Climate and human water use diminish wetland networks supporting continental waterbird migration

Abstract Migrating waterbirds moving between upper and lower latitudinal breeding and wintering grounds rely on a limited network of endorheic lakes and wetlands when crossing arid continental interiors. Recent drying of global endorheic water stores raises concerns over deteriorating migratory pathways, yet few studies have considered these effects at the scale of continental flyways. Here, we investigate the resiliency of waterbird migration networks across western North America by reconstructing long‐term patterns (1984–2018) of terminal lake and wetland surface water area in 26 endorheic watersheds. Findings were partitioned regionally by snowmelt‐ and monsoon‐driven hydrologies and combined with climate and human water‐use data to determine their importance in predicting surface water trends. Nonlinear patterns of lake and wetland drying were apparent along latitudinal flyway gradients. Pervasive surface water declines were prevalent in northern snowmelt watersheds (lakes −27%, wetlands −47%) while largely stable in monsoonal watersheds to the south (lakes −13%, wetlands +8%). Monsoonal watersheds represented a smaller proportion of total lake and wetland area, but their distribution and frequency of change within highly arid regions of the continental flyway increased their value to migratory waterbirds. Irrigated agriculture and increasing evaporative demands were the most important drivers of surface water declines. Underlying agricultural and wetland relationships however were more complex. Approximately 7% of irrigated lands linked to flood irrigation and water storage practices supported 61% of all wetland inundation in snowmelt watersheds. In monsoonal watersheds, small earthen dams, meant to capture surface runoff for livestock watering, were a major component of wetland resources (67%) that supported networks of isolated wetlands surrounding endorheic lakes. Ecological trends and human impacts identified herein underscore the importance of assessing flyway‐scale change as our model depictions likely reflect new and emerging bottlenecks to continental migration.


| INTRODUC TI ON
Water-limited ecosystems account for 40% of terrestrial land surfaces globally and support upwards of 2 billion people (Gilbert, 2011). Nearly half of these arid and semi-arid regions are made up of endorheic watersheds in which all runoff converges in terminal water bodies topographically landlocked from the ocean (Wada et al., 2011). Water scarcity places vital ecological and economic importance on endorheic watersheds as they are often associated with sizable lakes and wetland systems in otherwise arid landscapes. Endorheic water stores are driven predominantly by precipitation inputs and groundwater exchange that equilibrate through evaporation. Recent declines in global endorheic water storage (Wang et al., 2018) suggest increased evaporative demands, due to warming temperatures. More frequent droughts are threatening this delicate ecosystem water balance as growing human populations (Wada, van Beek, Wanders, & Bierkens, 2013) and intensifying climate change (Dai, 2013) are increasing water consumption and accelerating endorheic withdrawals (Wurtsbaugh et al., 2017).
Dewatering of endorheic watersheds has the potential to impact long-term population dynamics as carry-over effects driven by deteriorating migratory habitats reduce waterbird survivorship in subsequent life-history events (Hua, Tan, Chen, & Ma, 2015;Sedinger & Alisauskas, 2014).
Arid and semi-arid mid-latitudes of western North America are among the most important inland waterbird flyways in the Western Hemisphere (Oring & Reed, 1997;Wilsey et al., 2017).
In this region, migratory pathways are structured around endorheic watersheds that support large saline and freshwater lakes in addition to freshwater palustrine wetlands occurring along lake peripheries and throughout surrounding riparian drainages.
Snowpack accounts for 50%-70% of spring runoff that feed perennial stream flows supporting saline and freshwater lakes and a mosaic of emergent freshwater seasonally and semi-permanently flooded wetlands.
The monsoon region encompassed eight southern endorheic watersheds in the northern highlands of Mexico and the southwestern United States, hereafter 'monsoonal watersheds' (Figure 1; Table 1).
These watersheds receive their precipitation primarily during warm summer months (July-September) from convective thunderstorms influenced by complex interactions of subtropical ocean moisture and continental land masses (Adams & Comrie, 1997

| Estimating surface water trends
Lake and wetland surface water area was monitored annually from Training data for SMA were extracted from satellite imagery as spectral endmembers unique to individual images classified. Training site locations were representative of homogeneous land cover types mapped as water, wetland vegetation, upland, and alkali soil. Spectral endmembers for water were collected using image masks generated from 99th percentile normalized difference water index values (McFeeters, 1996). Mask extents were coincident with large deep water lakes within or proximal to endorheic watersheds. A similar masking approach was applied to collect wetland vegetation endmembers using normalized difference vegetation indices (Box, Holben, & Kalb, 1989).
Sampling was constrained to sites coincident with seasonally flooded wetlands and were representative of associated plant phenology. SMA requires minimal training data (Adams & Gillespie, 2006) which allowed upland and alkali soil endmembers to be generated from a small number of static plots within endorheic watersheds (n = 2; 0.5-1 km 2 ).
Upland plots were associated with homogenous shrublands characterized by low vegetative productivity and high soil exposure. Alkali soil plots were coincident with dry lake basins in areas of surface mineral deposits. Plot locations were identified using high-resolution (<0.5 m) multispectral satellite imagery or field reconnaissance.
Surface water extent within endorheic watersheds was averaged annually within offset 6 months of seasonal periods for the snowmelt (April-September) and monsoonal (October-March) regions (see Figure 1). Periods aligned broadly with known spatiotemporal waterbird breeding, migration, and wintering patterns (Baldassarre, 2014;Kushlan et al., 2002;Senner et al., 2016). Satellite data within annual periods of 6 months were averaged into single multi-spectral images and classified using SMA to produce seasonal estimates of

| Lake and wetland change
Changes to lake hydrology were assessed by summarizing SMA results along gradients of averaged variance in annual surface water area and the averaged proportion of basins covered with surface water between two periods 1984-1999 (P1) and 2000-2018 (P2; Figures S1 and S2; Table 2; Tables S1 and S2). We used these two periods primarily to capture the inter-annual variability of the climate driven by ENSO and PDO, the two main climate teleconnections that control climate in western North America. Two periods also gave a reasonable number of records to produce a statistically valuable results (n > 15 years). Maximum surface water area measured from 1984 to 2018 was used to calculate the proportion of lake coverage. Differences were plotted as change vectors for individual lakes grouped by snowmelt and monsoon regions. Wetland change was calculated as the difference in mean area between P1 and P2. Results were partitioned by snowmelt and monsoon watersheds.

| Estimating factors of human water use
Annual extent of irrigated agriculture and population density change  were utilized as an analog of human water use.

| Endorheic water balance and climate variables
Under equilibrium conditions, with no external modifications (e.g. no water withdrawal or inter-basin transfer of water), climate factors control the surface area of lakes through the balance between runoff, precipitation, and evaporation (Budyko, 1974;Mason, Guzkowska, Rapley, & Street-Perrott, 1994;Mifflin & Wheat, 1979).
Because lake area represents a natural equilibrium state between watershed runoff, precipitation, and evapotranspiration, it is also a measure of climate aridity where smaller lake area (for a given watershed) represents more arid conditions and larger lake area represents less arid conditions (Mason et al., 1994). These relationships are unique to endorheic watersheds because their lakes act as a combined 'rain gauge/evaporation pan' for the watershed which can be used to determine hydrologic changes that are driven solely by climate (i.e. changes in aridity). Therefore, when observed lake area is lower than that predicted by climate alone, it suggests that human modification of the water budget through basin withdrawals for agriculture and domestic use are present. This distinction makes it possible to attribute human versus climate driven hydrologic change within an endorheic watershed.
Using the SMA results, we explored patterns of lake and wetland surface water area as response variables to hydrologic processes within endorheic watersheds. Area of irrigated agriculture and population density were considered human-induced predictor variables because they modify the natural water balance through runoff withdrawal and interception. To estimate the attribution of climate versus direct human actions to changes in lake and wetland surface water area, we chose the predictor climate variables that directly affect endorheic water balance: runoff (RO), evapotranspiration (ET), and precipitation (PR); we additionally included snow water equivalent (SWE) as an important component of RO in the snowmelt region of the study (Dierauer, Whitfield, & Allen, 2018;Fritze, Stewart, & Pebesma, 2011).

| Statistical analysis
We attributed importance of climate (ET, PR, RO, and SWE) and human water use (irrigated agriculture and population density) to the prediction of lake and wetland surface water area using ran-domForestSRC regression tree analysis (Ishwaran & Kogalur, 2019), as a nonparametric measure of variable importance (VIMP). This approach is applicable to ecological systems with typically nonnormal distributions, which most of our variables showed through time (Cutler et al., 2007;De'ath & Fabricius, 2000;Zanella, Folkard, Blackburn, & Carvalho, 2017). RandomForestSRC allowed for a two-step method of randomization to de-correlate trees, which decreased variance and bias for a stronger representative model (Zhang & Lu, 2012 to all random forest analysis (Breiman, 2002). Model runs were conducted using 5,000 trees. Variable rankings were presented as boxplots for lakes and wetlands, partitioned by snowmelt and monsoonal regions.
We quantified change to climate variables (ET, PR, RO, and SWE) and lake surface area using the nonparametric Wilcoxon test (Siegel, 1957). Data were binned temporally from 1984 to 1999 (P1) and 2000 and 2018 (P2) to compare differences in long-term trends.
p value is considered a measure of significance strength in the difference between the two periods, but for convenience we used a p value of .05 to represent significant/nonsignificant change.
Boxplots were used to visualize variability and change and are provided in the Supporting Information (Figures S1-S10).

| RE SULTS
Detailed data supporting our analyses of waterbird flyway resilience in western North American are provided in Tables S1-S10 and

| Surface water trends in endorheic watersheds
Long-term monitoring of surface water revealed that lakes in snowmelt watersheds diminished at twice the rate (−27%, ~191 kha) as those found in monsoonal watersheds (−13%, ~5.3 kha; Figure 2a; Table 2). Whereas all lakes showed substantial interannual variability, patterns in long-term trends differed substantially between snowmelt and monsoonal regions.
Lake area declines were significant in 13 of 18 snowmelt watersheds ( Figure 2a; Figure S1; Table S1) with most annual trends exhibiting strong linear declines from 1984 to 2018 ( Figure S9).

2002).
Overall annual variability in lake surface water area differed between snowmelt and monsoonal watersheds as characterized regionally by patterns observed at Great Salt Lake and Laguna de Babicora (Figure 4). Surface water area at Great Salt Lake, for example, showed a strong linear decline over time, where long-term change (1984-2018) outweighed near-term variability (Figure 4a).
In contrast, annual surface water variability at Laguna de Babicora was much greater than the long-term change (Figure 4b), meaning that extensive surface water coverage or near dryness was equally likely from year to year so that long-term change was outweighed by near term variability. While monsoonal patterns were found to be highly dynamic, over the long term they remained relatively stable.
Snowmelt patterns generally showed long-term declines with relatively small short-term change.
Summer Lake was the only site where peripheral wetland area increased, but only by a small amount (+2%, 19 ha).
Human population density, a surrogate used for domestic/ industrial water consumption, increased by +41% (~1.3 million people) across all watersheds and by +40% (~1.2 million people) and +50% (~105 thousand people) in snowmelt and monsoonal F I G U R E 3 Change in mean endorheic lake (a) and wetland (b) surface water area between periods, 1984-1999 and 2000-2018. Results partitioned by snowmelt and monsoonal watersheds watersheds individually (Table 2). Growth was significant (p < .001) in all but the snowmelt-watershed Harney Basin, where populations declined significantly (Table S5). Despite growth, human population densities remained relatively low in most areas; however, changes were likely indicative of increasing domestic/industrial water demand throughout the study area.
Of climate variables examined as potential lake and wetland surface water predictors, ET was the only variable to increase significantly (+6%) across all snowmelt watersheds (Table 3; Table S6).
Linear time-series trends for ET were also statistically significant for the same watersheds ( Figure S13; Table S6). In contrast, changes to ET were insignificant in all monsoonal watersheds, increasing by only +1% overall (Table 3; Table S6). Increasing linear trends, however, were significant in seven of eight monsoonal watersheds ( Figure S13).

| Predictor variable importance
Variable importance analyses identified irrigated agriculture as the most important predictor of lake and wetland surface water, although its level of importance varied ( Figure 5). Irrigated agriculture was the dominant driver of wetland area with VIMP scores double that of climate variables (Figure 5b,d). Population density was also an important predictor of wetland surface water in monsoonal watersheds ( Figure 5d). While irrigated agriculture was the single most important predictor of lake surface area, individual climate variables were nearly as important (Figure 5a,c). SWE was inconsequential to monsoonal lakes and wetlands do to its rarity in the region (Figure 5c,d).

| D ISCUSS I ON
This study is the first that we are aware of to assess long-term lake and wetlands trends in endorheic watersheds spanning the United States and Mexico. Outcomes document strengthening patterns of landscape desiccation that could have major effects on waterbird habitats implicating a greater need for focused wetland conservation in this expansive region. Lake and wetland surface water loss was pervasive, driven largely by agricultural water use coupled with rising evaporative demands. Trends were particularly concerning in snowmelt watersheds that supported the vast majority of wetland resources where drying was significant in 16 of 18 watersheds.
While monsoonal watersheds represented a smaller proportion of total lake and wetland area, their distribution within highly arid regions of the continental flyway increases their value to migratory waterbirds. Less predictable annual patterns of inundation among these watersheds due to spatial variance in monsoonal rainfall were offset by broader habitat stability. However, long-term wetland inundation in the region remained relatively constant and generally highly productive when wet.
Wetland relationships with farming and ranching practices are complex and the simple interpretation that all irrigated agriculture is detrimental for wetlands is over generalized and likely misleading. In monsoonal watersheds, small earthen dams meant to capture surface runoff for livestock watering were a major component of wetland resources (67%). When flooded, dams acted as TA B L E 3 Summary of changing evapotranspiration (ET), precipitation (PR), runoff (RO), and snow water equivalent (SWE) means within snowmelt and monsoonal watersheds between 1984-1999 and 2000-2018   for endorheic lakes and wetlands within snowmelt and monsoonal watersheds (see Figure 1). Box: 25th, 50th (heavy vertical line), and 75th percentiles. Whiskers: 5th and 95th percentiles. Predictor variables: ET, evapotranspiration; IrrAg, irrigated agricultural area; PopDen, human population density; PR, precipitation; RO, runoff; SWE, snow water equivalent. VIMP is derived from Breiman-Cutler permutation; see Section 2 for details); VIMP is standardized for comparisons among all plots an artificial network of isolated wetlands known to support 31 species of waterbirds (Riojas-López & Mellink, 2005). We speculate through our results that increased wetland inundation within monsoonal watersheds was in-part due to new dam construction associated with growing human populations (+50%) and livestock ranching in the region. Rapid expansion of irrigated agriculture identified in monsoonal watersheds (+21%) was attributed to new and unregulated groundwater use in valley bottoms (Pool, Panjabi, Macias-Duarte, & Solhjem, 2014). Despite increasing hydrologic perturbation (e.g. damming and groundwater extraction) in many watersheds, there were no clear pattern modifications for decreasing areas of surface water. Area of lake and wetland coverage (natural and dammed) remained variable suggesting a stronger climatic influence.
A total of 61% of wetlands within snowmelt watersheds were affiliated with irrigated agriculture, mainly consisting of flooded riparian wet meadows which are a valuable waterbird habitat (Fleskes & Gregory, 2010;McWethy & Austin, 2009;Moulton, Carlisle, Brenner, & Cavallaro, 2013). Despite these contributions, only 7% of irrigated lands within snowmelt watersheds was associated with wetlands. Clearly, how and where irrigation is implemented has important consequences to wetland availability.
We lacked local irrigation and crop level data in watersheds to fully understand the mechanisms behind patterns of wetland drying that may be associated with agricultural practices. However, previous studies suggest changes to crop type (Bishop, Curtis, & Kim, 2010) and irrigation regimes (Hassanli, Ebrahimizadeh, & Beecham, 2009;Pfeiffer & Lin, 2014) can affect overall water use.
Sustained patterns of lake declines across snowmelt watersheds ( Figure 6) suggest a regional tipping-point in ecosystem water balance has been reached where increasing human and evaporative demands now consistently overdraw water supplies. Declining surface water trends raises concerns of trophic collapse within watersheds supporting productive saline lake food webs (e.g. Lake Abert, Mono Lake, and Great Salt Lake). Lower lake levels are resulting in increased salinity rates as freshwater inflows diminish (Larson et al., 2016;Moore, 2016).
Higher salinity can drastically reduce diversity and biomass of benthic macroinvertebrates that serve as critical food resource for waterbirds.
As water volumes continue to decrease, lakes can reach a point of infertility well before they dry completely (Herbst, 2006;Senner et al., 2018). Transition of some declining freshwater lakes to saline states (sensu Thomas, 1995-Walker Lake) may open habitat niches in snowmelt watersheds that offset losses in others. However, these lakes may also be vulnerable to collapse if freshwater inflows continue to decline.

| Implications to waterbird conservation
The results of this study have significant implications to waterbirds at both local and landscape levels because of (a) the overall impacts of water use and climate on lake and wetland area, (b) increasing lake salinity, and (c) broad scale loss of migratory connectivity along waterbird flyways. The significant decline in wetland area across snowmelt watersheds (−47%) is disconcerting. Comparatively, lake area in monsoonal watersheds were more stable with small overall increases in wetland extent. However, declining water quality has been identified as a conservation challenge in some monsoonal watersheds that may further deteriorate flyway resilience due to increased pollution from urbanization and industrialized agriculture (Benavides et al., 2008). Additionally, expanding agriculture in monsoonal watersheds is being driven by groundwater pumping (Pool et al., 2014) and long-term consequences of this practice on wetland F I G U R E 6 Transition in lake hydrology along gradients of averaged variance in annual surface water area and averaged proportion of lake basins inundated between two periods, 1984-1999 and 2000-2018. Data are partitioned by snowmelt (a) and monsoonal (b) watersheds. Change is measured as a vector depicting the rate and direction of transition between periods along a continuum of states from 'wet', 'dry', and 'annual surface water variance'. Maximum surface water area measured from 1984 to 2018 was used to calculate the proportion of lake inundated resilience remains unclear. Some lakes (McFarland & Ryals, 1991) and wetlands (Downard & Endter-Wada, 2013) rely on groundwater as a significant component of their water budget and groundwater declines could affect their long-term condition and functionality as waterbird habitat (Pritchett & Manning, 2012).
While groundwater connections are unknown, it was clear that monsoon rainfall patterns were an important driver of wetland area in monsoonal watersheds. Analysis of future climate in the Mexican Highlands predicts precipitation from 2030 to 2045 to be variable, but remains similar to patterns of recent decades (Verduzco et al., 2018). Although Cook and Seager (2013) similarly found that by 2080-2099 total precipitation would remain approximately the same, the timing of peak monsoon rainfall would shift from June-July to September-October. In addition, they found that despite stable precipitation trends, rising temperature and evapotranspiration would lead to increased annual drying. These changing monsoonal patterns coupled with increased evapotranspiration will likely shape lake and wetland resiliency in monsoonal watersheds by altering the timing and volume of runoff, influencing waterbird habitat availability and agricultural and urban water use.
In addition to loss and altered timing of runoff to wetlands in general, the specific loss of freshwater wetlands along the periphery and adjacent to lakes raises concern over potential decline of ecosystem diversity within some snowmelt watersheds (Table S10). In saline lakes particularly, waterbirds are reliant on adjacent freshwater wetlands to balance physiological demands of saltwater environments (i.e. osmoregulation), especially during brood rearing when chicks are freshwater dependent (Rocha et al., 2016;Wollheim & Lovvorn, 1995). Loss of peripheral wetlands was greater (−53%) than those in upper watersheds (−47%) and likely a result of compounding upstream water diversions.
In agriculturally dominated snowmelt watersheds, drought effects are not evenly distributed as water is allocated hierarchically on a 'first in time, first in right' basis where long-time irrigators are granted priority rights to consume water apportionments prior to junior users (Getches, Zellmer, & Amos, 2009). Depending on individual state laws, however, pumping of groundwater may or may not be regulated and can affect surface flows because of groundwater and surface water connectivity (Cooper, Sanderson, Stannard, & Groeneveld, 2006). Furthermore, legal rights to protect in-stream flows and associated ecosystem services were unrecognized as a beneficial use until the 1970s (Benson, Dan, Corbridge, Getches, & Bates, 2014) and are legally cumbersome in some states today (Szeptycki, Forgie, Hook, Lorick, & Womble, 2015), resulting in low priority and diminished water availability for maintenance of natural wetland systems.
Hydrologic resilience of western North American flyways has allowed pioneering waterbirds to leverage wetland availability to offset drought and maintain connectivity by adapting migratory pathways to shifting continental resource conditions (Albanese & Davis, 2013;Skagen, Granfors, & Melcher, 2008). Lake and wetland declines we identify in snowmelt watersheds may signal a loss of plasticity in migratory networks. Further impacts are expected as forecast of drought and water use demands intensify over coming decades (Dettinger, Udall, & Georgakakos, 2015). Drying of individual lakes has the potential of dramatically reconfiguring energetic demand of migration by increasing the flight distances between stopovers and reducing the total number of sites available to birds in water scarce landscapes (Haig et al., 1998

| Finding solutions
Sustainability of waterbird migration flyways in western North America will require adaptive changes to existing conservation priorities (e.g. North American Wetlands Conservation Act North American Wetlands Conservation Act, 1989) considerate of accelerating lake and wetland drying. To date, waterbird conservation has been structured around policies to protect land designated as wetlands (e.g. McBeth, 1997; Farm Bill swamp buster provisions) rather than the water supplies crucial to wetland hydrologic function (Downard & Endter-Wada, 2013). Evolutions in urban planning are offering solutions demonstrated by water efficiency programs and flexibility in water supply development used by the city of Los Angeles, California, for example, to reduce their reliance on Mono and Owens Lakes diversions by 60% (Hughes, Pincetl, & Boone, 2013). Forward looking voluntary and incentive-based approaches to agricultural water use could have similar effects, wherein producers are supported through government cost sharing of more efficient irrigation infrastructure and, in turn, are compensated for water savings designated for maintenance of wetland habitats (Castle, Beattie, Smith, Peternell, & Kowalski, 2016;Grafton et al., 2018). It is critical, however, that new water savings be redirected to ecosystem services (Kendy et al., 2018) as numerous studies indicate that irrigation efficiency often leads to planting of more waterintensive crops or expansion of agricultural areas (Batchelor et al., 2014;Scott et al., 2014).
We make our data available to landscape planners to promote ecosystem water balance of agriculture, urban, and waterbird migration. Applications may include targeted preservation of irrigation practices supporting wetlands that made up only 7% of the agricultural footprint in snowmelt watersheds, but were associated with the majority of freshwater emergent wetlands. Flood irrigation of these sites are often perceived as wasteful and singled out by water efficiency efforts as a means to generate agricultural water savings used to offset growing urban demands (Richter et al., 2017). Such practices, however, can unintentionally accelerate wetland loss and eliminate waterbird habitats that further degrade migratory flyways (Ward & Pulido-Velazquez, 2008). Consideration of the specific social, ecological, economic and hydrological contexts of watersheds and underlying aquifers will be necessary to accurately identify impacts and opportunities of various water management decisions. As noted in this study, the conservation value of these wetlands to waterbirds is manifested in the context of both local waterbird habitat needs and their contribution to processes supporting broader migratory connectivity. We encourage the use of our results to inform conservation solutions by means of collaborative and proactive decision-making among local and international stakeholders throughout western North American flyways.

ACK N OWLED G EM ENTS
We thank and acknowledge John Vradenburg (US Fish and Wildlife Service) for his initial review of this manuscript. We also thank members of the Intermountain West Joint Venture Science Advisory Committee for insight important in shaping this work. Views in this manuscript from United States Fish and Wildlife Service authors are their own and do not necessarily represent the views of the agency.
Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.

CO N FLI C T O F I NTE R E S T
The authors declare that there is no conflict of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request.