Evidence for soil water control on carbon and water dynamics in European forests during the extremely dry year: 2003

Transcription

1 Agricultural and Forest Meteorology 143 (2007) Evidence for soil water control on carbon and water dynamics in European forests during the extremely dry year: 2003 A. Granier a, *, M. Reichstein b,c, N. Bréda a, I.A. Janssens d, E. Falge e, P. Ciais f, T. Grünwald g, M. Aubinet h, P. Berbigier i, C. Bernhofer g, N. Buchmann j, O. Facini k, G. Grassi l, B. Heinesch h, H. Ilvesniemi m, P. Keronen n, A. Knohl c,o,b.köstner g, F. Lagergren p, A. Lindroth p, B. Longdoz a, D. Loustau i, J. Mateus q, L. Montagnani r,s, C. Nys t, E. Moors u, D. Papale b, M. Peiffer a, K. Pilegaard v, G. Pita q, J. Pumpanen w, S. Rambal x, C. Rebmann c, A. Rodrigues y, G. Seufert l, J. Tenhunen e, T. Vesala n, Q. Wang e a UMR INRA-UHP Forest Ecology and Ecophysiology, Champenoux, France b Department of Forest Environment Science and Resource, DISAFRI, University of Tuscia, Via Camillo de Lellis, Viterbo, Italy c Max Planck Institute for Biogeochemistry, Postfach , Jena, Germany d Department of Biology, University of Antwerp, Universiteitsplein 1, 2610 Wilrijk, Belgium e Pflanzenökologie, Universität Bayreuth, Bayreuth, Germany f LSCE, CE Orme des Merisiers, Bat 701, Gif sur Yvette Cedex, France g Department of Meteorology, Institute of Hydrology and Meteorology, Technische Universität Dresden, Dresden, Germany h Faculté des Sciences Agronomiques de Gembloux, Unité de Physique, B-5030 Gembloux, Belgium i UR EPHYSE-INRA Bordeaux, 69 route d Arcachon, Gazinet, France j Institute of Plant Sciences, Swiss Federal Institute of Technology (ETH), Zurich, Switzerland k Istituto di Biometeorologia CNR Via Gobetti, 101, Bologna, Italy l Environment Institute, JRC-Ispra, Ispra, Italy m Vantaa Research Centre, Finnish Forest Research Institute, P.O. Box 18, 01301, Finland n Department of Physical Sciences, P.O. Box 64, University of Helsinki, 00014, Finland o Department of Environmental Science, Policy and Management, Ecosystem Science Division, University of California, Berkeley, USA p Lund University, Department of Physical Geography and Ecosystem Analysis, Soelvegatan 12, Sölvegatan 12, S Lund, Sweden q Instituto Superior Técnico, Departamento de Engenharia Mecânica, Av. Rovisco Pais, Lisboa, Portugal r Agenzia Provinciale per l Ambiente, Via Amba-Alagi 5, Bolzano, Italy s Ripartizione Foreste di Bolzano, Via Brennero 6, Bolzano, Italy t INRA Biogeochemistry of Forest Ecosystems, Champenoux, France u Alterra, Postbus 47, 6700 AA Wageningen, The Netherlands v Plant Research Department, Risø National Laboratory, P.O. Box 49, 4000 Roskilde, Denmark w Department of Forest Ecology, P.O. Box 27, University of Helsinki, 00014, Finland x CEFE/CNRS 1919 Route Mende BP 5051 Montpellier, 34033, France y Estação Florestal Nacional, Departamento de Silvicultura e Produtos Florestais, Av. da República, Quinta do Marquês, Oeiras, Portugal Received 17 May 2006; received in revised form 12 December 2006; accepted 15 December 2006 * Corresponding author. Tel.: ; fax: address: agranier@nancy.inra.fr (A. Granier) /$ see front matter # 2007 Elsevier B.V. All rights reserved. doi: /j.agrformet

2 124 A. Granier et al. / Agricultural and Forest Meteorology 143 (2007) Abstract The drought of 2003 was exceptionally severe in many regions of Europe, both in duration and in intensity. In some areas, especially in Germany and France, it was the strongest drought for the last 50 years, lasting for more than 6 months. We used continuous carbon and water flux measurements at 12 European monitoring sites covering various forest ecosystem types and a large climatic range in order to characterise the consequences of this drought on ecosystems functioning. As soil water content in the root zone was only monitored in a few sites, a daily water balance model was implemented at each stand to estimate the water balance terms: trees and understorey transpiration, rainfall interception, throughfall, drainage in the different soil layers and soil water content. This model calculated the onset date, duration and intensity of the soil water shortage (called water stress) using measured climate and site properties: leaf area index and phenology that both determine tree transpiration and rainfall interception, soil characteristics and root distribution, both influencing water absorption and drainage. At sites where soil water content was measured, we observed a good agreement between measured and modelled soil water content. Our analysis showed a wide spatial distribution of drought stress over Europe, with a maximum intensity within a large band extending from Portugal to NE Germany. Vapour fluxes in all the investigated sites were reduced by drought, due to stomatal closure, when the relative extractable water in soil (REW) dropped below ca Rainfall events during the drought, however, typically induced rapid restoration of vapour fluxes. Similar to the water vapour fluxes, the net ecosystem production decreased with increasing water stress at all the sites. Both gross primary production (GPP) and total ecosystem respiration (TER) also decreased when REW dropped below 0.4 and 0.2, for GPP and TER, respectively. A higher sensitivity to drought was found in the beech, and surprisingly, in the broadleaved Mediterranean forests; the coniferous stands (spruce and pine) appeared to be less drought-sensitive. The effect of drought on tree growth was also large at the three sites where the annual tree growth was measured. Especially in beech, this growth reduction was more pronounced in the year following the drought (2004). Such lag effects on tree growth should be considered an important feature in forest ecosystems, which may enhance vulnerability to more frequent climate extremes. # 2007 Elsevier B.V. All rights reserved. Keywords: Drought; Europe; Carbon and water fluxes; Forest; Modelling; Water balance 1. Introduction Water and carbon fluxes and, as a consequence, productivity of terrestrial ecosystems are strongly influenced by drought. This is true among sites, when considering the relationship between average climate and forest productivity (Lieth, 1973; Gholz et al., 1990; Scurlock and Olson, 2002; Huxman et al., 2004), and also for the interannual variations within one site (grasslands: Meyers, 2001, forests: Granier et al., 2000b). Repeated droughts induce a reduction in leaf area index (Battaglia et al., 1998; Le Dantec et al., 2000) that in turn decreases GPP (Law et al., 2002; Hoff and Rambal, 2003). Due to recurrent severe droughts, the Mediterranean and dry-tropical vegetation show adaptations: species composition, increased root to shoot ratio and leaf thickness, drought-adapted physiology as enhanced osmotical adjustments, decreased vulnerability to cavitation (see a recent review on drought effects and adaptations by Bréda et al., 2006). However, even drought-adapted ecosystems are influenced by drought (coniferous: Goldstein et al., 2000, evergreen: Rambal et al., 2003; Reichstein et al., 2002b). In tropical rainforests, the dry season may have a strong influence on carbon fluxes (Vourlitis et al., 2001; Rascher et al., 2004). In boreal forests, Cienciala et al. (1998) pointed out a strong reduction in transpiration during dry years, while Krishnan et al. (2006) reported in a boreal aspen stand submitted to a 3-year long drought a decline in both growth and leaf area index that continued even after 2 years following the drought. Due to global change, more frequent and severe droughts are expected in some regions of the globe, mainly in the Northern hemisphere (Lawlor, 1998; Saxe et al., 2001; Meehl and Tebaldi, 2004; Schär et al., 2004 and International Panel on Climate Change, IPCC, 2001). The effects of such extreme events are, however, poorly documented because of their limited occurrence under past and actual climate (Innes, 1998). At the stand level, studies on the effect of drought on tree transpiration estimated from sap flow measurements are quite numerous (e.g., Bréda et al., 1993; Irvine et al., 1998). Granier et al. (2000c) derived responses of canopy conductance to drought for several temperate tree species. A strong reduction in tree and both canopy conductance for water vapour and standscale transpiration when soil water content decreases is generally found in most of the tree species. Under natural conditions, the impact of drought on carbon fluxes has been less frequently investigated than

3 A. Granier et al. / Agricultural and Forest Meteorology 143 (2007) that on water flux regulation. A network of eddycovariance flux towers revealed a pronounced relationship between NEE and annual rainfall (Grünzweig et al., 2003), and a decline in GPP with annual water balance calculated from integrated ET minus precipitation (Law et al., 2002). The drought and heat wave that occurred in summer 2003 in Europe was exceptional, both in duration and in its large distribution across Europe (Stott et al., 2004; Ciais et al., 2005) and provoked a dramatic reduction of the crop yield in France, Italy and South Germany (COPA-COGECA, 2003). The 2003 drought thus provided a good opportunity to examine the response of a wide range of terrestrial ecosystems to extreme climatic conditions, and thus to gain insights how the future climate, with potentially more intense and more frequent extreme climatic conditions, will alter ecosystem functioning (Schär et al., 2004). Especially, the Central European tree species like beech, which are typically not exposed to extreme summer drought, were expected exhibit a strong reaction during We analysed water and carbon fluxes data from 12 forest sites, covering oceanic to continental, and Mediterranean to boreal climates. The reader should bear in mind that, although we focused on drought, this drought was accompanied by a heat wave; at some places in Western Europe, air temperature was higher than 40 8C for several days during August, the monthly average was 5 6 8C higher than normal (Rebetez et al., 2006) in France and in Germany. The objectives of this work were: (i) to quantify the drought intensity and duration during 2003 in the 12 forest sites, plus 4 additional ones where above-canopy fluxes were not measured, (ii) to relate the measured carbon fluxes to the modelled soil water content and compare the response of the different species, (iii) to estimate the reduction in the annual 2003 net ecosystem exchange (NEE), the gross primary production (GPP) and the total ecosystem respiration (TER) as compared to a typical year. 2. Material and methods Most of the studied sites belong to the Carboeurope network aiming at measuring energy, carbon and water fluxes above different European vegetation types ( Climate and site data from four additional sites were also used for drought quantification. The main characteristics of the sites and references providing each site description are given in Table 1. Those sites belong to the most important European forest ecosystem types: temperate deciduous (Hesse, Sorø, Hainich, Vielsalm, Fougères, Nonantola), temperate coniferous (Tharandt, Loobos, Brasschaat), boreal coniferous (Norunda, Hyytiälä), mountain coniferous (Renon), Atlantic (Le Bray) and Mediterranean (San Rossore, Espirra, Puéchabon). They are even-aged stands, except Vielsalm (beech and Douglas-fir), Nonantola (mix of ash, oak, maple, willow plus other minor species), and Renon (spruce and pine). About half of the sites are plantations; most of them are mature or old stands Flux measurements and data processing CO 2 fluxes were estimated using the eddy covariance technique (see in Table 2 the main set up and sensor characteristics) and following the EUROFLUX methodology (Aubinet et al., 2000). Air temperature, CO 2 and water vapour concentration as well as the three components of the wind velocity were sampled at a 20 Hz frequency. Covariance of the vertical velocity component and of the CO 2 concentration was computed every half hour. Classical averaging, rotation and correction procedures (Aubinet et al., 2000) and quality tests (Foken and Wichura, 1996) were applied; we assumed that NEP NEE. The same procedure of gap filling was applied in all sites. Moreover, the same method for calculating ecosystem respiration (TER) and gross photosynthesis (GPP) from net CO 2 fluxes (NEE) was used (Reichstein et al., 2005). The partitioning of the observed NEE into gross primary production (GPP) and ecosystem respiration (TER) was achieved through an algorithm that first establishes a short-term temperature dependence of ecosystem respiration on air temperature from turbulent nighttime data and then uses this relationship for extrapolating respiration from nighttime to daytime. Day-to-day varying base rates of respiration were derived from the u * -filtered nighttime fluxes, where the u * threshold was derived specifically for each site-year according to Reichstein et al. (2005). By using short periods for deriving the temperature dependence, the algorithm avoids the confounding effect of covariance between general biological activity and temperature occurring at seasonal time-scales (cf. detailed discussion in paper cited above). Uncertainties of the changes of GPP and TER between the years were estimated as a combination of the uncertainties that arise from the eddy covariance measurements themselves, u * -filtering, the gap filling and the flux partitioning according to the following reasoning. We assume that potential

4 126 Table 1 Main characteristics of the investigated sites: location, elevation, tree species, tree age, mean annual temperature (T a ), annual rainfall (R): mean, 2003 rainfall and 2003 difference to the mean Site Country Species Latitude (8N) Longitude (8E) Age (year) Elevation (m) T a (8C) R (mm) Citation Mean 2003 Difference Fougères a,b,c,d France European beech Lebret et al. (2001) Hainich d Germany European beech Knohl et al. (2003) Hesse d France European beech Granier et al. (2000b) Sorø d Denmark European beech Pilegaard et al. (2001) Vielsalm Belgium Beech/Douglas-fir Aubinet et al. (2001) Nonantola c,e Italy Mixed deciduous Grassi et al. (2005) Espirra e Portugal Eucalyptus Rodrigues et al. (2005) Puéchabon c France Mediterrannean oaks Rambal et al. (2003) Le Bray France Maritime pine Berbigier et al. (2001) San Rossore Italy Maritime pine Tirone et al. (2003) Brasschaat b,c Belgium Scots pine Carrara et al. (2003) Loobos c,e Netherlands Scots pine Dolman et al. (2002) Hyytiälä d Finland Scots pine Rannik et al. (2002) Tharandt d Germany Spruce Grünwald (2003) Norunda Sweden Spruce/pine Lindroth et al. (1998) Renon Italy Mixed coniferous to Marcolla et al. (2005) The mean rainfall is calculated for the last years according to sites. a Not included in Carboeurope, no flux data. b No flux data in c Soil temperature not available. d Available soil water measurements. e More than 20% of missing eddy-flux data in A. Granier et al. / Agricultural and Forest Meteorology 143 (2007)

5 A. Granier et al. / Agricultural and Forest Meteorology 143 (2007) Table 2 Main characteristics of the eddy-covariance systems and set-up: measurement height, sensors and u * threshold (see text) Site EC measurement height (m) EC sensors u * threshold Hainich 44 Li6262 (Licor) + Solent R2 (Gill) 0.4 Hesse 23.5 Li6262 (Licor) + Solent R2 (Gill) 0.2 Sorø 43 Li6262 (Licor) + Solent R2 (Gill) 0.25 Vielsalm 40 Li6262 (Licor) + Solent R2 (Gill) 0.5 Nonantola 13 Li6262 (Licor) + Solent R2 (Gill) Espirra 33 Li7500 (Licor) + Solent R2 (Gill) 0.2 Puéchabon 12.2 Li6262 (Licor) + Solent R3A (Gill) 0.35 Le Bray 41 Li7500 (Licor) + Solent R2 (Gill) 0.4 San Rossore 24 Li6262 (Licor) + Solent R2 (Gill) Brasschaat 41 Li6262 (Licor) + Solent R2 (Gill) 0.2 Loobos 27 Li6262 (Licor) and after 2001 DOY Li7500 (Licor) + Windmaster Pro (Gill) Hyytiälä 23.3 Li6262 (Licor) + Solent R2 (Gill) 0.2 Tharandt 42 Li6262 (Licor) + Solent R2 (Gill) 0.3 Norunda 35, 70 and 100 Li6262 (Licor) + Solent R2 (Gill) 0.4 Renon 32 Li7500 (Licor) + Solent R3HS (Gill) 0.3 systematic errors that affect the absolute magnitude of the fluxes, as well as uncertainties by the u * -filtering do not affect estimates of between-year variability, since fluxes in different years should be affected similarly (see Morgenstern et al., 2004). Random errors of up to 50% for the half-hourly flux diminish by integration over a month or a year. For a more detailed uncertainty analysis the reader is referred to Papale et al. (2006). The uncertainty of the flux partitioning is largely determined by the uncertainty in the temperature sensitivity of the base respiration (E 0 ) when extrapolating from night to day. This uncertainty was estimated as the standard deviation of all E 0 estimates for 1 year (cf. Reichstein et al., 2005), assuming that the expected value of E 0 is constant over the year and all variability can be attributed to the estimation error. Clearly, since E 0 can vary through the year, this is a very conservative estimate of error. Errors for each year were summed for the difference between years, assuming that they are independent between years. These uncertainties remained between 4 and 17 g C m 2 month 1 for the summer months and between 25 and 95 g C m 2 year 1 for the whole year, but they do not include measurement uncertainties due to unfavorable conditions, even after u * -correction (e.g. advection at high u * ). Water vapour flux measurements were analysed here at six sites: Hesse, Hainich, Sorø, Tharandt, Hyytiälä and Loobos. At Hesse, additional measurements of sap flow in 10 trees of various diameters were performed (Granier et al., 2000a), allowing at estimating stand scaled tree transpiration (T) and deriving canopy conductance for water vapour (g c ) following the approach of Granier and Bréda (1996). Meteorological conditions were measured halfhourly above all stands including global, net and PPFD radiation, air temperature and humidity, rainfall, wind speed and direction. At some sites, soil water content was automatically monitored using the TDR technique, mostly in the upper soil layers: Hesse: 30 and 55 cm, Hainich: 30 cm, Sorø: 0 to 16 cm, Tharandt: 10 cm, Hyytiälä: 4 to 30 cm and 30 to 68 cm, Fougères: 30 and 55 cm Water balance modelling As soil water content was not measured in all the investigated sites (see Table 1), the water balance model BILJOU ( BILan hydrique JOUrnalier, Granier et al., 1999) was applied at the 16 sites using the abovecanopy measurements of daily climate (rainfall, global radiation, air temperature and humidity, wind speed). The model calculates daily: water fluxes (tree transpiration, understorey evapotranspiration, rainfall interception, and drainage) and soil water content at different depths. Tree transpiration (T) is calculated from the Penman Monteith equation (Monteith, 1965), under the big-leaf approximation. Stomatal regulation during water stress and leaf area index variation (the latter only in the broadleaved stands) is modelled according to Granier et al. (1999, 2000c). Soil water deficit (so called water stress later on) was assumed to occur in forests when the relative extractable soil water (REW) dropped below the threshold of 0.4 (Granier et al., 1999; Bernier et al., 2002) inducing stomatal regulation in forest trees (Lagergren and Lindroth, 2002, reported a or slightly lower threshold value).

6 128 A. Granier et al. / Agricultural and Forest Meteorology 143 (2007) REW t on day t is calculated from soil water content as follows: REW t ¼ EW t (1) EW in which EW t and EW are the actual (day t) and the maximum extractable soil water, respectively. EW is defined as the difference in soil water content between field capacity (soil water matric potential of 0.03 MPa) and wilting point (soil water matric potential of 1.6 MPa) in the entire root zone, which varies between 0.9 and 1.8 m according to the sites. We used soil water retention curves obtained at different depths in the soil to get field capacity and wilting point water content. Three variables are calculated to characterise water stress: start (i.e. the day of year when REW drops below 0.4), duration (i.e. the number of days when REW < 0.4) and intensity (i.e. I s = SUM (t = 1 365) max[0, (0.4 REW t )/0.4], which is dimensionless and ranges between 0 (no stress) and 365 (soil water reserve totally depleted during 1 year). I s reaches ca in the most severe observed water stress condition. Calculation of both stress duration and intensity were performed over the vegetation period: from budburst to leaf fall in the deciduous stands, or over the whole year in the coniferous and the evergreen Mediterranean stands. The site-related parameters of the model describe: (1) the stand structure and the tree phenology: maximum LAI (June July); for deciduous forests the dates of budburst and of complete leaf fall, and (2) soil properties according to a 1D multilayer sub-model (for each soil layer: the maximum extractable water, the root proportion, the bulk density, the soil water content at 1.6 MPa and the porosity). When data on the root distribution in the soil was lacking, an exponential decrease was assumed from soil surface to the maximum root depth, as defined by the bedrock depth, at the sites where this information was available. Fig. 1. Time-course in 2003 of relative extractable water, calculated from bulk soil water measurements using the TDR technique and of modelled REW (full lines) at six European sites. Depths of measurements were: Hesse (a) 30 cm (circles) and 55 cm (crosses), Hainich (b) 30 cm (circles), Sorø (c) 2 probes from 0 to 16 cm (circles and crosses), Tharandt (d) 10 cm (circles), Hyytiälä (e) 4to 30 cm (circles) and 30 to 68 cm (crosses), Fougères (f) 30 and 55 cm.

7 A. Granier et al. / Agricultural and Forest Meteorology 143 (2007) In all simulations (years 2002 and 2003), the model was initialised assuming that soils were saturated (at field capacity) on 1 January In sites where soil water content was monitored by TDR probes, the agreement between measured and modelled REW was good (Fig. 1), except at Tharandt (Fig. 1d) where measurements were only performed in the superficial soil layer (0 10 cm), while roots extend far below. At Hesse, where 6 soils layers were equipped with TDR probes from 10 to 160 cm, the comparison of REW (0 10 cm) to REW (0 160 cm), showed a discrepancy similar to that observed at Tharandt (data not shown). At Hyytiälä, the discrepancy between measurements and model for the first third of the year 2003 can be explained by the snow accumulation in winter and its melting in spring, phenomena that are not taken into account in the water balance model BILJOU. During the vegetation period (DOY ), measured and modelled REW agreed well Climatic conditions There is a large variation in the annual rainfall (longterm average for the last 30 years synoptic weather station data) in the Carboeurope sites we investigated, ranging from 510 to 1083 mm year 1. In most places, the annual rainfall in 2003 was lower than the long term mean; this reduction was 15% on average. Nevertheless, the difference in annual rainfall between the long-term average and the year 2003 was not larger than 325 mm. At Sorø and Puéchabon, the 2003 rainfall was higher than average. The rainfall deficit mainly took place during summer and autumn. The year 2003 was the driest for the last half century in some places (Hesse and Tharandt); at most of the other sites, 2003 was exceptionally dry, but some previous years were drier (1959, 1973, 1976) Tree growth Tree growth was measured in 2003 at only three sites (Hesse: beech, Tharandt: spruce, Norunda: spruce and Scots pine) from circumference measurements at breast height (1.3 m), manually (100 to 300 trees in the footprint of the towers) and at Hesse with automatic dendrometer bands. At Tharandt and Hesse, biomass increment was estimated from allometric relationships relating total tree biomass to circumference and tree height (J.M. Ottorini and N. Le Goff, personal communication). When used, dendrometer bands allowed estimating seasonal growth pattern and especially the date at which of radial growth stopped Plant area index At Hesse, in order to relate the day-to-day variation of transpiration and of the canopy conductance to water vapour to leaf development, we calculated the plant area index (PAI), from the measurement of intercepted radiation by the canopy. Although this method does not allow at calculating leaf area index because trunks and branches intercept radiation, especially during the nonleafed period, the rapid PAI variation during spring and fall are tightly due to leaf expansion and to leaf fall. PAI calculation was made under high diffuse-to-total radiation ratio (>0.5), with an extinction coefficient of 0.395, i.e. the mean value derived from the Beer- Lambert Law and maximum LAI as measured in the litter traps Modelling TER and GPP variation Both daily TER and GPP were fitted (software Statgraphics plus 1 4.1) using non-linear regressions in which the independent variables were climatic factors and REW. At all the 12 sites where CO 2 flux data were available and with less than 20% of missing data (see Table 3), air temperature and REW significantly explained the variation in TER (g C m 2 day 1 ). Unfortunately, soil temperature, which is often considered as the major driving variable of TER fluxes, could not be included in a general model of TER, since it was not measured at the same depth at all the sites (at 0.05 or 0.10 m); moreover, in some sites, this data was missing in the data base (Table 1). We therefore use the air temperature, which is well correlated to soil temperature (average r 2 = 0.85 where both were measured). The following model was fitted on the daily data for both years 2002 and 2003: TER ¼½1þaLnðREWÞŠ½b expðct a ÞŠ (2) where T a is the air temperature (8C), and a, b, c are the fitted parameters. As for TER, REW and T a explained the variation of GPP; global radiation was included in the model. In the deciduous stands, data were filtered by removing the periods when leaf area index was not at maximum, i.e. before DOY 140 or after DOY 280. The following multiplicative non-linear model of GPP was applied at the 12 sites, in which the first function of f 1 of REW is a non-rectangular hyperbola, which is often used to fit the leaf photosynthesis response (Thornley, 1998). The curvature coefficient was set to 1.4, from preliminary tests, as the best fit obtained on sites experiencing a

8 130 Table 3 Leaf area index in 2003 (LAI), maximum extractable water (EW), water stress duration (in days), beginning date of drought onset (day of year), drought intensity estimated with the water balance model, NEE, GPP and TER in 2002 and 2003 Site Tree density (n ha 1 ) Average tree height (m) LAI (m 2 m 2 ) Soil type Soil depth (cm) EW (mm) Stress duration in 2003 Date beginning stress in 2003 Stress intensity in 2003 Stress intensity in 2002 NEE GPP TER Fougères Alocrisol luvisol Hainich Cambisol Hesse Luvisol/stagnic luvisol Sorø Mollisol Vielsalm Cambisol Nonantola Eutric Vertisol Espirra Dystric Cambisol Puéchabon Rhado chromic luvisol Le Bray Hydromorphic podzol San Rossore Brasschaat Umbric regosol Loobos Podzolic Hyytiälä Haplic podzol Tharandt Loamy skeleton podzol Norunda Dystric regosol > Renon Haplic podzol The grey boxes indicate sites that where submitted to significant or severe water stress in A. Granier et al. / Agricultural and Forest Meteorology 143 (2007)

9 A. Granier et al. / Agricultural and Forest Meteorology 143 (2007) large range of REW variations. The second function ( f 2 ) is a rectangular hyperbola depending on global radiation, while the third function ( f 3 ) is a parabola describing the canopy photosynthesis limitation by temperature: GPP ¼ f 1 ðrewþ f 2 ðr g Þ f 3 ðt a Þ (3) f 1 ðrewþ ¼ fd þ erew ½ðd þ erewþ2 2:8deREWŠ 0:5 g 1:4 (3a) f 2 ðr g Þ¼ f R g ðg þ hr g Þ f 3 ðt a Þ¼i þ jt a þ kt 2 a (3b) (3c) where GPP is expressed in g C m 2 day 1, R g is the global radiation (W m 2 ) and [d...k] are the fitted parameters. 3. Results and discussion 3.1. Seasonal variation of water vapour and carbon fluxes An example (Hesse beech forest) of the time-course of water fluxes during 2003 for water vapour flux (E, evapotranspiration), as measured with eddy covariance, and stand-scaled sap flow (T) is presented in Fig. 2. Both E and T behaved similarly. Deviations from daily maximum evapotranspiration (ETM), as calculated with the Penman Monteith equation with calibrated maximum canopy conductance for water vapour for beech (Granier et al., 2000a), were observed starting DOY 178 and lasted until complete leaf fall around DOY 300. A parallel sharp decrease of both REW and canopy conductance for water vapour (g c ) was clearly visible, g c reaching very low values, less than 0.05 cm s 1, whereas it was comprised between 0.4 and 0.6 cm s 1 in The small rainfall events (DOY 183, 208, 228, 243) provoked an immediate, but temporary increase in g c and hence in tree transpiration, lasting for 4 5 days. Fig. 2. Top: variation of evapotranspiration (E, eddy covariance measurements), tree transpiration (T, stand-scaled sap flow), maximum evapotranspiration (ETM, Penman Monteith formula) and of plant area index (PAI, from radiation interception under diffuse radiation, see text) in 2003 at Hesse. The two horizontal grey lines indicate the PAI reduction due to the abnormal leaf fall in August Bottom: daily canopy conductance for water vapour, as calculated from the Penman Monteith formula inversion (squares) and the variation of modelled REW (grey line) and precipitation (Pi, left Y-axis).

10 132 A. Granier et al. / Agricultural and Forest Meteorology 143 (2007) Fig. 3. Variation of NEE (daily data) and that of modelled relative extractable water (REW) during 2003 in: (a) beech, (b) coniferous, and (c) Mediterranean stands. For more clarity, not all the sites are included in this figure. On the same graph, variation in the plant area index is also shown: an exceptional decrease of ca. 1 m 2 m 2 was observed during August (DOY ) that was due to abnormal premature leaf fall of green leaves, probably with embolized petioles. The 2003 time-course of daily-averaged NEE and modelled REW differed for beech, temperate coniferous, and Mediterranean (plus Le Bray) forests (Fig. 3). The three beech forests (Hesse, Sorø and Hainich, Fig. 3a) showed very similar seasonal time-courses and values of NEE. A rapid increase in carbon uptake occurred in spring starting around DOY 110 following bud break, and reached a maximum carbon fixation around DOY 170. At this date, NEE reached ca. 10 g C m 2 day 1. Then, carbon uptake declined to almost zero on DOY 220. Intermittent rainfall events that occurred after this date temporary stimulated the carbon uptake. In the coniferous stands (Fig. 3b), the pattern of variation in daily NEE was less comparable among sites, probably partly because of species differences (Pinus sylvestris and Picea abies). Nevertheless, the maximum carbon uptake occurred around DOY 180, followed by a decrease afterwards, to a minimum around DOY 240, thus similar to, but slightly later than the beech stands. Thereafter, NEE increased (i.e. absolute values of NEE decreased) towards zero, as for beech, and less fluctuated around zero. Note the abrupt increase in REW at Hyytiälä after DOY 280 resulting from heavy rains (84 mm in 5 days). Mediterranean sites also showed very similar seasonal variation in NEE, despite their different species composition: evergreen oak at Puéchabon and maritime pine at San Rossore and Le Bray (Fig. 3c). The main difference was observed in August, when sites typically became carbon sources. Generally, the highest CO 2 uptake was observed in the deciduous forests (ca. 10 g C m 2 day 1 ), while the coniferous forests showed medium values (NEE = 6 to 7 gcm 2 day 1 ). The lowest rates were measured in the Mediterranean stands (ca. 4 gcm 2 day 1 ), which also exhibited the lowest seasonal variation in NEE. In most of the investigated sites, there was a clear decrease in the CO 2 uptake after DOY , much sharper than in 2002 (Fig. 4) accompanying an increasing water stress severity (Fig. 3). After this period, NEE even became positive (i.e. source of CO 2 to the atmosphere) for some days at most of the sites. At Puéchabon and Le Bray, NEE was positive for more than 40 days, indicating that total ecosystem respiration exceeded gross photosynthesis. The seasonal variation in total ecosystem respiration (TER) and gross primary production (GPP) exhibited at all sites a sharp increase in spring, with maximum rates between May and August during wet years (e.g. in Fig. 4 at Hesse and Tharandt, the year 2002 during

11 A. Granier et al. / Agricultural and Forest Meteorology 143 (2007) Fig. 4. Variation of 24-h daily evapotranspiration (E, a and b), of gross primary production (GPP, c and d) and of total ecosystem respiration (TER, e and f) at Hesse (left panels) for beech and at Tharandt (right panels) for spruce in 2002 (closed circles) and 2003 (open circles). Also shown is the temporal variation of REW in 2002 and 2003 at the two sites (Hesse: panel g, Tharandt: panel h). The horizontal line indicates the REW value of 0.4. which REW remained above 0.4), corresponding to the period of high photosynthesis and warm temperatures. During 2003, TER showed a maximum around DOY 150 (4 8 g C m 2 day 1 ), which was more pronounced in the beech stands. Maximum GPP was also reached around DOY 150 at Hesse (15 g C m 2 day 1 ) and Tharandt (13 g C m 2 day 1 ). In 2003, TER and GPP were clearly reduced by drought thereafter. During the driest period, between DOY 210 and 250, TER was reduced by 30 32% in 2003 as compared to 2002; GPP was decreased in a larger proportion, by 36 50% and E by 35% versus 2002 annual carbon and water balances Variation in the annual NEE from 2002 to 2003 are compared in Fig. 5. At Le Bray, where 2002 reached the same drought intensity as 2003 (Table 3), the value of 2001 was used instead of The drought effect on

12 134 A. Granier et al. / Agricultural and Forest Meteorology 143 (2007) evaporative demand in 2003: the potential evapotranspiration (Penman formula) was on average 20% higher in 2003 than in Drought distribution in Europe Fig. 5. Variation in the annual NEE between 2003 and 2002 in 12 forest stands. The positive values indicate a decreased carbon fixation in 2003 compared to Top: variation among the sites. Bottom: the NEE variation as a function of water stress intensity (ISTRESS, dimensionless). At Le Bray, the reference (wet) year was 2001 because 2002 was a dry year. Renon is not shown due to incomplete data set in NEE was variable from site to site, most of them exhibiting a lower NEE in 2003 as compared to An increase (i.e. higher carbon uptake) in NEE from 2002 to 2003 was observed at Renon (113 g C m 2 year 1 ), Vielsalm (110 g C m 2 year 1 ) and Sorø (+63 g C m 2 year 1 ), a moderate decrease (i.e. a lower carbon uptake) of g C m 2 year 1 at Loobos, San Rossore, Hainich, Hyytiälä, Hesse and Puéchabon, and a stronger decrease of g Cm 2 year 1 at Tharandt and Le Bray. The three sites where an increase in NEE was observed were submitted to a moderate drought intensity in 2003 that did not reduce stomatal conductance to a large extend and for a limited period, while the generally observed higher radiation in summer 2003 enhanced photosynthesis. Besides stomatal closure caused by water stress (Fig. 2), the annual E was not reduced to a large extent in 2003 compared to 2002 ( 5to 10% at the driest sites versus 15 to 25% for both GPP and TER), due to the higher Water stress duration in 2002 and 2003, and the date of onset and the drought intensity (ISTRESS) in 2003 are shown in Table 3. In 2002, ISTRESS was null or low, except at the south-western sites (Le Bray, Puéchabon and Espirra) and the Nordic sites (Norunda and Hyytiälä), where the year 2002 was also relatively dry and even slightly drier than The largest values of ISTRESS in 2003 were reached at Hainich, Le Bray, Hesse, Tharandt, Puéchabon, Espirra, Nonantola and San Rossore, where water stress lasted for 3 4 months, while at Vielsalm, Fougères, Sorø, Hyytiälä, Norunda, Renon and Loobos it lasted for only 1 2 months. ISTRESS, onset and duration were well correlated to each other (r 2 = 0.80). The distribution of the ISTRESS indices in Europe is shown in the map of Fig. 6. The most severe drought was observed over a large southwest to northeast transect in Europe. The northwestern coast of France, the North Sea and the Baltic Sea area were less affected by drought. Over the investigated stands, the 2003 reduction in NEE was related to drought intensity (r 2 = 0.60, p = 0.003, Fig. 5) Relating fluxes to water stress intensity Although sites differed in the magnitude of soil water depletion, due to large differences in climate, soil and canopy properties, the time-courses of simulated REW in 2003 were very similar because the main rainfall events occurred quite simultaneously across the sites (around DOY 182, 208, 250 and 275, see Fig. 3). At all sites, extractable soil water dropped for a variable duration during the growing season below REW = 0.4, corresponding to the soil water depletion below the threshold inducing stomatal closure. In 2002 (data not shown), this threshold was either not reached or exceeded for only a few days, except at Le Bray, Espirra and Puéchabon where drought was severe. Large differences among sites were observed during the autumn and winter soil rewetting, mainly because of large differences in rainfall and also in physical soil properties; moreover in some sites (Tharandt, Sorø, Hesse) complete soil recharge did not occur before the end of 2003 or was not even achieved by then. This pattern was also exceptional, as in normal years complete soil recharge is achieved in October or November.

13 A. Granier et al. / Agricultural and Forest Meteorology 143 (2007) Fig. 6. Distribution of water stress intensity in 2003 over the 16 investigated forest sites of Carboeurope. Modelling the stand water balances allowed estimation of the temporal variation in soil water content at each site and hence the opportunity to relate water vapour and carbon fluxes to REW (Fig. 7). We remind that we used here the modelled REW because, except at few sites, there are no routine soil water content measurement in the whole root zone. The maximum water vapour and carbon fluxes (at high soil water content) varied among sites, as reported above (Fig. 3). Daily E, GPP and TER followed comparable patterns: they were reduced when REW decreased below threshold values (REW 0 ). Under declining REW, there was a more gradual and earlier decrease in GPP and E (REW 0 0.4) than in TER (REW 0 0.2). The large scatter of E was due to varying day-to-day weather condition (mainly radiation and VPD). In the six forest stands where vapour fluxes were studied, the decrease in REW from 1 to 0.4 did also not lead to significant change in E (Fig. 7). Below this REW threshold, there was a marked decrease in E in all the sites. One may also note that, for REW ranging from 0.8 to 1.0 (spring periods), the daily water vapour flux remained low as the result of non-complete leaf maturity and/or of temperature limitation for stomatal opening. As for E variability, the scatter in TER and GPP is well related to weather condition: for GPP, the values tending to zero correspond to low radiation and/or rainfall condition. Remarkably large TER fluxes are noticed at Norunda, which have on average an annual carbon balance close to zero (Lindroth et al., 1998) Carbon fluxes At the 12 sites where carbon flux data were available, we tested the influence of the major environmental variables (radiation, air temperature, vapour pressure deficit, modelled REW) on the daily values of TER and GPP, the latter during the growing season for the deciduous stands, using the models (2) and (3). For TER, fitted parameters and coefficients of determination are given in Table 4. All the coefficients of determination where higher than 0.5, except at Renon. The complexity of processes underlying TER probably explains these quite low coefficients of regression. In the site of Vielsalm where drought was very mild, the effect of REW on TER variation was not

14 136 A. Granier et al. / Agricultural and Forest Meteorology 143 (2007) Fig. 7. Variation of daily (24-h sums) evapotranspiration (E, top), total ecosystem respiration (TER, middle) and gross primary production (GPP, bottom) as a function of modelled relative extractable water in the soil (REW) in Evapotranspiration was measured in only 6 sites, carbon fluxes (GPP and TER) in 12 sites. Data of Vielsalm, where REW remained high during summer 2003, are not shown. For better clarity, data are restricted to the period between DOY 140 and 280. The sites Norunda and Renon, exhibiting extreme TER values are distinguished. Symbols are explained in Table 1. Table 4 Coefficients of the non-linear regression between TER, REW and air temperature (see Eq. (2)) Hainich Hesse Sorø Vielsalm Puécha bon Le Bray San Rossore Loobos Hyytiälä Tharandt Norunda Renon a b c r

15 A. Granier et al. / Agricultural and Forest Meteorology 143 (2007) Table 5 Coefficients of the non-linear regression between GPP, REW, global radiation and temperature (see Eq. (3)) Hainich Hesse Sorø Vielsalm Puéchabon Le Bray San Rossore Loobos Hyytiälä Tharandt Norunda Renon d e E f g h i j k r significant. The coefficient c, that describes the rate at which TER increases with temperature, is quite constant within individual tree species: in the 3 beech stands, in the 2 maritime pine stands, in the 2 Scots pine stands. Fits of GPP were generally better than that of TER. The regression parameters and coefficient of determination are given in Table 5. The fit was quite poor for Loobos (r 2 = 0.59), but coefficients of determination were higher than 0.70 at all the other sites, reaching 0.87 at Norunda. At Vielsalm, as for TER, the effect of REW on GPP variation was not significant. At some sites, the model could be improved to some extent when introducing the negative effects of increasing air vapour pressure deficit and/or the direct to diffuse radiation ratio on the GPP variation. Nevertheless, as the Fig. 8. Modelled ecosystem respiration (TER, top) and gross primary production (GPP, bottom) from Eqs. (2) and (3), of as a function of relative extractable water (REW) in 12 forest sites. Air temperature was set at 20 8C, global radiation at 20 MJ m 2 day 1. Note that GPP for Hesse and Sorø cannot be distinguished.

16 138 A. Granier et al. / Agricultural and Forest Meteorology 143 (2007) additional explained variance was limited and, moreover, only improved the fit at some sites, we preferred to use the simplest model (3). The variation of modelled TER and GPP as a function of REW is shown in Fig. 8 in the 12 sites. The maximum assimilation rates (GPP max ) under wet condition varied from 5.4 (Puéchabon) to 12.8 g Cm 2 day 1 (Hesse and Sorø). A general trend of increasing GPP max with LAI was found among sites (data not shown). Norunda exhibited very high rates of TER: under unlimited soil water and warm temperature, TER max reached 9.8 g C m 2 day 1. In the 11 other sites, TER max ranged between 3.0 and 7.5 g C m 2 day 1. Both GPP and TER decreased with REW. However, GPP was more responsive to increasing drought than TER; therefore NEE was reduced to a lesser extent than GPP. The three beech and the Mediterranean stands clearly exhibited a higher sensitivity of GPP to water stress than the temperate coniferous forests. The two models capture satisfactorily the GPP and TER variation (Fig. 9). However, modelled GPP was generally better correlated with observed data than TER. We can partly explain it by the use of air temperature instead of soil temperature in the Fig. 9. Plots of modelled versus measured GPP (left panels) and TER (right panels), daily values for 2002 and Sites are pooled as follow: beech stands (top), Mediterranean stands (middle) and temperate coniferous stands (bottom). The 1:1 line is drawn.

17 A. Granier et al. / Agricultural and Forest Meteorology 143 (2007) Fig. 10. Effect of drought on the reduction of gross primary production (GPP) and on ecosystem respiration (TER), daily estimates, when relative extractable water drops from 1.0 to 0.1, as calculated from models (2) and (3), with air temperature = 20 8C and global radiation = 20 MJ m 2 day 1. Letter B refers to broadleaved, M to broadleaved Mediterranean and C to coniferous stands. model. Moreover, because ecosystem respiration involves more processes than photosynthesis, each responding to specific and numerous drivers and thus may not be captured by an as simple approach as GPP. Moreover, the correlations between measured and modelled values were higher in the beech stands than in the other ecosystems types. From Eqs. (2) and (3) we calculated the theoretical daily reduction in TER and GPP for a drop of REW from 1.0 (i.e. at field capacity) to 0.1 (severe water stress), other factors (radiation, temperature) being set at optimum; the resulting decreases in GPP and TER are presented in Fig. 10. The broadleaved temperate and Mediterranean sites clearly exhibited stronger GPP reductions with drought (45 60%) than coniferous sites (16 38%). The drought-induced decrease in TER also showed, albeit less clearly, a difference among ecosystem types, i.e. a smaller reduction in TER in coniferous than in deciduous and Mediterranean forest ecosystems. At Le Bray and Hyytiälä, TER was more sensitive than GPP to drought TER; at Renon, it was the opposite Reductions in tree growth following the drought Tree growth measurements were performed at Hesse (beech) and Tharandt (spruce), both experiencing a severe and comparable water stress intensity in 2003, and at Norunda (spruce and Scots pine), under medium stress intensity (see Table 3). The seasonal increase in tree circumference at Hesse from dendrometer band measurements (data not shown) started earlier in 2003 than in 2002 as a consequence of hot spring and therefore earlier budburst. The circumference growth was substantially declined on DOY 180 and continued slowly until DOY 190, date of growth cessation, which is exceptionally early. REW reached 0.4 on DOY 178. When comparing , annual NEE was reduced by 18% ( 106 g C m 2 year 1 ) and 24% ( 151 g Cm 2 year 1 ) at Hesse and Tharandt, respectively, while the annual biomass increment was reduced in different proportions, i.e. by 22% and 44%, respectively. At Norunda, the 2003 growth reached only 64% of the mean 2004 and 2005 growth (two humid years, while 2002 was dry), pine trees being more affected by drought than spruces. Quantification of drought should be based on biological variables such as predawn leaf water potential (C p ), a measure of the water constraint that is actually experienced by the vegetation. At Hesse, C p dropped to 2.2 MPa in beech and in the accompanying hornbeam trees, indicating a severe water stress. This C p was lower than the theoretical value of soil water potential at the wilting point ( 1.6 MPa), indicating that these adult trees may extract water from very small soil micro pores, more actively than low vegetation, as was previously observed in Douglas fir (Aussenac et al., 1984), Scots pine (Sturm et al., 1998), Mediterranean oaks (Rambal, 1984) or sessile oaks (Bréda et al., 1995). Unfortunately, measurements of leaf water potential could not be routinely performed at all sites, let alone repeatedly. Therefore, our approach was to estimate soil water content in the root zone by modelling the water balance. The water balance model used here agreed well with soil water content measurements in the sites where those measurements were performed. In 2006, the duration of the water shortage period varied from 28 to 148 days among the sites, depending on stand LAI and soil extractable water. Both drought intensity and duration depended locally on the amount of rainfall during the June to August period. Rainfall events, even small ones, provoked a rapid increase in both evapotranspiration and carbon uptake due to stomata re-opening with free water becoming available for the superficial roots, even if the underlying bulk soil remained dry (Choisnel et al., 1995). Nevertheless, we observed here that small rain events during the summer of 2003 lead to only partial recovery of water vapour fluxes could indicate damages in the soilto-leaf conductivity, probably due to the fine root dying and/or xylem embolism in the most apical part of the trees. In autumn or winter, at some sites, the complete soil water recharge was not even achieved by the end of December. Maximum water fluxes mainly depend on the evaporative demand and on LAI, as previously reported for European beech (Granier et al., 2002) and for a wide