Interannual controls on riparian plant health in a dryland river

Riparian zones in drylands provide important refugia for plants but depend on groundwater and thus are subject to local temporal and spatial variability in abiotic controls. In lieu of costly field‐based sampling, we used readily available data to establish site–scale interannual relationships among riparian plant health and the abiotic factors that control their water balance for a historically persistent wetland adjoining the Santa Clara River in southern California, USA. Non‐linear generalized additive model (GAM) analysis of plant health, represented using the normalized difference vegetation index (NDVI), confirmed robust relationships among plant health and various geomorphological and hydrological factors over multi‐decadal timeframes, including years since last high‐flow event, intra‐year groundwater elevation changes and magnitude of 2‐year cumulative surface water inflows. Geomorphic controls are related to years with high flows that cause extensive scour and deposition that re‐set riparian plant communities. Relationships with dry‐season groundwater declines reflect direct plant access to sub‐surface moisture. Hydrological dependence via cumulative inflow magnitude indicates the dependency of groundwater elevations on sufficient winter recharge to prevent precipitous groundwater decline. GAMs‐based inflection point analysis of surface water inflows versus groundwater elevations confirmed that the cumulative magnitude of multi‐year inflows is critical in avoiding catastrophic groundwater declines and that large flood events drive groundwater recovery. We show that abiotic controls on plant health can be derived from readily available data and that non‐linear analysis better represents the complexity of these scalar controls. Our analysis has relevance for ecosystem management of human‐altered rivers and climate change adaptation.


| INTRODUCTION
Riparian zones in arid and semi-arid systems (drylands) provide important refugia for plant and animal species in otherwise fragmented, water-limited systems.The vegetation inhabiting these areas is highly dependent on shallow groundwater for survival (Pettit & Froend, 2018;Stella et al., 2013).In Mediterranean systems especially, groundwater is vital to riparian vegetation survival as minimal summer precipitation leads to seasonally dry conditions associated with intermittent streamflow (Zaimes, 2020).While riparian zones are more resistant to atmospheric changes in water availability (i.e., drought) than upland areas (Mayes et al., 2020), recent research indicates that increasing drought severity caused by anthropogenic factors and associated water table declines in groundwater dependent ecosystems (GDEs) results in extensive vegetation stress and mortality (Chiloane et al., 2022;Kibler et al., 2021).Precipitation, streamflow and evaporative demand also exert strong control on vegetation health and survival in dryland riparian areas and other GDEs (Torres-García et al., 2021;Williams et al., 2022;Williams et al., 2023).The relative influence of such abiotic forcing on vegetation health varies based on localized conditions, necessitating an analysis of these relationships at a fine spatial resolution (i.e., site-scale) to provide a holistic understanding of physical controls on plant health and guide management actions that generally occur at the site scale.
Seasonal atmospheric conditions and species functional traits determine spatially and temporally variable water source use by riparian vegetation (Pettit & Froend, 2018;Sargeant & Singer, 2016;Singer et al., 2013).Though ultimately dependent on taproot access to groundwater-based moisture, many riparian tree species have horizontally distributed roots near the land surface, which allow for the opportunistic use of shallow soil moisture when available (Dawson & Pate, 1996;Hultine et al., 2020).For example, isotope analyses revealed that along the intermittent San Pedro River in southeastern Arizona, USA, Fremont cottonwood (Populus fremontii) acquired up to 33% of its water from shallow soil moisture following summer precipitation events (Snyder & Williams, 2000).Along the perennial Bill Williams River (Arizona, USA), the same species showed an isotopic signature consistent with constant groundwater consumption (Busch et al., 1992), possibly to facilitate continual evaporative cooling for canopy thermal regulation (Hultine et al., 2020).In Mediterranean-climate regions, groundwater dependence of vegetation is higher during seasonally hot and dry summer months and in drought years, indicating temporal variation in the relationship between physical drivers and water use strategies is also common (Dawson & Pate, 1996;Rohde et al., 2021;Sarris et al., 2013).Further, reported rooting depths for phreatophytes are highly variable among and within species (ranging from <2 to >8 m, Hultine et al., 2020;Kibler et al., 2021;Stromberg, 2013), suggesting that riparian vegetation health and survival are strongly controlled by local and temporally variable physical factors.
Water limitation is becoming increasingly common in the southwestern United States due to increased temperatures and reduced frequency of precipitation events associated with climate change (Adams et al., 2009;Allen et al., 2015).Such climatic changes are expected to result in more frequent and intense water table declines, especially in southwest dryland systems where growing human populations place increasing demand on water resources and where dominant riparian tree species are highly sensitive to small fluctuations in groundwater elevation (Cooper et al., 2003;Hultine et al., 2010;Scott et al., 1999).Hotter temperatures will also increase plant transpiration, simultaneously increasing plant dependence on groundwater as water tables decline (Hultine et al., 2020;Seager et al., 2013;Williams et al., 2013).Riparian trees respond to short-term water limitations via stomatal closure, branch dieback, water transport failure, premature leaf-shedding and reduced leaf surface area (Amlin & Rood, 2003;Rood et al., 2000;Rood et al., 2003), resulting in a shared vegetation response of decreased 'greenness'.For riparian phreatophytes, root elongation can keep pace with gradual rates of groundwater decline (i.e., ≈0.5 m year À1 ; Scott et al., 1999), but declines >0.5 m year À1 are likely to result in severe drought stress (Williams et al., 2022).Although riparian foundational species such as cottonwoods and willows can exhibit strong recovery from seasonal water limitations (Amlin & Rood, 2003;Rood et al., 2000), when these effects are sustained over interannual periods, they can result in large-scale vegetation mortality (Kibler et al., 2021).Groundwater extraction for urban needs and intensive agriculture can further exacerbate groundwater declines in GDEs, increasing the risk to riparian vegetation (Rohde et al., 2021).
The Santa Clara River is the largest relatively unregulated river in southern California (Downs et al., 2013).It serves as a regional water source for agricultural and urban centres (Beller et al., 2016) and has a riparian zone of critical importance for regional biodiversity and several endangered bird species (Bennett et al., 2022;Parker et al., 2014).
The region experienced severe drought from 2012 to 2019, causing significant mortality of willow-cottonwood forests across the river channel (Kibler et al., 2021), challenging the resiliency and long-term vegetation health of the riparian areas.This study aimed to establish interannual relationships among riparian plant health and a suite of physical controlling factors as potentially explanatory variables at an extensive GDE in a mid-watershed reach.Traditional methods of characterizing interannual plant health using field-based sampling are cost intensive and temporally and spatially limited.Readily available satellite imagery-derived data for relative absorption and reflection were used instead.Such imagery can be applied to calculate metrics of plant health (Kumar et al., 2022), water availability (Nagler et al., 2021) and other useful indices of the photosynthetic status of plants in response to biotic and abiotic stressors (Rohde et al., 2021;Zhang et al., 2019).
Critically, here, remote sensing indices allow retrospective assessment of plant health over multi-decadal timeframes, allowing investigation of plant health over multiple wet and dry periods.The latter is critical given the increasing likelihood of sustained drought periods in the southwestern United States (Albano et al., 2022;Apurv & Cai, 2021).Following successful use in other studies (e.g., Chen et al., 2019), we represent plant health across multiple vegetation patches using publicly available normalized difference vegetation index (NDVI) values from 1985 to 2018 (Klausmeyer et al., 2019).Though more recently derived remote sensing metrics may perform marginally better than NDVI in certain applications (e.g., Kumar et al., 2022;Nagler et al., 2020;Tian et al., 2019), we focus here on NDVI as it remains a widely used, readily available and straightforward tool for land managers and policy holders.Physical control factors included hydrology (temporal measures of depth to groundwater and surface water inflows), meteorology (precipitation, evapotranspiration and drought severity indices) and geomorphology (time since the last major flood event resulting in fluvial re-working of the site).Physical control factors were drawn from multi-decadal data sets to match the period of the remote sensing record.We sought to identify regional relationships pertaining to a ca.200-ha site, hypothesizing that controls on the near-river riparian areas would be more closely related to hydrological and geomorphological factors associated with river flows, whereas more distant locations might be more strongly associated with meteorological attributes such as precipitation.Conscious of the fundamental importance of shallow subsurface water availability for GDEs, we also evaluated the responsiveness of groundwater elevations in this area of the Santa Clara valley to hydroclimatic variability using groundwater elevations and surface water inflows.This knowledge is used to explore the contemporary resiliency of the site and the likely implications of climate changes projected for the region.The outcomes have implications for informing riparian restoration strategies and sustainable groundwater management.

| Study area
The Santa Clara Valley in southern California rises in the mountainous areas of the Angeles and Los Padres National Forests and runs through multiple geological basins on its way to the Pacific Ocean, resulting in perennially flowing stretches interspersed with intermittent reaches where water flows through the sub-surface (Beller et al., 2016).The river experiences highly episodic precipitation events causing, on average, more than half the annual river flow volume to be discharged over a period of just 3-6 days (Downs et al., 2013).
High flows occur during large precipitation events in El Niño Southern Oscillation (ENSO) years, resulting in significant sediment transport and deposition (Warrick, 2002;Williams, 1979) and extensive channel re-working by fluvial scour that creates a broad floodplain of multiple sandy channels.
The ca. 200-ha study area is located in the mid-watershed (Figure 1).Part of the 'Sespe Cienega' of Beller et al. (2016), it is a historically persistent wetland whose existence is facilitated by a geological constriction of the Santa Clara valley that locally elevates groundwater levels.The predominantly right bank site, currently undergoing riparian restoration, extends from a highway and active fish hatchery downslope towards the Santa Clara River.It includes a series of former watercress beds and abandoned agricultural fields, stands of existing riparian vegetation, the active river bed and a narrow area of agricultural land on the left bank of the current channel.
The vigour and success of native riparian tree (phreatophyte) species (P.fremontii, P. trichocarpa, Salix laevigata, S. lasiolepis, Platanus racemosa) is determined largely as a function of water availability to the vegetation rooting zone (Williams et al., 2022).Consequently, the riparian vegetation on the site experienced significant dieback during the historic California drought from 2012 to 2019 (Kibler et al., 2021) due to precipitous declines in groundwater levels.

| Physical controls on plant health
Site-level controls on plant health should result from a water balance function of hydrological, geomorphological and meteorological F I G U R E 1 Location map indicating the spatial arrangement of data sources used to associate plant health (NDVI polygons) to physical controls.See Table 1 for corresponding naming designations for each polygon.Further details are provided in the text.controls.Specifically, we expected plant health (dependent variable) to fluctuate as some function of multiple independent variables related to surface water inflow, depth to groundwater, time since geomorphic disturbance, precipitation, evapotranspiration and drought.
Recognizing from field measurements, laboratory experiments and physiological modelling (Horton et al., 2001;Leffler et al., 2000;Rood et al., 2011;Stella et al., 2010) that plant health may have a highly non-linear and threshold-driven relationship to water availability (cf., Kibler et al., 2021), we used a generalized additive model (GAM) to evaluate potential non-linear relationships among plant health and the various abiotic environmental controls.GAMs are a semi-parametric (Wood, 2011) extension of generalized linear models wherein multiple smooth polynomial functions (splines) are fitted locally across the range of the explanatory data, allowing the (additive) fitting of nonlinear relationships.The modelling algorithm evaluates the estimated degrees of freedom (edf) needed for each parameter, which can be interpreted as the 'wiggliness' (non-linearity) of each modelling parameter's relationship to the response variable.Our analyses were performed in R (version 4.0.3)using the gam() and gamm() functions of R's mgcv package (Wood, 2004(Wood, , 2011)).The strength of equations resulting from these methods is measured using the coefficient of determination, adjusted for sample size and the number of explanatory variables (i.e., adjusted R 2 ).

| Dependent variable-plant health: NDVI
NDVI is based on a property of living vegetation whereby radiation in the red band of the visible spectrum is absorbed while nearinfrared (NIR) is reflected (Rouse et al., 1974).This allows NDVI to serve as a proxy for vegetation 'greenness'-its photosynthetic status (Xue & Su, 2017).NDVI is a frequently applied remote sensing metric for plant health due to its ease of interpretation, sensitivity to green vegetation and ability to maintain accuracy in areas of low vegetation cover (Xue & Su, 2017).Reliable ground-truthed relationships exist between NDVI and canopy structure, leaf area index and canopy photosynthesis (Xue & Su, 2017), but NDVI is not sensitive to changes in species composition.As such, local knowledge and ground-based observations are important complementary tools when evaluating NDVI data (G omez-Sapiens et al., 2020).Unsurprisingly in GDEs, NDVI is also robustly related to groundwater availability (Aguilar et al., 2012;G omez-Sapiens et al., 2020;G omez-Sapiens et al., 2021;Gong & Shi, 2004;Huntington et al., 2016;Jiang et al., 2006).
We used median dry season estimates (9 July-7 September) for NDVI computed for GDE polygons as part of The Nature Conservancy's 'GDE Pulse' project (Klausmeyer et al., 2019).The project used site surveys and Landsat satellite imagery from 1984 to 2018 to delineate polygons of similar vegetation composition, with dry season NDVI values calculated as the mean of all pixel values within each polygon.July-September values are used because vegetation in Mediterranean and dryland systems is expected to be increasingly dependent on groundwater during these months as the absence of precipitation leads to decreased shallow soil moisture content (Huntington et al., 2016;Sarris et al., 2013).

| Associative controls on plant moisture availability
We hypothesized that site-level controls on plant health result from a water balance function of hydrological, geomorphological and meteorological factors, as follows.

| Surface water inflow
Characteristic of intermittent rivers, water availability here is assumed to be closely linked to percolation of surface waters close upstream of the site, with the volumetric extent of percolation a function of the magnitude of surface water inflows received, especially close to the river.This fact is accentuated here because groundwater recharge is driven by 'losses' of surface flow into the upstream groundwater basin.
Inflow derives primarily from the upper Santa Clara River (USCR).

| Groundwater elevations
Groundwater elevation data were developed from four groundwater wells near the study site with long records (data records obtained from the Fillmore and Piru Basins Groundwater Sustainability Agency: https://fillmore-piru.gladata.com/#).One pair of wells ('Hatchery N' and 'Hatchery S') is situated in close proximity adjacent to the site's fish hatchery and is presumed indicative of groundwater conditions in the 'upper' part of the site and to be responsive to groundwater pumping at the hatchery (Figure 1).The second well pair ('Hopper' and 'Calvin') is located upstream close to the Santa Clara River and is presumed to closely reflect riparian groundwater conditions approaching the site (Figure 1).
Periods between well readings ranged from a few days to several years requiring, in some cases, lengthy straight-line interpolation between readings.To partially compensate, and facilitated because readings were often taken on different days, data were combined for each well pair to produce a synthetic record of greater resolution.
For each well pairing, three different groundwater measures were tested.They included the average July-September surface depthto-groundwater (mimicking the period for the NDVI data), the July-September net change in depth-to-groundwater and the average depth-to-groundwater for the 12 months from the preceding September.Thus, the first measure tests the association of plant health against the absolute dry season groundwater elevation, the second against changing dry season groundwater elevations and the third against a longer preceding period of absolute groundwater elevation.

| Time since the last major flood event
During high magnitude flood events on the Santa Clara River, large areas of the active channel bed are re-worked by erosion and deposition processes including, historically, a significant proportion of the study area.Ecological conditions are thus re-set by a combination of direct scour of existing vegetation and by creating post-flood bare surfaces ripe for colonization by propagules of native and non-native vegetation.As these changes will alter vegetation reflectance, NDVI values may thus show a recognizable progression in time since the last 'major' flood event.We developed variables indicating using two flow magnitudes to define 'major'.Based on instantaneous peak flow magnitudes recorded at the dominant USCR gauge, threshold magnitudes were chosen at 283 and 425 m 3 s À1 (10,000 and 15,000 cfs in native measurements).Following the second highest flood of record in 1983 (866 m 3 s À1 ), the lower threshold has been exceeded eight times while the upper threshold has been exceeded only twice.Thus, the measures record fundamentally different responses during the wet-phase ENSO period of the late 1980s and 1990s.Notably, both measures indicate the paucity of high flow magnitudes since the late 1990s.edu/) at 4-km spatial resolution (Daly et al., 1994).Monthly data were aggregated into averages for the 3-month dry season (July-September) and the preceding 12 months.Following exploratory analysis, four weakly correlated variables were retained.They included the 3-month Standardized Precipitation Index (SPI-3) based on a normal distribution of departures from a long-term average (McKee et al., 1993), the 3-and 12-month Standardized Precipitation Evapotranspiration Index (SPEI-3, SPEI-12) that combines monthly precipitation and average monthly temperature to produce a simple water balance (precipitation-potential evapotranspiration) (Vicente-Serrano et al., 2010) and a 3-month self-calibrated Palmer Drought Severity Index (sc-PDSI-3).The latter uses a locally calibrated version of the PDSI (Palmer, 1965;sc-PDSI, Wells et al., 2004), determining accumulated water excess or deficit based on precipitation, evapotranspiration and assumed loss to percolation (the water holding capacity of the top 250 cm of the soil acquired from the State Soil Geographic Data Base STATSGO).

| Meteorological variables
Based on the foregoing dependent and control variables (Table 1), numerous multi-variate GAMs were developed to evaluate non-linear associative relationships between plant health and abiotic environmental factors.

| Inflow controls on depths of shallow groundwater
As shallow subsurface water availability at the study site is fundamentally dependent on recharge resulting from percolation from surface water inflows (Beller et al., 2016), we tested groundwater responsiveness to hydroclimatic variability by comparing groundwater elevation data to the gauged daily mean surface water inflows.As groundwater elevations near the site are demonstrably subject to precipitous declines during drought (Kibler et al., 2021), our tests focused on establishing the threshold of inflow reduction likely to trigger rapid reductions in groundwater elevations.Groundwater well and flow gauging records allow a near 50-year period of analysis, which includes three significant drought periods.Knowledge of such thresholds provides potentially important foresight into impending periods of drastic groundwater decline that can trigger mass vegetation die-off.
GAM models were fitted with multiple splines across variable time periods to identify a statistically expressible inflection point between groundwater elevation and cumulative surface water inflows.Multiple polynomial GAMs were constructed for each of the four sample wells and the two synthetically averaged well records for increasingly lengthy cumulative flow periods (i.e., combined daily mean flow and 90-, 360-and 720-day cumulative daily mean flows).
Thresholds were determined as the inflection point in the first derivative of the groundwater depth (Simpson, 2018), using 200 evenly spaced points and calculating each derivative to a 99% confidence interval.
The SCR Southeast and Riparian polygons had mildly positive, but non-significant, trends in green vegetation over time.
There were strong NDVI associations (Pearson's correlation p < 0.01) among the eastern (88591,96439,111026,111028) and the south-western polygons (88611,96414,96409), and weaker but still significant relationships ( p < 0.05) between the 'upslope' northwestern (88611 and 111030) and central polygons on either side of the active channel (96409 and 111026).Such associations suggest that plant health is indeed responding relatively consistently to spatial variability in the abiotic stressors that influence water availability.As plant health is measured here by 'greenness', differences will also arise in response to intrinsic turnover in plant species over time, not least related to invasion and periodic eradication of Arundo donax (arundo; giant reed) (Table 2).While such intrinsic variability partially obscures relationships between physical drivers and plant health, we assume that soil moisture availability is a sufficiently strong influence on NDVI over multi-decadal time scales to outweigh confounding influences.

| Multivariate water balance controls on plant health
Based on the spatial associations, GAMs analysis was run on five representative polygons presuming non-linear relationships between plant health and abiotic controlling factors.To confirm, stepwise linear regression was also performed using transformed variables that meet assumptions for linear multiple regression.In every case, the best fit GAMs result was better than that obtained using the linear regression.
Table 3 indicates significantly associated variables, explained variance and variance improvement achieved using GAMs over the best fit linear regression equations for each polygon.Best fit improvements range from almost negligible (CSER West) to considerable (e.g., RipEast) illustrating varying degrees of non-linearity in the plant health-abiotic control relationship.For four of the five polygons, explained variance ranged from 57% to 88%.The GAMs approach also allowed definition of a best-fit relationship for the CSER Hatchery polygon, albeit with limited explanation (31%), whereas no statistically significant equations were returned using stepwise linear regression.This presumably reflects relatively consistent NDVI values over the study period (Figure 2).
Plant health is strongly associated with scour and deposition resulting from large flow events, groundwater levels and long-term hydrology.All NDVI polygons are statistically associated with the lower threshold of time since high flow disturbance variable (i.e., Time10k), four of the five polygons with a groundwater indicator T A B L E 1 Summary of data used in assessing physical control on plant health, Sespe Cienega site.

| Inflow controls on shallow groundwater depth
Groundwater elevations at the study site are highly dependent on recharge from upstream inflow.During wetter periods, depth to groundwater is frequently <1 m for the hatchery wells and about 1.In general, groundwater elevations at the Hopper and Calvin wells showed greater fluctuation than the hatchery wells (Figure 3) and recovered more rapidly following large floods, as might be expected given their close proximity to the river.
To identify the point at which reduced cumulative surface inflows lead to precipitous reductions in groundwater levels, we performed an inflection point analysis using GAMs for each well and for the synthetic paired well levels (Figure 4).All models achieved a significant fit (p < 0.001) with our four flow periods.However, variance estimates improve with longer cumulative flow periods with a maximum generally at the 720-day period (Table 4) except for the Calvin (slightly stronger relationships for the 360-day period), and this influences the Hopper-Calvin synthetic data.The best-fitting inflection points at the Hatchery indicate that a 720-day cumulative daily mean flow totalling ca.1400-1900 m 3 s À1 is required to prevent a precipitous drop in groundwater elevations.This suggests an equivalent daily mean flow in the range of 2-2.6 m 3 s À1 is required continuously over a 2-year period to maintain stable groundwater elevations.Summer baseflow is generally far below this threshold, indicating the critical importance of winter precipitation-driven recharge events in providing sufficient inflow to maintain groundwater levels.

| Physical controls on plant health
Our aim was to evaluate the influence of physical drivers on plant health (using NDVI) over decadal time scales at the site scale to    3).Such an outcome for a GDE in southern California is rational as California rivers have the highest interannual variability of streamflow in the USA (Dettinger, 2011).The study period included two periods of drought (1989-1991 and 2012-2018) and multiple flood events from the 1990s to 2005.The 2012-2019 drought was perhaps the most regionally significant in the last 800 years (Williams et al., 2020), while the flood sequence during our evaluation period was probably the most intense in the last 200 years (Downs et al., 2013).
Meteorological variables were not or were only mildly significant in models, suggesting marginal influence on intra-site scale plant health over multiple decades (see also Stella et al., 2013).Previous studies have similarly observed a stronger relationship between riparian vegetation vigour and local hydrologic variables compared to atmospheric drivers in Mediterranean climate riparian ecosystems (Rood et al., 2013;Schook et al., 2020;Valor et al., 2020;Williams et al., 2022), though meteorological variables such as temperature and precipitation play an important role in determining species distributions over broader spatial scales within dryland riparian systems of the western United States (Butterfield & Munson, 2016;McShane et al., 2015;Palmquist et al., 2018;Williams et al., in review).In addition to hydroclimatic factors, geomorphological influences such as the magnitude and timing of flood events strongly determine plant nutrient availability, species composition and successional processes within dryland rivers (Friedman et al., 2022;Merritt et al., 2010;Stella et al., 2013).For example, flood events of lesser intensity that do not While NDVI is a useful tool to evaluate trends in plant health, its detection of changes in species composition is not straightforward (He et al., 2009;McPartland et al., 2019;Pau et al., 2012), and its use as a proxy for post-disturbance ecosystem recovery should therefore be interpreted in coordination with ecological knowledge and sitebased observations.For example, Buma (2012) observed strong postfire NDVI recovery in a subalpine forest but discovered using ground surveys that this trend mostly reflected an increase in forb cover rather than forest recovery.Along similar lines, the 2005 flood event within the SCR that severely scoured and re-worked the left bank of  the active channel within our study area caused the SCR Southwest polygon to transition from mixed cottonwood-willow forest to riparian scrub, with a sparser canopy cover and thus inherently lower NDVI values (see Figure 2).This change in vegetation composition is visible both on the ground and in aerial images but would not be apparent based solely on NDVI data.Thus for this polygon, the only one with highly explained statistically significant NDVI losses over time trends in NDVI stem from geomorphological transformation and vegetation conversion.These observations highlight the need to recognize and mitigate potential limitations of remote sensing technology with complementary forms of investigation.
The dependence of GDE vegetation on groundwater elevation is reflected here at two temporal scales.First, there is a strong relationship with longer term (generally 2 years) cumulative inflows that fundamentally drive groundwater recharge (see below).Second, the frequent associations with summer-season changes in groundwater elevations likely reflect root access to soil moisture.Such access is often argued to be the greatest limiting factor for plant growth in semi-arid Mediterranean environments (Newman et al., 2006;Torres-García et al., 2021), but as the multi-decadal analysis demonstrates here, there is additionally an over-riding need for winter storms to drive groundwater recharge in areas such as southern California that are subject to high interannual flow variability.This phenomenon has also been documented along other intermittent and dryland rivers around the globe (Scanlon et al., 2006), including systems within Africa (Cuthbert et al., 2019;Lange, 2005), Brazil (Costa et al., 2013), China (Qin et al., 2012) and Australia (McCallum et al., 2014).However, within human-modified systems, groundwater recharge can become uncoupled from natural processes (Rohde et al., 2021).Such uncoupling is likely responsible for abiotic factors providing low explained variance of NDVI within the polygon nearest the hatchery (CSER Hatchery, Table 3).Due to its location, plant health in this polygon can more likely be attributed to overflow drainage from hatchery operations and agricultural return flows than abiotic conditions.Such anthropogenically augmented groundwater recharge can be an important water source for vegetation in heavily altered systems, such as the Colorado River Delta, where agricultural return flows are thought to be the main contribution to aquifer recharge that supports groundwater dependent vegetation (Cohen et al., 2001;Nagler et al., 2008).
Geographically, the analysis depicts a broad pattern of varying plant health dependencies trending from north-east to south-west across the study site.Regional factors that govern groundwater recharge from the  Second, the greater success of GAMs analysis than stepwise linear regression provides concurrence with research suggesting that plant health has a highly non-linear and threshold-driven relationship to water availability (e.g., Horton et al., 2001;Leffler et al., 2000;Rood et al., 2011;Stella et al., 2010).We interpret this result to reflect the intrinsic ability of curvilinear analyses to better integrate the combination of longer-and shorter-term abiotic control functions than a linear analysis.Curvilinear analysis generally added 20-30% further explained variance and, for one polygon (CSER Hatchery), allowed the formulation of a statistically significant relationship that was not forthcoming using linear analysis.

| Inflow thresholds and plant health resiliency to projected climate change
Regionally, the groundwater elevations critical for plant health are sustained primarily by the percolation of surface flows from upstream.
This is similar to many GDEs but felt very acutely in southern California where depths to groundwater are subject to progressive decline in the years following a large flood event and eventual precipitous decline during periods of drought.Large flood events act to recover and, if frequent, maintain groundwater levels close to the ground surface.Inflection point analysis indicated that flood events producing an equivalent daily mean inflow of 2-2.6 m 3 s À1 over a 2-year period should resist precipitous groundwater elevation declines.However, large floods here also act to scour and reset vegetation succession so there is a complex interannual relationship of riparian vegetation health to ENSO events (Andrews et al., 2004;Downs et al., 2013).
Our analysis of nearly 50 years of gauged discharges and groundwater well sampling includes one of the wettest decades of instrumented record, the most severe drought for centuries, increased groundwater extraction for agriculture and a progressive increase in summer season baseflows consequent on upstream urban expansion.
The period follows shortly after significant regulation of Piru Creek in 1955, a major upstream tributary, and more recently, a schedule of late summer flow releases to support downstream agriculture.While the wetland complex forming the study site has been historically persistent (Beller et al., 2016), under prevailing surface water inflow conditions (Figure 3), precipitous groundwater declines at the site are clearly possible but so is rapid recovery following groundwater recharge events (Figure 2 and Williams et al., 2022).Figure 5 illustrates the dramatic impact of flood events in bolstering 720-day cumulative inflows relative to the ≈2 m 3 s À1 threshold daily mean equivalent inflow required to prevent precipitous decline in ground water elevations (from Table 4).Large instantaneous peaks flows (i.e., >280 m 3 s À1 at the USCR gauge) generally produce a daily mean flow of around 85 m 3 s À1 (see Figure 5), which is sufficient to River, competing water uses for agricultural and municipal purposes is intense, making it difficult to balance human needs and conservation goals (Nagler et al., 2021;Nagler et al., 2022;Stromberg et al., 1996).
Regarding GDE resiliency in the face of climate change, downscaling the predictions of 32 Global Climate Models (Oakley et al., 2019) suggested that, overall, regional precipitation totals are not forecast to change much in coming decades, but there will more dry days, implying that precipitation will be more intense when it occurs and increasing inland temperatures.As such, plants will become increasingly dependent on groundwater reserves between precipitation periods.This impact will be accentuated by increasing precipitation volatility in 21st century California, with an increasing frequency of shifts from periods of multi-year dryness to extreme wetness (Swain et al., 2018).Multiyear drought followed by extreme wetness would result in longer periods of declining cumulative surface water inflows (indicated by DMF720) and thus an increasingly likelihood of inflows volumes falling below the threshold required to maintain 'close to surface' groundwater elevations at the site (Figure 5).Further, a shift to winter precipitation with less precipitation in the spring and fall would likely accentuate the statistical significance of summer season declines in groundwater elevation (e.g., HatchNSchanges), which, according to our analysis, would impose additional stress on riparian plant communities.
This general scenario will be exacerbated or mediated by changes in the frequency of large flood events that sustain groundwater levels near the site.An increasing frequency of high intensity ENSO-driven storm events or intense precipitation derived from 'atmospheric rivers' (ARs, see Gimeno et al., 2014, Huang et al., 2021) could bolster cumulative inflow totals and thus increase the resilience of plant health in intervening dry periods, unless long-lasting droughts also become more frequent.However, this ignores the ecologically complex relationship of plant health with geomorphological factors.An increasing frequency of large events will reduce the period since the last major flood (i.e., Time10k) and thus increase the proportional period in which plant health is dependent on site 'recovery' from whatever scour and deposition occurred in the previous event.Plant greenness would increasingly reflect the community of plants that colonize newly bare surfaces and thus could reflect the non-native invasive species that flourish under high disturbance.Conversely, if flood events become less common, a more static vegetation community could develop, but there is greater risk of dramatic groundwater declines resulting in plant stress and dieback.Unfortunately, predictions of changes in ENSO and AR frequency in California appear equivocal (compare Dettinger, 2011;Harris & Carvalho, 2018;Callahan et al., 2021).Further, there are distinct and varying geographical controls on plant health even at the site scale.For example, vegetation across the site was differentially affected by regional streamflow factors that govern groundwater recharge from the east and northeast, groundwater subsidies from hatchery outflow and irrigation return drainage, and proximity to the active channel.Such findings indicate that strategies for maintaining or restoring plant communities should vary spatially across this site and, we suspect, in many other GDEs.

| CONCLUSIONS AND IMPLICATIONS
Inflection point analysis of multi-decadal groundwater elevations in relation to surface water inflows identified a generally pronounced threshold between 2-year volumetric river-based inflows and groundwater elevations, below which groundwater levels drop precipitously and plant health suffers accordingly.As such, monitoring of 2-year cumulative surface water inflows could be a simple management tool for assuring plant health, both here and elsewhere.Maximum benefits could be achieved where, as here, supplemental waters were obtained from upstream reservoirs and from on-site wells ahead of critical groundwater declines.
Physical conditions throughout the site are resilient in the sense that groundwater recharges rapidly following a moderate to large magnitude flood event.Flow regulation may therefore have made this historical wetland more sensitive to drought-induced physiological plant stress than occurred historically, but the lack of pre-regulation gauging records limits inferences to the speculation that can be achieved from archive data sources (e.g., Beller et al., 2016).The future of this resilient but sensitive GDE under conditions of climate change is unclear because of uncertainties in the future frequency of hydrogeomorphic factors that both sustain and disturb the plant communities.More frequent high intensity ENSO events or ARs could assure plant health hydrologically (including for species dependent on hydrochory for natural recruitment) but would probably also increase the period of 'transitional' plant health related to geomorphic processes of flood scour and deposition.Fewer high magnitude events might allow plant communities to develop to their full potential but would also increase the likelihood of precipitous groundwater decline leading to plant stress and dieback.
Our analysis provides interannual insight into the underlying physical drivers of plant health for GDEs using readily available data.
Results emphasize the complex non-linear nature of abiotic controls acting on different time and space scales and the natural resilience of riparian wetlands in locations where groundwater levels recover quickly following a moderate to large flood flows.Analysis implies that restored wetlands can be persistent if multi-year water inflow shortages can be avoided and that, where flow augmentation is possible, simple flow monitoring strategies could be the basis for sustaining plant health under a climatologically uncertain future.
The USCR (drainage area 1670 km 2 ) is partially regulated by several dams on tributaries, and summer baseflows are enhanced by effluent discharges from treatment facilities constructed following rapid urban expansion upstream.Secondary inflow sources include the heavily regulated Piru Creek (1100 km 2 ) and the smaller, unregulated (and ungauged) Hopper Canyon Creek.Flow records from the United States Geological Survey (USGS) ratified gauges on the USCR and Piru Creek (retrieved from https://waterdata.usgs.gov/ca/nwis/)were combined to provide an estimate of daily mean inflows to the site since 1 October 1955.Linear regression of the combined daily mean flows against the individual gauges indicated an explained variance (R 2 ) of 0.95 with the USCR gauge and 0.09 with the climatologically wetter, but smaller and highly regulated Piru gauge, indicating that surface inflows are almost entirely related to flows received from the USCR.The timeframe over which surface water reductions might influence plant health is unclear.Combined daily mean flows were aggregated over four time periods including an average inflow for July-to-September of that year and cumulative inflows over the preceding 90-, 360-and 720-day periods.The first measure thus provides a temporal match to the high summer season period over which the NDVI is averaged and tests plant health as it responds to shortterm water availability.The latter measures represent various periods of cumulative water stress on plant health (following Bellot & Ortiz de Urbina, 2008) and test whether vegetation is responsive to longer term water availability.
Daily mean flow (DMF) discharge, from combined USCR & Piru flow, cfs July-Sept average DMF July-Sept average of cumulative totals for previous 90-day period July-Sept average of cumulative totals for previous 360-day period July-Sept average of cumulative totals for previous 720-day period Time since last major flood event Years since exceeding an instantaneous peak flow magnitude Threshold of 283 m 3 s À1 (10,000 cfs) Threshold of 425 m 3 s À1 (15,000 cfs) Groundwater elevation Sampled surface depth to groundwater data, interpolated for two pairs of nearby wells Hatchery N + S well: year average (Sept-Sept) Hopper + Calvin wells: July-Sept average Hopper + Calvin wells: change in level July-Sept Hopper + Calvin wells: year average (Sept-Sept) Precipitation Standardized precipitation index (SPI) Average of preceding 12 months (SPI-3) Evapotranspiration Standardized precipitation evapotranspiration index (SPEI) 3-month average of July-Sept months (SPEI-3) 12-month: average of preceding 12 months from September (SPEI-12) Drought severity Self-calibrated Palmer drought severity index (scPSDI) 3 month average of July-Sept months (scPDSI-3)(either HatchNSchanges or C-H GW12) and three polygons with longterm (two-year) cumulative inflows (DMF720).Meteorological variables were associated with two of the polygons but usually less strongly than the other significant abiotic variables (Table3).Geographically, plant health in the central and eastern polygons is very well explained by abiotic variables (RipEast, adj.R 2 = 0.876, SCR SE, adj.R 2 = 0.820), with associations to 2-year cumulative inflows, flood flow disturbances, summer changes in groundwater elevation at the Hatchery wells and a meteorological variable.Plant health in three of the four north (i.e., right) bank polygons is linked to the dynamics of the Hatchery groundwater well, whereas the south bank SCR SW polygon and, surprisingly, Hatchery polygon are not.Perhaps befitting its location in the bed of the active channel, the SCR SW polygon is associated with flood flow disturbances and cumulative inflows but not with groundwater elevation.
5and 3 m from the upstream floodplain well (Hopper and Calvin, respectively).But, during extended drought periods, groundwater elevations are capable of precipitous decline (Figure3).Over the last 50 years, there have been three distinct periods of drought-associated groundwater drawdown (Figure3): first, drawdown that began in 1975 and continued until rapid recovery with the early 1978 flood (i.e., encompassing the 1976-1977 drought); second, an extended period of drawdown occurred from the mid-1980s until progressive recovery during 1991 that was completed by the 1992 flood (i.e., encompassing the 1989-1990 drought); and, finally, a very extended period of drawdown from 2012 associated with the recent 'mega drought', which may be one of the most regionally significant in history(Williams et al., 2020), until recovery during 2019.This last recovery was without a corresponding large flood event but rather was facilitated by above average (124% of normal) winter rains in 2018-2019 and extended water releases from the upstream reservoir.

T A B L E 3
Best fit GAMs on NDVI polygon water balance associations, fitting smoothed spline functions ('s') for each of the variables.Threshold for variable inclusion is p = 0.05, except for CSER West where a second equation form is provided where the p value for Time10k is 0.06.The final column indicates the adjusted R 2 improvement over that achieved using stepwise linear regression.

F
I G U R E 2 Temporal trends in NDVI for each of the eight polygons.identify critical factors in the resiliency of riparian vegetation to drought.Trends in NDVI from 1985 to 2018 were variable among polygons (Table 2, Figure 2) but included consistent losses in plant health during the 1990-1991 period and a generally rapid decline in health since 2013, both intense drought periods.Geographical correlations between NDVI polygons and statistically significant non-linear associations with abiotic factors over time indicate that plant health responded consistently to spatial and temporal fluctuations in the physical stressors that influence water availability and was not greatly confounded by plant species turnover.Over decadal timeframes, explanatory variables often reflected interannual influences related to years since the last significant ENSO-generated flood event (LogTime10k), the magnitude of 2-year cumulative surface water inflows (LogDMF720) and intra-year changes in depth to groundwater (HatchNSchanges).Explained variance in four of the five polygons was strong (57-88%) signifying robust local physical control on plant health (Table

F
I G U R E 3 Groundwater depths from surface for four long-term wells (two within the Sespe Cienega [Hatchery N and Hatchery S] and two a short distance upstream [Hopper and Calvin]), compared to river inflows from combining the USGS gauges on Piru Creek and the Santa Clara River just upstream of the property.Well records extend generally back to 1972, but the Hopper well is extended back to 1969 to illustrate the effect of the large flood in 1969.lead to extensive scouring are likely to provide debris, moisture and nutrients to riparian soil that can benefit plant health (Camporeale et al., 2019).In contrast, high flows can result in near complete scouring of vegetation and concomitant reductions in NDVI.Subsequent recolonization normally leads to increasing NDVI values over time as vegetation development increases canopy cover.

F
I G U R E 4 Curvilinear (GAMs) relationships between best-fitting cumulative daily mean inflows (see Table4 ) and sampled groundwater elevations at the four wells (Hatchery N [a], Hatchery S [b], Hopper [c], Calvin [d]), and two synthetic well pairings (Hatchery N + S, [e], Hopper-Calvin [f]).Inflection points are where the red line breaks to black.Waiting on a couple of metric versions for these plots.
east (upper SCR) and northeast(Piru Creek and Hopper Canyon Creek)    are progressively overlain by (i) groundwater subsidies from Hatchery outflows and irrigation drainage and (ii) proximity to the active channel which reflects direct hydrogeomorphic control by the SCR.Polygons to the east of the site (Riparian East and SCR SE), beyond the influence of hatchery outflows, demonstrate very highly explained variability in plant greenness (82-88%, Table4) related to multi-decadal geomorphic disturbance, cumulative inflows and shorter-term water availability.To the west, plant health in the CSER West and Hatchery polygons appears to relate most closely to drainage emanating from the fish hatchery with some underlying 'natural' physical control related to flood disturbance and groundwater elevations.As noted above, plant health in the far southwest of the site has been low since 2005 as a function of geomorphic processes that caused wholesale changes in the community composition and a subsequent shift (deterioration) in growing conditions for the plant community.Lower explained variance values in the western polygons (31-62%, significantly raise groundwater elevations.The 'rescue' of groundwater elevations by large flood events is demonstrated very clearly in2004-2005.Groundwater elevations very close to the threshold for precipitous decline were reversed by two large flood events in January and February 2005.These events filled Piru Reservoir and, along with several smaller flood events over the next 5 years, facilitated late summer agricultural flow releases from Piru Dam through 2012 that bolstered cumulative inflow totals (Figure5).Once drought and dwindling storage caused managed flow releases to be discontinued, groundwater elevations dropped dramatically until 2017.Frequently occurring smaller magnitude high flow events can also reverse groundwater declines such as in 2017 and 2019 although such events are historically rare for the SCR(Downs et al., 2013).Flood events not only underpin groundwater recharge along intermittent rivers but may stimulate the recruitment of native pioneer tree species.Flow requirements for recruitment can similarly be quantified via threshold analysis.Zamora-Arroyo et al. (2001) determined that the minimum discharge necessary for the recruitment of willowcottonwood forests along the lower Colorado River was a modest flow of 3 Â 10 9 m 3 over a 3-month period that coincides with seed release timing (i.e., February-April).Like our findings, their results have clear implications for the management of flow releases from upstream dams to support native riparian forests but within dryland riparian systems such as the Santa Clara River and lower Colorado Investigation here focused on the role of multi-decadal physical factors in controlling riparian plant health in the Sespe Cienega GDE in the Santa Clara Valley, California.The study period of 30-50 years encompasses the typical extreme interannual variability of streamflow in California.Such variability results in strong variations in photosynthetic status among the NDVI polygons and in time, with plant health decreasing during drought periods and improving rapidly following flood events.Curvilinear statistical associations on plant health indicate that site-level water balances are controlled both by long-and short-term abiotic attributes.Statistically significant factors often include years since the last major flood, 2-year cumulative surface water inflows and annually averaged and summer season changes in groundwater elevations.The robust relationships (57-88% explained variance for four of the five analysed polygons) achieved using readily available data suggest that at interannual scales, the abiotic requirements for plant health are sufficiently consistent between polygons to override biological factors.The results stress the importance of flood scour and subsequent deposition on successional processes that root access to soil moisture is a critical limiting factor on plant growth in semi-arid Mediterranean environments and that periodic high flow events are critical for multi-year recharge of groundwater stores.F I G U R E 5 Variability in cumulative surface inflows from the Upper SCR and Lower Piru Creek in relation to combined daily mean flows (secondary axis) and the threshold for precipitous decline in groundwater levels in the synthetic record for the Hatchery N + S well pairing (R 2 = 0.827).
Long-term meteorological data sets were obtained from the West Wide Drought Tracker (WWDT; https://wrcc.dri.edu/wwdt/,Abatzoglou et al., 2017) and the Parameter-elevation Regressions on Independent Slopes Model (PRISM) Climate Mapping Program produced by Oregon State University (http://www.prism.oregonstate. T A B L E 2 Summary statistics and vegetation characteristics for eight NDVI polygons within and bordering the Sespe Cienega site between 1985 and 2018.Based on geographical correlations, three polygons (grey text) were removed from further analysis.