filatova (Eds.) Identifying optimm strategies for agricltral management considering mltiple ecosystem services and climate change A. Holzkämper a, P. Calanca a, J. Fhrer a a Agroscope Reckenholz-Tänikon Research Station ART, Air Polltion and Climate Grop, Reckenholzstrasse 191, CH-8046 Zrich, Switzerland, annelie.holzkaemper@art.admin.ch Abstract: In many Eropean regions, the likely increase in water shortage and extreme weather events dring the coming decades may case more freqent crop loss, yield instability, and make cltivated areas less sitable for traditional crops. Possible adaptation measres are for example changes in crop choice, rotation and a more widespread adoption of irrigation. However, increased water se for irrigation may lead to conflicts with other ecosystem fnctions. Hence, measres to minimize prodctivity losses and preserve other ecosystem services sch as water reglation, erosion control and ntrient cycling need to be developed. In this paper, we present an approach for identifying optimm adaptation strategies for agricltral management considering mltiple ecosystem services and climate change. The method is based on a mlti-objective spatial optimization rotine which integrates the crop model CropSyst for evalating the effects of climate and management changes on yields, water consmption, soil erosion and ntrient leaching. The method is illstrated with reslts from a preliminary model test, where we maximize crop prodction, while meeting a constraint on agricltral water se for two climate scenarios. Trade-offs between maximm crop prodction and minimm water se are investigated. Keywords: Agricltral management; ecosystem services; mlti-objective spatial optimization; climate change adaptation 1. INTRODUCTION Agricltre is an important economic sector which is going to be strongly affected by climate change. In cool temperate regions of Erope, climate change dring the next decades is expected to have positive effects on agricltre throgh higher crop prodctivity and expansion of sitable areas for crop cltivation, and introdction of adapted crop species and new varieties [IPCC 2007a]. However, increasing water shortage and extreme weather events sch as smmer droghts dring the cropping season may also case more freqent yield loss and instabilities, and make areas less sitable for traditional crops [Olesen and Bindi 2002]. The droght risk on the Swiss Central Platea may increase from abot 15% to over 50% with ftre climate change [Calanca 2007]. In Swiss agricltre, this trend is expected to have negative impacts on prodctivity and to increase prodction risks by the end of the centry [Fhrer et al. 2006, Torriani et al. 2007, Finger and Schmid 2008]. Hence, adaptation strategies for agricltral water resorce management are needed to cope with the expected change in climatic conditions, taking into accont possible increases in costs for spplemental water. These may inclde adjstments of crop rotations (e.g. shifting from high to low water-demanding crops) and of prodction intensities, se of redced (or no) tillage, integration of cover crops, adoption of irrigation with efficient technologies and choice of water sorces (srface water or grondwater), retention of water in reservoirs (e.g. rainwater harvesting with cisterns), introdction of sitable landscape elements to redce rnoff, or changes in stocking rates and livestock types. Farmers who have sfficient access to capital and technologies shold be able to continosly adapt their farming system by changing the mix of crops, adopting irrigation

3 and adjsting fertilization and plant protection [Easterling and Apps 2002]. However, in connection with climate change this might intensify existing impacts on the environment and lead to new conflicts between ecosystem services [MA 2005, Schröter et al. 2005, IPCC 2007b]. For example, increased water se for irrigation cold conflict with water demands for domestic or indstrial ses, and lead to negative ecological implications [Bates et al. 2008]. Also, soil loss throgh erosion may increase de to climate change, an effect which cold be aggravated throgh changes in land management [Lee et al. 1999, O'Neal et al. 2003]. To prevent contined degradation of natral resorces, policy will need to spport farmers adaptation while considering the mltifnctional role of agricltre [Olesen and Bindi 2002, Betts 2006]. Hence, effective measres to minimize prodctivity losses and preserve finite natral resorces need to be developed at all decision levels, and scientists need to assist decision makers in this process [Salinger et al. 1999, Salinger et al. 2002]. Mlti-objective optimization methods in connection with biophysical models have shown great potential for addressing sch isses of opposing management goals [Ines et al. 2006, Bryan and Crossman 2008, Higgins et al. 2008, Sadeghi et al. 2009, Meyer et al. 2009, Whittaker et al and Latinopolos 2009]. Bryan and Crossman [2008] developed an optimization-based regional planning approach to identify geographic priorities for ongrond natral resorce management actions that most cost-effectively meet mltiple natral resorce management objectives. Higgins et al. [2008] applied a mlti-objective integer programming model, with objective fnctions representing biodiversity, water rnoff and carbon seqestration. Sadeghi et al. [2009] applied an optimization approach to maximize profits from land se, while minimizing erosion risk. Meyer et al. [2009] copled SWAT (Soil and Water Assessment Tool) with an optimization rotine to determine optimm farming system patterns to redce nitrogen leaching while maintaining income. Similarly, Whittaker et al. [2009] applied SWAT in connection with a Pareto-optimization approach considering profits from land se and chemical polltion from farm prodction. Latinopolos [2009] applied optimization to a problem of water and land resorce allocation in irrigated agricltre with respect to a series of socio-economic and environmental objectives. Sch approaches can be very sefl to spport the development of regional land se adaptation strategies. However, they have not been sed yet in combination with scenarios of climate change. In this paper we present an approach for identifying optimm adaptation strategies for agricltral land management with respect to mltiple ecosystem services. These inclde not only food prodction bt also water reglation, soil protection and ntrient cycling. We present an approach for mlti-objective spatial optimization based on a genetic algorithm and discss ways to accont for ncertainties in climate projections. The methodology will be applied to identify optimm land management patterns in a small catchment in the Swiss Pre-Alps. Preliminary reslts are presented to illstrate the optimization method and possible otcomes. 2. METHOD To investigate optimm agricltral management adaptations to climate change, a variety of management options have to be considered, inclding crop/rotation choice, irrigation levels, fertilization levels, soil and reside management. The diversification of crops may decrease the risk of crop failre throgh extreme weather events and pests. Rotations may be adapted by shifts in sowing dates. Earlier sowing dates can be advantageos for smmer crops nder increased temperatre conditions, as plants can tilize the higher soil moistre dring spring, ths decreasing the risk of water stress in the growth cycle. On the other hand this may increase the risk of crop failre throgh late frosts. For winter crops later sowing dates cold be beneficial to avoid damages that can occr dring the cold period if crops are too far developed. As CO 2 concentration and temperatres increase, fertilization and irrigation reqirements may also increase. Soil and reside management can have impacts on evaporation and ths on the soil water balance.

4 2.1 Optimization For identifying optimm patterns of agricltral management considering mltiple ecosystems and climate change, we apply a mlti-objective spatial optimization rotine based on the tool developed by Holzkämper and Seppelt [2007]. The approach is based on a genetic algorithm library by Wall [1996]. The genetic algorithm had proven to be highly sitable for addressing complex combinatorial problems in many previos applications [Ko et al. 2000, Ines et al. 2006, Whittaker et al. 2009, Li 2009]. It is an iterative search algorithm that is based on the principles of evoltion [Goldberg 1989]. A soltion is represented as a genome. The optimization starts with an initial poplation of genomes. With each iterative step the genomes of the poplation are evalated with a defined objective fnction and the fittest genomes are chosen to be recombined. The newly generated soltions or offspring genomes also are evalated and the least fittest genomes are exclded from the poplation to maintain the original poplation size. Genome definition In order to represent the land management pattern within the optimization, we first divide the stdy area into a nmber of decision nits. Decision nits are defined as parcels of agricltral land within which soil conditions are assmed to be homogenos and which are thoght to be managed as entire nits. The decision nits can be derived by clipping a raster layer of soil types with a layer of cropland (Figre 1). The reslting raster of soil types nder cropland can be sed as the basis for the derivation of an ID-map where niqe identifiers are assigned to all cropland clsters with homogenos soil type and a certain minimm size. The niqe identifiers in the map are related to places in a one-dimensional array that defines the genome in the genetic algorithm. Each place in the array defines the management in one decision nit of the stdy area. Figre 1. Data processing for derivation of decision nits in a test stdy area: The procedre is based on a layer that defines all cropland (first map); the map of soil types is clipped with the cropland layer to reslt in the second map; niqe identifiers are assigned to all cropland clsters with a homogenos soil type and a minimm size of 25 ha (third map); a one-dimensional array genome is derived from the ID map and initialized with different management types (e.g. consisting of combined crop and irrigation options). Objective fnction

5 Management options intended to increase agricltral prodctivity can have varios positive or negative impacts on other ecosystem services sch as water reglation, erosion control/sediment retention and ntrient cycling. Ths, in order to identify management patterns that maximize food prodction, while minimizing environmental management impacts, we need to define an objective fnction that combines all these ecosystem services (1) Food prodction P is maximized; water consmption for irrigation I is minimized to spport the water reglation fnction and avoid conflicts with other water sers; erosion control is considered throgh the minimization of soil loss E; and ntrient leaching N is minimized to represent the ecosystem service of ntrient cycling. As the objective fnction combines for variables with different ranges and nits, the variable vales need to be normalized and weighted. 1 I' w 1 E' w 1 N' J max w P' w (1) P I E In (1) P, I, E and N denote normalized vales of total food prodction, water consmption for irrigation, soil loss throgh erosion and ntrient leaching within the stdy region. By sbstracting the vales of I, E and N from 1, we allow for all ecosystem service goals to be maximized. The weights are defined so that w P + w I + w E + w N = 1.Ths, the optimization criterion J ranges from 0 to 1. One possibility to obtain dimensionless variables with vales between 0 and 1 is to define: i1 Vi Vmin, i i1 i1 V ' (2) V V max, i i1 min, i where V represents any of P, I, E or N; is the nmber of decision nits within a stdy area; and for each decision nit i V max,i and V min,i denote the maximm and minimm of V, respectively, over the set of climatic conditions and management options envisaged for the stdy. These vales have to be determined prior to the optimization for all possible combinations of management, climate and soil type, and cold be stored for convenience in a look-p table. The objective fnction is sbject to constraints regarding P, I, E and N to take accont of landscape planning goals for the region (e.g. minimm amonts of specific crops, maximm irrigation, maximm soil loss, maximm ntrient leaching). These constraints will have to be defined in consltation with regional stakeholders. The optimization will be performed mltiple times sing different weighting combinations to derive a series of optimm trade-off soltions. The trade-off soltions can be presented to regional stakeholders as a basis for discssing the perspectives of adaptation strategies for agricltral land management. N 2.2 Simlation model Soltion of the optimization problem reqires evalation of each of the terms in (1) as a fnction of climate and management. To accomplish that, we se the generic crop model CropSyst [Stöckle et al. 2003]. CropSyst simlates the soil water bdget, soil-plant nitrogen bdget, crop phenology, canopy and root growth, biomass prodction, crop yield, reside prodction and decomposition, soil erosion by water, and salinity on a daily basis. It is driven by daily weather data and reqires specification of crop and soil characteristics, as well as management options inclding crop rotation, cltivar selection, irrigation, nitrogen fertilization, tillage operations, and reside management. 2.3 Climate change scenarios Daily weather data at the local scale cannot directly be extracted from the otpt of Global Climate Model (GCMs) or even Regional Climate Model (RCMs) simlations. One reason

6 for that is the coarse spatial resoltion of crrent GCMs and RCMs. This is for example the case with the models sed to create the IPCC AR4 scenarios ( Therefore, downscaling techniqes need to be applied. Stochastic weather generators are among the most convenient downscaling tools for applications at the scale of small catchments. 3. PRELIMINARY MODEL TESTS To illstrate the implementation of the presented approach and its possible otcomes, we condcted preliminary model tests in the agricltral areas of the Broye catchment, which covers an area of 392 km 2 (57% agricltral land se; [BFS 2001]) and is located in Western Switzerland. 3.1 Genome definition For this simplified test stdy, we considered only a limited nmber of management options, i.e. crop choice and irrigation level. Crop choice inclded winter wheat, winter barley, maize, potato and canola. Irrigation options inclded five different levels of maximm allowable depletion to trigger irrigation, i.e. fractions of 0.7, 0.8, 0.9, 0.95, 1 of the minimm plant available water reqired for optimm prodction. Ths, 25 combinations of crop choice and irrigation level are possible and each place in the genome is filled with one ot of these 25 combinations. The genome represents the pattern of decision nits shown in Figre Objective fnction The optimization goal was to maximize food prodction P in terms of yields from all decision nits i in tons, sbject to a constraint c on water consmption for irrigation water I from all decision nits i in m 3. J max P i (3) i1 sbject to i1 I i c By shifting the constraint vale, the trade-off between food prodction and water reglation can be investigated. The optimization was performed with a steady state genetic algorithm of the following specifications: nmber of generations = 1000; genome size = 24, poplation size = 50; proportion of replacement = 0.25; selection rotine = rolette wheel; mtation probability = As the genetic algorithm is a stochastic search techniqe, it prodces slightly diverging reslts with each optimization rn. To take this stochasticity into accont, each optimization rn was repeated 100-times. 3.3 Climate scenarios The stochastic weather generator LARS-WG [Racsko et al. 1991; Semenov and Barrow 1997] was sed to generate 50 years of daily weather data for assming a climate change signal consistent with the IPCC-AR4 scenarios HADCM3-SRA2 (scenario A) and IPCM4-SRA2 (scenario B). The weather generator was calibrated with data from the climate station of Payerne, which was assmed to be representative for the stdy region

7 (data cortesy of the Swiss Federal Office of Meteorology and Climatology, MeteoSwiss. See also Optimization reslts Preliminary optimization reslts for the two climate scenarios are shown in Figre 2. If no water is available for irrigation, the optimm management pattern wold be only barley nder scenario A. Under scenario B the optimm management pattern wold contain 8000 ha of barley, bt also 2000 ha of maize. As the irrigation constraint is released, the maize areas increase nder both climate scenarios, while area with barley decreases. Area with potato increases slightly with increasing availability of irrigation water, bt seems to decrease as water availability increases frther. In contrast to these reslts, the optimm management identified nder crrent climatic conditions wold be only maize irrespective of the irrigation constraint. A B MAIZE MAIZE BARLEY BARLEY POTATO POTATO 0e+00 2e+08 4e+08 6e+08 8e+08 1e+09 0e+00 2e+08 4e+08 6e+08 8e+08 1e+09 Irrigation [m 3 ] Irrigation [m 3 ] Figre 2. Crop area in optimization reslts with varying constraints to irrigation and scenario A and scenario B, respectively (black line indicating median, grey lines indicating 10 th and 90 th percentiles of otpts from the 100 optimization rns). Comparing the water consmption for irrigation on the two most dominant soil types for these optimization rns (Figre 3), we can see a trend towards higher irrigation on loam than on sandy loam in both climate scenarios A and B. Irrigation [mm] [mm] A Irrigation [mm] B loam sandy loam 2 10 loam sandy loam 2 10 Figre 3. Boxplots of distribtions of water consmption for irrigation on loam and sandy loam derived from optimization reslts considering climate scenarios A and B with an irrigation constraint of m 3 (A = scenario HADCM3_SRA2, B = scenario IPCM4_SRA2; boxes extend from 25 th to 75 th percentiles with horizontal line indicating the median, whiskers extend to most extreme vales or a maximm of 1.5 of the interqantile range, points beyond whiskers are sspected otliers).

8 4. CONCLUSIONS AND OUTLOOK In this paper we presented an approach for identifying optimm adaptation strategies for agricltral management. The method is based on a genetic algorithm, which allows for an efficient optimization of highly complex spatial allocation problems. Thereby, the integration of the process-based simlation model CropSyst allows for the consideration of complex interactions between crop management and site conditions. Throgh the consideration of different climate scenarios, the ncertainties of the climate scenarios can be taken into accont. Preliminary test reslts showed how optimm soltions can vary in terms of composition and configration depending on a constraint for irrigation and depending on the climate projection. In the frther implementation of the presented approach, the spatial heterogeneity of the stdy area will be considered in more detail and a greater variety of management options and objectives will be taken into accont (as otlined above). For erosion and ntrient leaching barrier effects can be taken into accont, relating the objective fnction to spatial interactions between decision nits. For qantifying food prodction it might be more adeqate to se calories rather than dry matter yields, as yields do not necessarily refer to food energy. Frthermore, a measre of interannal yield variability needs to be integrated into the objective fnction to allow for the minimization of prodction risks. For a more comprehensive consideration of ncertainties in climate projections, the objective fnction cold be defined as a weighted linear combination of objective vales as calclated in (1) based on different climate scenarios, where the weights represent the probabilities of the climate scenarios. REFERENCES Bates, B., Z. W. Kndzewicz, S. W, N. Arnell, V. Brkett, et al., Climate Change and Water. Technical Paper of the Intergovernmental Panel on Climate Change. Geneva, IPCC Secretariat: 210, Betts, R., Implications of land ecosystem-atmosphere interactions for strategies for climate change adaptation and mitigation. 1st International Integrated Land Ecosystem Atmosphere Processes Stdy Science Conference ( ileaps), Bolder, CO, BFS, Bodenntzng im Wandel: Arealstatistik der Schweiz. Nechâtel, Bryan, B. A. and N. D. Crossman, Systematic regional planning for mltiple objective natral resorce management. Jornal of Environmental Management 88(4): , Calanca, P., Climate change and droght occrrence in the Alpine region: How severe are becoming the extremes? Global and Planetary Change 57(1-2): , 2007 Easterling, W. and M. Apps, Assessing the conseqences of climate change for food and forest resorces: A view from the IPCC. International Workshop on Redctin Vlnerability of Agricltre and Forestry to Climate Variability and Climate Change, Ljbljana, SLOVENIA, Springer, 2002 Finger, R. and S. Schmid, Modeling agricltral prodction risk and the adaptation to climate change. Agricltral Finance Review 68(1): 25-41, Fhrer, J., M. Beniston, A. Fischlin, C. Frei, S. Goyette, K. Jasper and C. Pfister, Climate risks and their impact on agricltre and forests in Switzerland. Climatic Change 79(1-2): , Goldberg, D. E., Genetic Algorithms in Search, Optimization and Machine Learning. Reading, M.A., Addison Wesley, Higgins, A. J., S. Hajkowicz and E. Bi, A mlti-objective model for environmental investment decision making. Compters & Operations Research 35(1): , Holzkämper, A. and R. Seppelt, A generic tool for optimising land-se patterns and landscape strctres. Environmental Modelling & Software 22(12): , Ines, A. V. M., K. Honda, A. Das Gpta, P. Droogers and R. S. Clemente, Combining remote sensing-simlation modeling and genetic algorithm optimization to explore water management options in irrigated agricltre. Agricltral Water Management 83(3): , 2006

9 IPCC Climate Change 2007: Impacts, Adaptation and Vlnarability. Contribtion of Working Grop II to the Forth Assessment Report of the Intergovernmental Panel on Climate Change. M. Parry, O. Canziani, J. Paltikof, P. van der Linden and C. Hanson. Cambridge, United Kingdom and New York, NY, USA, Cambridge University Press, 2007a. IPCC Climate Change 2007: Mitigation. Contribtion of Working Grop III to the Forth Assessment Report of the Intergovernmental Panel on Climate Change. B. Metz, O. Davidson, P. Bosch, R. Dave and L. Meyer. Cambridge, United Kingdom and New York, NY, USA, Cambridge University Press, 2007b. Ko, S. F., G. P. Merkley and C. W. Li, Decision spport for irrigation project planning sing a genetic algorithm. Agricltral Water Management 45(3): , Latinopolos, D., Mlticriteria decision-making for efficient water and land resorces allocation in irrigated agricltre. Environment, Development and Sstainability 11: , Lee, J. J., D. L. Phillips and V. W. Benson, Soil erosion and climate change: Assessing potential impacts and adaptation practices. Jornal of Soil and Water Conservation 54(3): , Li, Y., Atomatic calibration of a rainfall-rnoff model sing a fast and elitist mltiobjective particle swarm algorithm. Expert Systems with Applications 36(5): , MA, Ecosystems and Hman Well-being: Synthesis. Washington, DC, Island Press, Meyer, B. C., J. M. Lescot and R. Laplana (2009). Comparison of Two Spatial Optimization Techniqes: A Framework to Solve Mltiobjective Land Use Distribtion Problems. Environmental Management 43(2): , O'Neal, M. R., M. A. Nearing, R. C. Vining, J. Sothworth and R. A. Pfeifer, Climate change impacts on soil erosion in Midwest United States with changes in crop management. Meeting of the Soil-Erosion-Network, Tcson, AZ, Olesen, J. E. and M. Bindi, Conseqences of climate change for Eropean agricltral prodctivity, land se and policy. Eropean Jornal of Agronomy 16(4): , Racsko, P., L. Szeidl and M. Semenov, A Serial Approach to Local Stochastic Weather Models. Ecological Modelling 57(1-2): 27-41, Sadeghi, S. H. R., K. Jalili and D. Nikkami, Land se optimization in watershed scale. Land Use Policy 26(2): , Salinger, M. J., M. V. K. Sivakmar and R. Motha, Redcing vlnerability of agricltre and forestry to climate variability and change: Workshop smmary and recommendations. International Workshop on Redctin Vlnerability of Agricltre and Forestry to Climate Variability and Climate Change, Ljbljana, SLOVENIA, Salinger, M. J., C. J. Stigter and H. P. Das, Agrometeorological adaptation strategies to increasing climate variability and climate change. International Workshop on Agrometeorology in the 21st Centry - Needs and Perspective, Accra, Ghana, Schröter, D., W. Cramer, R. Leemans, I. C. Prentice, M. B. Arajo, et al., Ecosystem service spply and vlnerability to global change in Erope. Science 310(5752): , Semenov, M. A. and E. M. Barrow, Use of a stochastic weather generator in the development of climate change scenarios. Climatic Change 35(4): , Stöckle, C. O., M. Donatelli and R. Nelson, "CropSyst, a cropping systems simlation model." Eropean Jornal of Agronomy 18(3-4): , Torriani, D. S., P. Calanca, S. Schmid, M. Beniston and J. Fhrer, Potential effects of changes in mean climate and climate variability on the yield of winter and spring crops in Switzerland. Climate Research 34(1): 59-69, Wall, M., GALib A C++ Library of Genetic Algorithm Components. Version 2.4 from Whittaker, G., R. Confesor, S. M. Griffith, R. Fare, S. Grosskopf, J. J. Steiner, G. W. Meller-Warrant and G. M. Banowetz, A hybrid genetic algorithm for mltiobjective problems with activity analysis-based local search. Eropean Jornal of Operational Research 193(1): , 2009.