Increasing stomatal conductance in response to rising atmospheric CO₂

Abstract

Background and Aims

Studies have indicated that plant stomatal conductance (g_s) decreases in response to elevated atmospheric CO₂, a phenomenon of significance for the global hydrological cycle. However, g_s increases across certain CO₂ ranges have been predicted by optimization models. The aim of this work was to demonstrate that under certain environmental conditions, g_s can increase in response to elevated CO₂.

Methods

Using (1) an extensive, up-to-date synthesis of g_s responses in free air CO₂ enrichment (FACE)experiments, (2) in situ measurements across four biomes showing dynamic g_s responses to a CO₂ rise of ~50 ppm (characterizing the change in this greenhouse gas over the past three decades) and (3) a photosynthesis–stomatal conductance model, it is demonstrated that g_s can in some cases increase in response to increasing atmospheric CO₂.

Key Results

Field observations are corroborated by an extensive synthesis of g_s responses in FACE experiments showing that 11.8 % of g_s responses under experimentally elevated CO₂ are positive. They are further supported by a strong data-model fit (r² = 0.607) using a stomatal optimization model applied to the field g_s dataset. A parameter space identified in the Farquhar–Ball–Berry photosynthesis–stomatal conductance model confirms field observations of increasing g_s under elevated CO₂ in hot dry conditions. Contrary to the general assumption, positive g_s responses to elevated CO₂, although relatively rare, are a feature of woody taxa adapted to warm, low-humidity conditions, and this response is also demonstrated in global simulations using the Community Land Model (CLM4).

Conclusions

The results contradict the over-simplistic notion that global vegetation always responds with decreasing g_s to elevated CO₂, a finding that has important implications for predicting future vegetation feedbacks on the hydrological cycle at the regional level.

INTRODUCTION

Water loss through plant stomata – small pores on the surface of leaves through which gas exchange between plants and the atmosphere takes place – is an unavoidable trade-off in the exchange for CO₂, the substrate for photosynthesis. Decreased stomatal conductance (g_s), via physiological (stomata responding dynamically to environmental stimuli) and/or morphological changes (via alteration in stomatal density and size) has been observed in elevated carbon dioxide (CO₂) environments in both laboratory and free air CO₂ enrichment (FACE) studies (Farquhar and Sharkey, 1982; Woodward, 1987; Drake et al., 1997; Ainsworth and Rogers, 2007; Leuzinger and Körner, 2007). However, recent studies suggest that rising atmospheric CO₂-induced decreases in g_s may be offset by contemporaneous increases in leaf area index (LAI) during the course of a growing season (Piao et al., 2007; Wu et al., 2012; Niu et al., 2013; Frank et al., 2015; Schymanski et al., 2015). Thus, despite significant improvements in our understanding of plant–atmosphere interactions in recent years, the net stomatal conductance response of the entire global vegetation system to rising anthropogenic CO₂ remains unclear.

In addition, little is known regarding the physiological response of plants to increasing CO₂ across multiple biomes, and in varying temperature and humidity regimes. For example, FACE studies are predominantly limited to the mid-latitudes of the northern hemisphere (Fig. 1), biasing our understanding of plant responses to these regions. Moreover, disparate vegetation responses in dry and drought-prone environments have been reported (Choat et al., 2012; Limousin et al., 2013; Zhou et al., 2013; De Kauwe et al., 2015; Mencuccini et al., 2015). It is therefore critical to improve our understanding of these responses to better predict future freshwater cycling, especially in regions vulnerable to drought and desertification in the 21st century (Lawrence et al., 2011).

Fig. 1.

The location of FACE studies included in our assessment. Fifty-one FACE studies are shown (most overlap on this scale). Most FACE studies are located in northern hemisphere locations between 30 and 60°N. FACE studies which did not, to our knowledge, document g_s changes were not included. See Materials and Methods for all cited studies used.

Here we demonstrate that g_s can in some cases increase in response to increasing atmospheric CO₂. This is shown using (1) in situ measurements of 51 woody plant taxa across four biomes showing dynamic g_s responses to a CO₂ rise of ~50 ppm, which represents the change in this greenhouse gas over the past three decades, (2) an extensive, up-to-date, synthesis of g_s responses in FACE experiments, (3) both the stand-alone and the Community Land Model version 4 (CLM4)-integrated application of the Farquhar–Ball–Berry (FBB) photosynthesis–stomatal conductance model and (4) the Medlyn et al. (2011) optimal stomatal model.

MATERIALS AND METHODS

Synthesis of free air CO₂ enrichment (FACE) studies

A literature review was undertaken of studies that specifically focused on the effect of elevated CO₂ on plant stomatal conductance (g_s) in FACE experiments. A total of 51 studies were included in the database (in alphabetical order: Adachi et al., 2014; Ainsworth and Rogers, 2007; Ainsworth et al., 2003; Bader et al., 2010; Bhattacharya et al., 1994; Borjigidai et al., 2006; Bryant et al., 1998; Calfapietra et al., 2005; Chen et al., 2014; Ellsworth, 1999; Ellsworth et al., 1995, 2012; Garcia et al., 1998; Ghini et al., 2015; Grant et al., 1999; Gunderson et al., 2002; Hamerlynck et al., 2000, 2002; Hao et al., 2013; Hättenschwiler et al., 2002; Herrick et al., 2004; Herrick and Thomas, 1999, 2003; Hileman et al., 1992, 1994; Huxman and Smith, 2001; Ji et al., 2015; Keel et al., 2006; Leakey et al., 2006; Lee et al., 2001; Marchi et al., 2004; McElrone et al., 2005; Naumburg and Ellsworth, 2000; Naumburg et al., 2003, 2004; Neal et al., 2000; Nijs et al., 1997; Noormets et al., 2001; Nowak et al., 2001; Pataki et al., 2000; Pearson et al., 1995; Rogers et al., 2004; Ruhil et al., 2015; Shimono et al., 2010; Singsaas et al., 2000; Tricker et al., 2005; Wall et al., 2000, 2001; Wechsung et al., 2000; Wullschleger et al., 2002; Yoshimoto et al., 2005). The FACE synthesis was built on the original data set by Ainsworth and Rogers (2007). Values reported in tables and in the text were taken directly from publications, whereas results in graphs were digitized. Individual independent observations were obtained following the longest period of CO₂ exposure reported in each study (independent = plant; repeated = species). Studies that examined multi-factorial designs could have contributed several observations for each response variable (drought, nitrogen enrichment, etc.). The mean, standard deviation (s.d.) and the effect size of the treatment (Ne) and of the relative control treatment (Na) were recorded. If standard error (s.e.) was reported we transformed these according to s.e. = s.d.*[(n − 1)/2]. Database records typically included the year and month the data were collected, GPS site locations, ambient CO₂, elevated CO₂, study organism (incl. varieties), plant functional type (PFT), photosynthetic pathway and other experimental treatments (e.g. nitrogen fertilization). Stomatal conductance measurements from 52 different species, within seven PFTs (C₃ crops, C₃ forbs, C₃ grasses, C₃ herbs, C₃ shrubs, C₃ conifer trees and C₃ broadleaved trees) were included in the analysis. The ranges of ambient and elevated CO₂ between studies were 350–411 and 538-680 ppm, respectively. A kernel density estimation was used to visualize the stomatal conductance data by estimating the unknown probability of the data, based on a sample of points taken from that distribution.

Dynamic g_s responses to CO₂ change (across four biomes)

Assessment of the dynamic stomatal responses to increasing CO₂ across four different biomes (including a tropical seasonal biome which had been subjected to drought) was achieved during a 10-week scientific expedition to North and Central America in summer 2014. A total of 51 woody tree and shrub species were measured with a CIRAS-2 gas analyser (PP-Systems, Amesbury, MA, USA) attached to a PLC6 (U) cuvette fitted with a 1.7-cm² measurement window and a red/white light LED unit.

Measurements were carried out (see Fig. 3) at two boreal forest sites [16 species, Bird Creek (60°58′N, 149°28′W) and Kenai (60°33.3′N, 151°12.8′W), Alaska, USA], one temperate deciduous forest site [11 species, Smithsonian Environmental Research Centre (38°53′N, 76°32′W), Maryland, USA], two tropical seasonal forest (wet) sites [15 species, Cambalache (18°27′N, 66°35′W) and Guajataca (18°24′N, 66°58′W), Puerto Rico] one of which had undergone a long drought period (Cambalache), and one tropical seasonal forest (dry) site [nine species, Guanica (17°93′N, 66°92′W), Puerto Rico]. See Supplementary Data Table S1 for a complete species list.

Fig. 3.

Dynamic g_s responses to a subtle CO₂ increase across four biomes observed in field conditions compared with modelled responses. (A) Percentage change in g_s during the transition from 354 (sub-ambient) to 400 ppm (modern ambient) atmospheric CO₂, which is representative of the atmospheric changes that have occurred over the past ~25 years. The boxes signify the distribution of the 25–75 % quartiles, with median and average values represented by a vertical line and an open square within the box, respectively. Whiskers indicate the distribution of the 5–95 % quartiles. Solid boxes represent the field measurements. Striped boxes represent the modelled percentage responses of g_s using the Farquhar–Ball–Berry model and the A, T_v and e_a/e_i values measured in the field. Different lower-case letters denote statistically significant differences between biomes (P ≤ 0.05). Asterisks indicate within-biome statistically significant differences between the conductance values at 354 and 400 ppm CO₂. n = 24–66 independent measurements depending on biome (see Table S1 for species list). (B) Percentage change in transpiration between 354 and 400 ppm atmospheric CO₂. (C) Locations of expedition sites visited during this study. See Table S1 for geographical coordinates and site information.

Stomatal responses were assessed on an average of four individuals per species between 0900 and 1300 h. A sun-exposed branch was sampled following standard protocols (Dang et al., 1997; Koch et al., 2004; Berveiller et al., 2007; Domingues et al., 2010; Rowland et al., 2015) from each individual using either a pruner (shrubs) or a pole with a scythe fitted on its top (trees) and was immediately recut under water. Following this, a fully expanded leaf from each branch was enclosed in the cuvette of the gas analyser, which was running at a sub-ambient ~year 1990 reference CO₂ concentration of 354 ppm (Betts et al., 2016). Stomatal conductance at sub-ambient CO₂ concentration was recorded upon stabilization of its value, which typically took less than 15 min. Subsequently, reference CO₂ was established at 400 ppm (year 2016 values) (Betts et al., 2016) and the leaf was left to equilibrate for at least 15 min before g_s at modern ambient CO₂ was recorded. Randomization of the sequence of the two treatments was ensured; overall about 65 % of the measurements started at 400 ppm (386.6 ± 0.5) and were reduced to 354 ppm (342.4 ± 0.5), while the remaining measurements (35 %) started at 354 ppm and were increased to 400 ppm. On several occasions the reversibility of the CO₂ effects on g_s was tested. This was done by measuring g_s at a starting CO₂ concentration of 400 ppm, after which CO₂ was reduced to 354 ppm for several minutes, before it was returned to the initial concentration of 400 ppm. The final g_s values at 400 ppm were the same as those initially recorded (data not shown).

Stomatal responses to a subtle increase in CO₂ were estimated as the percentage change in the g_s values between sub-ambient CO₂ and modern ambient CO₂. Air flow, light intensity and incoming mole fraction of water during the measurements were maintained at 200 cm³ min⁻¹, 1000 μmol m⁻² s⁻¹ and 80–90 % of ambient, respectively. Since ambient and leaf temperatures varied significantly between the beginning and the end of the daily measurement time window in all biomes, the measurements were taken at the calculated mean and biome-specific leaf temperature at 0900 h. Calculation was performed early on the first measurement day at each site by running the gas analyser at the set points mentioned above (i.e. 1000 µmol m⁻² s⁻¹ of light, 80–90 % of ambient water vapour, 400 µmol mol⁻¹ CO₂, no temperature control) and by recording the leaf temperatures of at least ten leaves belonging to ten different species growing at the site. Differences in g_s responses between biomes were tested on the normal data using analysis of variance (ANOVA). Moreover, a linear model was used to test for the correlation of g_s to vapour pressure deficit (VPD) and leaf temperature and the modelled and observed g_s data. Mixed effects models were used to test which variables best explain the observed changes in g_s and the best model was selected following Akaike’s Information Criterion (AIC).

Farquhar–Ball–Berry model (combined photosynthesis and g_s)

The model relates g_s to net leaf photosynthesis, scaled by the relative humidity at the leaf surface and the CO₂ concentration at the leaf surface (Collatz et al., 1991; Sellers et al., 1996). It solves the following three equations:

$$g_s=mA\frac{e_a}{e_i}\frac{p_a}{c_a}+b$$ (1)

$$A=\frac{g_s}{1.65}\frac{(c_a-c_i)}{p_a}$$ (2)

$$A=\min (w_c,w_j,w_e)$$ (3)

where g_s is the stomatal conductance to water vapour, A is the photosynthetic uptake flux of CO₂, c_a and c_i are partial pressures of CO₂ just outside and inside the stomata, respectively, p_a = 10⁵ Pa is atmospheric pressure, e_a and e_i are the water vapour pressures just outside and inside the stomata, respectively (the latter computed as the saturation vapour pressure at leaf temperature T_v), and m and b are empirical constants taken as m = 6 and b = 3 × 10⁴ µmol m^–2 s^–1. The uptake flux is taken to be the minimum of three rate-limiting processes for C₃ plants: Rubisco limitation, w_c = V_cmax (c_i– Γ^*)/(c_i+ K_c+ o_iK_c/K_o); light limitation, w_j = α PAR (c_i– Γ^*)/(c_i+ 2Γ*); and export limitation w_e = 0.5 V_cmax. In these expressions K_c and K_o are Michaelis–Menten constants for CO₂ and O₂, respectively, which vary with leaf temperature T_v (expressed in °C) as $K_c=K_{c25}{a}_{kc}^{(T_p-25)/10}$ and $K_o=K_{o25}{a}_{ko}^{(T_v-25)/10}$ where K_c25 = 30 and K_o25 = 30 000 are reference values while a_kc = 2.1 and a_ko = 1.2. The CO₂ compensation point is taken as Γ* = 0.105 o_iK_c/K_o with o_i the partial pressure of oxygen. PAR = 1000 µmol m^–2 s^–1 is the photosynthetically active radiation flux falling on the leaf, and α = 0.06 is the quantum efficiency of photosynthesis. Finally, V_cmax is the temperature-dependent maximum carboxylation rate modelled following Katul et al. (2010) as V_cmax = V_cmax25e^{0.88(Tv–25)}/(1 + e^{0.29(Tv–41)}) where V_cmax25 = 60 µmol m^–2 s^–1 is the maximum carboxylation rate at 25 °C. Given values of c_a, e_a, T_v, PAR and V_cmax25, the equations are solved numerically using an iterative method to yield c_i, A and g_s.

Optimization model

For the comparison of our field data with the optimum g_s model of Medlyn et al. (2011) we used measured values of A, c_a and VPD and Biome-specific g_l values and the version of the model equation from Lin et al. (2015).

$$g_s=1.6(1+\frac{g_l}{\sqrt{D}})\frac{A}{c_a}$$ (4)

where D is VPD (kPa), and g_l is the model coefficient.

The Community Land Model version 4 (CLM4)

The Community Land Model version 4 (CLM4), released in 2010 (Oleson et al., 2010; Lawrence et al., 2011) was used in this study. Land cover and atmospheric weather conditions serve as boundary conditions for CLM4. Grid cells in CLM4 may include vegetation, wetlands, lakes, glacier and urban regions. CLM4 can be used in conjunction with the other models in the Community Earth System Model (CESM), or independently (stand-alone), as is the case here. This is referred to as an I-compset. Specifically, we have used the I-compset with an f19g16 resolution and CLM4 satellite phenology. This simulation has the carbon and nitrogen cycling (biogeophysics ‘CN’) turned off. CLM4 parameterizes stomatal responses via an FBB scheme as described above.

CLM4 uses atmospheric boundary conditions for integration. We use the QIAN atmospheric input data set, for 1972–2004 (Qian et al., 2006). This is a global forcing dataset for the period 1948–2004 with 3-hourly temporal and T62 spatial resolution (1.875°). The dataset was developed by combining analyses of monthly precipitation and surface air temperature with intra-monthly variations from the National Centers for Environmental Prediction – National Center for Atmospheric Research (NCEP–NCAR) re-analysis (Qian et al., 2006). Using the I-compset we performed experiments at 350, 400 and 700 ppm. Results are provided as climatological mean values over the forcing period (1974–2004). Atmospheric forcing, as per Qian et al. (2006), is identical between each of the 350, 400 and 700 ppm runs.

RESULTS

Free air CO₂ enrichment studies (FACE)

To investigate the range of responses of g_s across global sites (Fig. 1) we performed a synthesis of data from 51 FACE studies. Of the 1313 independent measurements across 52 species, 88.2 % of the measurements showed a decrease in g_s in response to elevated CO₂ (Fig. 2). However, 11.8 % of the measurements showed an increase in g_s (Fig. 2). Such increases have gone largely unreported in the past, with most meta-analyses focusing on the overall mean negative response (decrease) of g_s to increasing CO₂ concentration (e.g. Ainsworth and Rogers, 2007). Overall, g_s decreased by ~19 % on average across all FACE studies (Fig. 2).

Fig. 2.

FACE synthesis of g_s responses to increasing CO₂ concentration. Kernel density probability distribution of the percentage change of g_s to increasing CO₂ concentration. Each colour represents a different plant functional type (PFT). The percentage g_s change is expressed as the delta change of g_s between ambient and high CO₂ treatments. Solid lines are median (blue) and mean (red) change in g_s across all PFTs. The dashed line is the zero percentage change mark. See Materials and Methods for details of the synthesis and cited FACE studies used.

Field survey of g_s responses to a 50 ppm CO₂ rise

A total of 51 C₃ tree and shrub species (n = 209) were sampled during the in situ CO₂ gas exchange measurements across four biomes (Fig. 3). Measurements reveal significant variation in the dynamic g_s responses to an ~50 ppm CO₂ increase, which was selected to represent anthropogenic climate change over the past 25 years (from 354 to 400 ppm) across the different biomes (Fig. 3). The species of the boreal, temperate deciduous forest and tropical seasonal forest (moist) biomes displayed an overall negligible response to increasing CO₂ (Fig. 3). In contrast, the species of the tropical seasonal forest (dry) and, to an even greater extent, the species of the tropical seasonal forest (drought), which had been subjected to a 1-month-long drought period prior to the measurements, displayed statistically significant mean increases in g_s in response to a 50 ppm rise in CO₂ (6.8 and 11.1 %, respectively) (Fig. 3). The grouping of stomatal responses between wet [i.e. boreal forest, temperate deciduous forest and tropical seasonal forest (moist)] and dry regions [i.e. tropical seasonal forest (dry) and tropical moist seasonal forest (drought)] is also clearly reflected in the corresponding changes in plant transpiration; decreasing and increasing mean transpiration are observed, respectively (Fig. 3).

Field g_s data – model comparison

Our finding that g_s can respond positively to increasing CO₂ is supported by the theoretical predictions of the combined FBB photosynthesis and g_s model. The model simulations, under an ~50 ppm CO₂ rise scenario, demonstrate that increases in atmospheric CO₂ drive increases in g_s (Fig. 4) under conditions of high VPD (expressed as e_a/e_i in the model) and medium to high leaf temperature (T_v). The dependence of g_s responses to increasing CO₂ on air moisture and leaf temperature is also observed in the field gas analysis data by positive correlations between g_s responses and VPD and leaf temperature (Fig. 5). This was also confirmed using mixed effects models, which showed that the measured relative changes in g_s are best explained when the relative changes in A and e_a/e_i are used as fixed factors (AIC = 1633.8, χ² = 4.0348, P = 0.044). The FBB simulations provide a theoretical underpinning for the field observations by demonstrating that plants can increase g_s as a response to increasing CO₂, while simultaneously optimizing water use efficiency (WUE) (Fig 4). In the model, increases in WUE are observed across all values of T_v and humidity. However, increases in WUE are highest in the parameter space where leaf humidity is low (dry regions) and T_v is high (warm–hot regions). A second simulation shows that the model produces an even higher g_s increase in response to a doubling of CO₂ (to 700 ppm) in dry and warm–hot regions of the parameter space (not shown).

Fig. 4.

Results from the Farquhar–Ball–Berry combined photosynthesis and g_s model. (A) A–c_i response curves at two different leaf temperatures, as indicated in the key. (B) A–c_a response curves at two different temperatures and humidities (see key in C). (C) Sensitivity of A to c_a, normalized by A/c_a, as a function of c_a at two different temperatures and humidities, as indicated in the key. (D) Predicted g_s at c_a = 350 ppm as a function of leaf temperature and humidity. (E) Predicted percentage change in g_s when c_a changes from 350 to 400 ppm, with zero contour highlighted by solid black line. (F) Predicted percentage change in water use efficiency (WUE) when c_a changes from 350 to 400 ppm. Symbols in all panels indicate three selected cases: high temperature, low humidity (circles); high temperature, high humidity (squares); and low temperature, low humidity (triangles).

Fig. 5.

Gas analysis relationship between g_s and vapour pressure deficit and leaf temperature. Linear relationship and 95 % confidence bands (dotted lines) between the percentage change in g_s during the transition from 354 (sub-ambient) to 400 ppm (modern ambient) atmospheric CO₂ and (A) VPD (kPa) (y = 5.94x − 5.24, r² = 0.21, P < 0.01) and (B) leaf temperature (°C) (y = 0.63x − 12.82, r² = 0.14, P < 0.01). Data represent species averages with an average number of four individuals measured per species.

To test how well the field infrared-gas-analyser measured g_s is described by the FBB model, as well as the optimal g_s model of Medlyn et al. (2011), we used the recorded values of photosynthesis (A), T_v and water vapour concentration to calculate the model-implemented g_s of all 51 taxa analysed. For the Medlyn et al. (2011) model we used published g_l values by Lin et al. (2015) for evergreen and deciduous trees and shrubs. Here g₀ values of 20 mmol m⁻² s⁻¹ are used. The comparison of modelled and recorded data revealed that the FBB model can accurately predict the observed g_s, with the regression between estimated and observed g_s falling very close to the 1: 1 line (Fig. 6). Furthermore, the model-implemented g_s responses are strikingly similar to those observed in the field (Fig. 3). A similarly good fit was found when observed g_s values were plotted against the optimal g_s model of Medlyn et al. (2011) (Supplementary Data Fig. S1).

Fig. 6.

Comparison of measured and modelled g_s values under 354 and 400 ppm of atmospheric CO₂. Relationship (0.95x+6.8, r² = 0.431, solid line) between measured and modelled g_s values. Stomatal conductance was modelled using the Farquhar–Ball–Berry model and the A, T_v and e_a/e_i values measured in the field. The dashed line represents the 1: 1 relationship. Mixed effects model results showed that the relative changes in g_s are best explained when the relative changes in A and e_a/e_i are used as fixed factors (AIC = 1633.8, χ² = 4.0348, P = 0.044).

The Community Land Model – a spatial investigation of global g_s

To gain a deeper understanding of the land–vegetation system response to increases in CO₂ at a spatial global scale, we performed simulations using the CLM4 land–vegetation model. The FBB model is also used for the parameterization of CLM4. Simulations of the same CO₂ increases in CLM4 resulted in a similar pattern of g_s responses (Fig. 7). In response to a 50 ppm CO₂ increase the CLM4 simulation produces predominantly negative changes (decreases) in g_s (Fig. 7). An ~3.2 % annual global climatological maximum decrease in g_s is simulated (Table 1). However, positive g_s responses are also simulated, with a maximum increase of ~4.9 % (Fig. 7, Table 1). A second annual global simulation, forcing the system with a doubling of CO₂ (to 700 ppm), resulted in a larger ~16.8 % global climatological maximum decrease in g_s (Fig. 7). As in the 50 ppm scenario, positive g_s responses were also simulated across the low latitudes, this time with higher maximum positive changes of ~18.9 % (Fig. 7, Table 1). There was a clear seasonal latitudinal and regional trend in the magnitude of g_s change between months in the simulation (Fig. S2). For example, positive g_s increases (to 50 ppm) were mostly observed in the months between December and May in Central Africa and between June and October in South Africa. In contrast, positive g_s increases in Central America were observed in the months between January and June and in South America between June and November. Interestingly, the g_s increases were accompanied by increases in soil moisture (Fig. 8, Table 1). Annual modelled regions experiencing an increasing g_s response to CO₂ include Mexico, the Galapagos Islands, Dominican Republic, Columbia, Venezuela, Brazil, Bolivia, Sudan, South Sudan, Somalia, Tanzania, Democratic Republic of Congo (D.R.C.), Angola, Namibia, Botswana and Indonesia (Fig. 7, Table 2). Similar to our field observations, areas that showed positive g_s increases were situated in hot and dry biomes (Table 2).

Fig. 7.

Annual g_s response to increasing CO₂ in the CLM4 land–vegetation model. Negative and positive g_s responses to increasing CO₂ in CLM4, for (A) a 400 ppm and (B) a 700 ppm scenario, relative to 350 ppm. Modelled regions experiencing positive g_s responses for both A and B include parts of Central America, South America, Africa and Asia (see Table 2 for more detail). Note that the majority of the land surface experiences decreases in g_s in response to increasing CO₂.

Table 1.

Community Land Model maximum annual increases/decreases and percentage of grid cells showing increases/decreases or no change in g_s and soil moisture worldwide

CO2 (ppm)	Variable	Max. decreases	Max. increases	Percentage no. of grid cells
				Increase	Decrease	No change
400–350	Stomatal conductance (mmol m–2 s–1)	0.00075 (3.15 %)	0.00004 (4.92 %)	1.94	64.22	33.83
	Soil moisture (kg m−2)	0.1 (0.21 %)	1.1 (2.3 %)	48.55	0.15	51.33
700–350	Stomatal conductance (mmol m–2 s–1)	0.00004 (16.82 %)	0.00001 (18.94 %)	1.45	65.81	32.74
	Soil moisture (kg m−2)	2.6 (5.6 %)	0.01 (0.02 %)	80.87	0.03	19.11

Fig. 8.

Detailed analysis of Community Land Model grid cells showing positive g_s responses under a 400 and 700 ppm CO₂ scenario. Percentage change of soil moisture and g_s for a 400 ppm (solid lines) and a 700 ppm (dashed lines) scenario, relative to 350 ppm. Only grid cells that showed positive increases in g_s are used for this analysis (geographical areas coloured in red and orange in Fig. 7).

Table 2.

Countries and associated biomes that showed annual positive increases in g_s under a 50 ppm increase in CO₂

Continent	Country	Biome
Central America	Mexico	Tropical & Subtropical Dry Broadleaved Forest
South America	Galapagos Islands	Mediterranean Forests, Woodland & Shrub
South America	Dominican Republic	Tropical & Subtropical Dry Broadleaved Forest
South America	Columbia	Tropical & Subtropical Dry Broadleaved Forest & Deserts & Xeric Shrublands
South America	Venezuela	Deserts & Xeric Shrublands
South America	Brazil	Deserts & Xeric Shrublands
South America	Bolivia	Tropical & Subtropical Grasslands, Savannas & Shrublands
Africa	Sudan	Tropical & Subtropical Grasslands, Savannas & Shrublands
Africa	South Sudan	Tropical & Subtropical Grasslands, Savannas & Shrublands
Africa	Somalia	Tropical & Subtropical Grasslands, Savannas & Shrublands
Africa	Tanzania	Tropical & Subtropical Grasslands, Savannas & Shrublands
Africa	D.R.C.	Tropical & Subtropical Grasslands, Savannas & Shrublands
Africa	Angola	Tropical & Subtropical Grasslands, Savannas & Shrublands
Africa	Namibia	Tropical & Subtropical Grasslands, Savannas & Shrublands
Africa	Botswana	Tropical & Subtropical Grasslands, Savannas & Shrublands
Asia	Indonesia	Tropical & Subtropical Dry Broadleaved Forest

DISCUSSION

Overall, our results clearly demonstrate that in dry, warm environments, or during drought periods, plants can respond to increases in CO₂ by increasing their g_s, while, crucially, maximizing the increase in their WUE (Figs 3, 4 and 7) compared to plants growing in the cooler moist conditions of the temperate latitudes. Implementation of the FBB model clearly shows a region of parameter space where CO₂, g_s and WUE increases can coincide (Fig. 4). The FBB model, when supplied with independently measured values of V_cmax, was able to accurately predict field observations, including the unexpected increases in g_s at high T_v and high VPD (Figs 3 and 6), a region of parameter space not often explored in standard gas analysis protocols, which typically run under standardized temperatures and VPD of 22 °C and 1 kPa, respectively. Although the measured g_s responses are small and difficult to capture under field conditions, Figs 3 and 6 show excellent agreement between modelled and observed values and strongly support our claims.

For a more mechanistic understanding of the g_s responses documented above, we turn to a more detailed analysis of the FBB model. Firstly, we note that in the light-saturated conditions we are exploring here, A is Rubisco-limited and is thus expected to increase with temperature. In the particular formulation used here (see Materials and Methods), V_cmax increases roughly exponentially with temperature at temperatures below ~35 °C, leading to a strong steepening of the A–c_i response curve as temperature increases (Fig. 4). This steepening carries over to the A–c_a response, as shown in Fig. 4; this figure also shows that higher humidity yields greater A at a given temperature and c_a, because greater humidity promotes stomatal opening (Fig. 4) and thus greater c_i, enhancing photosynthesis. Furthermore, we note that eqn (1) in the model (see Materials and Methods) implies that the sensitivity of g_s to ambient CO₂, dg_s/dc_a, at fixed temperature and humidity is given by:

$$\frac{c_a^2{e}_a}{mA{p}_a{e}_i}\frac{d{g}_s}{d{c}_a}=\frac{c_a}{A}\frac{dA}{d{c}_a}-1$$ (5)

Thus, increasing g_s in response to increasing c_a is possible when the first term on the right-hand side is greater than one, i.e. when the relative change in A is greater than the relative change in c_a. This condition can be met when temperature is high and humidity is low (as exemplified by the solid circles in Fig. 4): in that case, dA/dc_a is high while A is low, bringing dg_s/dc_a above zero (Fig. 4). When both temperature and humidity are high (squares in Fig. 4), A is large enough to make the first term on the right less than one; conversely, when both temperature and humidity are low (triangles in Fig. 4), A is low but dA/dc_a is also low, and the first term on the right is still less than one.

In summary, the FBB model predicts dg_s/dc_a> 0 at high temperature and low humidity under light-saturated conditions because high temperature promotes a strong gain in A per unit increase in c_i (or c_a), while low humidity keeps the base value of A low. Naturally, different model formulations would give quantitatively different results; in particular, the threshold values of temperature and humidity required for dg_s/dc_a> 0 are likely to be strongly model-dependent. However, the qualitative nature of the result appears robust, because increasing V_cmax with increasing temperatures and stomatal opening with increasing humidity are both well-known features of plant physiology. Note in particular that the optimization models of Medlyn et al. (2011) also predict increasing g_s as humidity increases (or VPD decreases), and would thus give qualitatively similar behaviour to the empirical Ball–Berry closure reported here (Fig. S1).

It is surprising that the possibility of g_s increasing as a response to rising CO₂ under these particular climatic conditions has not been highlighted before. As implied above, optimization models also predict similar increases within the CO₂ envelope tested in the present study (i.e. 354–400 ppm CO₂) (Arneth et al., 2002; Konrad et al., 2008; Medlyn et al., 2011, 2013). For example, the optimization model of Konrad et al. (2008) demonstrates that the inflection point between rising and falling g_s response to CO₂ is dependent on the ‘cost of water’ (Fig. 4 in their article). In particular, high cost of water shifts the inflection point to higher values, which are similar to those used in the present study. These predictions fit well with both our measured and modelled g_s responses.

It is intriguing that a substantial number of the FACE studies (see Materials and Methods) also report increases in g_s under super-ambient CO₂. These increases in g_s are generally not discussed, or are disregarded as methodological artefacts (Gunderson et al., 2002). Due to a lack of standardized FACE protocols, the exact reasons why positive g_s responses are observed across these studies remain largely unclear. Possible reasons for the observed increases might include: (1) differences in the climatic and/or cuvette measurement conditions; (2) differences in soil nutrient and water status; (3) differences in the signal to noise ratio with regard to g_s (i.e. species with low g_s show a greater propensity for erroneous measurements); and (4) studies do not consistently record the time when measurements are taken, despite literature which shows that g_s responses to CO₂ are highly dependent on the time of day (Konrad et al., 2008). Unfortunately, FACE studies inherently include a range of weather regimes/cuvette conditions and measurement times, which are inconsistent amongst studies and typically unreported. It is therefore not possible to assess the role of these conditions with regard to the reported g_s increases. Secondly, nutrient concentrations and soil water content naturally vary between sites, but are inconsistently documented across studies (e.g. Naumburg et al., 2003) making direct comparison unfeasible at this time. Regarding the potential low signal to noise ratio of the species that display increases in g_s as a response to increased CO₂, our meta-analysis of FACE studies showed that there is no significant difference in the g_s values between species that show either positive or negative responses to CO₂ (F = 1.663, P = 0.198). The same was found for the g_s responses of different PFTs, with the exception of shrubs (F = 4.122, P < 0.001). Thus, the observed positive g_s responses in FACE studies may arise for several reasons. It is likely that at least some of them are due to warm, dry conditions, as demonstrated by our field data (Figs 3 and 5) and model comparisons (Fig. 6 and Fig. S1).

Positive g_s responses have the potential to alter regional or even global hydrological and carbon cycles, and other ecological processes. We acknowledge that there are limitations in assessing long-term g_s trends through field measurements, as they cannot account for long-term water availability changes resulting from the CO₂ effects on g_s. Several studies have shown that decreasing soil moisture can elicit greater stomatal closure under elevated CO₂ than ambient CO₂ (Leakey et al., 2006; Piao et al., 2007; Gray et al., 2016). Similarly, increases in LAI have been shown to reduce soil moisture, thus indirectly affecting g_s (Field et al., 1995; Wenfang et al., 2013). Our global simulations using the CLM can only partially test for this, as LAI was not simulated here. It also needs to be noted that current CLM parameterizations do not account for many morphological plant responses to elevated CO₂ (e.g. changes in stomatal density). Keeping these reservations in mind and although predictions of future g_s are somewhat beyond the scope of the present study, Fig. 8 shows that in regions where g_s is predicted to increase in response to a 50 and 350 ppm CO₂ rise, soil moisture also increases (in this instance the increased soil moisture may be caused by water savings due to suppressed g_s in prior months, and may in fact cause the annual mean increase of g_s. at these locations). Coupled with potential increases in LAI in response to elevated CO₂ (Piao et al., 2007; Wu et al., 2012; Niu et al., 2013; Frank et al., 2015; Schymanski et al., 2015), regionally increasing g_s may act to offset the much studied effects of decreasing g_s, such as increasing river runoff (Gedney et al., 2006; Betts et al., 2007; de Boer et al., 2011; Gopalakrishnan et al., 2011; Lammertsma et al., 2011), or even drive enhanced drought and desertification in certain regions (Dai, 2013). Areas that were predicted by the CLM to show increases in g_s with elevated CO₂ (~50 and 350 ppm) are located in hot and dry biomes (Fig. 7 and Table 2). A monthly analysis of g_s for the CLM also suggests that the relative timing of temperature and relative humidity is important in driving the g_s increases, which leads us to expect increases in g_s in monsoonal regions (Fig. S2). However, due to other confounding factors (e.g. vegetation types and/or soil moisture) this expectation is not always met (e.g. India) and requires further investigation, which is beyond the scope of the current study. Continued land–vegetation model development based on field data at the biome (and community–species) level, as well as further Earth System Model inter-comparison studies, will be required to assess the implications of this shift in our understanding of vegetation responses to elevated CO₂, and for improved prediction of the global hydrological cycle, particularly in dry and warm–hot regions.

We have demonstrated that increases in g_s can occur under elevated CO₂ in environments that are hot and dry (high VPD). Our field observations across several global biomes are in excellent agreement with predictions from optimization models and fall within a previously unrecognized parameter space within the FBB model. The implications of our findings are of global significance for future modelling of soil–vegetation–climate feedbacks, as the FBB model is also implemented in the CLM. Although most of the global vegetation responds by decreasing g_s under elevated CO₂, biomes that already experience drought conditions are likely to show increases in g_s. It remains to be seen how these increases will affect soil–canopy–atmosphere climate feedbacks in the future, particularly in areas that are already expected to be more threatened as a result of predicted changes in climate.

SUPPLEMENTARY DATA

Supplementary data are available online at www.aob.oxfordjournals.org and consist of the following. Table S1: Species list and site descriptions. Fig. S1: Comparison of measured and modelled g_s values under 354 and 400 ppm of atmospheric CO₂ using the optimal g_s model of Medlyn et al. (2011). Fig. S2. Stomatal conductance response to increasing CO₂ in the CLM4 land–vegetation model for each month of the year. Negative and positive g_s responses to increasing CO₂ in CLM4 (400 ppm relative to 350 ppm).