Plant genetic variation drives geographic differences in atmosphere–plant–ecosystem feedbacks

Why this research Matters The objective of this study was to understand how genetic variation in a riparian species, Populus angustifolia, affects mass and energy exchange between the land and atmosphere across ~1,700 km of latitude of the western United States. To examine the potential for large‐scale land–atmosphere feedbacks in hydrologic processes driven by geographic differences in plant population traits, we use a physical hydrology model, paired field, and greenhouse observations of plant traits, and stable isotope compositions of soil, stem, and leaf water of P. angustifolia populations. Populations show patterns of local adaptation in traits related to landscape hydrologic functioning—a 47% difference in stomatal density in greenhouse conditions and a 74% difference in stomatal ratio in the field. Trait and stable isotope differences reveal that populations use water differently which is related to historical landscape hydrologic functioning (evapotranspiration and streamflow). Overall, results suggest that populations from landscapes with different hydrologic histories will differ in their ability to maintain favorable water balance with changing atmospheric demands for water, with ecosystem consequences.

or due to adaptations to differing historical abiotic conditions. For instance, the ability of plants to use water to produce biomass depends strongly on soil water availability which varies significantly across the landscape and is also affected by temperature (Beier et al., 2012;Rodriguez-Iturbe & Porporato, 2005). If atmospheric demand for water increases (i.e., high atmospheric vapor pressure deficit), which is predicted to occur globally, plants must prevent excessive water loss (Grossiord et al., 2020). Rapid responses to high vapor pressure deficits include adjusting stomatal aperture, while longer-term responses include altering the density, distribution, and size of stomatal pores (Bertolino, Caine, & Gray, 2019;Cowan & Farquhar, 1977;Hetherington & Woodward, 2003;McAdam & Brodribb, 2014;Oren et al., 1999). Genetically based variation stomatal density or size (Mitton, Grant, & Yoshino, 1998) result in variations of maximum stomatal conductance, affect a plant's ability to manage limited resources, and affect large-scale ecosystem processes (Novick et al., 2016). Transpiration directly supports primary productivity, biomass accumulation, and carbon assimilation, thus is directly related to carbon, water, and energy fluxes on the landscape (Hetherington & Woodward, 2003;Kominoski et al., 2013;Sposito, 2017). Here, we consider variation in historic water cycling on the landscape and examine local adaptation of plant populations to understand ecological and evolutionary linkages on a landscape scale.
Using the Budyko physical hydrology model, paired field and greenhouse observations of P. angustifolia traits, and stable isotope compositions of soil, stem, and leaf water, this study considers the potential for large-scale land-atmosphere feedbacks in hydrologic processes driven by geographic differences in plant population traits. With the observation that the supply and atmospheric demand for water, as well as water use differ on the landscape across populations of P. angustifolia, we test the following specific hypotheses: (a) Populations of P. angustifolia show genetic divergence in stomatal density, stomatal distribution, stomatal size, and aboveground biomass, (b) consistent with patterns of genetic divergence and local adaptation, stomatal traits are related to hydrologic variables on the landscape, (c) populations draw water from different sources (e.g., stream water or precipitation), and (d) populations vary in water use given atmospheric demands. Overall, results show that divergent plant populations have evolved in response to geographic variation in dryness.

| Building site-level energy and water budgets
We built energy and water budgets using Budyko model parameters describing how precipitation (P) is recycled to the atmosphere via actual evapotranspiration (AET) or held on land as streamflow (Q) across a continuum of humid to arid systems (PET/P; Figure 1b; Budyko, 1974). The theoretical model (note, the modeled curve is not depicted in Figure 1b) provides expectations for the energy balance and water use based on physical processes (i.e., evapotranspiration consumes heat as latent energy flux during the phase change of liquid water to vapor) and the assumption that P = Q + AET (Budyko, 1974;Trenberth, Fasullo, & Kiehl, 2009;Wang & Dickinson, 2012). Values derived from real data represent longterm patterns describing how water actually cycles on the landscape, taking into account more than physical processes-in other words, landscape variation in interactions between soil, vegetation, and atmospheric conditions. The dryness index (PET/P) on the xaxis of the model represents the aridity of the climate, with values greater than 1 indicating arid climates whereby plants are limited by water rather than by energy. The evaporative index (AET/P) on the y-axis describes how precipitation is distributed on land, or the percentage of P recycled back to the atmosphere through AET. An energy limit exists where AET = PET (i.e., demand limit; at which atmospheric demand for water is met), and a mass limit exists where AET = P (also known as a water limit, or supply limit; i.e., 100% of P is partitioned back to the atmosphere) (Budyko, 1974;Creed et al., 2014;Jones et al., 2012). In this paper, we are explicitly interested in unique long-term patterns of water cycling on the landscape which capture landscape heterogeneity, and not in the theoretical predictions ( Figure 1b; Gentine et al., 2012;Troch et al., 2013). We extracted mean annual precipitation from WorldClim (Fick & Hijmans, 2017), and mean annual PET and AET from the CGIAR-CSI GeoPortal (Trabucco & Zomer, 2009) using geo-referenced locations of our collection sites.

| Study species and sites
To understand plant-energy-water relationships we used a dominant riparian tree species: Populus angustifolia James (Rood, Nielsen, Shenton, Gill, & Letts, 2010), the narrowleaf cottonwood, that is widely distributed along the Rocky Mountains from northern Mexico to southern Canada  and span large precipitation, temperature, stream flow, and soil water gradients. Cottonwoods, Populus ssp., are an ideal study system for examining these relationships as they show intraspecific variation in physiological and morphological responses to changes in the water cycle, including groundwater, precipitation, and streamflow (Rood, Braatne, & Hughes, 2003). Furthermore, Populus ssp. are foundation species in riparian ecosystems in the western U.S. contributing greatly to ecosystem transpiration, but have been generally labeled as "drought sensitive" species that are declining in recent years (Kominoski et al., 2013;Schaeffer, Williams, & Goodrich, 2000). It is also clear that these riparian forests do not receive enough precipitation during the growing season to support the levels of transpiration to meet atmospheric demand (Flanagan, Orchard, Logie, Coburn, & Rood, 2017;Scott, Shuttleworth, Goodrich, & Maddock III, 2000;Yang, Rood, & Flanagan, 2019). Populations under such environmental constraints are ideal for identifying genetic divergence in response to varying hydrologic dynamics.
We have established field sites along 17 rivers in the western United States that span significant environmental gradients and nearly 1,700 kilometers of latitude ( Figure 1a). In 2012, over 525 genotypes of P. angustifolia were collected and geo-located from multiple (minimum three, maximum five) sites along each river, including at the highest and lowest elevations. The collected trees have been established in a greenhouse at the University of Tennessee and all tree replicates were tagged with a number and randomized in the common environment to minimize microspatial variation in light or temperature (details in Ware, Van Nuland, et al., 2019). This is a conservative experimental approach to examining genetic variation at multiple genetic hierarchies, including provenance, population, site, and genotype, that reduces observer sample bias. No plants were water limited in the greenhouse, and temperature conditions were maintained between 65 and 75 degrees Fahrenheit. Testing for variation in trait measurements in the common environment and relating these traits to environmental parameters allows us to infer patterns of local adaptation (Kawecki & Ebert, 2004;Leimu & Fischer, 2008). We refer to populations as groupings of all genotypes from sites along each river, resulting in 17 river populations.
These 17 populations vary locally and regionally, grouping into three genetically distinct provenances which have been geographically isolated by large landscape features including the Great Basin, the Rocky Mountains, and the Mogollon Rim (Figure 1a; Evans et al., 2015). The blue line represents a water limit (AET = P), at which 100% of water supplied to the landscape as precipitation (P) is cycled back to the atmosphere through evapotranspiration (AET). The red line represents an energy limit (AET = PET), at which the amount of water recycled to the atmosphere through evapotranspiration (AET) meets the atmospheric demand for water (PET). Genetic provenance is represented by color as in panel A. Points are plotted using observed values of mean annual P, PET, and AET from georeferenced locations of field collection sites. Inset boxplots show provenance differences in the two Budyko parameters, with letters referring to statistically significant differences between provenances from post-hoc Tukey Contrasts [Colour figure can be viewed at wileyonlinelibrary.com] affect stomatal traits (e.g., Hamanishi, Thomas, & Campbell, 2012;Pearce, Millard, Bray, & Rood, 2005), we checked for ontogenetic differences in traits between older clones from which biomass was derived and their respective younger clones (from the same "source" tree in the field). Seven of the same genotypes were measured for stomatal traits in the "older" (2012) trees in October 2017. A two-tailed unpaired t test on stomatal density and stomatal distribution showed no difference between the two age groups (p = .56, p = .39, respectively).

| Aboveground biomass
In the field, aboveground biomass estimates of P. angustifolia genotypes were made in 2012 by measuring tree circumference (m) which was used to calculate DBH (cm): DBH = 100 × circumference/3.14).
We estimated biomass (kg) using an allometric equation for Populus

| Stomatal traits
We measured three traits related to stomatal function: density; distribution, and size. Stomata control the movement of gases in and out of the leaf (e.g., carbon dioxide for photosynthesis, water via transpiration). Variation in the size and the density of stomata as well as the location on leaf surfaces (i.e., adaxial (top), abaxial (bottom)) reflect ways that plants can control water loss, and thus are important to plant function (Aasamaa, Sõber, & Rahi, 2001;Bertolino et al., 2019;Cornelissen et al., 2003;Hetherington & Woodward, 2003;Sack, Melcher, Liu, Middleton, & Pardee, 2006).
Leaves were collected in the field in June 2017 from two genotypes along (3 sites) six rivers distributed across the three genetic provenances (Provenance 1: Blue River, NM and Oak Creek, AZ; Provenance 2: San Miguel River, CO and Indian Creek, UT; Provenance 3: Weber River, UT and Snake River, WY). These collections resulted in six genotypes per river, or 12 genotypes per genetic provenance, and were the same genotypes that were visited in 2012 collection described above. We chose three leaves from the terminal shoots of lower exterior branches of each tree to minimize intra-canopy and age variation in stomatal density (Sack et al., 2006). Impressions of the leaf epidermis were made on the adaxial and abaxial side of each leaf using clear nail varnish and tape, then individually arranged on glass slides. Counts were made in the software ImageJ (Schneider, Rasband, & Eliceiri, 2012) from light microscopy photographs with a 10X objective. We calculated the total number of stomata per area by adding the number of stomata on both leaf surfaces (henceforth "stomatal density"), and we calculated the relative placement of stomata by calculating the ratio of adaxial density to abaxial density (henceforth "stomatal ratio"). These methods resulted in six impressions per genotype, or 216 total impressions. Finally, we made 20 measurements of stomatal pore length on each photograph in ImageJ (Schneider et al., 2012). As stomatal density on photographs was often higher than 20, we overlayed a grid in ImageJ and randomly selected a row across which to begin measurements. If density was too low to obtain 20 measurements, more often on the adaxial impressions, we measured the pore length of every present stoma. These methods resulted in 120 pore length measurements per genotype. We repeated the same measures from leaves collected from the same genotypes of trees growing in the common environment, described above, although we lost three genotypes from the San Miguel and the Weber Rivers and one genotype from both the Blue River and the Snake River. Greenhouse measurements therefore consisted of a leaf collected from three clonal replicates of 28 genotypes.

| Water stable isotope measurements
We analyzed river water, stem, leaf, and soil samples for stable isotope measurements (δ 18 O and δ 2 H) to determine plant water source.
In June 2017, at the mid-elevation site along each of the six rivers, we collected stem and leaf samples for stable isotope analysis from two unique genotypes of P. angustifolia. Soil samples were collected from underneath each genotype at a depth of approximately 10 cm.
River water samples were collected below the surface of the water in each of the six rivers. All samples were kept on dry ice until de-

| Statistical Analyses
All analyses were performed using the statistical software R (version 3.6.1; R Development Core Team, 2016). To confirm our observations that water availability and the atmospheric demand for water vary across the range of P. angustifolia, we built linear models predicting variation in the two axes of the Budyko water budget (dryness index and evaporative index) with population. Separate models were built for the 17 populations examined for biomass, the six-population subset used for stomatal measurements, and the three genetic provenances (with population as a random effect; R package lme4 (Bates et al., 2015)). Hypothesis testing for each linear model was done by marginal sums of squares ANOVA in the R package car (Fox et al., 2018) and the null hypothesis was rejected at an α = 0.05.
To test the hypothesis that stomatal and growth traits from P. angustifolia genetic provenances reveal patterns of local adaptation, we ran linear mixed effects models with biomass, stomatal density, and stomatal ratio as response variables, provenance as a fixed effect, and population (river) as a random effect in the lme4 R package (Bates et al., 2015). For stomatal density and ratio, genotype was also included as a random effect. Models were compared to null models with random effects only using likelihood ratio tests and by comparing AIC values. Post-hoc pairwise differences comparisons were made of provenance-level means with Tukey contrasts using the ghlt function in R package multcomp (Hothorn, Bretz, & Westfall, 2008) with the null hypothesis rejected at an α = 0.05.
To test the hypothesis that water-regulation and functional traits are related to hydrologic variables on the landscape, we used restricted estimated maximum likelihood (REML) linear mixed models (R package lme4 (Bates et al., 2015)). We included water budget parameters as fixed effects and we included genetic provenance in models as a random effect to remove "blocked" variation that can be attributed to genetic grouping. Response variables included P. angustifolia greenhouse biomass, stomatal density, stomatal ratio, and stomatal pore length measurements. Hypothesis testing for each linear model was done by marginal sums of squares ANOVA in the car R package (Fox et al., 2018) and the null hypothesis was rejected at an α = 0.05.
To test the hypothesis that populations draw water from different sources, we used stable isotope values to calculate deuterium excess values (d-excess) as d-excess = δ 2 H -8 × δ 18 O (Dansgaard, 1964).
This metric represents deviations from the average global relationship between δ2H and δ18O in precipitation, the global meteoric water line (GMWL; Craig, 1961). Because the global relationship varies across latitudes and continents (for example; (Sprenger, Leistert, Gimbel, & Weiler, 2016), we also calculated local meteoric water lines (LMWL) that are regionally specific. We used precipitation isotopic signatures for the month of June (when samples were col-

| Water supply (precipitation), atmospheric demand for water (potential evapotranspiration), and water use (actual evapotranspiration) differ across the range of Populus angustifolia
Populus angustifolia riparian forests across the western United States

| Populations of P. angustifolia show patterns of genetic divergence in traits related to the water cycle
Biomass: In the field, we find that biomass is lowest in Provenance 1 compared to Provenances 2 and 3 which do not show significant differences (Figure 2a, Table 1). Conversely, in the common environment we find that Provenance 1 had the highest biomass (µ p1 = 294.8 g), while Provenance 3 had the lowest average aboveground biomass (µ p3 = 89.7 g). Overall, this represents a 69% genetically based difference in biomass across the three provenances.
These results demonstrate a pattern of genetic divergence at the provenance level (Figure 2b, Table 1) and environmental constraints on biomass production in the field within the range of Provenance 1, likely related to limitations in the supply of water and plant strategies to mitigate water limitation. A post-hoc Tukey test reveals significant differences between Provenances 1 and 2, and Provenances 1 and 3 (Table 1). Provenances 2 and 3 show marginally significant differences in biomass (p = .09; Table 1) although these provenances have the lowest sample size (N p2 = 213 and N p3 = 75, respectively).
To check if these differences could be explained by growth duration (e.g., Evans et al., 2016), we also ran models including growing season length in the greenhouse (recorded as the number of days between first bud break in the spring and plant senescence in the fall), and latitude as a proxy for growing season length in the field.
Our findings did not change with consideration of these variables.
Stomatal Traits: While the average stomatal density does not differ between provenances in the field (Figure 2c; Table 1), Provenance 1 shows 47.3% difference in stomatal density in the common environment compared to Provenance 3 (µ p1 = 118.1; µ p3 = 80.2; p < .001), and a 23.7% increase relative to Provenance 2 (µ p2 = 95.5; p = .062) (Figure 2d; Table 1). Overall, this represents nearly a doubling of the total number of stoma on leaf surfaces across the provenances. Additionally, we find that provenances differ in the field in F I G U R E 2 Genetic provenances of Populus angustifolia differ in traits relating to water-use and ecosystem function. Letters refer to statistically significant differences between provenances from post-hoc Tukey Contrasts. Note the difference in y-axis scale between field and greenhouse biomass. showing a species average of stomatal ratio to be about 0.32 (Pearce et al., 2005). In field conditions, Provenance 1 has more stomates on the abaxial leaf surface, and although the greenhouse trend reflects this field trends, the only emergent significant difference is between Provenances 1 and 2 ( Figure 2f; Table 1). Finally, we found no significant differences between the three provenances in adaxial stomatal pore length. However, while abaxial stomatal pore length did not differ between provenances in the field (Figure 2g; Table 1), with an average length of 32.4µm, we did find differences in the greenhouse.
Similar to the patterns of stomatal distribution in the greenhouse, abaxial stomatal pore length in the greenhouse of Provenance 1 (µ p1 = 29.0 µm) was significantly smaller than Provenance 2 (µ p2 = 33.8 µm; p = .009), and marginally different from Provenance 3 (µ p3 = 32.8 µm; p = .0596) (Figure 2h; Table 1). Conforming to trends commonly found in the literature (e.g., Brodribb, Jordan, & Carpenter, 2013), our data show significant negative logarithmic relationships between stomatal density and stomatal pore length in the field and in the greenhouse, although this relationship depends on leaf surface.  Figure 3b). Furthermore, higher biomass plants generally have higher water demands that may be reflected in stomatal density. We show that stomatal density is positively correlated to biomass (g) in greenhouse plants, accounting for 80% of the variation (Figure 3c; R 2 = .80, p = .016). In the field, stomatal ratio appears to be positively related to biomass (kg) of field plants (R 2 = .55, p = .089). Also adhering to expectations, we find that lc-excess in the soil is significantly correlated with streamflow in the month prior to collection (May, R 2 = .38, p = .033) and marginally correlated with mean annual streamflow (R 2 = .297, p = .066). Although we do not have deep groundwater samples for our sampling locations, groundwater is known to consistently plot along the LMWL (Sprenger et al., 2016).

| Water s t ab le isotope compositions
We acknowledge that throughfall water may already be enriched when it reaches the nonsaturated soil zone and that tree cover TA B L E 1 Summary of the linear mixed effects model rankings for determining importance of provenance for biomass, stomatal density, distribution (ratio), and size (abaxial pore length) in the field [F] and in the greenhouse [GH]. River is included as random effect for all models. Genotype is also included as a random effect for stomatal models may decrease fractionation processes in soil (Sprenger et al., 2016).  Table 2). Although this regression is significant and shows good fit (Table 2; R 2 = .77), regressions of stem isotope compositions split by provenance each show a stronger fit (respectively by provenance, R 2 = .90, .95, .96), and different slopes (respectively by provenance, 3.5, 10.4, 3.9; Figure S2). Despite this, lc-excess values of stem water do not significantly differ between provenances ( Figure 4c). As expected, the slope for leaf isotopic composition is lower than all others, as leaves experience substantial isotopic enrichment during evapotranspiration ( Figure 4a, Table 2). Values of lc-excess in leaves are significantly higher in Provenance 3 samples (Figure 4d), supporting local adaptation patterns found in leaf traits (Figure 2) as well as the relationship between AET and stomatal ratio (Figure 5a).
Stomatal ratio in the field is significantly correlated with leaf lc-excess (R 2 = .53, p = .0076). Deuterium-excess (d-excess) values showed the same patterns as line conditioned excess values (lc-excess).
Our results suggest that P. angustifolia populations may draw water from different water sources (stream, soil, or precipitation) and/or may have locally adapted rooting structures as lower lc-excess values often correlate with shallower soil-water use (Sprenger et al., 2016). Previous research on Populus in Arizona shows that some trees can opportunistically use precipitation water when it is available but rely on groundwater or streamwater during dry periods (Snyder & Williams, 2000). We acknowledge that drawing comparative inferences from soil and plant stable isotope data across space is cautioned (Goldsmith et al., 2019), as is assuming plant accession to specific water sources based on matching isotope compositions (Zhao et al., 2016).

| Populations' role in the water c ycle varies on the landsc ape
The evaporative index (AET/P) represents the percentage of precipitation water recycled to the atmosphere through plants.
Stomatal ratio, in field and greenhouse plants, is related to the evaporative index (AET/P) such that stomatal ratios are lower (more stomates on the bottom of leaves) when a higher percentage of available water is cycled back to the atmosphere in a given location water is "lost" or held in another pool that is not captured by this model (e.g., in plant biomass) while negative values indicate that water is supplied to the system by means other than precipitation (e.g., snowmelt).

| Interacting global change gradients
Atmospheric hydrologic model parameters that capture variation in water-energy interactions across landscapes do a 30% better job of explaining patterns of plant biomass than temperature and precipitation in statistical models ( Figure S3), consistent with predictions that, although derived from temperature, PET should select more strongly than temperature on water-use traits and plant biomass (Siepielski et al., 2017;Wright et al., 2004). While impossible to simultaneously consider all interacting gradients across a landscape, these hydrologic variables do capture nuances in climatic interactions that independent gradients of temperature and precipitation do not: For example, physical water-energy interactions on the landscape vary across factors such as soil type, vegetation type and cover, and other biotic factors that the metrics in this   (Guisan & Zimmermann, 2000) and thus are critical to the functioning and persistence of ecosystems. A 2016 review of plant distribution models revealed that temperature and water-related variables appear in 88.5% of models, but that water-related variables that depend on temperature (e.g., evapotranspiration, moisture deficit) appeared in <20% of the models (Mod, Scherrer, Luoto, & Guisan, 2016). Our results and this identified gap in modeling distributions highlight the importance of including variables that more accurately represent the availability of water in ecosystems and demands for water from the atmosphere. Understanding complex interactions of global change gradients is a significant challenge for modeling the evolutionary (e.g., plant adaptation) and ecosystem consequences (e.g., plant function) of climate change.

| Evolution
Variation in water-use traits will determine plant response to changing water availability on the landscape. We show that stomatal traits and plant biomass have evolved among genetic groups of P. angustifolia across a landscape gradient of dryness (PET/P). Plants derived from more arid regions (higher dryness index values) produced more biomass in the greenhouse and biomass was positively related to stomatal density ( Figure 3c). These results conform to those found previously in P. trichocarpa, P. balsamifera, and P. angustifolia (Guy & Gornall, 2007;Soolanayakanahally et al., 2009). Interestingly, Kaluthota et al., 2015 found that differences in density between provenances were not related to aridity (Kaluthota et al., 2015) confusing the relationship we found that supports predictions that plants with high stomatal conductance in dry conditions may demonstrate rapid opportunistic biomass production (rate of photosynthesis) during infrequent or short periods of water availability (Hetherington & Woodward, 2003;Snyder & Williams, 2000). Conversely, populations derived from regions with historically high water supply may be less able to control water use and be at higher risk to drought-induced mortality (Dudley, 1996), although experiments are necessary to confirm these predictions (e.g., Barton, Jones, Edwards, Shiels, & Knight, 2020). Numerous other physiological studies on Populus species show that water stress through reductions in precipitation, groundwater, or streamflow, can lower leaf gas exchange, water potentials, xylem cavitation, stomatal conductance, and net photosynthetic rates (Horton, Kolb, & Hart, 2001;Rood et al., 2003;Tyree, Kolb, Rood, & Patino, 1994), resulting in morphological changes such as lower biomass production, increased branch sacrifice and crown reduction, leaf size, or stomatal size and number (Dunlap & Stettler, 2001;Rood et al., 2003;Rood, Patiño, Coombs, & Tyree, 2000). On the other hand, inundation with water, as would occur with flooding, has been shown to lower net photosynthetic rate, stomatal conductance, transpiration, and growth in Populus (Amlin & Rood, 2001;Rood et al., 2010). Varying responses to these two extremes of water stress, drought, and flooding, emphasize the need to consider population-level responses to multiple aspects of the water cycle. Global variation in plant growth is predominantly attributed to temperature and water (Babst et al., 2019;Bates et al., 2008;Jones et al., 2012;Lytle & Poff, 2004;Milly, Dunne, & Vecchia, 2005;Poff & Zimmerman, 2010). As temperature increases, trees are becoming increasingly limited by water as the atmospheric demand for water (PET) increases (Babst et al., 2019;Novick et al., 2016). In relation to landscape water supply and demand, we show biomass and stomatal traits differ between field and greenhouse trees, suggesting that plasticity in these correlated traits may also vary on the landscape.
Although, whether there are genetically based differences in phenotypic plasticity requires further study (e.g., Barton et al., 2020) of population tolerance to environmental conditions as well as their capacity to display a range of phenotypes (Nicotra et al., 2010). If plastic, variation in these traits could affect population responses to a changing climate-either buffering against rapid environmental change or assisting in adaptation (Chevin, Lande, & Mace, 2010;Lande, 2009;Nicotra et al., 2010); could modify the strength and direction of plant-atmosphere feedbacks.

| Feedback
Much variation in ecosystem function depends on the metabolicoften adaptive-characteristics of individual organisms, which are governed by laws of mass and energy balance (Brown et al., 2004).
Above, we discussed how large-scale mass-energy relationships of the water cycle drive the evolution of plant populations to control water use (Figure 3a (Eagleson, 1978;Eagleson & Tellers, 1982;Gentine et al., 2012).
Similarly, soil moisture in zones of hybrid Populus (cross between parent species P. angustifolia and P. fremontii) was found to be lower than in adjacent zones dominated by the parent species (Schweitzer, Martinsen, & Whitham, 2002), reinforcing that genetically based differences in transpiration rates (Fischer, Hart, Whitham, Martinsen, & Keim, 2004) and water-use traits (shown here) can be the basis for discovering feedbacks between population genetic variation and long-term variation in ecosystem fluxes of energy and water across large landscapes.

| Implications
Increased drought conditions are predicted to become more widespread and more severe in many geographic locations (Famiglietti, 2014;Georgakakos et al., 2014;Milly et al., 2005).
The western United States is currently experiencing a 1000-year drought threatening the most diverse ecosystems in the desert (riparian ecosystems) with widespread mortality (Gitlin et al., 2006;Kominoski et al., 2013). Occurring at the terrestrial-freshwater interface (Naiman & Décamps, 1997), riparian ecosystems are likely to be affected by changes to many aspects of the water cycle, such as streamflow or the atmospheric demand for water, as well as precipitation (Lytle & Poff, 2004;Milly et al., 2005;Perry, Andersen, Reynolds, Nelson, & Shafroth, 2012;Poff & Zimmerman, 2010;Rood et al., 2003). Although threatened, these systems may be "hotspots" for adaptation to climate change as they historically have been highly exposed to extremes of these various climatic stimuli (Capon et al., 2013). We demonstrated that biomass and stomatal traits, estimates of carbon acquisition, primary productivity, and water-use efficiency (Cornelissen et al., 2003), differ across populations of an foundational riparian tree. These adaptations are important for the plant and the entire ecosystem to deal with drought (Aasamaa et al., 2001;Cornelissen et al., 2003;Hetherington & Woodward, 2003;Sack et al., 2006). In drought conditions, the ability of plants to control water may alter feedbacks to the atmosphere (AET; Figure 5a-c), while the ability of plants to obtain water from different sources may alter stream flow (Q; Figure 5b,d) and the greater stream ecosystem.

| Final conclusions
Integrating ecohydrology and landscape-level genetic variation using the theoretical Budyko Curve allowed us to consider fluxes of energy and matter, interacting climatic gradients, and population genetic structure together to understand linkages between largescale hydrologic processes and evolutionary processes. and provide information about where populations and watersheds may be at risk and where ecosystem processes may be stable.

CO N FLI C T O F I NTE R E S T
The authors declare no conflict of interest.
[Correction added on 11 June 2021, after first online publication: Conflict of Interest statement added to provide full transparency.]

ACK N OWLED G EM ENTS
We

AUTH O R S ' CO NTR I B UTI O N S
SLJB and JKB conceived of the manuscript. SLJB collected data, analyzed data, and wrote manuscript. LOM, IMW, JAS, and JKB all assisted with data collection and provided significant editorial and analytical advice.