Response of ecosystem intrinsic water use efficiency and gross primary productivity to rising vapor pressure deficit

Elevated vapor pressure deficit (VPD) due to drought and warming is well-known to limit canopy stomatal and surface conductance, but the impacts of elevated VPD on ecosystem gross primary productivity (GPP) are less clear. The intrinsic water use efficiency (iWUE), defined as the ratio of carbon (C) assimilation to stomatal conductance, links vegetation C gain and water loss and is a key determinant of how GPP will respond to climate change. While it is well-established that rising atmospheric CO2 increases ecosystem iWUE, historic and future increases in VPD caused by climate change and drought are often neglected when considering trends in ecosystem iWUE. Here, we synthesize long-term observations of C and water fluxes from 28 North American FLUXNET sites, spanning eight vegetation types, to demonstrate that ecosystem iWUE increases consistently with rising VPD regardless of changes in soil moisture. Another way to interpret this result is that GPP decreases less than surface conductance with increasing VPD. We also project how rising VPD will impact iWUE into the future. Results vary substantially from one site to the next; in a majority of sites, future increases in VPD (RCP 8.5, highest emission scenario) are projected to increase iWUE by 5%–15% by 2050, and by 10%–35% by the end of the century. The increases in VPD owing to elevated global temperatures could be responsible for a 0.13% year−1 increase in ecosystem iWUE in the future. Our results highlight the importance of considering VPD impacts on iWUE independently of CO2 impacts.


Introduction
Given that plants are constantly challenged to maximize carbon (C) uptake and minimize water loss by regulating stomata, an important metric of ecosystem sensitivity to climate is the intrinsic water use efficiency (iWUE), defined here as the ratio of C assimilation (A n ) to stomatal conductance (g s ). The iWUE reflects plants' adaptability to changing environmental conditions (Farquhar et al 1982), and because it ultimately governs the coupling between carbon assimilation and stomatal conductance, it has received considerable attention (Keenan et al 2013, Lévesque et al 2014, Frank et al 2015, van der Sleen et al 2015. Recently, increasing iWUE has been observed in many ecosystems. This phenomenon has largely been attributed to rising atmospheric CO 2 concentrations (Franks et  Original content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence.
Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. Mastrotheodoros et al 2017) and theoretical models (Knauer et al 2017b). However, homeostatic iWUE has also been reported (Marshall and Monserud 1996, Wu et al 2015, Wieser et al 2016, and models and experiments frequently show diverse patterns of iWUE response to rising CO 2 (Keenan et al 2013, Knauer et al 2017b. While many factors could explain this disagreement, among them is the possibility that CO 2 may not be the only slowly-evolving driver affecting iWUE. The balance between vegetative carbon uptake and water loss is also affected by moisture deficits in the soil and the atmosphere. Theory predicts, and some data confirm, that iWUE should increase when water availability is low (Gomes et al 2009, Schultz and Stoll 2010, de Santana et al 2015. The extent to which changes in atmospheric evaporative demand (i.e. vapor pressure deficit; VPD) can affect iWUE independently of changes in soil moisture (Schultz and Stoll 2010, Lévesque et al 2014, Tomas et al 2014, Kwon et al 2017 is less understood. Unlike the total water use efficiency (WUE=assimilation/transpiration), the iWUE describes plant responses to rising VPD without being confounded by the abiotic increase in evaporative potential when VPD is high. Consequently, even though total WUE could decline during drought, iWUE would increase. Quantifying the sensitivity of iWUE to VPD is particularly important for understanding the response of gross primary productivity (GPP) to VPD, as VPD has been rising in many places, and is expected to increase further in most biomes as temperatures continue to rise (Ficklin and Novick 2017).
Plant physiological theory offers some important clues as to how iWUE should respond to rising VPD. Because the intercellular CO 2 concentration (c i ) is strongly controlled by stomatal conductance (Farquhar and Wong 1984), and because the relationship between A n and c i is hyperbolic, the relationship between A n and g s is also hyperbolic. As VPD initially begins to rise, g s will be reduced, but the rate of and biochemical potential for photosynthesis may be unaffected because of the sufficiently atmospheric CO 2 supply. In this case, iWUE is expected to increase. However, if VPD-driven reductions in g s (and thus c i ) become more severe, reducing photosynthetic rates, or if coincident reductions in soil moisture limit biochemical capacity, then iWUE may stabilize or even decrease as VPD rises (Yi et al 2019). The primary goal of our study is to characterize these responses over broad gradients in climate and ecosystem types, and to project the magnitude of VPD-driven effects on iWUE into the future.
Meeting this objective has many practical applications. It could help to explain why temporal trends in iWUE inferred at the ecosystem scale from flux towers are higher than expected when considering CO 2 effects alone (Keenan et al 2013, Mastrotheodoros et al 2017. VPD-effects on iWUE could also partially explain the increasing decoupling of growth and iWUE as observed from tree rings (Andrer-Hayles et al 2011, Peñuelas et al 2011). A VPD-driven increase in iWUE should not increase tree growth, as overall it will likely be accompanied by lower GPP (Peters et al 2018). Finally, our results are relevant for drought monitoring and prediction indices that incorporate information about the difference between potential and actual evapotranspiration (Narasimhan andSrinivasan 2005, Otkin et al 2013), which is largely driven by stomatal conductance during dry periods (Novick et al 2016a). Therefore, our analysis can clarify how drought-driven declines in conductance translate to declines in GPP, which may ultimately form an important indicator of agricultural drought.
Here, we assessed the extent to which iWUE responds to increasing VPD versus declining soil moisture by leveraging long-term eddy covariance observations of ecosystem-scale C and water exchange from 28 sites representing a broad range of biomes and climate regimes. These data are collected at a timescale (i.e. half-hourly or hourly) over which VPD and soil moisture are typically decoupled (Novick et al 2016a), allowing the influence of each on iWUE to be independently assessed. We then projected changes in iWUE to the end of this century by using a VPD dataset downscaled from 15 General Circulation Models (under Representative Concentration Pathway 8.5 (RCP 8.5)-the highest greenhouse gas emission scenario). We also present a theoretical and mathematical framework that facilitates the mechanistic interpretation of results, which is critical for future efforts to represent these dynamics in ecosystem models.

Materials and methods
Tower-derived estimates of ecosystem intrinsic water use efficiency Intrinsic water use efficiency at the leaf level (iWUE, μmol mol −1 ) is defined as the ratio of carbon assimilation (A n , μmol m −2 s −1 ) to stomatal conductance (g s , mol m −2 s −1 ): In this study, the ecosystem-scale intrinsic water use efficiency is estimated at the canopy scale as: noting that iWUE is estimated at the half-hourly or hourly resolution. The high frequency (half-hourly or hourly) data used in this study were obtained from 28 FLUXNET2015 Tier1 dataset sites located in North America (http://fluxnet.fluxdata.org/data/ fluxnet2015-dataset/) (see table S1 in the supplementary information-SI is available online at stacks.iop. org/ERL/14/074023/mmedia), which span eight vegetation types and a range of dryness index (DI=potential evapotranspiration/precipitation; see DI calculation later in this section), noting that each site has at least three years of records. The GPP is gross primary productivity estimated from tower-derived net carbon flux using two partitioning approachesthe so-called 'nighttime approach' (Reichstein et al 2005) and 'daytime approach' (Lasslop et al 2010); the half-hourly or hourly GPP products were provided by FLUXNET. G s (m s −1 ) is surface conductance, which is determined by inverting the Penman-Monteith equation (Monteith 1965) into: is the slope of saturated vapor pressure against temperature, R n −G (W m −2 ) is available energy-the difference between net radiation (R n ) and ground heat flux (G), ρ (kg m −3 ) is air density, is latent heat flux, and g a (m s −1 ) is aerodynamic conductance, quantified by using the Monteith-Unsworth model (Monteith and Unsworth 1990) as where P a (Pa) is air pressure, T a (K) is air temperature, and R (=8.314 J mol −1 K −1 ) is the gas constant. The FLUX-NET2015 dataset includes information on the uncertainty of every half-hourly or hourly water vapor flux measurements (given as a mean and standard deviation), which we propagated through to our derivation of G s . Specifically, for every half-hourly or hourly estimate of LE, we generated 10 000 estimates drawn from a normal distribution with the mean and standard deviation proposed by FLUXNET. Thus, we also derived 10 000 estimates of G s , using the mean of this distribution in our analysis.
The Penman-Monteith equation, which describes ecosystem evapotranspiration, relies on a big-leaf assumption, and in many applications G s is assumed to be equal to the canopy-level surface conductance. However, ecosystem evapotranspiration includes a significant contribution from soil evaporation (typically on the order of 15%-30% for a range of biomes (Wilson et al 2001, Oishi et al 2010, Wang et al 2013, Sulman et al 2016). In modeling studies, this problem is often met with a two-source (i.e. canopy and soil) modeling approach where both canopy and soil evaporation are independently simulated from Penman-Monteith type equations (Shuttleworth and Wallace 1985). However, in data driven applications like this one, when surface conductance is derived by inverting the Penman-Monteith equation from measured ecosystem scale LE observations, the derived G s is influenced by evaporation from soil and precipitation interception, and should not be assumed to be representative of canopy stomatal conductance alone (Schulze et al 1994, Kelliher et al 1995, Baldocchi and Meyers 1998, Knauer et al 2017a. To minimize the contribution of surface evaporation to iWUE, and to control for confounding effects from variable leaf area, radiation, and windspeed, the data were subjected to the following set of filters: (1) we only used original measurements (quality control flag=0) or gap filled data of good quality (quality control flag=1) as determined by the FLUX-NET2015 protocols; (2) the analysis was conducted for periods of relatively stationary Normalized Difference Vegetation Index (NDVI) (https://modis.ornl.gov/ fixedsite/) with NDVI>80%NDVI max (the peak of the annual NDVI series). The NDVI product has a spatial resolution of 250 m, and we used the pixel where the tower resides; (3) the analysis was constrained to 10:00-15:00 local time and only when incoming short-wave radiation (R s ) is higher than 50% of the peak of its annual series (R smax ), i.e. R s >50%R smax . Note that a discussion of uncertainty introduced by the selection of these two thresholds is provided later in this document; (4) data were filtered to exclude records within two days following rain events (>1 mm) to minimize the contamination of rain on LE measurements; (5) the wind speed was constrained to exceed 1 m s −1 to select for periods with negligible leaf boundary layer resistance.
Linking the dynamics of intrinsic water use efficiency to vapor pressure deficit and soil moisture To disentangle the response of iWUE to changes in soil moisture and VPD, the iWUE determined from equation (2) was averaged within discrete soil moisture and VPD bins. Though soil moisture can sometimes be made at different depths among FLUXNET sites, these measurements are commonly used to characterize the water status of the surface (∼the upper 30 cm) soil at FLUXNET sites in cross-site syntheses. Following the approach of Novick et al (2016a), the soil moisture bins were set as 15, 30, 50, 70, 90, 100% percentiles of all records, and VPD was binned every 0.2 kPa. We determined a reference GPP (GPP ref ) and reference conductance (G sref ) under conditions of VPD ∼1 kPa and soil moisture >90% of the sitespecific saturation value. We then normalized the tower-derived GPP and G s by the reference values to facilitate a cross-site comparison of the relationship of each to VPD. To evaluate the uncertainty associated with the R s and NDVI filters (NDVI>80%NDVI max and R s >50%R smax ), we ran the following two tests: To avoid confounding effects introduced by covariation of VPD and CO 2 concentrations, we repeated the analysis but filtered the data into three bins associated with unique ranges of CO 2 concentration (365 to 375 ppm, 370 to 380 ppm, and 375 to 385 ppm). This approach allows us to evaluate if the response of iWUE to VPD is robust to changes in CO 2 concentration.
Using dryness index to facilitate cross-site trend analysis We linked cross-site trends in the dynamics of iWUE to the Dryness Index (DI), defined as the ratio of mean annual potential evapotranspiration (PET) to mean annual precipitation (P): We used the Priestley-Taylor equation (Priestley and Taylor 1972) to determine PET, which is well-correlated with PET derived from the more mechanistically complete but data-demanding Penman-Monteith equation (Novick et al 2016a). The Priestley-Taylor equation takes the form: where PT a =1.26 is the Priestley-Taylor coefficient, and l is the latent heat of vaporization.
The DI is usually used to quantify the 'dryness' of a landscape. At sites where DI is lower than 1, annual precipitation is higher than potential evapotranspiration, indicating excess water for soil moisture recharge and runoff generation. In such conditions, plants are not limited by water availability (except in seasonally dry climates). When DI is higher than 1, most of the annual precipitation is turned into evapotranspiration, leaving plants limited by water availability. We calculated the DI using the long-term mean annual precipitation (see table S1) reported by each site to the FLUXNET Database (http://fluxnet.ornl.gov/) instead of calculating mean annual precipitation from the FLUXNET2015 high frequency dataset, which is sensitive to missing records. Note that for US-AR1 and US-AR2 sites, where no annual precipitation was provided in the FLUXNET Database, we used the high frequency precipitation records in the FLUXNET2015 dataset.
Historic and projected VPD Bias-corrected and downscaled (∼4 km grid) General Circulation Model (GCM) projections for each site in the US were extracted from the Multivariate Adaptive Constructed Analogs-version 2 (MACA (Abatzoglou and Brown 2012)) data warehouse (http://maca. northwestknowledge.net/). Specifically, daily maximum and minimum air temperature and maximum and minimum relative humidity from 15 GCMs (table S2) were extracted from the Coupled Model Intercomparison Project-Phase 5 (CMIP5 (Taylor et al 2012)) archive for Representative Concentration Pathway 8.5 (the highest greenhouse gas emission scenario) from 1950-2099. First, the daily maximum and minimum saturated vapor pressures were calculated using the maximum and minimum air temperatures, respectively, and then averaged to calculate the daily saturated vapor pressure. Second, the actual vapor pressure was calculated by multiplying the daily saturated vapor pressure and the daily relative humidity estimated from daily maximum and minimum relative humidity (Allen et al 1998). Finally, daily averaged VPD was calculated by subtracting the daily actual vapor pressure from the daily saturated vapor pressure. To account for the uncertainty of air temperature and relative humidity in the projections, VPD of the 15 different GCMs were averaged for each site in this analysis. The VPD generally exhibits an increasing trend in all sites (figure S1) as a result of the rising temperature (figure S2) and relatively steady or slightly declining relative humidity (figure S3).
Quantifying the sensitivity of iWUE and GPP to VPD When VPD is relatively low and conductance is relatively high, a positive relationship between iWUE and VPD is expected, as conductance declines swiftly with rising VPD, but assimilation is relatively insensitive to the changes in conductance, since intercellular CO 2 concentrations remain high. However, as VPD continues to increase and conductance continues to decrease, reductions in intercellular CO 2 will begin to limit photosynthesis, either alone or together with other non-stomatal limitations if soil moisture also decreases (Lawlor andTezara 2009, Novick et al 2016b). Thus, the relationship between iWUE and VPD may be linear, hyperbolic, or even unimodal depending on how hydrologic stress limits the transfer of CO 2 from the air to the chloroplast. Therefore, we fit the iWUE-VPD relationship to a second order polynomial curve because it can accommodate the entire range of responses that may be expected across the wide range of sites considered here: where a, b and c are parameters that were optimized by using the least square method and listed in table S3. It follows that the sensitivity of iWUE to VPD (= iWUE VPD ¶ ¶ / ) may be expressed as 2aVPD+b. The change in iWUE as VPD rises can mitigate, or buffer, drought-driven reductions in GPP caused by reduced conductance or decreased photosynthetic capacity. To quantify this effect across sites, we expressed GPP as the product of iWUE and G s while neglecting the CO 2 effects The full derivative of GPP with respect to VPD is thus: is the sensitivity of GPP to VPD linked to stomatal closure, whereas G s iWUE VPD ¶ ¶ is the sensitivity of GPP to VPD linked to rising iWUE.
Here, the relationship between G s and VPD was assumed to have the form proposed by Oren et al (1999): where G sREF is the reference conductance when VPD=1 kPa, and m describes the sensitivity of conductance to VPD; both parameters were optimized by using the least square method and listed in table S4. The F-test was used to evaluate the significance of regressions.

Results and discussion
Response of intrinsic water use efficiency to hydrologic stress As hydrologic stress developed at each site, we observed a nearly universal initial increase in iWUE that was primarily driven by rising VPD (see figures 1(a)-(d) for representative sites with contrasting DI; figures S4 and S5 in the SI for all sites). Within each site, the initial increase of iWUE to VPD was observed across all soil moisture conditions, indicating a robust positive control of VPD on iWUE. Additionally, no detectable trend was found in the relationship between iWUE and soil moisture under the same VPD levels (figure S6), underscoring the critical role of VPD in regulating iWUE during drought. Notably, the increases in iWUE were driven by decreases in G s (figures 1(i)-(l); S7 for all sites), rather than by changes in GPP (figures 1(e)-(h); S8 for all sites). Because the partitioning method did not strongly affect results when comparing figures S4 with S5 in SI, here and throughout, we focus on trends in GPP estimated using the so-called 'nighttime approach' by Reichstein et al (2005). The uncertainty associated with the R s and NDVI filters (NDVI>80%NDVI max and R s >50%R smax ) is negligible, because when we changed the threshold of NDVI and R s filters, similar results were achieved (see figures S9 and S10). Therefore, the filter choice does not appear to strongly influence results and we only focus the analysis under the filter scheme of NDVI>80%NDVI max and R s >50%R smax . In many biomes, the increase in iWUE with rising VPD did not continue unabated, and a near-hyperbolic relationship emerged; in a few cases, a unimodal relationship between iWUE and VPD was observed  To describe the full range of the iWUE response to VPD at each site, we parameterized a second order polynomial model for the relationship between iWUE and VPD regardless of soil moisture conditions (see method section and table S3 for details). In addition, to better understand the initial iWUE increase with rising VPD, we estimated the slope of the iWUE-VPD relationship at a relatively low reference VPD (VPD=1 kPa). Our results confirm that an initial positive relationship between iWUE and VPD is common across sites (figure S11(a)). Next, by analyzing the slope of the iWUE-VPD relationship when VPD was at its site-specific maxima across all sites, we further showed that most sites continue to exhibit an increasing iWUE trend in response to VPD even when VPD is high ( figure S11(b)). To understand the influence of vegetation type versus climate on iWUE-VPD relationship, we averaged both slopes by the IGBP vegetation type (see figure S12), and found the initial slope of iWUE to VPD in Closed Shrublands is the lowest, while that of Mixed Forest is the highest (figure S12(a)). Though an initial increase of iWUE is found across all vegetation types, when VPD is at the site maxima, the slope becomes substantially lower in most vegetation types and even negative in grasslands ( figure S12(b)), however, in Closed Shrublands and Deciduous Broadleaf Forests the slope increases. While the initial slope of the iWUE-VPD relationship exhibits no clear dependence on DI (figure S13(a)), the slope calculated at the site VPD maxima showed a weak negative correlation with DI ( figure S13(b)). In other words, the iWUE tends to decrease under high VPD conditions at arid sites.
The initial iWUE increase in response to rising VPD reflects the relatively slower decrease in GPP compared to the rapid decrease in G s as VPD rises. To reiterate, these different patterns between GPP and G s can be understood by considering the leaf-level processes that control gas exchange. When internal CO 2 concentration is high (as is the case when G s is high), photosynthesis is less sensitive to small changes in internal CO 2 concentration and will therefore show relatively little response to a decrease in stomatal conductance, leading to the increasing iWUE at low to moderate VPD shown in figure 1. As stomata close while VPD further increases (Oren et al 1999), leaf internal CO 2 concentration continues to decline; at these lower intercellular CO 2 concentrations, photosynthesis declines rapidly with decreasing CO 2 availability, and iWUE no longer rises with increasing VPD (Farquhar and Sharkey 1982). However, if conditions associated with high VPD suppress photosynthetic capacity, such as high temperatures (Yamori et al 2014) or low leaf water potentials (Lawlor and Tezara 2009) that develop in dry conditions, photosynthetic capacity can be inhibited, lowering the demand for CO 2 and stabilizing the atmosphere-leaf CO 2 concentration gradient to lower values. When inhibition of photosynthesis develops, iWUE begins to decrease, as seen at high VPD and low soil moisture conditions (e.g. US-Var in figure 1; also US-SRM and US-AR1 in figure S4).
Assessing the sensitivity of GPP to VPD The contrasting response of iWUE and G s to VPD raises the question: to what extent does high VPD ultimately limit GPP? The sensitivity analysis described in equations (8)-(10) allows us to quantify when and where VPD-driven reductions in conductance translate into reductions in GPP. We found that stomatal conductance declines monotonically with increasing VPD (figures 1(i)-(l); S7 for all sites); thus, the iWUE G VPD s ¶ ¶ component of equation (9) is always negative (blue line in figure 2), thereby driving a decrease in GPP. However, G s iWUE VPD ¶ ¶ is typically positive under low VPD levels (magenta line in figure 2). Or in other words, GPP is less sensitive to VPD than is conductance (figure 1). Overall, GPP exhibits a net decline in response to increasing VPD (green line in figure 2), but of a magnitude that is less severe than that predicted from stomatal closure alone regardless of local climate (figures 2(b)-(f) with contrasting DI).
Projecting intrinsic water use efficiency into the future VPD is already rising in many places (Ficklin and Novick 2017), driven by increases in the saturated vapor pressure that are not matched by concurrent increases in actual vapor pressure. To quantify the potential for varying VPD to drive both historic and future changes in iWUE, we used site-specific relationships between iWUE and VPD (figure S4 and table S3) and estimated an upper limit on the extent to which rising VPD due to climate change could affect future relationships between photosynthesis and stomatal conductance. Using daily VPD records and projected VPD trends from a downscaled and bias-corrected General Circulation Model ensemble (15 models; table S2) together with the aforementioned iWUE-VPD relationship, we estimated historic changes in iWUE and projected iWUE values into the future for each site (figure 3; see figure S14 for the variation resulting from the 15 GCM products at each site). Because these projections are based on iWUE-VPD relations, they do not include any CO 2 effect.
Most sites showed an increasing trend in iWUE relative to the average of 1950-1960. In particular, our results suggest that rising VPD will increase iWUE by 5%-15% (relative to the 1950-1960 baseline) in many sites by 2050, and by 10%-35% in the majority of sites by the end of the century (figure 3). These increases in iWUE are not as large as those expected from future elevated CO 2 , which are on the order of 50%-70% for CO 2 concentrations of 540-570 ppm suggested by tree ring core analysis in Free-Air CO 2 enrichment (FACE) experimental conditions (Battipaglia et al 2013). However, at least in some sites, the magnitude of projected increases in iWUE due to VPD alone is still substantial. Our results also indicate a 0.13% year −1 increasing rate is expected for iWUE in the future owing to the rising VPD associated with the elevated global temperatures.
Excluding the CO 2 contamination effect To ensure that our results (figures 1 and S4) are not biased by co-occurring changes in CO 2 , we repeated the analysis using data constrained to a narrow range of CO 2 . After reducing the effects of CO 2 , our results exhibited the same hyperbolic iWUE-VPD relationship (figures S15-17). These results are consistent with previous results (   are higher than expected when considering CO 2 effects alone. Specifically, Mastrotheodoros et al (2017) used a model to show that consideration of changes in humidity and temperature better reconcile apparent iWUE trends with those predicted from CO 2 effects alone. Our study now offers an empirical analysis in support of those model predictions. This result implies that attributing the historic increase in iWUE to rising CO 2 alone is challenging, at least at the fine timescale (∼hourly) of the data used in this study, and that VPD in addition to rising CO 2 is likely to be a significant driver of historic and future increases in iWUE. Finally, our results show that increased iWUE driven by higher VPD usually corresponds with a reduced GPP (figures 1(e)-(h)), different from increased iWUE under elevated CO 2 that results from increased GPP. Thus, an increase in iWUE does not necessarily translate into increased carbon uptake as simulated by land surface models (Peters et al 2018) or increased tree growth as reported by tree ring data (Lévesque et al 2014, van der Sleen et al 2015, or in other words, whether higher iWUE corresponds to higher GPP depends on the driver of iWUE changes.

Conclusion
Intrinsic water use efficiency has long been recognized as a key parameter linking the global C and water cycles. However, the effect of atmospheric demand on ecosystem iWUE has rarely been quantified when analyzing historic iWUE data or predicting iWUE trends into the future. We show that, when VPD is relatively low, iWUE increases with rising VPD across a wide range of biomes. Whether this increase persists under high VPD depends on climate regime, among other factors. Our work suggests that the historic sensitivity of iWUE to rising CO 2 (Feng 1999, Keenan et al 2013 may be overestimated in places where VPD has also been rising and may explain previously contradictory results of spatial heterogeneity in iWUE patterns (Marshall and Monserud 1996, Keenan et al 2013, Wu et al 2015. The increase in iWUE may not necessarily translate into increase in tree growth, as a VPD-driven increase in iWUE is not accompanied by an increase in GPP. In fact, GPP tends to decline with rising VPD, but as our results show, the magnitude of this decline is less substantial when compared to VPDdriven declines in conductance.