Characterisation of ecosystem water-use efficiency of european forests from eddy covariance measurements

Abstract. Water-use efficiency (WUE) has been recognized as an important characteristic of vegetation productivity in various natural scientific disciplines for decades, but only recently at the ecosystem level, where different ways exist to characterize water-use efficiency. Hence, the objective of this research was (a) to systematically compare different ways of calculating ecosystem water-use efficiency (WUEe) from eddy-covariance measurements, (b) quantify the diurnal, seasonal and interannual variability of WUEe in relation to meteorological conditions, and (c) analyse between-site variability of WUEe as affected by vegetation type and climatic conditions, across sites in European forest ecosystems. Day-to-day variability of gross primary productivity (GPP) and evapotranspiration (ET) were more strongly coupled than net ecosystem production (NEP) and ET, obviously because NEP also depends on the respiration that is not heavily coupled to water fluxes. However, the slope of daytime NEP versus ET (mNEP) from half-hourly measurements of a single day may also be used as a WUEe-estimate giving very similar results to those of the GPP-ET slope (mGPP), since the diurnal variation is dominated by GPP. Since ET is the sum of transpiration (linked to GPP) and evaporation from wet vegetation and soil surfaces (not linked to GPP) we expected that WUEe is increasing when days after rain are excluded from the analysis. However only very minor changes were found, justifying an analysis of WUEe related to vegetation type. In most of the studied ecosystems the instantaneous WUEGPP was quite sensitive to diurnally varying meteorological conditions and tended to decline from the morning to the afternoon by more than 50% because of increasing vapour pressure deficits (VPD). Seasonally, WUEGPP increased with a rising monthly precipitation sum and rising average monthly temperatures up to a threshold of 11, 14 and 18°C in boreal, temperate and Mediterranean ecosystems, respectively. Across all sites, the highest monthly WUEGPP-values were detected at times of positive anomalies of summer-precipitation. During drought periods with high temperatures, high VPD, little precipitation and low soil water content, the water-use efficiency of gross carbon uptake (WUEGPP) tended to decrease in all forest types because of a stronger decline of GPP compared to ET. However the largest variation of growing season WUEGPP was found between-sites and significantly related to vegetation type: WUEGPP was highest in ecosystems dominated by deciduous trees ranging from 5.0 g CO2 kg H2O−1 for temperate broad-leaved deciduous forests (TD), to 4.5 for temperate mixed forests (TM), 3.5 for temperate evergreen conifers (TC), 3.4 for Mediterranean broad-leaved deciduous forests (MD), 3.3 for Mediterranean broad-leaved evergreen forests (Mbeg), 3.1 for Mediterranean evergreen conifers (MC), 2.9 for boreal evergreen conifers (BC) and only 1.2 g CO2 kg H2O−1 for a boreal wetland site (BT). Although vegetation type and meteorology co-vary, the WUEGPP variation was hardly related to meteorology, as we could show by comparing similar meteorological conditions only. Furthermore we compared across-site WUEGPP only under conditions when the 10% high GPP rates were exhibited. The between site differences remained, and at all sites ecosystem reached higher WUEGPP levels under this condition. This means when vegetation is most productive usually it also maximises the amount of carbon gained per water lost. Overall our results show that water-use efficiency exhibits a strong time-scale dependency in the sense that at longer time-scale meteorological conditions play a smaller role compared to shorter time scale. Moreover, we highlight the role of vegetation in determining carbon-water relation at ecosystem level. Consequently, all predictions of changing carbon-water cycle under changing climate should take into this role and the differences between vegetation types. These results show the strong time-scale dependency of water-use efficiency

Interactive Discussion soil water content, the water-use efficiency of gross carbon uptake (WUE GPP ) tended to decrease in all forest types because of a stronger decline of GPP compared to ET. However the largest variation of growing season WUE GPP was found betweensites and significantly related to vegetation type: WUE GPP was highest in ecosystems dominated by deciduous trees ranging from 5.0 g CO 2 kg H 2 O −1 for temperate broad-5 leaved deciduous forests (TD), to 4.5 for temperate mixed forests (TM), 3.5 for temperate evergreen conifers (TC), 3.4 for Mediterranean broad-leaved deciduous forests (MD), 3.3 for Mediterranean broad-leaved evergreen forests (Mbeg), 3.1 for Mediterranean evergreen conifers (MC), 2.9 for boreal evergreen conifers (BC) and only 1.2 g CO 2 kg H 2 O −1 for a boreal wetland site (BT). Although vegetation type and meteorol-10 ogy co-vary, the WUEGPP variation was hardly related to meteorology, as we could show by comparing similar meteorological conditions only. Furthermore we compared across-site WUE GPP only under conditions when the 10% high GPP rates were exhibited. The between site differences remained, and at all sites ecosystem reached higher WUE GPP levels under this condition. This means when vegetation is most productive 15 usually it also maximises the amount of carbon gained per water lost.
Overall our results show that water-use efficiency exhibits a strong time-scale dependency in the sense that at longer time-scale meteorological conditions play a smaller role compared to shorter time scale. Moreover, we highlight the role of vegetation in determining carbon-water relation at ecosystem level. Consequently, all predictions Introduction

Conclusions
References Tables  Figures   Back  Close Full Screen / Esc

Printer-friendly Version
Interactive Discussion assimilate a certain amount of carbon has received attention both from plant physiological and from more applied scientific disciplines such as applied hydrology, irrigation science, agronomy and agroecology (de Wit, 1958). As shown in Fig. 1, water-use efficiency (WUE) can be calculated in very different ways depending on the temporal and spatial scales of interest, as well as on the sci-5 entific question of interest. Although various definitions of WUE are applied in different scientific disciplines, the common characteristic is that WUE is always a ratio of carbon gain to water loss.
Biologists and plant physiologists consider WUE usually on leaf and plant scales and are mainly interested in relations between total or above ground biomass (B), 10 stem biomass (S) and net CO 2 assimilation (A) to transpiration (T) or evapotranspiration (ET). ε may represent both T and ET (Loomis and Connor, 1992;Denmead et al., 1993). ET is defined as the total water vapour flux between the canopy and the atmosphere consisting of evaporation from soil (E), plant transpiration and evaporation of the intercepted fraction. Agricultural scientists appreciate WUE mostly as a relation 15 of yield (Y) or B to ε, the total water provided to the crop, including precipitation (W) or the amount of irrigated water (W U ), which is mainly of interest for irrigation purposes (Jones, 2004). To estimate the WUE of whole ecosystems (WUEe) geoscientists and ecologists commonly use the ratio of the main ecosystem fluxes such as net primary production (NPP), net ecosystem production (NEP), or gross ecosystem production 20 (GEP) to the water losses (ET or T) (Law et al., 2002;Reichstein et al., 2002).
In general, major ecozones with typical dominant plant functional types are often characterized by differing water availability (Woodward, 1987). Thus we expect ecosystems to have different water-use efficiencies because of inherent physiological variation in leaf gas exchange characteristics and because of differences in environmental 25 conditions among habitats (Farquhar et al., 1989). Therefore comparative studies of WUEe are crucial to our understanding on how future climate change accompanied by hydrological changes will affect the carbon and energy budgets of ecosystems.
The application of the eddy covariance technique for the continuous determination Interactive Discussion of carbon and water fluxes is deployed as a network throughout the world thus enabling scientists to study both, the temporal and the large-scale spatial variability of WUEe. However, one major problem is that both fluxes directly measured by the eddy covariance method (NEP and ET), are not directly related to the canopy function (GPP and T), but are confounded by respiration and evaporation from soil and wet surfaces, 5 respectively. Consequently, there are different ways to calculate WUEe that have potentially different interpretations. These different options have not been systematically compared across sites until now. Hence, in this study we take advantage of a large and harmonised flux dataset from the Carboeurope-IP project, covering a large range of forest types and climate zones 10 and aim at: (1) comparing and interpreting different ways of calculating ecosystem water-use efficiency from eddy covariance data, and (2) summarizing the overall magnitude as well as the temporal and spatial variability of WUEe and the relevant driving factors (e.g. meteorology and vegetation type) for European flux sites. Interactive Discussion deciduous (MD) or (8) Mediterranean broad-leaved evergreen (Mbeg). The detailed characteristics and codes for the sites are shown in Table 1. The carbon, water and energy exchange between the atmosphere and the vegetation was measured with the eddy covariance technique (for details see e.g. Aubinet et al., 2000) from towers above the vegetation canopies. Three-axis sonic anemometers measured wind speed and virtual temperature, infrared gas analyzers measured concentrations of water vapor and CO 2 and a suite of software was needed for real-time and post-processing analysis. The data were quality checked, and data gaps due to system failure or data rejection were filled using standardized methods (for details see Papale et al., 2006;Reichstein et al., 2005a;Moffat et al., 2007) to provide complete and standardized data sets. The measured NEP fluxes has been partitioned in the two main components GPP and Reco using the method described by Reichstein et al. (2005b) that is based on night-time data extrapolation using nonlinear regressions with temperature.

15
As mentioned in Sect. 1, water-use efficiency can be calculated in very different ways depending on the scientific discipline and the temporal and spatial scale of interest. For calculating WUEe we analyzed both, the ratio of daily integrals between NEP (g CO 2 ) and ET (kg H 2 O) as well as the ratio between GPP (g CO 2 ) and ET (kg H 2 O), calculated from half-hourly measurements, which provide a temporally highly resolved basis for 20 calculating WUE NEP and WUE GPP (g CO 2 kg H 2 O −1 ).
Considering only days with active canopy during the growing season we used a filter to exclude daily NEP-, GPP-and ET-values, when mean daily latent heat flux (LE)<20 W m −2 , global radiation (Rg)<150 W m −2 and mean air temperature <0 • C. Moreover we made use of a gap-filling quality measure of aggregated daily values 25 that indicates which fraction of the data was original or most reliably filled (cf. Reichstein et al., 2005a). Only if more than 80% of the half-hourly data were original and reliably filled the data were used in the analysis. An analysis was performed of how the exclusion of rainy days and the first, second and third day after, respectively, affects WUEe. Grelle et al. (1997) highlighted the need for eliminating days with high evaporation and interception causing underestimated WUEe which occur basically on day after precipitation events.
After these preparatory steps WUEe was calculated in different ways: 1. As ratio between daily integrals of GPP and ET (WUE GPP ) 2. As ratio between daily integrals of NEP and ET (WUE NEP ) 3. As slope from a linear regression between half-hourly values of GPP and ET calculated for each day (cf. Fig. 2 between 04:00 and 21:30 and (2) to estimate the mean daily WUEe (m GPP , m NEP ) as the slope between GPP and ET or NEP and ET, respectively. Furthermore the ratio of daily sums of GPP, NEP and ET were used to estimate daily WUE GPP and WUE NEP . The monthly WUE GPP and WUE NEP was calculated by the ratio of monthly integrals of GPP divided by the respective H 2 O flux integrals (WUE GPPmonth =GPP month /ET month ).

20
The seasonal trend of WUEGPP is characterized by the mean monthly WUEGPP. The seasonal progress of GPP and ET is represented by normalized data such that the mean monthly maximum equals 100%. The standard error, SD x , where σ n is the standard deviation and n is the number of months, was calculated for each calendar month. The mean annual or long-term WUEe is the median of all 25 available daily averages.

Comparison of different calculations of WUEe
Since NEP and GPP are differently related to ET an evaluation of these two parameters has to be done. The filtered (see Sect. 2 for details) NEP-, GPP-and ET-values for FI-Hyy are shown in Fig. 3, and demonstrate a much higher correlation between ET and 5 GPP than between ET and NEP, a pattern which occurred at all investigated sites. Furthermore, we expected to observe lower WUE NEP -and WUE GPP -values on precipitation days and the day after due to enhanced evaporation from surfaces that has never been part of the plant metabolism (cf. Grelle et al., 1997). However, a comparison of WUE GPP -values calculated from all data or data excluding rainy and subsequent 10 days, revealed that the evolution of WUE GPP after rain events is quite conservative (Fig. 4). Only coniferous forests and the DE-Hai site (TD) show a WUE GPP -decrease up to the second day after the rain event. The same results were found for WUE NEP (not shown).
In a further step we performed a correlation analysis of different forms of daily WUEevalues, such as m GPP , m NEP , which represent the slope between half-hourly carbon fluxes, WUE GPP and WUE NEP , and ET for each day (see Sect. 2). The plots for three coniferous sites of the boreal (FI-Hyy), temperate (DE-Tha) and Mediterranean zone (IT-SRo) are shown in Fig. 5a-d and demonstrate that correlations are remarkably high between the daily slopes from half-hourly measurements, m NEE  Table 2). In boreal ecosystems the highest correlations are found in m GPP :WUE GPP  Interactive Discussion WUE GPP -values are always highest in comparison to WUE NEP -, m GPP -and m NEPvalues. A summary of the slopes and r-values for all sites is presented in Table 2.
The mean long-term WUE NEP -, m NEP -, m GPP -and WUE GPP -values, summarized in Fig. 6, shows highest medians for WUE GPP at all sites and lowest medians for WUE NEP at most of the sites. Highest WUE GPP -values of 4 g CO 2 kg H 2 O −1 and more arise in 5 temperate mixed (TM) and temperate broad-leaved deciduous forests (TD). In contrast, WUE NEP -values are only 50% and less of WUE GPP -values.
Remarkably high WUE NEP -values can be observed particularly in some temperate (DE-Hai, FR-Hes) and Mediterranean broad-leaved deciduous forests (IT-Col, IT-Ro2, IT-PT1), which may indicate lower respiration-assimilation ratios in these ecosystems.

10
The Water-use efficiencies calculated from diurnal slopes (m GPP , m NEP ) are quite similar, however WUEe from m GPP is always slightly higher than WUEe from m NEP . The reason for this similarity with the slopes is that the slope of NEP versus ET is much less influenced by respiration, because the diurnal variation is dominated by the GPP, while the ratio of NEP to ET is influenced by more important background values 15 of respiration.

Seasonal water-use efficiency
Seasonal patterns of WUE GPP GPP and ET are shown for selected sites from Table 1 of each vegetation functional type in Fig. 7.
In all ecosystems we found a remarkable high correspondence between the monthly 20 amount of GPP and ET which both peak between May in Mediterranean conifers (MC) and August in temperate mixed forests (TM). As a result of year-round active assimilation in temperate conifers (TC), Mediterranean conifers (MC) and Mediterranean broadleaved evergreen forests (Mbeg), WUE GPP can be estimated for all seasons. Even some maritime influenced boreal conifers show a surprisingly long potential assimila-25 tion period due to the fact that they can even assimilate during days when mean daily temperatures are slightly below 0 • C. This characteristic was also reported by Suni et al. (2003) and Sevanto et al. (2006). However, boreal ecosystems and temperate de- Unlike the summer maximum of GPP and ET, WUE GPP can peak, site-dependently in any season. A summer maximum of WUE GPP can only be observed for the boreal wetland site (FI-Kaa), even though the seasonal course is flat compared to the other sites. All forest sites show a decrease of WUE GP P during summertime due to a stronger de-5 cline in GPP than in ET. The more southernly the site is located, the longer the summertime depression lasts. This summertime depression lasts longer, the more southernly the site is located. This trend can even be observed in boreal conifers, where the mean long-term WUE GPP -max appears in September. This is very interesting since many investigations on leaf-level (e.g. Monteith, 1995;Maroco et al., 1997) showed that stomatal closure due to heat stress reduces transpiration more than assimilation and hence induces higher water-use efficiencies. For these reasons many ecosystems tend to develop two peaks in seasonal WUE GPP . Temperate ecosystems exhibit for example with the exception of very oceanic-exposed conifers (e.g. NL-Loo), one maximum in June and one in September/October. During summer months WUE GPP normally decreases 15 by about 1 g CO 2 kg −1 H 2 O and reaches a relative minimum in August. This progress is most pronounced in evergreen Mediterranean ecosystems (MC, Mbeg), where the long hot and dry summer season leads to a massive shift of the two WUE GPP -peaks to early spring (February-March) and late fall (October-December). During summer months WUE GPP normally decreases by more than 2 g CO 2 kg H 2 O −1 and reaches its 20 absolute minimum in July or August. Mediterranean deciduous forests (MD) achieve their maximum WUE GPP -values later, in May-June and September-October due to foliation and non year-round assimilation restricted to the growing season.
On a monthly time scale mean WUE GPP tends to increase with rising precipitation sum and rising temperatures until average temperatures reach a threshold of 11, 14 25 and 18 • C in most of the boreal, temperate and Mediterranean ecosystems, respectively. Hence, highest monthly WUE GPP -values due to relatively higher GPP-and lower ET-values usually arise in wet summer-months with high precipitation sums and low mean VPDs. The higher GPP during these months can be the result of an increase Interactive Discussion of diffuse radiation that might stimulate assimilation (Alton et al., 2007;Knohl and Baldocchi, 2008). Similar long-term trends, fluctuations and estival decreases were also published for WUEe from the ratio above-ground production to accumulated transpiration in willow plantations in southern Sweden by Lindroth et al. (1994). Thus, the seasonal progress of WUE GPP differs very strongly depending on site, 5 climate, vegetation type including the understory and growing season length, which lasts at our investigated sites between six and twelve months. In some deciduous broad-leaved forests understory plays a major role and is largely responsible for the demonstrated WUEe during early spring and late autumn months. The mean annual WUE GPP is between 1. In order to exclude direct climatic effects on WUEe we analyzed WUE GPP also under similar meteorological conditions but at different sites. The range of mean daily temperature (T), mean daily global radiation (Rg) and mean daylight VPD was set to 20 17.5-22.5 • C, 200-300 W m −2 and 5-15 hPa, respectively. The resulting WUE GPP values correspond well with the mean annual WUE GPP , indicating that water-use efficiency is more driven by vegetation functioning than by climate forcing. In an alternative approach, where WUE GPP was multiplied by VPD to account for direct VPD effects, Beer et al. (2007) also found strong between-site variability of this meteorology-adjusted 25 WUE GPP . Also the differences between the vegetation types remain similar, when sites are compared with respect to their WUE GPP under conditions when their maximum seasonal GPP is reached. Moreover, all sites reach their highest water-use efficiency when also their overall productivity is highest, indicating that these high carbon up-

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version Interactive Discussion take rates are not only caused by increased stomatal conductance (which affects both carbon and water fluxes), but also increased carboxylation efficiency. Few published data of long-term WUEe makes a thorough comparison with other studies difficult. Similar or some lower WUE GPP -patterns due to plant functional types were reported by Law et al. (2002) who estimated 3.2 g CO 2 kg H 2 O −1 for tem-5 perate deciduous forests, 2.4 g CO 2 kg H 2 O −1 for temperate evergreen conifers and 1.5 g CO 2 kg H 2 O −1 for tundra vegetation. However, their results are based on much shorter time series. Highest WUE GPP -values for temperate deciduous forests were also estimated by Reichstein et al. (2007a). Furthermore there are some studies which do not represent ecosystem WUE GPP but water-use efficiency of specific tree species.

15
Moreover WUE GPP also varies from year to year due to meteorological conditions, however diverse conditions affect WUE GPP differently depending on site. The highest annual WUE GPP -values due to higher GPP-and lower ET-values usually arise in wet years with a high annual precipitation sum and a low mean annual VPD. Only the mountainous Mediterranean deciduous broad-leaved site IT-Col shows a negative correlation 20 between the annual precipitation sum and the mean annual WUE GPP (not shown). On an annual time-scale the temperature effect does not seem to be significant. While the boreal coniferous site SE-Nor shows the highest WUE GPP in the warmest years, the Mediterranean coniferous site IT-SRo shows the highest values in the coldest years. All other sites do not show dependencies between mean annual temperature and mean Furthermore the interannual WUE GPP -fluctuations under similar meteorological conditions (Temperature: 17.5-22.5 • C, Global radiation: 200-300 W m −2 , VPD: 5-15 hPa) are, as expected, not lower but showed consistently comparable values (data not 5 shown), suggesting that instantaneous meteorological conditions were not responsible for between-year variability of water-use efficiency. However, higher fluctuations can be observed when WUE GPP was only calculated for days with the 10% highest GPP per site, i.e. under "optimal" conditions. This might indicate that interannual variability of WUE is caused by the overall vegetation state and its development, respectively.

Instantaneous water-use efficiency
Unlike the seasonal cycle of WUE GPP , the instantaneous WUE GPP which was estimated for every half hour shows a similar diurnal cycle trend for different sites, with a primary WUE GPP -maximum in the early morning and a secondary maximum in the evening (see Fig. 8 for temperate deciduous and mixed forests and reaches at all sites a minimum between 1 and 2 p.m. This WUE GPP -decrease is largely attributed to an increase in solar radiation (max. between 1 and 2 p.m.) during daytime, which causes a shift between the earlier GPP-and the later ET-maximum and was detectable in all investigated ecosystems. The deferred increase of VPD, reaching its maximum at 4 p.m. seems to play a minor part in affecting water-use efficiency during the course of the day. The hyperbolic relationship between WUE GPP and VPD during the daylight period is shown for the three coniferous forests FI-Hyy, DE-Tha and IT-SRo in Fig. 9. These results confirm the effects of VPD on stomatal conductance, which is known on leaf level (Schulze and Hall, 1982) and also at ecosystem scale (Law et al., 2002). Increasing VPD until afternoon causes stomatal closure, thus decreasing stomatal conductance.
Since stomatal conductance impacts both GPP and transpiration, but it is only the water flux which is proportional to VPD, the overall effect of VPD on GPP is higher than on transpiration. This leads to a decreasing WUE GPP . However, we find a hysteresis, where WUE GPP is higher in the evening than under similar VPD conditions during the day after VPD decreases again (cf. color scheme in Fig. 9). Since radiation is lower 10 after 5 p.m., this phenomenon could be explained by two alternative hypotheses: (1) transpiration is lower because of lower leaf water potential and stomata react to these leaf water changes, and (2) leaf temperatures and thus the VPD leaf−to−air gradient is smaller in the evening when incoming shortwave radiation is less intense.
Similar trends but with some higher values of instantaneous WUEe were also de-15 tected for a short rotation Salix viminalis plantation by Lindroth and Cienciala (1996) and for boreal conifers in Norunda, Sweden, by Morén et al. (2001).

Concluding discussion
We identified and compared different methods to calculate ecosystem water-use efficiency (WUEe) at temporal scales of hours, days, months and years for European 20 forest sites representing different biomes. The interpretation of daily-integral ratios of NEP versus ET are hampered by the fact that NEP is strongly influenced by respiration, which may obscure the coupling of canopy carbon and water fluxes. GPP derived from flux-partitioning algorithms should be preferred when calculating meaningful ecosystem level water-use efficiency, even though GPP is not directly measured and any possible biases in the flux-partitioning algorithm translate into WUEe biases. The slopes of half-hourly measured daytime NEP/ET-(m NEE ) and GPP/ET-ratios (m GPP ) were conspicuously consistent, offer an alternative for calculating ecosystem water-use efficiency and may be considered as the most direct physiological indicator of canopy water-use efficiency, in particular since both NEP and ET are directly measured. However, since m NEE and m GPP do not cross the origin (intercept>0), the slopes do not inform about the longer-term ratio of carbon gains versus water losses.

5
Moreover, even the interpretation of WUE GPP , as well as m NEE and m GPP in terms of canopy function, are in principle complicated by evaporation from soil and wet surfaces which is included in the ET and not related to the canopy function and the coupling of carbon and water fluxes therein, although we did not find changes after excluding rainy days. In this respect, a larger use of sap-flow measurements at flux sites could pro-10 vide further insight of GPP and transpiration coupling. However, our analysis reveals that the soil and wet surface evaporation effect on WUEe calculation is not so large to compromise cross-site comparisons of WUEe from eddy covariance.
With between-site variability of WUEe being significantly larger than interannual variability, distinct groups of vegetation types with very similar mean annual WUE GPP -15 values can be identified. The highest WUE GPP with values of 4.5 to 5.0 g CO 2 kg H 2 O −1 are found in temperate mixed and temperate deciduous forests. A second group includes the boreal and temperate conifers as well as the Mediterranean ecosystems with values between 3 and 3.5 g CO 2 kg H 2 O −1 , while the boreal wetland site reached only values of around 1 g CO 2 kg H 2 O −1 , probably caused by a combination of low photo-20 synthetic capacities and open water evaporation. While it is known that WUEe strongly depends on the atmospheric conditions in the boundary layer (e.g. VPD), interestingly between site differences cannot be explained by such factors as we show by comparing WUEe under similar meteorological conditions. Hence, WUEe can be largely considered as a site or vegetation characteristic, hinting to adaptation of vegetation to 25 their growing environment. Nevertheless, an abstraction of WUEe from meteorological conditions, the intrinsic water-use efficiency calculated as Assimilation/stomatal conductance (A/g) by Schulze and Hall (1982)  We also find very consistent behaviour of seasonal WUEe across sites: apart from some high-boreal and high-oceanic conifers, a decrease during summer months due to arising drought stress is observed. However, on a seasonal time scale the negative correlation between water-use efficiency and drought characteristics is no implicitness, and should be subject to additional scientific attention since investigations on a fast 5 growing willow plantation in Sweden has shown increasing water-use efficiencies in response to reduced water availability and high VPD (Linderson et al., 2007).
In comparison to the seasonal trend, WUE GPP during the course of the day shows very similar changes with lowest values at times of highest VPD in the afternoon at all sites. This emphasizes the dependency of WUEe on daily varying meteorological 10 conditions.
In summary, water-use efficiency is highly vegetation-dependent but also subject to short-term variation of meteorological conditions. Highest WUE GPP -values arise at times of low VPDs under warm and humid conditions. While it is difficult to predict directly from our observations how WUEe will alter under global climate change, these 15 generalized findings could be used as evalutation benchmarks for process-oriented dynamic global vegetation models (DGVMs) and Landsurface schemes (LSMs) which are used to predict trajectories of carbon and water cycle under a changing climate. In particular the role of vegetation compared to long-term meteorological conditions should be more strongly emphasized. Extensions of our analysis to other regions world 20 wide and corroboration of our results with simultaneous sap-flow measurements will provide further insight in the coupling of carbon and water fluxes in forest ecosystems. Introduction      , as long-term average (all available data) and seasonal maximum (including month of year) for different functional vegetation types. The standard deviation (SD) is only reproduced if more than two sites within one vegetation type and if more than two years for calculating mean interannual variability were available, respectively. All daily WUEe-data were excluded if mean daily latent heat flux (LE) <20 W m −2 and less than 80% of the half-hourly NEE-data of a single day were original or reliably filled.