Seasonal and Species‐Level Water‐Use Strategies and Groundwater Dependence in Dryland Riparian Woodlands During Extreme Drought

Drought‐induced groundwater decline and warming associated with climate change are primary threats to dryland riparian woodlands. We used the extreme 2012–2019 drought in southern California as a natural experiment to assess how differences in water‐use strategies and groundwater dependence may influence the drought susceptibility of dryland riparian tree species with overlapping distributions. We analyzed tree‐ring stable carbon and oxygen isotopes collected from two cottonwood species (Populus trichocarpa and P. fremontii) along the semi‐arid Santa Clara River. We also modeled tree source water δ18O composition to compare with observed source water δ18O within the floodplain to infer patterns of groundwater reliance. Our results suggest that both species functioned as facultative phreatophytes that used shallow soil moisture when available but ultimately relied on groundwater to maintain physiological function during drought. We also observed apparent species differences in water‐use strategies and groundwater dependence related to their regional distributions. P. fremontii was constrained to more arid river segments and ostensibly used a greater proportion of groundwater to satisfy higher evaporative demand. P. fremontii maintained ∆13C at pre‐drought levels up until the peak of the drought, when trees experienced a precipitous decline in ∆13C. This response pattern suggests that trees prioritized maintaining photosynthetic processes over hydraulic safety, until a critical point. In contrast, P. trichocarpa showed a more gradual and sustained reduction in ∆13C, indicating that drought conditions induced stomatal closure and higher water use efficiency. This strategy may confer drought avoidance for P. trichocarpa while increasing its susceptibility to anticipated climate warming.


Introduction
Riparian woodlands are productive and biodiverse ecosystems in dryland regions that owe their persistence in the landscape to groundwater (Bateman & Merritt, 2020;Sabathier et al., 2021;Singer et al., 2014;Stella et al., 2013).Most dryland riparian trees are assumed to be phreatophytes that rely on consistent root access to groundwater to tolerate seasonally intermittent precipitation and high vapor pressure deficits (Hultine et al., 2020); as such, they are highly susceptible to groundwater decline (Rood et al., 2003;Stromberg et al., 1996).In dryland regions, climate change is expected to increase the frequency and severity of drought events, accompanied by warmer temperatures (Li et al., 2021), contributing to more severe drought-induced groundwater decline (Dragoni & Sukhija, 2008).The effects of these hydroclimatic stressors are exacerbated by human land conversion and groundwater extraction, which collectively threaten the persistence of these ecosystems in their current state and distribution (Krueper, 1993;NRC, 2002;Stella & Bendix, 2019).
The population decline of cottonwoods and poplars (Populus spp.) is of particular concern, as these are foundation species in dryland riparian communities, and interacting stressors have diminished their range and abundance throughout western North America over recent decades (Braatne et al., 2007;Howe & Knopf, 1991;Lite & Stromberg, 2005).Groundwater decline in particular is a well-documented threat to their persistence (Kibler et al., 2021;Rood et al., 2003;Scott et al., 1999;Williams et al., 2022).Cottonwoods have been described as both obligate (Andersen, 2016;Busch et al., 1992;Rood et al., 2013) and facultative phreatophytes (Hultine et al., 2010;Rood et al., 2011;Stromberg et al., 1996), depending on the species and the specific hydroclimatic regimes in which they were established (Hultine et al., 2020).Species that inhabit more arid regions have prodigious water requirements to meet greater atmospheric demand, which intensifies their dependence on groundwater (Hultine et al., 2020;Voltas et al., 2015).Consequently, increasing evaporative demand associated with climate change represents a compound stressor for cottonwoods and other phreatophytes that increases their susceptibility to drought and associated groundwater decline (Williams et al., 2022).
Stomatal closure is the most immediate drought-coping mechanism for plants to regulate transpirational water loss, which helps maintain stem water potential above a critical threshold and prevent xylem embolism that can cumulatively induce hydraulic failure (Farooq et al., 2012;Martin-StPaul et al., 2017;Pirasteh-Anosheh et al., 2016).At the same time, stomatal closure leads to higher internal leaf temperatures and consequently increases the risk of heat stress when drought is accompanied by high temperatures (Blasini et al., 2022).However, groundwater provides a perennial water source that can buffer dryland riparian trees from the effects of atmospheric drought and rising temperatures (Wang et al., 2023).Therefore, consistent access to groundwater may represent the only feasible means of persistence for riparian trees in regions where droughts and warming are becoming more extreme with climate change.
The effects of climate change may differentially affect dryland riparian tree species based on their water-use strategies, drought susceptibility, and the extent of their groundwater dependence.In this context, carbon and oxygen stable isotopes in tree rings can be used to investigate historical patterns of drought response and source water use to infer future responses of dryland riparian woodlands to climate change and groundwater decline.Carbon isotope discrimination (∆ 13 C) derived from tree rings provides a retrospective measure of canopyintegrated leaf gas-exchange (Cernusak et al., 2013;Farquhar et al., 1989;Francey & Farquhar, 1982), making it a useful tool to evaluate past drought responses of plants (Klein et al., 2013).Within the same tree-ring series, stable oxygen isotope ratios (δ 18 O) can be used to infer changes in source water use (Ehleringer & Dawson, 1992;Sargeant et al., 2019) and utilized in coordination with ∆ 13 C to investigate plant ecohydrologic responses (Altieri et al., 2015;Battipaglia & Cherubini, 2022;Gessler et al., 2018;Moreno-Gutiérrez et al., 2012).
The environmental processes that determine ∆ 13 C and δ 18 O composition in tree-rings derive from different fractionation mechanisms.∆ 13 C reflects the ratio of leaf internal CO 2 concentration to atmospheric CO 2 concentration (C i /C a ), and the ∆ 13 C signal in tree rings is therefore determined by the balance between the rate of assimilation (i.e., photosynthesis) and leaf gas-exchange (i.e., stomatal conductance) over the growing season, where an increase in photosynthesis or a decrease in stomatal conductance results in lower values of ∆ 13 C, and vice versa.For vegetation in dryland systems that are not light-limited, stomatal conductance is the dominant influence on C i /C a , and decreases in ∆ 13 C are often attributed to stomatal closure in response to drought stress (McCarroll & Loader, 2004).
For δ 18 O, the leaf water pool signal that is imprinted in tree-ring cellulose reflects to a great extent the isotopic signal of source water, which does not fractionate during root water uptake; therefore, variation in δ 18 O values over time can indicate a relative shift in tree water sources (Dawson & Ehleringer, 1998).Water in the vadose zone (shallower soil layers) is subject to evaporative loss making it isotopically enriched relative to groundwater (Dawson & Ehleringer, 1998;Singer et al., 2013).This pattern allows for differentiation among plant water sources at different depths along the soil profile and can be useful for evaluating tree dependence on groundwater, which is typically depleted in δ 18 O, versus shallow moisture sources (Sargeant & Singer, 2016).In addition, δ 18 O in plant tissues reflects changes in vapor pressure deficit because evaporative demand causes enrichment of leaf water that is also recorded in tree-ring cellulose (Barbour, 2007).Consequently, enriched (higher) values of δ 18 O are often associated with stomatal closure in response to greater evaporative demand (Barbour & Farquhar, 2000;Scheidegger et al., 2000), yet this signal can be overpowered when changes in source water are considerable (Barbour et al., 2004;Sarris et al., 2013;Treydte et al., 2014).Additional influences on tree-ring δ 18 O include diffusion of water vapor from the air into leaves during humid conditions, mixing of leaf water with unenriched stem water (the Péclet effect), and post-photosynthetic processes that may complicate interpretations of leaf-level processes (Barbour, 2007;Gessler et al., 2014;Roden et al., 2000).Recently developed mechanistic models can account for many of these influences and allow for the estimation of source water δ 18 O based on tree-ring δ 18 O, facilitating direct comparisons with endmember δ 18 O composition to better infer plant water sources (Sargeant et al., 2019).
In this study, we used isotope dendrochronologies to build upon Williams et al. (2022)'s investigation of riparian cottonwood (Populus trichocarpa and P. fremontii) responses to the extreme 2012-2019 drought, along the largest free-flowing river in southern California.Previously, we found that ∆ 13 C and annual growth of cottonwoods along the semi-arid Santa Clara River exhibited a coordinated drought response, determined by the severity of multi-year groundwater decline, and that trees subjected to faster rates of groundwater decline showed greater stomatal sensitivity to increasing atmospheric demand.Here, we incorporate semi-annual δ 18 O data from the same trees and time period to evaluate seasonal changes and species differences in water-use strategies during the drought.Our goals were to determine (a) how riparian tree water sources and water-use strategies changed seasonally and interannually in response to drought conditions, and (b) whether water-use strategies and drought susceptibility differed between cottonwood species with overlapping distributions.Whereas the previous study used tree-ring ∆ 13 C and annual growth to evaluate drivers and indicators of drought stress shared among dryland riparian phreatophytes, the present study employs a dual isotope (∆ 13 C and δ 18 O) approach to investigate the extent to which riparian cottonwood species are obligate versus facultative phreatophytes and to infer potential differences in climate change vulnerability for species with differing water-use strategies.

Site and Species Description
The Santa Clara River drains a watershed of 4,204 km 2 in southern California (Figure 1) and produces an average annual water yield greater than 0.124 km 3 (Birosik, 2006;Brownlie & Taylor, 1981).Mean annual precipitation varies from 35 cm where the river meets the Pacific Ocean to <20 cm at the easternmost extent of the catchment (Beller et al., 2016;Downs et al., 2013).Precipitation is generally confined to the rainy season (October -March) and conditions become increasingly arid over the course of the growing season.Groundwater recharge occurs at the highest rates during episodic flooding events in the winter (Andrews et al., 2004;Cayan et al., 1999), and the water table can decline markedly during extended periods of low rainfall (Kibler et al., 2021).
The Santa Clara River is an intermittent river composed of gaining (wetter) and losing (dryer) reaches, which largely determine vegetation composition (Beller et al., 2016).Wetter reaches are composed of dense willowcottonwood forests that host a significant proportion of regional biodiversity (Bennett et al., 2022;Hall et al., 2020).Two cottonwood species are foundational components of riparian forests in the catchment and region, with their distributions related to a gradient of increasing aridity with distance from the coast (Orr et al., 2011; Figure 1).Black cottonwood (P.trichocarpa, Torr.& A. Gray) is abundant where the coastal influence is strong but is largely absent in more arid upstream reaches.Fremont cottonwood (P.fremontii, S. Watson) is increasingly abundant with distance from the coast and becomes the dominant tree species in the warmer, interior portion of the catchment.This pattern matches the distributions of these species throughout North America, with P. trichocarpa found in wetter and more temperate regions and P. fremontii in more arid landscapes (Braatne et al., 1996;Cooke & Rood, 2007).
The meteorological drought in southern California that extended from 2012 to 2019 was the most extreme on record (Robeson, 2015;Warter et al., 2021).It was exacerbated by consecutive years of high temperature anomalies (Figure S1 in Supporting Information S1; Luo et al., 2017;Mann & Gleick, 2015), which combined with the drought caused severe soil moisture and groundwater depletion (Warter et al., 2021).These stressors triggered extensive mortality within riparian woodlands of the Santa Clara River during the peak of the drought (2013-2016) (Kibler et al., 2021), when drought conditions within the study area were classified from severe to exceptional (U.S. Drought Monitor, 2021), followed by a gradual recovery up until 2019.

Data Collection
Increment cores were collected in the summer of 2019 from 114 cottonwood trees (P.trichocarpa and P. fremontii) at seven sites along a 50-km stretch of the lower Santa Clara River (Figure 1; Table 1; Williams et al., 2022).The sampling sites spanned from Ventura County to Los Angeles County (river kilometers 18-71), including the transition zone where the marine influence on atmospheric conditions (i.e., fog) dissipates (Figure 1).

Water Resources Research
10.1029/2023WR035928 WILLIAMS ET AL.

Response Variables (δ 18 O and ∆ 13 C)
Cores were prepared and annual growth rings were measured, crossdated, and detrended using splines fit to each tree's growth series (Williams et al., 2022).For each of the seven sites, a subset of six trees (42 trees total) with the  strongest correlation to site-averaged ring-width chronologies was chosen for carbon and oxygen stable isotope analyses.For each tree, 10 annual rings (2010-2019) were selected for isotopic analysis, spanning pre-drought (2010-2011), peak drought (2013-2016), and drought recovery (2017-2019) periods.As described in Williams et al. ( 2022), annual rings were separated into earlywood and latewood segments and reduced to α-cellulose for a total of 20 samples per tree for carbon and oxygen stable isotope analysis.A description of isotope analyses, precision, and standards is included in Supporting Information S1 (Text S1 and Table S1), and all tree-ring stable carbon and oxygen isotope data used for this study are publicly available (Williams, 2022).
Mass spectrometer measurements of δ 13 C were converted to carbon isotope discrimination (∆ 13 C) to reflect plant-level processes (Farquhar et al., 1989): Smoothed δ 13 C atmosphere data were obtained from ScrippsCO 2 station in La Jolla, CA (Keeling et al., 2001).

Environmental Drivers
Daily climate data, including precipitation (PPT), maximum temperature (T max ), and maximum vapor pressure deficit (VPD max ) were obtained from PRISM Climate Group at 4 km resolution (Oregon State University; Daly et al., 2008); and monthly self-calibrated Palmer Drought Severity Index (sc-PDSI) data were obtained separately for the same pixels (WestWide Drought Tracker; Abatzoglou et al., 2017;Palmer, 1965).PDSI is a commonlyused drought index that has shown strong correlations with measured growing season soil moisture content (Dai, 2011;Dai et al., 2004;Wang et al., 2015), and its self-calibrated variant (sc-PDSI) accounts for differences in climate regime among pixels to facilitate site comparisons (Wells et al., 2004).Depth to groundwater (DTG) was calculated for each site based on groundwater elevation data from nearby wells obtained from the California Department of Water Resources' California Statewide Groundwater Elevation Monitoring online portal (CDWR SGMA Data Viewer, 2020), as described in Williams et al. (2022).

Data Analyses
To evaluate the regional response of cottonwood trees in the lower Santa Clara River to environmental drivers, annual values of tree-ring cellulose δ 18 O from 2010 to 2019 were averaged across trees from all sites (42 trees), with earlywood and latewood fractions analyzed separately.Correlations were calculated between δ 18 O and environmental drivers, which included a mix of atmospheric drivers (T max , and VPD max ) and moisture related drivers (PPT, sc-PDSI and DTG).Correlations between ∆ 13 C and environmental drivers for the same trees and time period can be found in Williams et al. (2022)   correlations between δ 18 O and environmental drivers were evaluated using the non-parametric Spearman's rankorder correlation.
∆ 13 C residuals (∆ 13 C res ) were calculated for each species to account for significant species differences in baseline (pre-drought) ∆ 13 C values that could obscure interpretations of drought responses, as these baseline differences could be due to physiological (e.g., Sparks & Ehleringer, 1997) or anatomical (e.g., Ponton et al., 2001) species differences unrelated to the environmental drivers evaluated in this study.Values of ∆ 13 C res were calculated separately for each species as departures from baseline (pre-drought) conditions, which we define as the mean of the 2010-2011 ∆ 13 C values across all trees within each species.This approach allows for the evaluation of shared species responses to environmental drivers and facilitates the more direct comparison of species differences in drought responses.Specifically, ∆ 13 C res was calculated by subtracting the mean baseline ∆ 13 C values of each species from annual ∆ 13 C values of each tree within their respective species.Therefore, positive ∆ 13 C res values represent increased 13 C discrimination relative to baseline conditions.For trees under high-light conditions in this water-limited system, increased 13 C discrimination suggests greater stomatal conductance, usually associated with negligible water limitation.Conversely, negative ∆ 13 C res values indicate decreased 13 C discrimination, suggesting reduced stomatal conductance, which is commonly induced by drought stress in this biome.
To evaluate the influence of evaporative demand on the correlation between carbon and oxygen isotope ratios, Spearman's rank-order correlations were calculated between annual δ 18 O and ∆ 13 C res values across all 42 trees and plotted as a function of annual, site-averaged VPD max for the 2010-2019 period.Correlations were calculated separately for earlywood and latewood isotope values and compared to average early season (February -June) and late season (March -September) VPD max , respectively.Separate linear trendlines were then fit to earlywood and latewood values.Statistical analyses were carried out using R version 4.2.1 (R Core Team, 2022).
To retrospectively infer water sources used by trees included in this study, we employed Sargeant et al. ( 2019)'s Identification of Source-water Oxygen isotopes in trees Toolkit (ISO-Tool).ISO-Tool applies an inverse of the Barbour et al. (2004) model of biochemical fractionation to estimate source water used by plants based on observed tree-ring δ 18 O values, climate inputs, and plant physiological variables, as described in Supporting Information S1 (Text S2, Table S2).We compared the resulting source water δ 18 O estimations (δ 18 O mod ) with observed groundwater (shallow well) and soil moisture (20-100 cm depth) δ 18 O data collected from three of the study sites (Figure 1) between April 2018 and July 2019 (Figures 5e and 5f; Text S3 and Table S3 in Supporting Information S1).To validate model results, we compared δ 18 O mod from 2018-2019 at these three sites with xylem δ 18 O data collected by Kui and Kibler (2023) from separate Populus spp.trees during the same time period and at the same sites (Text S2 and Figure S2 in Supporting Information S1).

Interannual Trends and Seasonal Patterns in Environmental Drivers and Response Variables
Values of sc-PDSI and VPD max climate data were aggregated across all seven sites to provide a regional context of drought conditions and atmospheric demand during the study period (Figures 2a and 2b).Mean annual sc-PDSI was similar between early and late season (Figure 2a; paired t-test: T 9 = 0.06, p = 0.95), which is expected given the modeled lag in soil moisture incorporated in this drought index.In contrast, VPD max showed significant seasonal differences (paired t-test: T 9 = 14.8, p < 0.001) and was on average 0.4 kPa higher in the late season, which is characteristic of the Mediterranean climate in southern California.During the pre-drought period (2010-2011), average sc-PDSI across sites was 0.36 (mean of early and late season), indicating relatively favorable moisture conditions for the region, and site-averaged VPD max was 1.5 kPa (early season) and 2.0 kPa (late season).Following the onset of the drought in 2012, sc-PDSI declined sharply and remained below 3.7 during the peak of the drought (2013-2016), signifying severe to exceptional drought conditions (Figure 2a).VPD max concurrently increased and reached its most extreme in 2015, at 2.0 kPa (early season) and 2.3 kPa (late season) (Figure 2b).Values of sc-PDSI generally increased during the drought recovery period (2017-2019), coinciding with strong groundwater recharge across sites (Williams et al., 2022).Sc-PDSI experienced a minor decrease in 2018 before rebounding to 0.3 in 2019 (Figure 2a).VPD max steadily declined during the drought recovery period and fell below pre-drought values in 2019, reaching 1.3 kPa (early season) and 1.9 kPa (late season) (Figure 2b).Together, these climate trends depict relatively benign baseline (pre-drought) conditions from 2010-2011, followed by severe water limitation and high atmospheric demand during the peak of the drought (2013-2016), and a gradual return to pre-drought conditions during the drought recovery period (2017-2019).

Water Resources Research
10.1029/2023WR035928 WILLIAMS ET AL. δ 18 O data from both species were combined across trees from all seven sites (42 trees total) to evaluate common responses of cottonwoods to environmental drivers within the lower Santa Clara River and for comparison with the regionally-averaged ∆ 13 C presented for the same trees and time period in Williams et al. (2022) and reproduced in Figure S3 in Supporting Information S1. ∆ 13 C and δ 18 O were highly responsive to drought conditions, with interannual trends of ∆ 13 C most closely following those of PDSI (Williams et al., 2022;Figure S3 in Supporting Information S1), and those of δ 18 O more strongly mirroring VPD max (Figures 2b and 2c).Both earlywood and latewood δ 18 O exhibited strong enrichment from their pre-drought averages (32.7 and 31.5‰,respectively), reaching their highest point during the peak of the drought, in 2014 (33.9 and 32.6‰).Values of δ 18 O showed a marked decline during the drought recovery period, with the most depleted values of the study period observed in 2019 (31.4 and 31.2‰).
While interannual trends in δ 18 O were positively related to changes in VPD max , we observed pronounced seasonal differences in δ 18 O that ran counter to the expected climate response; δ 18 O was consistently lower in latewood of every year, despite VPD max being significantly higher in the late season (Figures 2b and 2c).In fact, the average difference between annual earlywood and latewood δ 18 O ( 1.2‰) was more than double the average annual change between sequential years in either earlywood (0.5‰) or latewood (0.4‰).The fact that seasonal changes in δ 18 O were greater than interannual changes, and in an opposing response to VPD max , indicates that the consistent intra-annual decrease in δ 18 O is unlikely attributable to seasonal shifts in atmospheric conditions (Figures 2b and 2c).

Correlations Between Environmental Drivers and Response Variables
In prior work, we established that the trees included in this study showed strong and significant correlations between carbon isotope discrimination (∆ 13 C) and nearly all environmental drivers, with the strongest (p < 0.01) relationships for DTG (r < 0.9), T max (r < 0.85), and VPD max (r < 0.83) (Williams et al., 2022).For oxygen isotopes, there were also significant relationships with these environmental variables, but the correlations varied seasonally, with earlywood δ 18 O generally more sensitive to hydroclimate drivers than latewood (Table 2).The strongest relationship was between earlywood δ 18 O and VPD max (r = 0.89), with latewood only slightly less correlated (r = 0.82).Similarly, DTG was linked positively and significantly with δ 18 O, and the earlywood correlation (r = 0.82) was stronger than latewood (r = 0.63).The most notable seasonal differences in δ 18 O correlations were observed for temperature and precipitation (Table 2).T max was significantly correlated with earlywood δ 18 O (r = 0.87), but not with latewood.Precipitation showed opposite seasonal relationships with δ 18 O; earlywood was significantly correlated only with current-year (February-June) precipitation (r = 0.61), and latewood significantly correlated (r = 0.72) only with prior winter precipitation (prev.Oct-cur.March).Notably, winter precipitation is largely responsible for the bulk of annual groundwater recharge in this system.Together these relationships indicate that water availability and atmospheric demand had a combined influence on both δ 18 O and Δ 13 C, but that seasonal differences in climate sensitivity were more pronounced for δ 18 O.
Of all atmospheric drivers, VPD max exhibited the strongest combined correlations with both δ 18 O and Δ 13 C (Table 2; Williams et al., 2022: Table 2), suggesting a set of similar physiological mechanisms by which atmospheric dryness influenced both proxies.Although earlywood δ 18 O and Δ 13 C were both significantly correlated with VPD max on an interannual basis, the coordinated physiological response of trees to VPD max (via correlations between carbon and oxygen isotopes across individuals in the same year) was lower in earlywood compared to latewood (Figure 3).In fact, there were no years when δ 18 O and ∆ 13 C res were significantly correlated across all 42 trees for earlywood (Figure 3).In contrast, isotopic coordination was strongly related to VPD max in the late season, when VPD max was 0.4 kPa higher, on average.In years with higher late-season VPD max , latewood δ 18 O and ∆ 13 C res values among all trees showed stronger (negative) correlations, indicating a more pronounced coordinated response of these variables with greater evaporative demand (Figure 3).

Species Differences in Response Variables and Apparent Water-Use Strategies
We observed apparent differences between P. trichocarpa and P. fremontii in their water-use strategies and drought responses, demonstrated by: (a) contrasts in δ 18 O and Δ 13 C values before the drought (Figure 4 higher Δ 13 C for both earlywood (Welch's t-test, t 21 = 6.20, p < 0.001) and latewood (Welch's t-test, t 27 = 7.78, p < 0.001) by an average of 1.9‰ (mean of earlywood and latewood), indicating higher C i /C a for this species under relatively non-stressed conditions (Figures 4a and 4b).During this same time period, earlywood δ 18 O was similar between species (Welch's t-test, t 26 = 0.36, p = 0.72), while latewood δ 18 O of P. fremontii was slightly, but significantly, depleted compared to P. trichocarpa (Welch's t-test, t 26 = 2.31, p = 0.03) by an average of 0.5‰ (Figures 4c and 4d).
As the drought began, P. trichocarpa showed a gradual decrease in ∆ 13 C res that did not exceed 0.5‰ per year (Figure 5a).In contrast, ∆ 13 C res for P. fremontii trees remained at pre-drought levels (2010-2011) through 2013, until experiencing a precipitous decline of 2.1‰ (mean of earlywood and latewood) in 2014 (Figure 5b).The recovery of P. fremontii was similarly more dynamic than that of P. trichocarpa, where mean ∆ 13 C res began to increase (+0.5‰) in 2015 for P. fremontii but continued to decline until 2016 for P. trichocarpa.Together, these results indicate a more delayed but severe regulation of leaf gas-exchange for P. fremontii.
Modeled estimates of source water δ 18 O (δ 18 O mod ) revealed a shared seasonal pattern of source water shifting, as well as species differences in water sources.Latewood δ 18 O mod was consistently depleted relative to earlywood, indicating that both species shifted toward the use of more depleted moisture sources in the late season.In this system, groundwater δ 18 O is more depleted (mean = 7.3‰ ± 0.6‰ SD) than soil moisture ( 4.0‰ ± 3.4‰) (Table S3 in Supporting Information S1), and generally closer to the range of latewood δ 18 O mod , suggesting that trees used a greater proportion of groundwater in late season (Figures 5e and 5f).In addition, the variation among years for latewood δ 18 O mod values was smaller than earlywood in both P. trichocarpa (1.2‰ latewood vs. 2.6‰ earlywood) and P. fremontii (2.8‰ latewood vs. 4.0‰ earlywood), as would be expected with more consistent consumption of deeper water sources in the late season.Comparing the two species, P. fremontii exhibited consistently lower δ 18 O mod than P. trichocarpa, by an average of 1.9‰ (mean of earlywood and latewood), and P. fremontii earlywood δ 18 O mod even overlapped with the observed groundwater signal in some years (Figure 5f).In contrast, δ 18 O mod of P. trichocarpa was consistently more enriched than the groundwater signal in earlywood of every year, and also in latewood of some years (Figure 5e).Together these source water estimates indicate that both species shifted toward the use of more depleted water sources over the growing season, but that P. fremontii used proportionally more depleted water sources.

Discussion
An important finding of this study was the seasonal and species-level variation in groundwater dependence for riparian cottonwoods, which are foundation species in dryland ecosystems (Patten, 1998;Stromberg, 1993).We applied tree-ring stable isotopes in a novel context to evaluate the timing and drivers of tree ecophysiological function, as well as their vulnerability to drought and climate change.By analyzing early and late season oxygen and carbon stable isotopes in tree rings separately and in a dual-isotope approach, we were able to resolve key temporal strategies that varied by species for accessing more depleted (deeper) water sources to mitigate atmospheric drought conditions and soil-moisture limitation.Our results revealed cottonwoods in this system to be facultative phreatophytes that used shallow soil moisture advantageously but relied on deeper moisture sources to maintain physiological function during seasonally dry conditions and long-term drought.By comparing species with overlapping distributions across a pronounced aridity gradient, our results provide implications as to how increasingly severe droughts and warmer temperatures associated with climate change may differentially affect the distribution and survival of each species.

Seasonal Shifts in Groundwater Dependence
We observed a distinct seasonal pattern in δ 18 O that suggests trees shifted water sources over the course of the growing season.As expected, both earlywood and latewood δ 18 O were significantly positively correlated with VPD max in their respective seasons (Table 2) and shared the same interannual trends (Figures 2b and 2c).S3 in Supporting Information S1).
However, we observed seasonal changes in δ 18 O that were surprisingly opposite that of VPD max ; latewood δ 18 O was significantly depleted relative to earlywood in nearly every year despite VPD max being significantly higher in the late season of every year (Figures 2b and 2c).The magnitude of seasonal shifts in δ 18 O were, on average, more than double the interannual variation for either earlywood or latewood alone, suggesting a stronger mechanism driving intra-annual changes in δ 18 O.Studies have shown that changes in source water can exert greater control over the δ 18 O signature of tree-ring cellulose than leaf (or needle) water enrichment and can effectively overshadow the leaf water signature (Barbour et al., 2004;Sarris et al., 2013;Treydte et al., 2014).Analogously here, the δ 18 O data suggest that interannual trends were largely reflective of changes in leaf water enrichment in response to fluctuating atmospheric conditions, but the pronounced and consistent seasonal shift to more depleted δ 18 O likely resulted from trees utilizing water in deeper soil layers late in the growing season when soil moisture in the vadose zone was scarce.Within dryland systems, deep soil moisture is generally supported by groundwater via the capillary fringe that links saturated and unsaturated soil layers (Camporeale et al., 2019), hence the seasonal use of more depleted source waters indicates a reliance on groundwater to maintain physiological function during summer drought.Source water modeling results further support these inferences as earlywood δ 18 O mod was enriched relative to groundwater in most years, and latewood δ 18 O mod was close to, or within, the range of observed groundwater δ 18 O in every year (Figures 5e and 5f).
Our observations are consistent with similar accounts of plants increasing groundwater use seasonally in semiarid and Mediterranean climates, where precipitation is strongly limiting during the growing season (Antunes et al., 2018;Barbeta et al., 2015;Eggemeyer et al., 2008;Mayes et al., 2020;Phelan et al., 2022;Snyder & Williams, 2000;Voltas et al., 2015;Zimmerman et al., 2023).Many species that inhabit these systems exhibit dimorphic root systems (both shallow, lateral roots and deep taproots), allowing them to exploit seasonal differences in water and nutrient availability (Antunes et al., 2018;Dawson & Pate, 1996;Snyder & Williams, 2000).Even with perennial groundwater availability, it may be advantageous for plants to maintain shallow lateral roots to exploit the higher nutrient concentrations that are characteristic of shallow soil layers (Evans & Ehleringer, 1994;Gebauer & Ehleringer, 2000), especially considering that these nutrients may not be labile and thus unavailable later in the growing season when soil moisture is scarce.Conversely, trees may also need to use a considerable portion of groundwater even when soil moisture is available in cases when high evaporative demand increases tree water requirements (Flanagan et al., 2019).Thus, shifting source water over the course of the growing season may increase the survival and productivity of riparian phreatophytes in dryland regions.

Greater Coordination of δ 18 O and Δ 13 C Responses With Increased Atmospheric Demand
Annual comparisons of δ 18 O and ∆ 13 C res correlations across trees from all sites revealed a coordinated response to changes in VPD max (Figure 3).Evaporative demand strongly influences stomatal conductance, which in turn mediates leaf water enrichment and leaf internal CO 2 concentration.Consequently, VPD max (i.e., evaporative demand) should exert strong control over both δ 18 O and Δ 13 C when stomatal conductance is the primary mechanism influencing these variables.Consistent with this reasoning, δ 18 O and ∆ 13 C res showed stronger correlations in years with higher VPD max , and this relationship was more pronounced in the late growing season, when significantly higher VPD max would presumably induce a stronger stomatal response (Figure 3).Interestingly, Guo et al. (2022) showed that stomatal conductance of P. fremontii along the arid San Pedro River in Arizona was largely unresponsive to changes in VPD under favorable water status (i.e., DTG <1 m).We found that many of the years with the strongest correlations between δ 18 O and ∆ 13 C res occurred during the peak of the drought and in its immediate aftermath.This pattern suggests that water limitation increased the stomatal sensitivity of trees to VPD max , as plants may demonstrate differential responses to changes in evaporative demand based on their water status (Williams et al., 2022).
The less coordinated response of earlywood δ 18 O and Δ 13 C res to changes in atmospheric demand (Figure 3) could also be the result of post-photosynthetic processes or seasonal differences in source water.In a recent review, Siegwolf et al. (2023) note the potential for reallocation of stored carbohydrates to contribute to a dampening effect on 13 C that could obscure dual isotope (δ 18 O and Δ 13 C) interpretations in earlywood.In addition, the early season use of shallow soil moisture, which is more isotopically variable, may have overshadowed the influence of leaf water enrichment and resulted in a partial decoupling of tree-ring δ 18 O and stomatal conductance.Sargeant and Singer (2016) showed that the δ 18 O of tree-ring cellulose exhibited greater variability when P. nigra trees used shallow soil moisture compared to groundwater, which had a more consistent isotopic composition.Therefore, the greater use of deeper (and more isotopically stable) moisture sources during the hot, dry summer Water Resources Research 10.1029/2023WR035928 likely resulted in a more consistent δ 18 O signal of source water and allowed for a more coordinated response of ∆ 13 C res and δ 18 O that reflected leaf-level processes.Furthermore, greater drought stress during the late season could similarly reduce the influence of source water relative to leaf-level enrichment, as stomatal closure weakens the Péclet effect (Siegwolf et al., 2023).In this context, our interpretations are consistent with assumptions and predictions under the dual isotope approach of Scheidegger et al. (2000), in which concurrent enrichment of δ 18 O and reduction of Δ 13 C under increasing VPD max signify reductions in stomatal conductance (as opposed to directional changes in net photosynthesis).

Species Differences in Water Sources
Species differences in groundwater reliance are consistent with their distributions along the Santa Clara River, with P. trichocarpa found in cooler coastal reaches and P. fremontii in more arid inland reaches (Orr et al., 2011).Source water modeling suggests that P. fremontii utilized shallow moisture sources early in the growing season, but relied on groundwater as its main water source during dry summer months (Figure 5f).P. fremontii is generally distributed in low-lying floodplains of more arid regions with historically shallow and stable water tables, which promotes high groundwater reliance for this species (Hultine et al., 2020;Rood et al., 2003).However, previous studies from Arizona (USA) indicate that the extent of such groundwater reliance varies with hydroclimatic conditions.For example, along the perennial Bill Williams River, which regularly experiences extreme summer temperatures, Busch et al. (1992) showed that P. fremontii relied almost exclusively on groundwater to facilitate high transpiration, ostensibly as a means of canopy cooling.In contrast, Snyder and Williams (2000) found that P. fremontii along an intermittent segment of the San Pedro River predominately consumed groundwater but advantageously sourced 33% of its xylem water from soil moisture following summer monsoon rains.This latter observation is more similar to our own findings, where greater water table variability and less extreme temperatures compared to the Bill Williams River encouraged P. fremontii to rely on groundwater during seasonal drought, but advantageously use shallower moisture sources when available.P. trichocarpa also shifted toward the use of more depleted source water during summer drought, but δ 18 O mod suggests this species used a lesser proportion of groundwater compared to P. fremontii.P. trichocarpa is commonly found in more temperate regions (Rood et al., 2003), where milder temperatures facilitate lower water requirements.Hence, P. trichocarpa may use a lesser proportion of groundwater due to hydroclimatic adaptations that confine its distribution to cooler, more coastal reaches of the Santa Clara River (Orr et al., 2011).However, both species shifted toward the use of deeper (groundwater-supported) moisture sources to mitigate drought conditions, implying a shared susceptibility to forecasted trends of long-term groundwater decline in dryland regions (Ghazavi & Ebrahimi, 2018;Hanson et al., 2012).Some P. fremontii values of δ 18 O mod were more depleted than the groundwater δ 18 O signal observed in 2018 and 2019 (the only years for which concurrent field data were available; Figure 5f), which could be due to spatial and/ or temporal heterogeneity of δ 18 O in precipitation and groundwater throughout the floodplain.The combination of continental and elevation effects on isotopes in rain water (Dutton et al., 2005) could presumably generate more depleted source water δ 18 O at P. fremontii sites, which are upstream (further inland) and higher in elevation than P. trichocarpa sites.Furthermore, we observed precipitation δ 18 O to be highly variable and more depleted than the groundwater signal by nearly 5‰ during some rain events from 2018-2019 (Figures 5e and 5f), highlighting the possibility of interannual variability in δ 18 O of precipitation-driven groundwater recharge (Figures 5e and 5f).Alternatively, this observation could be the result of greater post-photosynthetic oxygen exchange in P. fremontii than accounted for in source water modeling, which was hypothesized by Stolar (2019) but to our knowledge has not been documented.Nevertheless, the consistency between δ 18 O mod values at the Fillmore Cienega site from 2018-2019 and separate measurements of cottonwood xylem δ 18 O at the same site and time period provide support for source water modeling results (Figure S2 in Supporting Information S1).

Species Differences in Leaf Gas-Exchange and Drought Response
P. fremontii exhibited significantly higher Δ 13 C than P. trichocarpa during both the pre-drought (Figure 4) and drought recovery periods (Welch's t-tests, t 20 > 4.79, p < 0.001), which could be due to physiological or anatomical species differences related to their distributions along the river.For example, unlike P. trichocarpa leaves which contain little to no adaxial stomata (Al Afas et al., 2006;Ferris et al., 2002), the leaves of P. fremontii have stomata on both the adaxial and abaxial surfaces (Blasini et al., 2022).This adaptation reduces constraints on leaf gas-exchange to allow for heat dissipation and may also increase C i /C a (Drake et al., 2019)  increase carbon isotope discrimination (Bradford et al., 1983).Furthermore, research has shown that a multitude of traits related to leaf gas-exchange (stomatal conductance, transpiration, sapwood area, specific leaf area, etc.) are highly variable for P. fremontii and influenced by local climate conditions, with populations in more arid regions displaying traits that facilitate greater leaf gas-exchange (Blasini et al., 2022) that could also influence Δ 13 C.However, leaf-level comparisons of Δ 13 C mechanisms between these species are not available, precluding the determination of whether this contrast is the result of morphophysiological differences, or a consequence of unrelated post-photosynthetic processes (Gessler et al., 2014).
P. fremontii maintained ∆ 13 C at pre-drought levels even as drought conditions worsened, up until 2014 when trees experienced a precipitous reduction in ∆ 13 C (Figure 5b) concurrent with a period of accelerating groundwater decline (Williams et al., 2022).This indicates that trees were resistant to moderating leaf gas-exchange rates during the onset of drought stress, which illustrates a water-use strategy whereby trees prioritized maintaining photosynthetic processes over hydraulic safety.Alternatively, P. fremontii may have exhibited a delayed drought response due to other factors, such as stronger xylem structure or more extensive rooting architecture than P. trichocarpa, that allowed this species to tolerate dry conditions for longer before succumbing to drought stress.Previous studies have demonstrated that P. fremontii exhibits minimal stomatal regulation, operates under a narrow hydraulic safety margin, and is highly vulnerable to xylem cavitation (Leffler et al., 2000;Pockman & Sperry, 2000).Our findings demonstrate that sampled (surviving) P. fremontii trees in this study did strongly regulate stomatal conductance to prevent water loss at a certain threshold of drought stress (Figure 5b).Yet, the timing of the precipitous reduction in ∆ 13 C for P. fremontii trees also corresponded with sharp increases in mortality for other trees in these stands (including P. fremontii) during the same period due to groundwater decline (Kibler et al., 2021), highlighting the risks associated with such a delayed drought response.Interestingly, mean ∆ 13 C of P. fremontii sampled for this study began to recover in 2015, coincident with the peak die-off (Kibler et al., 2021), which could be due to surviving trees experiencing a combination of leaf shedding and branch die-back that increased their root:shoot ratio and allowed them to regain favorable water status, as has been observed for P. fremontii in other systems (Rood et al., 2000); this was supported by field evidence of crown dieback for sampled trees (Text S4 and Figure S4 in Supporting Information S1).
P. trichocarpa showed an earlier and more gradual stomatal response to drought stress compared to P. fremontii (Figures 5a and 5b).Furthermore, P. trichocarpa trees maintained low ∆ 13 C values throughout the peak of the drought (Figure 5a), indicating water limitation induced persistently higher water-use efficiency (Gornall & Guy, 2007).This conservative strategy may be necessary for P. trichocarpa in dryland regions, considering this species is also highly vulnerable to xylem cavitation (Fichot et al., 2015).In line with this reasoning, Bassman and Zwier (1991) found stomatal regulation of P. trichocarpa clones in Washington that were subjected to experimental drought to be highly variable and influenced by their climate of origin.In that study, dry-adapted clones exhibited high water use efficiency and stomatal sensitivity in response to declining stem water potential, whereas moist-adapted clones maintained high stomatal conductance despite increasing soil water deficits.Schulte et al. (1987) similarly observed stomatal regulation of P. trichocarpa to be conditioned by previous exposure to water limitation.Therefore, in the semi-arid climate of the Santa Clara River, P. trichocarpa may have adapted greater stomatal sensitivity to limit transpirational water loss as a drought avoidance strategy.P. trichocarpa also exhibited stronger negative ∆ 13 C correlation with T max and VPD max compared to P. fremontii (Table S4 in Supporting Information S1), indicating greater stomatal sensitivity to these environmental drivers.However, stomatal closure can lead to higher internal leaf temperatures (Blasini et al., 2022;Kibler et al., 2023) and consequently increase susceptibility to heat stress (Lipiec et al., 2013;Teskey et al., 2015).Therefore, the tendency of P. trichocarpa to close stomata early in response to water limitation may confer greater drought avoidance (Martin-StPaul et al., 2017) while ultimately precluding its establishment in warmer (inland) segments of the river, where it would be even more susceptible to temperature increases anticipated with climate change.

Conclusions
Our results demonstrate that riparian cottonwoods in this system were generally facultative phreatophytes that used shallow soil moisture advantageously but shifted toward deeper water sources (i.e., groundwater) when soil moisture was limiting and when increased water consumption was required to meet greater atmospheric demand.We observed clear species differences in water-use strategies and drought responses related to their distribution along the Santa Clara River.P. trichocarpa was confined to milder coastal reaches, and P. fremontii inhabited more arid river segments, where higher atmospheric demand encouraged greater groundwater reliance.In

Water Resources Research
10.1029/2023WR035928 WILLIAMS ET AL.
response to drought, P. trichocarpa moderated leaf gas-exchange earlier to increase water-use efficiency, whereas P. fremontii exhibited a more delayed stomatal response and prioritized maintaining physiological function, until a critical point.The greater water-use efficiency of P. trichocarpa likely allowed trees to limit water loss, while increasing the risk of carbon limitation and heat stress.In contrast, the more delayed stomatal response of P. fremontii likely helped tolerate greater evaporative demand, while increasing its groundwater reliance.The contrasting water-use strategies and drought responses of these species imply that hydroclimatic stressors associated with climate change may differentially affect their persistence and regional range limits.Our findings suggest that the greater groundwater reliance and more arid distribution of P. fremontii may increase its susceptibility to severe drought-induced groundwater decline, whereas P. trichocarpa may become progressively limited in range due to climate warming.

Figure 1 .
Figure 1.Site locations within the lower Santa Clara River, California.Sites containing Populus trichocarpa labeled in black: 1. Hanson, 2. South Mountain Road, 3. Hedrick Ranch Lower, 4. Hedrick Ranch Upper, and 5. Taylor.Sites containing P. fremontii labeled in red: 6. Fillmore Cienega and 7. Newhall Ranch.Groundwater observation wells associated with P. trichocarpa sites shown in black triangles and those associated with P. fremontii sites shown in red triangles.Locations of groundwater (GW) and soil moisture δ 18 O sampling from 2018 to 2019 labeled with pink asterisks.Pixel coloring denotes average number of days with fog cover in June from 2010 to 2019, retrieved from Google Earth Engine (Gorelick et al., 2017).Sources: terrain map from Esri World Terrain Bae (ArcGIS Pro 3.1), floodplain boundaries from Stillwater Sciences (2019), watershed boundary from the National Hydrography Dataset (NHDH_CA), via The National Map.

Figure 2 .
Figure2.(a) Time series of self-calibrated Palmer Drought Severity Index (sc-PDSI), averaged across all seven sites within the lower Santa Clara River, California.Climate data were aggregated over the early season (February-June) to coincide with tree earlywood production (green), and late growing season (March-September) to coincide with tree latewood production (orange).(b) Time series of maximum vapor pressure deficit (VPD max ) during the early growing season (green) and late growing season (orange), averaged across all sites.(c) Regional isotope dendrochronologies of δ 18 O in earlywood (green) and latewood (orange) averaged across trees of both species from all seven sites (42 trees total).Bars display ±1SE.Gray shading denotes the peak drought period (2013-2016).
); (b) the magnitude of their change from baseline values during the drought; and (c) the immediacy of their drought response and recovery (Figures5a and 5b).In pre-drought years (2010-2011), P. fremontii had significantly

Figure 3 .
Figure 3. Correlations between annual values of δ 18 O and ∆ 13 C residuals (∆ 13 C res ) as a function of maximum vapor pressure deficit (VPD max ).Pairs of ∆ 13 C res and δ 18 O values for all trees at all sites within the lower Santa Clara River, California (42 trees) were compared to calculate annual values of correlation, with each point representing a year from 2010 to 2019.Correlations were calculated separately for earlywood (green) and latewood (orange).Values of VPD max were aggregated across all sites and averaged over the months of February-June for the earlywood comparison and March-September for the latewood comparison.Separate trendlines are displayed for earlywood and latewood with standard error shaded.Horizontal dashed line depicts the threshold below which all points show a significant correlation between ∆ 13 C res and δ 18 O ( p < 0.05).

Figure 5 .
Figure 5. Species-level responses for Populus trichocarpa trees (n = 30) at five sites (panels a, c, e) and P. fremontii trees (n = 12) at two sites (panels b, d, f) along the lower Santa Clara River, California.Panels (a) and (b): isotope dendrochronologies of earlywood Δ 13 C residuals (∆ 13 C res ; green) and latewood ∆ 13 C res (orange) averaged across all trees within each species.Bars display ±1 SE.Panels (c) and (d): isotope dendrochronologies of earlywood δ 18 O (green) and latewood δ 18 O (orange)cellulose averaged across all trees within each species.Panels (e) and (f): modeled source water δ 18 O (δ 18 O mod ) for earlywood (green) and latewood (orange) calculated as a single mean value for each site and averaged across all sites containing each species.Points in panels (e) and (f) denote soil moisture (20-100 cm depth), rain, and groundwater δ 18 O observations.Blue dashed lines in panels (e) and (f) denote the mean of groundwater δ 18 O observations ±1 SD, and dotted lines denote ±2 SD (TableS3in Supporting Information S1).

Table 1
Site Information for Populus trichocarpa and P. fremontii Trees Sampled Along the Lower Santa Clara River, California a Mean maximum vapor pressure deficit (VPD max ) was calculated using daily summer (March -September) values from 2010-2019, retrieved from PRISM Climate Group (Oregon State University).b Mean depth to groundwater (DTG) was calculated by averaging annual median-values of all observations from 2011-2019.c Full time series of DTG for each site can be found in Williams et al. (2022) Figure 4. Water Resources Research 10.1029/2023WR035928 WILLIAMS ET AL.

Table 2
Correlations Between δ 18 O and Environmental DriversNote.δ 18 O data were averaged across all trees of both species at all seven sites (42 trees) along the lower Santa Clara River, California.Environmental drivers and response variables were averaged across sites to calculate regional means.Significance codes: p < 0.1 (*), p < 0.05 (**), p < 0.01 (***).a Atmospheric drivers: T max = maximum temperature, VPD max = maximum vapor pressure deficit.b Moisture-related drivers: sc-PDSI = self-calibrated Palmer Drought Severity Index; DTG = Depth to Groundwater; PPT = precipitation.c Annual values for environmental drivers were averaged (summed for precipitation) over the time period noted in the "Months" column.d Water-year winter precipitation values were also summed from October of the previous year to March of the current year.
and consequently WILLIAMS ET AL.