How drought severity constrains gross primary production(GPP) and its partitioning among carbon pools in a Quercus ilex coppice?

. The partitioning of photosynthates toward biomass compartments plays a crucial role in the carbon (C) sink function of forests. Few studies have examined how carbon is allocated toward plant compartments in drought-prone forests. We analyzed the fate of gross primary production (GPP) in relation to yearly water deﬁcit in an old evergreen Mediterranean Quercus ilex coppice severely affected by water limitations. Carbon ﬂuxes between the ecosystem and the atmosphere were measured with an eddy covariance ﬂux tower running continuously since 2001. Discrete measurements of litterfall, stem growth and f APAR al-lowed us to derive annual productions of leaves, wood, ﬂowers and acorns, and an isometric relationship between stem and belowground biomass has been used to estimate perennial belowground growth. By combining eddy covariance ﬂuxes with annual net primary productions (NPP), we managed to close a C budget and derive values of autotrophic, heterotrophic respirations and carbon-use efﬁciency (CUE; the ratio between NPP and GPP). Average values of yearly net ecosystem production (NEP), GPP and R eco were 282, 1259 and 977 g C m − 2 . The corresponding aboveground net primary production (ANPP) components were 142.5, 26.4 and 69.6 g C m − 2 for leaves, reproductive effort (ﬂowers and fruits) and stems, respectively. NEP, GPP and R eco were affected by annual water deﬁcit. Partitioning to the different plant compartments was also impacted by drought, with a hierarchy of responses going from the most affected – the stem growth – to the least affected – the leaf production. The average CUE was 0.40, which is well in the range for Mediterranean-type forest ecosystems. CUE tended to decrease less drastically in response to drought than GPP and NPP did, probably due to drought acclimation of autotrophic respiration. Overall, our results provide a baseline for modeling the inter-annual variations of carbon ﬂuxes and allocation

2010). Estimations and simulations of carbon uptake by forest ecosystems have been greatly improved in recent decades, but, unfortunately, how this assimilated C is transferred from the atmosphere to the terrestrial biomass remains a challenging issue. Luo et al. (2011) highlighted a lack of mechanistic understanding on this question and suggested developing "generalizable models of C allocation to biomass growth of plant parts, respiration, nonstructural C reserve, reproduction and defense". A recent synthesis has demonstrated that the partitioning of gross primary production (GPP), among aboveground and belowground production and respiration, can vary greatly across biomes, according to climate and fertility (Litton and Giardina, 2008). However, a more detailed understanding of how environmental factors affect the distribution of C among the different tree parts at the ecosystem scale remains a crucial step to improving the accuracy of local and global vegetation models (Fatichi et al., 2013;Leuzinger and Quinn Thomas, 2011).
Understanding C allocation patterns appears particularly important in drought-prone areas, such as those with a Mediterranean-type climate, which are particularly vulnerable to ongoing climate change (Giorgi, 2006). According to global and regional climate models, Mediterraneantype ecosystems (MTEs) will suffer longer and more intense droughts as a result of (1) increasing temperature and decreasing rainfall (Hoerling et al., 2011), (2) a change in largescale circulation conditions (Kjellström et al., 2013), and (3) the persistence of heat wave anomalies (Jaeger and Seneviratne, 2011). In MTEs, drought is already the prevailing constraint on the net ecosystem productivity (NEP) (Allard et al., 2008;Grünzweig et al., 2003). This sink strength is likely modified by the differential sensitivity to water limitation of leaf photosynthesis and whole-tree respiration, and of the C allocation to short-and long-lived pools. The representation of C use in models currently lacks consensus and is achieved by a plethora of concurrent approaches (Franklin et al., 2012). This modeling deficiency seems to be due to the difficulty in interpreting this information in generic schemes that are valid under a wide range of conditions, particularly water limitation.
So far, studies addressing the question of C use in MTEs have relied on the coupling of field data of standing biomass and growth compartments with simulation models. Pioneering works started in the 1970s onwards (Eckardt et al., 1975). Oechel and Lawrence (1981) applied the process-based model MEDECS to eight woody Mediterranean species growing in California chaparral and Chilean matorral. The model scaled up leaf level respiration and assimilation, together with stem respiration to yield yearly C budgets using a radiation transfer scheme. The hierarchy of C allocation to leaves, stems and roots followed species-specific rules and a phenological calendar. From this modeling exercise, the authors deduced changes in C use that deeply modified the respiratory costs in response to changes in air temperature. Yet, the effect of drought on C use remains more difficult to understand and simulate.
Forests and woodlands dominated by the evergreen oak Quercus ilex L. (hereafter Q. ilex) occupy large areas in the surrounding area of the Mediterranean Sea (Quézel and Médail, 2003) and are emblematic of the MTEs. Due to its resprouting nature, Q. ilex can remain in the same place for hundreds of years, and populations display minimal changes in stool number per area. Very high survival rates and fast recovery of its foliage after complete dieback (Lloret et al., 2004) reflect its high ability to survive in damping climate extremes (Misson et al., 2011). In contrast, co-occurring obligate seeders are subjected to all the vicissitudes of regeneration, and are particularly affected by drought mortality at the seedling stages and by wildfires (Ackerly, 2004;Zavala, 1999). The growing interest in resprouting ability as a major plant functional trait is reflected in a number of recent contributions aimed at understanding the biogeography at developing functional models of resprouting species (Clarke et al., 2010;Vesk and Westoby, 2004;Vilagrosa et al., 2014). Resprouters have the particularity to store considerable amounts of C belowground at the cost of high maintenance respiration (Iwasa and Kubo, 1997). Characterizing the ecosystem C use for such species is important for managing and predicting the response of Mediterranean forests to the ongoing climate changes.
The functioning of Quercus ilex stands in the south of France was simulated by Hoff et al. (2002) and Hoff and Rambal (2003) using the Forest-BGC model. C use rules in this simple model are implemented so as to follow an optimal trajectory: trees use C first into leaves and fine roots for maximizing productivity, while minimizing water limitation; finally, stems appear as an end product built with the remaining C. Other modeling exercises with Q. ilex ecosystems also retained water-related constraints for their C use rules. Gracia et al. (1999) developed a dynamic growth model, where the partitioning of growth between leaves and perennial wood compartments is performed so as to fulfill the assumptions of the pipe model theory (Shinozaki et al., 1964; see also Mäkelä, 1986 for substantial accounts;Valentine, 1985); i.e., so as to maintain the sap area / foliage area ratio constant. Gracia et al. (1999) also constrain growth to fine roots to follow the functional balance hypothesis (Brouwer, 1962). Both of the abovementioned modeling exercises yielded credible results when validated against yearly variations of radial growth. Fortunately, the increasing availability of longterm field measurements of productivity and eddy covariance fluxes can now help to refine these previous modeling hypotheses.
In this study, our main objectives were: (1) to evaluate the fraction of GPP partitioned to aboveground and belowground parts in a Quercus ilex forest by comparing different ecosystems across a range of climate, management and drought resistance of dominant species, and (2) to assess how year-toyear variation in drought severity impacts the partitioning of Biogeosciences, 11, 6855-6869, 2014 www.biogeosciences.net/11/6855/2014/ GPP between production and respiration, and among aboveground and belowground C pools. For these purposes, we used long-term data of eddy covariance fluxes and primary productivity of aboveground components (leaves, flowers, fruits and stems), plus punctual data of root biomass taken from literature and our own excavation of four Q. ilex trees. and Juniperus oxycedrus compose a sparse shrubby layer with a percentage cover lower than 25 % and a height less than 2 m. The area has a Mediterranean-type climate. Rainfall mainly occurs during autumn and winter, with about 80 % taking place between September and April. The mean annual precipitation is 916 mm, with a range of 556-1549 mm recorded over the 1984-2011 period. Mean annual temperature over the same period was 13.0 • C, with a minimum in January (5.5 • C) and a maximum in July (22.9 • C). The rocky soil is formed on Jurassic limestone; on average, the volumetric fractional content of stones and rocks is about 0.75 for the top 0-50 cm and 0.90 below. The stone-free fine fraction of the soil is a homogeneous silty clay loam (USDA texture triangle) within the top 0-50 cm layer (38.8 % clay, 35.2 % silt and 26 % sand). The fine fraction fills up the space between stones and rocks, and provides a source of water throughout the long dry summers for the deep-rooted Q. ilex (Rambal, 2011). The highly permeable soil prevents any surface runoff to occur even for high intensity rain events.

Water limitation: soil water balance model and drought index
Soil water storage integrated over the rooting depth, that is ca. 4.5 m (Rambal, 2011), has been measured during the vegetative periods of 1984-1986 and since July 1998 onwards, at approximately monthly intervals, using a neutron moisture gauge (see Hoff et al., 2002). Discrete measurements were interpolated at a daily time step with a soil water balance model proposed in Rambal (1993) and further used in Grote et al. (2009). The drainage curve relating deep drainage to soil water storage depends on the stone content over the whole-soil profile (Rambal, 1990). The model was driven by daily values of incoming solar radiation, minimal and maximal temperature and rain amount. Soil water storage and soil water potential were related by a Campbell-type retention curve (Campbell, 1985) whose parameters are strongly dependent on soil texture (see details in Rambal et al., 2003). Comparison of measured against simulated values of soil water storage (in mm) and predawn leaf water potential (in MPa) displayed very good agreement. Leaf water potential values came from discrete measurements performed on the study site (see Limousin et al., 2012 for a substantial account). For soil water storage, reduced major axis (RMA) regressions yielded SWS sim = α rma SWC obs +β rma with α rma ± standard error (SE) = 0.94 ± 0.03, β rma ± SE = 6.0 ± 4.4, R 2 = 0.93, F = 1137, p <0˙.0001 and n = 91; for the predawn potential, ψ pdsim = α rma ψ pdobs + β rma with α rma ± SE = 0.93 ± 0.05, β RMA ± SE = −0.09 ± 0.09, R 2 = 0.840, F = 273.3, p < 0.0001 and n = 54. The continuous daily course of relative water content, RWC, was derived from SWS sim divided by the soil water storage at a field capacity that we chose to fix at 205 mm. This value corresponds to that observed after 2 days of free drainage in a cool, wet period after a substantial rain event. For characterizing the whole-year water limitation, we calculated the water stress integral (WSI) as the yearly sum of ψ pdsim . For days with RWC ≥ 1, ψ pdsim is fixed to −0.03 MPa. The WSI is expressed in MPa day.

Drought frequency analysis
The return periods for drought events were calculated using a monthly 239-year precipitation historical data set (1762-2011) for downtown Montpellier. This data set was scaled to our experimental site using overlapping precipitation data from 1984 to 2011. As shown by Rambal and Debussche (1995) and López-Moreno et al. (2009), the coefficient of variation for precipitation is regionally conserved and was used to fit theoretical lognormal distribution functions for extreme precipitation events at our site. Return periods were calculated as 1/p, where p is the probability of occurrence (Rambal and Debussche, 1995).

Carbon fluxes and ancillary data
Daily climate data, further used as model inputs for a water budget model, came from a weather station located 200 m away from the flux tower.
Eddy covariance fluxes of CO 2 , sensible heat, latent heat and momentum were measured continuously since 2001 at the top of a 12 m high tower that was approximately 6 m above the canopy. Our eddy covariance facility included a three-dimensional sonic anemometer (Solent R3, Gill Instruments, Lymington, England) and a closed path infrared gas analyzer (IRGA; model LI 6262) (Li-Cor Inc., Lincoln, Nebraska, USA), both sampling at a rate of 21 Hz. Flux data were processed with protocols defined within the Carbo-Europe network (www.carboeurope.org, Aubinet et al., 2000). Processing schemes of FLUXNET have been used for filling data gaps and partitioning NEP into GPP and ecosystem respiration R eco (Papale, 2006;Reichstein et al., 2005). The half-hourly fluxes were summed at yearly time steps for further analysis. Photosynthetically active radiation PAR top was recorded at the top of the flux tower. Photosynthetically active radiation (PAR) was recorded at the top of the flux tower, PAR top . The fraction of PAR absorbed by the canopy (f APAR ) was derived from 14 PAR sensors randomly set up in understorey locations and measuring PAR below : (1)

Leaf production and other growth components
ANPP stem was estimated from yearly measurements of stem DBH and the allometric relationship between stem biomass and stem DBH. ANPP leaf and ANPP reprod were derived from monthly litter falls measured on 26 litter traps of 0.141 m 2 area. ANPP reprod comprised flowers and acorns. ANPP leaf was derived by estimating yearly changes of leaf mass at peak leaf area index plus the amount of leaves lost as litter.
Leaf production in year t occurred from May to June, and M leaflitter was calculated as the sum of monthly values of leaf litter fallen from August t − 1 to July t. M leaflitter was corrected for mass loss at abscission, using the results of Cherbuy et al. (2001): Peak LAI = PAI−SAI was estimated from continuous measurements of half-hourly f APAR between 11.00 a.m. and 1.00 p.m. from day of year (DOY) 205 to 225. We first derived the plant area index (PAI) by using Beer's law with an extinction parameter equal to k/ sin β. The parameter k was set to 0.72 as in Rambal et al. (2003), and β is the solar elevation angle. The stem area index (SAI) was estimated by image processing of hemispheric photography. It was assumed constant for the whole period and equal to 0.5 (Poncelet's unpublished data). LAI was converted to leaf mass with a canopy-averaged leaf mass per area of 215 g m −2 (see Rambal et al., 1996). The below-canopy PAR sensor network was set up in 2001, so the leaf production for 2001 was not available. Even though Q. ilex is a strong emitter of terpenoids (Staudt et al., 2002), biogenic volatile compound emissions are relatively minor C sources and they were neglected here. So, the aboveground net production was computed as In 2005, we observed a massive outburst of Lymantria dispar. Grazing from caterpillars drastically impacted the leaves, so we decided to exclude data from 2005 in our calculations. Data for the belowground perennial components were obtained by excavating four stumps at our site and from literature values published by Canadell and Roda (1991) and Djema (1995) for Q. ilex coppices growing in northeast Spain under similar climate conditions. We compiled 19 biomass values for root crown, roots greater than 5 cm, and roots ranging from 1 to 5 cm diameter. The whole perennial belowground compartment is the sum of root crown and large roots.
We obtained an isometric relationship between stem and belowground biomass, with a slope equal to 1.068 ± 0.1235 (s x,y = 62.2, n = 19, p <0˙.001) (Appendix Fig. A1). All these data came from excavations in very stony soils and only concerned the top 0-1 m layer. A significant part of the root system was not extracted, because we observed that tap roots are able to uptake soil water at depths ranging between 4 and 5 m (Rambal, 2011). We thus applied a conservative correction factor of 10 % to account for the missing root part. Our belowground / aboveground ratio could be considered constant whatever the stool size, so we propose an isometric partition of C between these two perennial compartments. We postulate that the error we made in estimating BNPP coarse is equivalent to the one we made in evaluating the change in stem biomass: with BNPP representing belowground net primary production. Fine root production was taken from literature values. López et al. (2001a) extensively monitored fine root productivity in a Q. ilex coppice. They found annual fine root production over the 0-60 cm soil layer to be quasi identical to leaf production (average fine root / leaf production ratio over 2 years was 1.04). We correct this value for the whole profile using a ratio of 1.25, based on the distribution of fine roots over the soil profile proposed by Jackson et al. (1997) for sclerophyllous shrubs and trees, and the increase in fine root turnover rate with depth (López et al., 2001b): Biomasses were converted to C using tissue-specific C contents whenever available; otherwise, 0.48 was used as a default.

Carbon budget estimate
The different components of the carbon budget were related to each other according to three identities considered here as yearly sums (Fig. 1): R a is the autotrophic respiration, including both growth and maintenance components, with R aa and R ab standing for  the aboveground and belowground parts, respectively. R h is the heterotrophic respiration. Uncertainty estimation of fluxes were around 20 g C m −2 yr −1 , 30 g C m −2 yr −1 and 40 g C m −2 yr −1 for NEE, GPP and R eco , respectively (Misson et al., 2010; see also Stauch et al., 2008).
Total belowground carbon flux (TBCF) was defined as carbon allocated belowground by plants to coarse and fine roots production, root respiration and root exudates and mycorrhizae. TBCF is either respired by microbes or roots (measured as soil-surface CO 2 efflux) or stored in soil as organic matter in the litter layer or in living and dead roots. Growth respiration was calculated using the yield of growth processes Y (Thornley, 1970). This yield is the amount of biomass increment per unit of C substrate used in growth processes. It was expressed in g C of new biomass (g C of substrate used in the growth processes) −1 . For Q. ilex in Puéchabon, the Y parameter has been estimated to 0.8 g C appearing in new biomass per g of C substrate utilized (Rambal et al., 2004). In Eqs. (6), (7) and (9), we neglect nonstructural C storage above or belowground. In the carbon budget, we wrote an equation in which C balance is 0, independently of the water limitation, and consequently the storage of nonstructural C pool remains constant (see Ryan, 2011;Sala et al., 2010;Wiley and Helliker, 2012 for the role of nonstructural carbohydrates in coping with drought).

Environmental conditions and exceptional years
Over the study period ( 1780, 1817, 1929, 1945 and 1995, but all these years displayed less severe droughts than 2006. So, over the 2001-2011 period, we observed a very large range of water limitation from well-watered conditions to severe drought. There was no significant covariation between mean annual temperature and WSI.

C fluxes and production
The mean gross C input (GPP) was 1259 g C m −2 yr −1 and its coefficient of variation (CV), or between-year variation, was 13.3 %. For NEP, the mean value was  Figure 3. Ordinary least-square regression lines between the WSI and gross primary production GPP (light grey circle) and NEP (dark grey circle). WSI is expressed in MPa day, and both GPP and NEP are expressed in g C m −2 yr −1 . Data from 2005 that were not used in the analysis were also plotted (empty square for GPP and empty triangle for NEP).

281
.7 g C m −2 yr −1 with a larger CV of 33.5 %, and for R eco , it was 977.2 g C m −2 yr −1 , with a CV = 8.9 % (Fig. 2). The average LAI was 2.25 ± 0.2, which corresponds to a supported leaf mass of 231.7 g C m −2 (n = 10) with a coefficient of variation of CV = 9 % (Fig. A2.). Our calculation of the leaf production yields an average value of 142.5 g C m −2 yr −1 (n = 9) with a large CV of 28.5 %. The leaf production ranged from 202.8 ± 77.1 g C m −2 yr −1 in 2006, the year after the Lymantria dispar outburst and heavy grazing, to 69.6 ± 58.2 g C m −2 yr −1 the following year in 2007. The reproductive effort, ANPP reprod , evaluated in pooling flowers and acorns, displayed the greater between-year variation, with a 42.5 % CV, and a mean value of 26.4 g C m −2 . The components of ANPP reprod were, on average, 11.0 g C m −2 yr −1 for flowers (CV = 48.5 %) and 15.4 g C m −2 yr −1 for acorns, which displayed the largest variation (CV = 87.8 %). Summing leaves plus flowers and acorns, we obtained an average 169.6 g C m −2 yr −1 , which accounted only for 16.9 % of the yearly GPP.

Relationships between production components and water limitation
Significant linear declines of GPP, NEP and R eco with increasing drought severity were observed across years (Table 1; Fig. 3). Respectively, 72 and 80 % of the variance in GPP and NEP was explained by the WSI. The slopes of the GPP and NEP vs. WSI lines were 1.91 ± 0.43 and 1.15 ± 0.20, respectively, which means that we project a decline of GPP of 191 g C m −2 yr −1 and of NEP of 115 g C m −2 yr −1 for an increase in drought severity of 100 MPa day, expressed in terms of WSI. The sensitivity to drought of R eco was lower than that for the two other components of the whole-ecosystem C budget, with a lower slope of 0.77 ± 0.32 associated with a lower explained variance of 42 %. Among the aboveground tree compartments, the most affected by drought was the stem (Fig. 4), with dANPP stem / dWSI = 0.42 ± 0.10 (Table 1; Fig. 4). According to the linear equation fitted between ANPP stem and WSI, the predicted allocation of C to the stem ranged from 120.9 g C m −2 for a hypothetic wet year that underwent a WSI of −100 MPa day (WSI in 2004 equaled −112.6 MPa day), to 0 in a severely dry year with a WSI of −390 MPa day. Reproduction was also affected by water stress, with dANPP reprod / dWSI = 0.10 ± 0.04 (Fig. 5). In contrast, no significant relationship was found between WSI and ANPP leaf (p = 0.54; Table 1, Fig. 6b). ANPP leaf was, however, significantly related to the WSI of the previous year, with a slope of 0.41 ± 0.15 and an explained variance of 52 % (Fig. 6a).

Relationship between carbon-use efficiency (CUE) and water limitation
By combining the latter results with Eqs. (6) to (9), a model of C use changes not in drought severity can be proposed. Figure 7a depicts the changes of GPP and NPP, and of the above and belowground compartments with WSI. Carbon-use efficiency (CUE = NPP / GPP) is also presented. For WSI declining from −100 MPa day in a wet year to −400 MPa day in a particularly dry year, NPP and CUE decline from 621.4 to 339.4 g C m −2 yr −1 and from 0.419 to 0.373, respectively. Figure 7b depicts the declines of R eco and NEP with WSI and the corresponding changes of the ratios of autotrophic respiration to GPP (R a / GPP) and heterotrophic respiration to whole-ecosystem respiration (R h /R eco ). The R a / GPP ratio increased from 0.581 to 0.627 for a change of WSI from Table 1. Parameters of the linear ordinary least-square regression lines between the WSI in MPa day and components of the ecosystem yearly C budget and aboveground components of the productivity. α OLS is the slope of the Y vs. X relationship. GPP, R eco and NEP are gross primary productivity, ecosystem respiration and net ecosystem productivity, respectively, in g C m −2 yr −1 . The components of the aboveground productivity for leaves, reproductive effort and stem ANPP leaf , ANPP reprod and ANPP stem are also expressed in g C m −2 yr −1 .  −100 to −400 MPa day. For the same decline in WSI, the ratio of R h /R eco increased from 0.192 to 0.321, with R h slightly increasing from 205.1 to 268.1 g C m −2 yr −1 .

Carbon-use efficiency in a Mediterranean coppice -management and drought-adaptation constraints on carbon allocation rules
CUE, the ratio of net primary production (NPP) to gross primary production (GPP), describes the capacity of forests to assimilate C from the atmosphere into terrestrial biomass. CUE of forests has been assumed, by some authors, to be a constant value of 0.47 ± 0.04 (Gifford, 2003;Waring et al., 1998), which supposes that tree respiration is a constant fraction of GPP. Contrary to this assumption of con-stancy, substantial variations in CUE have been reported in forest ecosystems. Medlyn and Dewar (1999) demonstrated that CUE likely ranges between 0.31 and 0.59, and a more recent synthesis by DeLucia et al. (2007) showed that the slope of the relationship between NPP and GPP (CUE) was 0.53, ranging from 0.23 to 0.83 among forest types. CUE decreased with increasing age, and a substantial portion of the variation among forest types was caused by the ratio of leaf mass / total mass. For a ratio of leaf mass / total mass of 0.03, corresponding to our Q. ilex forest, DeLucia et al. (2007) predicted a CUE of 0.38, similar to the mean of 0.40 obtained here, and the same value that Oechel and Lawrence (1981) obtained for Californian and Chilean shrub and tree species.
With the process-based simulation model Gotilwa applied to a Q. ilex coppice in northeastern Spain, Gracia et al. (1999) predicted a CUE of 0.41. In contrast, Luyssaert et al. (2007) derived a surprisingly high value of 0.54 from a global database for their so-called "Mediterranean warm evergreen" biome ( Table 2). The low ecosystem CUE observed at our site (around 0.40) could be due to the ancient management of the ecosystem as a coppice. The large belowground biomass and respiratory maintenance costs associated with this management system may alter C use rules and constrain CUE compared to more productive high forests (Salomón et al., 2013). Furthermore, relatively high R aa (see below) could be associated with the role of aboveground organs in storing nitrogen and nonstructural carbohydrates. 1-year old leaves act as reservoirs contributing to spring shoot growth (Cherbuy et al., 2001), while stumps and stems contain a large amount of parenchyma, helping the tree to resprout after perturbations. Accurately quantifying the relative importance of respiratory sources is an important step towards understanding the whole C budget. Under the steady-state assumption of Eq. (9) (Raich and Nadelhoffer, 1989), our values of GPP, ANPP and R aa resulted in TBCF = 670 g C m −2 yr −1 . R aa was 460 g C m −2 yr −1 , a value estimated from leaf respiration and stem CO 2 efflux measurements made at our www.biogeosciences.net/11/6855/2014/ Biogeosciences, 11, 6855-6869, 2014  Waring et al. (1998) Broad range of forests (BRFs) 0.47 ± 0.04 Medlyn and Dewar (1999) BRFs 0.31-0.59 Gracia et al. (1999) Quercus ilex coppice 0.41 De Lucia et al. (2007) BRFs 0.53(0.23-0.83) Luyssaert et al. (2007) Mediterranean warm evergreen 0.54 Litton and Giardina (2008) BRFs 0.43 Luyssaert et al. (2009) Temp. and boreal forests 0.51 ± 0.02 Piao et al. (2010) BRFs ( year yielded a TBCF of 630 g C m −2 yr −1 . With our estimate of BNPP = 270 g C m −2 yr −1 , the R ab ranged between 360 and 400 g C m −2 yr −1 . Finally, we could deduct an R h ranging between 210 and 230 g C m −2 yr −1 by summing the three respiration components to reach the wholeecosystem respiration R eco . For comparison, the metaanalysis of Litton and Giardina (2008) report a TBCF of 705 g C m −2 yr −1 and a BNPP of 334 g C m −2 yr −1 , and Rodeghiero and Cescatti (2006) measured, in a more mesic Quercus ilex coppice in which the soil respiration is very high (1079 g C m −2 yr −1 ), a TBCF of 564 g C m −2 yr −1 with the two belowground respiration components R ab and R h being equal.

Sensitivity of carbon use and partitioning to between-year variation in water limitation
To characterize year-to-year variations in drought severity, we used a long-term cumulated water stress index: the WSI. This concept likely originated in Schulze et al. (1980a, b), who related changes in normalized maximal assimilation rates and daily carbon gain with the sum of water stress obtained by cumulating daily pre-dawn water potentials from the day of the last rainfall to the day under consideration. Later, Wullschleger and Hanson (2006) did the same with transpiration rates from trees growing in a throughfall displacement experiment. This cumulated water stress, called water-stress integral, or WSI by Myers (1988), has been applied to predict growth processes occurring at longer timescales, such as canopy development, litter fall dynamic and tree radial growth (Benson et al., 1992;Raison et al., 1992a, b). In our study, we demonstrated that WSI was significantly related to the current year reproductive effort, secondary growth and all ecosystem C fluxes (see also Arneth et al., 1998), and that it was useful in explaining how the previ-ous year's drought limitation affected the leaf production in the subsequent year. GPP, R eco and NEP were largely impacted by water limitation. The decline of GPP with drought has been observed in our site at different time and space scales. At a seasonal timescale, Limousin et al. (2010) intensively discussed how leaf photosynthetic limitations were related to predawn water potential. At a daily timescale, GPP estimated from eddy correlation fluxes was related to predawn water potential . The ANPP components have also been shown to be impacted by drought severity, with a hierarchy of responses going from the more affected -the stem -to the less affected -the leaves ( Table 1). The larger sensitivity of stem growth validates the hypothesis of the Forest-BGC model (Hoff et al., 2002) in which trees allocate C first to leaves and fine roots, for maximizing productivity while minimizing water stress, and then to stems; and then to stems which appear as an end product built with the remaining C. The reproductive effort also declined significantly with increasing drought, although it represented a smaller C use. Acorn production -the larger component of reproduction -has been shown to be influenced by water availability during the fruiting process, in particular during the initial (spring) and advanced (summer) stages of the maturation cycle (Pérez-Ramos et al., 2010).
The leaf production was not related to the current year's WSI but to the previous year's WSI. Limousin et al. (2012) observed that, in Q. ilex, the leaf litterfall was also positively correlated with the previous year's WSI so that more leaves were shed and replaced following wet years than following dry years. This phenomenon might be explained by the cost-benefit hypothesis (Chabot and Hicks, 1982;Kikuzawa, 1991): if the leaf carbon assimilation is reduced by water limitation during a dry year, the leaf life span should increase for the leaf lifetime carbon gain to pay back the leaf construction cost, and thus fewer new leaves need to be produced to maintain the LAI. This results in an alternation of years with high leaf production / shedding following wet years and years of opposite characteristics, as commonly observed in evergreen species and, in particular, Q. ilex (Montserrat-Marti et al., 2009;Ogaya and Penuelas, 2006;Rapp, 1969). Such a mechanism may also contribute to maintain the water transport capacity of Q. ilex under long-lasting drought, as proposed by Martin-StPaul et al. (2013). Current-year drought causes GPP and, less so, NPP to decline, so that CUE declines slightly. In a wet year following a drier one, CUE could decline because ANPP stem would be ruled by current climate and would be high, corresponding to a wet year, while leaves would be produced in fewer amounts due to the 1-year-lag effect of drought on leaf production, and CUE could decline. Further research could be necessary to quantify such a carry-over effect on CUE. Perhaps the strategy of Q. ilex to buffer the hydraulic system from climatic extremes has a penalty on CUE. Based on the responses to drought of the different compartments and on the assumptions stated above (see Sect. 2), we calculated the yearly CUE response to drought (Fig. 7a). CUE slightly decreased with drought from 0.419 at WSI = −100 MPa day to 0.373 at WSI = −400 MPa day. Interestingly, CUE declined at a slower rate than GPP and NPP in response to water deficit (Fig. 7a). Maseyk et al. (2008) reported a constant CUE of 0.4 in a Pinus halepensis forest growing in a semi-arid Mediterranean-type climate and proposed that acclimation of maintenance respiration to dry conditions could help in maintaining CUE and keeping productivity relatively high under such water limited climate. Recent studies at our site showed that respiration rates declined exponentially in both leaves and stems as tree water availability decreased through summer months (Rodríguez-Calcerrada et al., 2011. Based on the relationships between leaf / shoot predawn water potential and leaf / stem respiration, we calculated that stem and foliage CO 2 efflux declined by 4.7 and 7.1 %, respectively, for an increase of drought severity of WSI = 100 MPa day. Altogether, acclimation of leaf, stem and root respiration to plant water deficit buffers NPP sensitivity to drought and contributes to keeping CUE relatively constant across years of widely different rainfall and vegetation stress. The ultimate reasons for such reduction in respiration rates are still unclear, but it appears that reduced demand of respiratory products from growth and maintenance processes may cause a down-regulation of mitochondrial activity (Atkin and Macherel, 2009).
Besides reductions in autotrophic respiration, changes in R h contribute to complicating our understanding of the impact of drought on the whole-ecosystem C sink strength. In trees, acclimation refers to strictly physiological processes, while soils changes in R h refer to ecosystem-level phenomenon potentially driven by multiple mechanisms, including substrate depletion, changing microbial community composition and physiological changes. Substantial questions remain about its response to soil water status, the interactions with substrate quality, and the role of the top soil dryingrewetting cycles (Wei et al., 2010). The course of soil water content at timescales shorter than the season is not necessarily correlated to the WSI. In Mediterranean-type ecosystems, R h is likely more influenced by an unpredictable supply of substrate to the rhizosphere than by changes in the microbial community or its efficiency (Curiel Yuste et al., 2014). Finally, we suggest, as Hopkins et al. (2013) did, that substrate availability sensu lato, including GPP and storage of nonstructural C pool (neglected here), may be the ultimate driver of the two respiration fluxes.

Conclusions
Comparative measures of ecosystem fluxes and production components across 11 yr of contrasting water limitations in a Q. ilex coppice help to better understand how Mediterraneantype forest ecosystems will respond to the ongoing climate change and to better project future C sequestration capacity.
We observed a clear effect of water availability in limiting all the ecosystem fluxes GPP, R eco and NEP, and that the drought-induced decline in R eco dampens the decline of the ecosystem C sequestration under drought conditions. In parallel, all the growth components were found to be affected by water limitation, with a partition of GPP into tissues that tends to minimize the negative impacts of drought on growth. An important result is that all the changes followed the same trajectory as water stress varied over a large range of conditions, from a wet year to a dry year occurring only once every 67 yr. We did not observe any tipping point or discontinuity in the C partitioning pattern. On average, only 40 % of the carbon assimilated as gross photosynthesis was used to construct new tissues, with the remaining 60 % being respired back to the atmosphere as autotrophic respiration. This low ecosystem CUE could be inherited from the ancient manage-ment of the ecosystem as a coppice and its large amount of standing belowground biomass.
There are several ecological issues that question the values of the estimated C fluxes and their changes in increasing drought severity. It appeared, in our case, that autotrophic respiration by trees and heterotrophic respiration by soil microorganisms are primarily responsible for mediating the larger part of the carbon exchanges between the biosphere and atmosphere. Climate changes and projected increasing dryness have the potential to influence the activity of trees regulating exchanges among the carbon pools. Functional "down-regulation" or acclimation of plant respiration could reduce the respiratory autotrophic loss of ecosystems, but, unlike plant components, the existence of this phenomenon in heterotrophic respiration remains more controversial (Harmon et al., 2011;Wieder et al., 2013). Current models can simulate GPP relationships with autotrophic fluxes in a warmer environment (Piao et al., 2010;Wythers et al., 2013), yet the parameterization of models able to capture the apparent respiratory acclimation of both R a and R h to water limitation of ecosystems is an emerging challenge for the modeling and flux research communities. We suggest that both communities adopt a bottom-up approach to advance our understanding at tissue, tree and ecosystem scales to increasingly larger timescales and space scales.   Figure A2. Time course of the peak LAI derived from continuous measurements of half-hourly f APAR between 11.00 a.m. and 1.00 p.m. from DOY 205 to 225. The stem area index (SAI) was estimated by image processing of hemispheric photography and assumed constant for the whole period and equal to 0.5. The relationship between leaf area index and water stress integral (WSI) is statistically non-significant.