A Global Analysis of Groundwater Recharge for Vegetation, Climate, and Soils Open Access

Groundwater sustains the lives of one quarter of the human population (Ford and Williams, 1989; White et al., 1995) and is vital for industrial, agricultural, and recreational activities and for the health of other species and ecosystems (Postel and Carpenter, 1997; Jackson et al., 2001). Its importance is most apparent in arid and semiarid regions, where a paucity of surface waters often leads to greater groundwater exploitation. Given the increasing use and scarcity of groundwater in many locations, and its relatively slow replenishment, sustainable groundwater use and management are necessary to meet the needs of people and ecosystems (Shiklomanov, 1997; Shah et al., 2000; Vörösmarty et al., 2000).

The relationships between groundwater recharge and physical variables have long been of scientific and practical interest, traceable back to ancient Roman times (Dr. Nitish Priyadarshi, earthday.ning.com/profiles/blogs/1734264:BlogPost:24384 [verified 23 Oct. 2011]). Previous studies have identified climatic and geologic factors as major environmental controls on the rate of groundwater recharge. In general, recharge increases with the amount and intensity of rainfall, which influence how much water enters the soil and rocks (Lvovitch, 1970; Freeze and Cherry, 1979; Bredenkamp, 1988; Edmunds, 2001a; Jan et al., 2007; Stonestrom et al., 2007). In contrast, recharge typically decreases with increasing PET, an expression of the amount of energy available to evaporate water (Thornthwaite et al., 1957). Once in the soil, the movement of water is influenced by soil texture and structure, with sandier soils tending to have greater rates of recharge and more clayey soils having increased tortuosity and more limited water movement (Athavale et al., 1980; Kennett-Smith et al., 1994). Such general relationships are already important for use in some global models (e.g., Döll et al., 2003).

One aspect of recharge that is less well understood and rarely incorporated into global land-surface models is the effect of vegetation on recharge (Jackson et al., 2000; Gerten et al., 2004). Examples of key uncertainties include the primary effect of vegetation type compared with the physical climate and soil variables, as well as how changes in vegetation interact with climate and soils to alter recharge. Although considerable research has examined physical factors as controls of recharge, earlier work has rarely emphasized the effects of vegetation (but see Lull and Munns [1950] and also Petheram et al. [2002] for a review of Australian studies). Several studies have included vegetation in attempts to model groundwater recharge at various scales (i.e., Finch, 1998; Keese et al., 2005; Döll and Fiedler, 2008), although most have found or assumed the relationship to be of secondary importance compared with the effects of physical factors such as climate and soil on recharge.

Plants often mediate water fluxes between the soil and the atmosphere through the uptake of soil water by roots and through evapotranspiration (ET) from leaves, with plant traits such as rooting depth, leaf area, and phenology affecting the magnitude and duration of these fluxes (Skiles and Hanson, 1994; Neilson, 1995; Milly, 1997; Kergoat, 1998; Peel et al., 2001; Jackson et al., 2008). Pervasive land-use and land-cover changes from anthropogenic and natural forces could have large consequences for groundwater recharge and potentially for downstream effects such as salinization (Walker et al., 1999). Building on earlier studies of land use and recharge, including studies in Australia (Petheram et al., 2002) and in arid and semiarid regions (Scanlon et al., 2006), we examined the relative importance of vegetation in the relationship between recharge and physical factors.

We compiled a new global synthesis of groundwater recharge rates and data for different climates, soils, and vegetation types to understand how different vegetation types affect recharge. We hypothesized that vegetation would exert as strong an influence on recharge as climate and soils do. Moreover, we expected strong interactions among plant, climate, and soil factors that would create predictable patterns of recharge under different vegetation types. Among vegetation types, we emphasized croplands, grasslands, and forests because shifts among these common land covers represent most of the ongoing land-use changes today (Meyer and Turner, 1994; Klein Goldewijk and Battjes, 1997). To test the synthesis data and to compare recharge under paired vegetation types, we also collected new field data from paired land uses across precipitation gradients in central Argentina and the southwestern United States. We applied our findings to examine how land-use and land-cover changes may affect recharge across climatic and soil factors.

Because θ, θ_s, and K_s are not always reported in published studies, we estimated K_s in our regression model of recharge based on the soil texture information that the studies provided more frequently (see below). Furthermore, because ET depends on the available soil moisture, we used the energy available for evaporation or PET as a proxy for ET. Although our approach was statistical, we chose the predictors for the regression model based on this conceptual framework. The seasonal amplitude of rainfall and synchrony of rainfall with PET are both additional important considerations in the water balance because these factors affect the downward soil water flux bypassing root uptake (Milly, 1994; Potter et al., 2005). The predictors we chose for our model were precipitation, PET, K_s, and the seasonality of rainfall in addition to vegetation type.

We examined studies of recharge and physical variables associated with land use or vegetation type, identified using literature searches involving the keywords “groundwater recharge,” “deep drainage,” or “residual flux,” henceforth collectively referred to as recharge (Petheram et al., 2002; Scanlon et al., 2006). From tables, digitized graphs, and text, we recorded recharge estimates, precipitation during the study period or the reported long-term mean (P), PET, soil texture (clay and sand contents or textural classes), K_s, vegetation type, species present, and the amount of irrigation (I), when present. In studies where recharge estimates included data from multiple years and locations, such as those using permanent boreholes for the same vegetation and soil type, we used the mean of the estimates. Across the data set, 46% of the data points came from Oceania, 19% from North America, 15% from Asia, 10% from Africa, 6% from Europe, and 3% from South America (see Fig. 1).

Because we wanted to compare the effects of biological and physical variables on recharge, we excluded data from sites with significant sources or sinks for runoff, such as sinkholes, playas, and streams. Studies that estimated recharge for <1 yr were also excluded from our analysis. Due to the large number (>2500) of studies in the search, we sorted the results by relevance in the Web of Knowledge (Thomson Reuters, Philadelphia, PA) and included studies until fewer than two out of 10 additional studies yielded data on the following variables: recharge, vegetation type, precipitation, irrigation, and soil texture or K_s.

For the vegetation analyses, we divided plant types into five broad categories: cropland, grassland, woodland, scrubland, and no vegetation. Annual agricultural fields were classified as croplands; grasslands and pastures as grasslands; forests and woodlands as woodlands; scrublands, heathlands, shrublands, steppes, fynbos, and savannas as scrublands; and areas with sparse or no vegetation as “NoVeg.”

Most studies did not provide data for PET or the seasonality of rainfall, and these variables were therefore obtained from the high spatial resolution (10′ by 10′) Climate Research Unit global data set (New et al., 2002; csi.cgiar.org/cru/index.asp [verified 24 Oct. 2011]), using locations of sites given in the studies. We calculated PET using the Penman–Monteith equation from the monthly climate data set. We defined two variables associated with the seasonality of rainfall (Milly, 1994; Potter et al., 2005): (i) the difference between the maximum and minimum mean monthly rainfall (amplitude); and (ii) the number of months between the maximum mean monthly temperature and precipitation (phase). Water input, the aridity index (AI), and potential water excess (PWE) were calculated as P + I, (P + I)/PET, and P + I − PET, respectively, to identify the climatic index with the strongest associations with recharge.

We estimated K_s using soil texture classes (Rawls et al., 1982). Where different soil horizons existed within the depth of soil examined, the estimated K_s for the top layer was used. To ensure that our estimates of PET and K_s were reasonable, we compared them with values of PET and K_s from the subset of studies where they were reported. Our estimates matched well with reported the PET and K_s values across the studies [n = 220, 71; R² = 0.71, 0.70; P < 0.0001, 0.0001, for PET and log(K_s), respectively].

Proportional recharge (P/WI) between each pair of vegetation and soil types was compared using a Kruskal–Wallis test. Proportional recharge was used for this analysis instead of recharge because it allowed comparisons after controlling for the effect of WI. A nonparametric test was used because the data were not normally distributed. Grouping the data into two soil texture categories was done for some analyses to more easily examine the effects of soil texture on recharge: “clays” were defined as soils whose estimated K_s was <0.25 m d⁻¹ (silt loam and more clayey soils) and “sands” were texture classes with higher values of K_s.

We tested all climate variables (WI, AI, PET, PWE, and seasonality) and models (linear, logarithmic, exponential, and sigmoidal) to determine the best predictor of recharge. This approach was taken to choose a single best predictor variable to easily represent and compare the synthesis data with the new field data (see below). Due to the relatively low sample size (n < 50) and limited ranges in climatic variables (e.g., WI = 159–937 mm yr⁻¹) for scrublands and NoVeg, those two vegetation types were excluded from the curve-fitting and regression analyses (see below). All of the models tested were susceptible to the influence of relatively few data points at the most humid end of our data range (n = 5 for the perhumid region data); we therefore limited our curve-fitting and regression analyses to a data set without these extremely humid regions.

We tested for effects of WI, PET, vegetation type, K_s, seasonality of rainfall, and accompanying interactions on recharge using multiple regression analyses. Because of heteroscedasticity, we logarithmically transformed recharge and examined appropriate models to relate recharge to each of the predictor variables. The Breusch–Pagan test was used to test for homoscedasticity, and logarithmic transformation of recharge gave the most homoscedastic relationships with the predictor variables out of all the transformations of recharge values (untransformed, natural logarithm, and square root). We examined appropriate models (linear, exponential, and logarithmic) to relate recharge to the predictor variables individually and found that a logarithmic model explained the most statistical variation in the logarithmically transformed recharge using WI and PET and that a linear model maximized the fit of the logarithmically transformed recharge with K_s and the seasonality of rainfall (amplitude and phase). Thus, we logarithmically transformed WI and PET to linearize them with respect to the logarithmically transformed recharge for the multiple regression. We used WI and PET instead of PWE or AI for our multiple regressions to tease out the relative importance of WI and PET. Stepwise regression with whole effects was used to determine which main and interaction terms to retain in the model and to determine the relative importance of each term for recharge.

Finally, to test the reliability and predictive capability of our regression model, we used threefold cross-validation, in which a model based on a subset of the data is tested against the remainder of the data (Kohavi, 1995).

In addition to the literature synthesis, we collected an extensive new field data set as an independent test of our global data set, using paired comparisons of adjacent vegetation types in Argentina and the United States. In Argentina, we located six sites in the Pampas on relatively level landscapes across a precipitation gradient that ranged from 382 to 1215 mm yr⁻¹. Where available, rainfed cropland and woody plant invasion (WPI) plots were paired with an adjacent or nearby (<1 km) natural grassland plot at each site. Cultivation and WPI plots correspond to cropland and woodland vegetation designations, respectively, in our literature synthesis (Tables 1 and 2).

We also selected five sites along a precipitation gradient (407–860 mm yr⁻¹) in the southern Great Plains of the United States. Land uses selected as paired plots were natural grasslands, rainfed croplands, and irrigated croplands. In both the U.S. and Argentina, most plots had >30 yr of relatively continuous land-use history (Table 1). Landowners or farm managers were surveyed for land-use history at each site, including cropping schemes (species and rotations) and fertilizer, pesticide, and irrigation inputs. Tree stand ages were verified with aerial photos or tree ring cores taken during our sampling campaign (2008–2010). Precipitation data were obtained from long-term (>30-yr) records maintained by weather stations onsite by the farm managers or from separate stations 1 to 30 km away (Instituto Nacional de Tecnología Agropecuaria, www.inta.gov.ar/index.asp [verified 23 Oct. 2011]; National Climatic Data Center, www.ncdc.noaa.gov/oa/ncdc.html [verified 23 Oct. 2011]).

In addition to our new field data, we also estimated additional recharge rates based on soil Cl⁻ data from five paired grassland and woody-encroached sites located across a precipitation gradient in the southwestern United States. Detailed descriptions for these sites are available in Jackson et al. (2002) and McCulley et al. (2004). We collectively refer to these and the southern Great Plains sites as our southwestern U.S. sites.

At the Argentinean sites, soil samples were taken by augering three to eight boreholes 6 to 9 m deep or to the depth of groundwater, as well as four to six shallow cores (30 cm deep), at each land-use plot. Augered samples were taken every 20 cm to the 1-m depth, then every 30 cm to the 4-m depth, then every 50 cm. The soil samples were homogenized and subsampled in the field and then frozen until analysis.

At our U.S. sites, we used a direct-push mechanical coring rig (Geoprobe Systems, Salina, KS) for five to eight cores per plot to a 8.5-m depth. At only one plot near San Angelo, TX, were soil samples not retrieved to 8.5 m because of indurated caliche found around the 5-m depth that blocked further coring. The soil cores were weighed in the field, subsampled for soil moisture and bulk density using intact cores and for elemental analysis using homogenized soil cores, then shipped to Duke University for analysis.

In the laboratory, the soil samples were oven dried for gravimetric moisture content and analyzed for chemical constituents. Dried and homogenized soil samples were mixed with double deionized water at a 1:1 (w/w) ratio and shaken for 4 h. The mixture was centrifuged, the supernatant filtered, and the filtrate analyzed for anion contents (Cl⁻, Br⁻, NO₃⁻, SO₄²⁻, and PO₄³⁻) by ion chromatography (Dionex ICS-2000, Sunnyvale, CA). The Cl⁻ concentrations in the soil pore water were calculated by dividing the soil Cl⁻ contents (mg Cl kg⁻¹ soil) by the gravimetric soil moisture. Soil texture was determined by the pipette method (Klute, 1986) and ranged from sand to clay (Table 1).

For the global synthesis data, we used recharge predicted from the best-fit curves to calculate the absolute and relative differences in recharge among vegetation types.

Vegetation and soil types had strong effects on the proportional recharge (R/WI) globally (χ² = 73.7 and 13.9, respectively; P < 0.0002). On average, proportional recharge was 0.18, 0.11, 0.08, 0.06, and 0.05 under NoVeg, croplands, grasslands, woodlands, and scrublands, respectively (P < 0.0005 for all pairwise comparisons except scrublands to woodlands and grasslands to croplands; Table 2). Sandy soils had 50% more proportional recharge as clayey soils, on average.

Potential water excess fitted to an exponential model was the best single predictor of recharge across the data set (Fig. 2). Recharge increased with PWE for croplands, grasslands, and woodlands in both sands and clays (average R² = 0.52, P < 0.0001 for all vegetation–soil types; Fig. 2). Differences among vegetation types were evident in the fitted curves. For example, at a PWE of −250 mm yr⁻¹ in clays, the predicted recharge under croplands, grasslands, and woodlands were 112, 61, and 35 mm yr⁻¹, respectively (n = 220, 138, and 109, respectively).

Water inputs in the multiple regression explained 29% of the statistical variation in recharge across the data set (P < 0.0001; Table 3). Other significant variables in the order of decreasing importance were vegetation type (16%, P < 0.0001), PET (12%, P < 0.0001), and K_s (6%, P < 0.0001), with amplitude and phase (seasonality) of rainfall contributing a statistically significant but minor 1% of the variation (P < 0.0001; Table 3). Recharge increased with increasing WI, K_s, and seasonality of rainfall and decreased with increasing PET. Overall, recharge was greatest under croplands, about two and 15 times greater than under grasslands and woodlands, respectively (P < 0.0001; Table 3, Veg term).

The interaction terms of vegetation type with climate or soil variables collectively explained an additional 8% of the variation in recharge (Table 3; Fig. 3). Of all the vegetation types, cropland recharge increased the most with WI, but grassland recharge increased the most with increasing K_s and decreasing PET. In contrast, woodland recharge was the least sensitive to K_s and PET, indicating that recharge under different vegetation types responded differently to climate and soil factors. These responses accentuated the differences in recharge among vegetation types in humid regions and in sandy soils (Fig. 3a, 3b, and 3c). The cross-validation analysis of the regression model and the data set produced comparable results, giving confidence in the model's reliability (Supplementary Fig. 3).

Our new field data set from central Argentina and the southwestern United States independently confirmed the strong differences in recharge for croplands, grasslands, and woodlands that the global synthesis revealed. Croplands had significantly lower average soil pore water Cl⁻ concentrations below the root zone, while woodland plots had significantly higher soil pore water Cl⁻ compared with their grassland pairs (Table 4; Supplementary Fig. 4; signed Wilcoxon test; P < 0.0020 and 0.0039 for grassland–cropland and grassland–woodland comparisons, respectively). This result indicated that the greatest recharge occurred under croplands, intermediate recharge occurred under grasslands, and the lowest recharge occurred under woodlands. This strong biological control over soil water fluxes is in close agreement with our global review (Fig. 4; Table 4). Crop cultivation using groundwater as an irrigation source resulted in a very high net discharge of groundwater (Table 4).

Our field data set also confirmed the interactive effect of vegetation and climate on recharge that the global synthesis revealed. For our global synthesis, absolute differences in recharge among vegetation types using PWE as the best-fit predictor variable were small in arid climates and grew with increasing PWE and were larger in sandy soils than in clays between grassland and woodland (Fig. 3 and 4). Relative differences were largest in arid climates and in clays (Supplementary Fig. 5), however, suggesting that proportionately greater hydrologic effects of land use change may occur in more arid regions and in clayey soils. Similarly to the global synthesis, our new field estimates of recharge gains or losses due to land-use conversions of natural grasslands increased in magnitude with PWE (Fig. 4), revealing interactions between land use and the abiotic environment in determining recharge. As in the global synthesis, our field-based estimates of relative changes in recharge showed an increasing importance of vegetation effects toward lower precipitation and higher clay content areas, suggesting that while land-use changes have the potential to change recharge by large amounts in humid regions and coarse-textured soils, vadose zone processes may be particularly sensitive to land-use changes in relatively arid areas and fine-textured soils (Fig. 4; Supplementary Fig. 5).

Although the role of vegetation in global terrestrial water fluxes is well recognized (Hutjes et al., 1998; Kucharik et al., 2000; Arora, 2002; Jackson et al., 2005), this synthesis is, to our knowledge, the first attempt to quantify the relative importance of vegetation on recharge rates globally. Vegetation was the second most powerful predictor of recharge after WI, explaining about 1.3 and three times as much variation in recharge as PET and K_s, respectively, indicating that vegetation type is often more important for determining recharge than most physical variables (Table 3). As a result, vegetation should be one of the key components of analyses or models addressing scales sufficiently large to include multiple vegetation types.

The treatment of vegetation parameters in global land-surface models are sometimes cursory and are rarely process based with regard to recharge (Gerten et al., 2004). Our global synthesis should help parameterize such models and could contribute as inputs or could be used for independent testing of global water balance or climate models. For example, studies modeling the reciprocal effects of groundwater on climate (e.g., Niu et al., 2007) may benefit from better constrained estimates of recharge under different vegetation types.

Changes in recharge with land-use changes in our field data followed the patterns of recharge observed under different vegetation and soil types across our global synthesis (Fig. 4; Supplementary Fig. 5). Overall, agreement between the field and synthesis results suggests that vegetation is responsible for a large portion of the variation in recharge and that distinct patterns of recharge among vegetation types are typically clear and reproducible when covarying site factors such as soil properties are controlled for. Agricultural conversion of grasslands or woodlands would therefore probably bring about greater recharge, whereas woody plant invasion or afforestation into grasslands or croplands would probably reduce recharge. These hydrologic changes may be especially severe for land-use changes to and from woodlands because the capacity of woody plants to limit recharge leads to large differences in recharge between woodland and the other vegetation types (Fig. 2; Table 3). Loss of renewable water yield to planted or invading woody plants could be detrimental to groundwater-dependent communities, both human and natural, across long time scales. In contrast, cultivation generally increases recharge but may pose a risk of salinization or degradation of groundwater quality in some regions through associated leaching of salts through the vadose zone (Smettem, 1998; Boumans et al., 2005; Jobbágy and Jackson, 2007; Scanlon et al., 2007a). Such disruptions to the hydrologic cycle should be recognized in land management and policy decisions.

The effect of vegetation on recharge was further evident along the entire climate gradient and across soil types (Fig. 2). In our synthesis, we observed large absolute differences in recharge among vegetation types in mesic regions (high WI, low PET) and in sands (high K_s) but larger relative differences in arid climates and in clays (Fig. 4; Supplementary Fig. 5). Relative differences between grasslands and the other vegetation types in clay soils, for example, were as much as −70 and 250% (woodlands and croplands, respectively) in arid climates compared with only −20 and 60% in humid areas (Supplementary Fig. 5). Although the large absolute differences in recharge among vegetation types in humid climates highlight the importance of land-use changes on water yields in these climates, large relative differences in drier climates forecast proportionately important hydrologic changes in arid regions, as observed previously for stream flow (Farley et al., 2005). Mirroring the synthesis data, the observed 70% reduction in grassland recharge with woody plant invasion and >500% gain in recharge with cultivation of arid grasslands with clayey soils in our field data indicate that near-complete loss of groundwater recharge or flushing of accumulated vadose zone solutes may be possible with land-use changes (Fig. 4; Supplementary Fig. 5). The different responses of recharge among vegetation types to climate and soils warrant careful consideration of these interactions to avoid adverse hydrologic consequences of land-use changes.

Vegetation type explained a similar amount of variation in recharge as important physical variables did, and its interactions with physical variables contributed additional explanatory power. Recharge was correlated with high K_s (Table 3), but we observed this effect primarily in grasslands, which have relatively shallow root systems (2.5 m; Canadell et al., 1996). The analogous increases in woodland recharge were less pronounced. Woody plants grow deeper roots in areas with sandy soils (high K_s), in part to capture water throughout the soil profile (Schenk and Jackson, 2002, 2005); these deep woody roots may limit recharge despite higher K_s. In croplands, with the shallowest expected rooting depth (generally <2 m), recharge was generally higher than for other vegetation types but did not vary substantially with K_s. This result may be due to particular management practices in croplands, such as tillage increasing deep drainage and weakening the overall positive effect of K_s on recharge under croplands (Daniel 1999; Scanlon et al., 2008). Interactions between vegetation and physical variables such as K_s and PET collectively explained >8% of the statistical variation in recharge and helped identify potential mechanisms responsible for differences in recharge among vegetation types.

Irrigation is often used to enhance crop productivity and to meet increasing food demands given decreasing available productive land area (Kendall and Pimentel, 1994), but it also causes a large net discharge of groundwater, as we observed at our southwestern U.S. sites. Assuming that rainfed croplands represent the upper limits for recharge and irrigation uses groundwater, we consider the difficult issues of irrigation and sustainable groundwater use from a land management perspective. We observed from our global synthesis that, despite being the land use with the highest recharge, rainfed cultivation allowed only marginal recharge compared with the net discharge (irrigation − recharge) of irrigated cultivation (Fig. 5). Across a gradient of water availability, the area of rainfed cultivation needed to sustainably supply groundwater for 1 ha of irrigated agriculture decreases from 70 ha in arid climates to 0.5 ha in humid climates (Fig. 5), providing first-order approximation of the irrigated/rainfed cropland ratio necessary for sustainable groundwater management. It points to challenges associated with providing enough groundwater for irrigated crops, especially in more arid regions where the lack of rainwater results in both larger irrigation needs and lower recharge rates under rainfed cultivation.

For the range of PWE and the recharge values analyzed, the exponential model gave the best overall fit but should not be extrapolated beyond the ranges presented in this study. For instance, with inclusion of the very limited data from perhumid regions, the sigmoidal model gave the best overall fit (data not shown), indicating that the increase in recharge with PWE may taper off in very humid regions due to the increasing importance of runoff on the water balance (Milly, 1997). Moreover, irrigation reported at most of the sites were often estimates without long-term monitoring, introducing uncertainty in our observed relationship between recharge and WI. The average uncertainty associated with irrigation from studies that reported ranges of irrigation was about 190 mm yr⁻¹. Although the effect of this uncertainty on the estimated parameters of our multiple regression were not statistically significant (data not shown), the large explanatory power of WI in our model highlights the importance of obtaining the best possible irrigation and precipitation data for recharge predictions.

In conclusion, vegetation and its interactions with other factors have a strong effect on groundwater recharge, explaining ∼24% of the global variation in recharge—more than other variables except WI. An average of 11% of WI becomes recharge under croplands, whereas only 8 and 6% do under grasslands and woodlands, respectively. Vegetation types had predictable effects on groundwater recharge, and the differences in recharge among vegetation types also varied predictably across the climate and soil variables. Independent field estimates of recharge under paired land-use plots confirmed our global synthesis results and provided a direct test of the relationships between vegetation and recharge. Significant gains and losses in recharge are possible with conversion to crops and to forests, respectively, and absolute changes in water yield accompanying land-use changes are likely to be larger in humid or sandy areas. Proportionately large relative hydrologic consequences result from land-use changes in arid or clayey regions, however, as observed previously for stream flow (Farley et al., 2005). Quantifying and predicting changes to water yield from land-use changes are necessary steps for sustainable and holistic management of water resources; our results highlight the importance of land-use change for the vadose zone and groundwater resources.

This work was supported by the National Science Foundation (DEB no. 0717191, IOS no. 0920355, and GRFP no. 2006044266). We wish to thank members of the Jackson lab for helpful comments on the manuscript, Nancy Scott and Chinling Chen for their assistance with the database, and Ricardo Andres, Matt Cleary, Laura Beth Konopinski, and others for their outstanding help in the lab and the field. We also thank many landowners and personnel at the following research centers who provided access to the sites and logistical support: Grupo Estudios Ambientales in San Luis, Argentina; Institute for Agricultural Plant Physiology and Ecology at Universidad de Buenos Aires (UBA); Estancia San Claudio maintained by UBA; Western Kansas Agricultural Research Center in Tribune; Oklahoma Panhandle Research and Extension Center in Goodwell; Texas AgriLife research and extension centers at Vernon and San Angelo; and USDA-ARS Grassland Soil and Water Research Laboratory in Temple, TX.