Integrated water resources trend assessments: State of the science, challenges, and opportunities for advancement

Water is vital to human life and healthy ecosystems. Here we outline the current state of national‐scale water resources trend assessments, identify key gaps, and suggest advancements to better address critical issues related to changes in water resources that may threaten human development or the environment. Questions like, “Do we have less suitable drinking water now than we had 20 years ago?” or “Are flood events more common now than they were in the past?” prompted improvements in data, trend estimation methods, and modeling frameworks to track changes in, and better understand how land use and climate influence four water resources domains: surface and groundwater quantity and quality. However, continued advancement in trend assessments to better address issues related to changes in water availability is needed. Areas of need include more timely and efficient delivery of water resources trend results and improved capacity to estimate trends at unmonitored locations. Additional integration pieces include increased understanding of groundwater–surface water interactions, incorporation of both quantity and quality trends into water availability estimates, and the refinement of trend metrics to account for the competing needs of society and ecological integrity. Coupled with improved driver attribution studies, these components will better inform current and future water resources management.

the Gravity Recovery and Climate Experiment satellite data, which are available for the time period 2002 onward, have been used to estimate trends in global groundwater storage (Jakeman et al., 2016;Wada et al., 2010).The integration of remote sensing data with empirical data and hydrologic model output has been essential for accurately assessing changes in groundwater quantity at regional scales (Rateb et al., 2020).
Remote sensing data have also been used to estimate 34-year trends  in surface water quality at a continental scale by examining water color changes over time (Gardner et al., 2021), and trend assessments have benefitted from data repositories, like Aquasat, that have combined in situ water quality data with satellite observations (Ross et al., 2019;Topp et al., 2021).

| Methodological Advancements
Trend studies have commonly employed parametric, linear regressions or nonparametric, order-statistic techniques, such as the Mann-Kendall and seasonal Kendall test, to quantify changes in all four water resources domains.The nonparametric Theil-Sen slope is often used to determine annual rates of change in monotonic trends, since environmental data frequently arise from non-normal distributions (Helsel et al., 2020).
For surface and groundwater trends, the Regional Kendall test (Helsel & Frans, 2006), which is a modification of the Seasonal Kendall test, has been used to analyze changes in surface water quantity trends and contaminant concentrations for aquifers or groundwater management areas (Lindsey & Rupert, 2012;Yue et al., 2002).This technique, which accounts for the spatial correlations among sites, can provide information on the overall spatial coherence of trends in a regional aquifer system (Lindsey & Rupert, 2012;McMahon & Chapelle, 2008) or a groundwater management area (Chaudhuri & Ale, 2014) as opposed to individual wells.
The aforementioned statistical methods are commonly used, but they are limited by the assumptions of linear or monotonic gradual trends.
New applications of established statistical methods to account for abrupt changes and non-monotonic trends have proliferated in the past two decades.Mathematical functions (polynomials of order two or greater) that accommodate nonmonotonic temporal trajectories over an entire period of analysis or nonmonotonic responses to physical covariates can be employed (Fleming & Dahlke, 2014).A variety of methods can be used to account for abrupt step changes in streamflow (Ryberg, Hodgkins, & Dudley, 2020), including a combination of the Mann-Kendall trend test and the Pettitt change-point test (Rougé et al., 2013;Sadri et al., 2016).The Pettitt test has also been successfully implemented at a national scale in a groundwater quality context to track the reversal of contaminant concentrations following legislation and management actions (Frollini et al., 2021).
Trend methods have evolved to also quantify changes in metrics that are not just focused on the central tendency (mean or median values) and can be useful to represent changes across a distribution, including low and high values (Hecht & Vogel., 2020).The application of these types of methods is most commonly found in relation to surface water quantity trends, which have employed both quantile regression (Luce & Holden, 2009) and Quantile-Kendall approaches (Bhaskar et al., 2020;Rodgers et al., 2020) to examine changes in the distribution of streamflow, or the percentiles of flow distributions, over time.
Another limitation of traditional statistical methods like Mann-Kendall to streamflow trend analysis is that the statistical significance of such trend tests is affected by the serial correlation in hydrologic time series data.Methods have been developed that adjust the p-value to account for short-term and long-term persistence (Sagarika et al., 2014).Short-term persistence is accounted for by "pre-whitening" the data, F I G U R E 1 Recent water resources trend results typically are reported separately for these four water resources domains.
which involves the removal of serial correlation, then performing the trend test on the uncorrelated residuals (Hamed, 2008); although in some cases, pre-whitening historical data can alter the variability and can cause an underestimation in the significance of trend, particularly those related to extreme streamflow events (Razavi & Vogel, 2018).The technique to account for the effect of long-term persistence involves the modification of trend tests using the Hurst coefficient (Hodgkins et al., 2019;Matalas & Sankarasubramanian, 2003).
Enhanced methods have also been developed to inform policy makers and the public about surface water-quality improvements in response to wastewater treatment plant upgrades or changes in agricultural practices.Long-term water quality load and concentration trends that correct for random variation in streamflow allow for a more focused consideration of the effects of management actions.The Weighted Regression on Time Season and Discharge (WRTDS) model (Hirsch et al., 2010) was in part developed to produce these types of results in the form of flow-normalized concentrations and loads.In addition, with method advancements in both the WRTDS and the Water Quality Trend (QWTREND) models (Vecchia & Nustad, 2020), it is possible to address abrupt shifts, or step changes, in water quality records due to adjustments in laboratory methods or sample collection procedures, which had previously limited the length of datasets used in trend analysis.
In addition to the daily discharge estimates and discrete water quality samples that are commonly used in surface water quality trend analysis, high frequency, sensor-derived datasets are now available (Abbott et al., 2016;Krause et al., 2015).Additional advancements to address the serially correlated residuals in high-frequency time-series data are required for use in a surface water quality trend analysis context.Generalized additive models, which can address residuals with high serial correlations, have been used to incorporate high-frequency time-series data streams into water quality trend analyses (Yang & Moyer, 2020).Another promising approach is the use of Kalman filtering techniques (Zhang & Hirsch, 2019), which locally weight residuals to minimize the autoregressive structure in high-frequency observations.Altogether, the increases in available data and trend methodology advancements have allowed for a proliferation of water resources trend results across the four water resources domains.

| Summary of recent findings from national-scale water resources trend studies
Here, we summarize select findings related to the four water resources domains, with a focus on studies published from 2013 onward.Recent surface water quantity studies typically capture changes during the time period from 1916 to 2020 (Table 1).Using modeled streamflow estimates, Rice et al. (2015) found positive trends in average conditions (e.g., annual mean and variance) in the Northeast and Southeast conterminous United States (CONUS).However, many streamflow trend studies are focused on the analysis of change in extreme conditions.Trends based on the evaluation of empirical streamflow measurements showed no coherent change in CONUS flood dynamics (Archfield et al., 2016).Exceptions included the Northeast, which showed increasing flood frequency, but decreasing flood peaks, duration, and volume, while the opposite pattern appeared for parts of the central CONUS.Using changes in gage height as an indicator of flooding, Slater and Villarini (2016) found that flood risk increased in the upper Midwest/Great Lakes region and decreased on the Gulf Coastal Plain, the southeastern U.S., and California.Basins with a significant portion of streamflow as snowmelt have also been a focus of high-flow-related trend studies, because observed changes in snowpack accumulation and ablation can affect the magnitude and timing of spring peak flow.A recent study found that the majority of watersheds had a shift toward earlier snowmelt peak flows, with larger and more significant shifts in the eastern U.S. than the western U.S. (Dudley et al., 2017).
On the other end of the streamflow spectrum, increasing occurrences of streams with no flow in the southern and western U.S., decreasing magnitude of the lowest annual flows in the southeastern and parts of western U.S., and increasing magnitudes of the lowest annual flows in the northeastern and midwestern U.S. have been documented (Dethier et al., 2020;Dudley et al., 2020;Zipper et al., 2021).A recent droughtfocused study showed comparable trend results to the low-flow studies, indicating that the number of dry days and the deficit (i.e., flow volume below a particular threshold) have increased in the southern and western U.S. (Hammond et al., 2022).Results from Zou et al. (2018), which measured changes in the areal coverage of lakes, rivers, and reservoirs, derived from remote sensing imagery, corroborated the regionally diverse trend results noted in the aforementioned flood and drought studies.In summary, while flood trends show fragmented spatial patterns across parts of the southeastern, Midwest, and northeastern U.S., multiple studies have documented spatial coherence in the magnitude and duration of changes related to both low-flow and drought conditions in the western U.S.
Groundwater quantity trend studies typically capture changes during the time period from 1900 to 2018 and have focused on assessing water levels, depletion rates, and storage to identify where the rate of groundwater withdrawal exceeds the rate of recharge.A regional study of groundwater level trends in the Glacial Aquifer system in the Northeast and Upper Midwest CONUS showed that most sampled wells demonstrated increasing groundwater levels, and most of those increases were in the eastern part of the system, with mixed results in the central and western parts (Hodgkins et al., 2017).A national-scale study indicated that groundwater levels declined across much of the country between 1949 and 2009 (Russo et al., 2014), and modeled groundwater storage estimates for a similar time period indicated an acceleration of groundwater depletion rates over time (Konikow, 2015).However, trend results derived from a combination of modeling and remote sensing data for the time period 2002 to 2017 showed that most CONUS aquifers had stable or increasing storage (Rateb et al., 2020).At a regional scale, the studies had comparable results in that the largest water level and storage declines occurred in the California Central Valley and Arizona Alluvial Basins as well as the Central and Southern High Plain and Texas Aquifers.
The pollution of freshwater bodies used for drinking water, recreation, and the sustenance of aquatic ecosystems is one of the most important environmental concerns in the U.S. (Keiser & Shapiro, 2019), and surface water quality studies typically capture changes during the time period from 1950 to 2018.There are hundreds of potential water quality constituents of interest that occur in elevated concentrations and have known adverse effects on human or aquatic life (Olker et al., 2022), but the constituents commonly included in surface water quality trend assessments have been limited to those that have long-term (>10 years) data records.Using in situ water quality monitoring data, widespread decreases in sediment concentrations (Li et al., 2020;Murphy, 2020) and widespread increases in both salinity and alkalinity (Kaushal et al., 2018) have been documented.Regional differences in salinity trends have also occurred, with the greatest number of increases in the Northwest and some decreases in arid western states.Coherent patterns in national-scale nutrient trends were related to land use, with decreases in urbanized watersheds and no significant change in agricultural ones (Stets et al., 2020).Using remote sensing data, Gardner et al. (2021) showed that more than one-third of rivers in CONUS had significant color changes.While the use of remote sensing data increased the spatial coverage of rivers included in the trend analysis, it was not possible to determine which water quality parameter contributed to the change, as color can be indicative of primary productivity regimes, suspended sediment, or light availability (Julian et al., 2013).Shoda et al. ( 2019) not only reported results related to major nutrient and major ion trends, but also compared the trends to environmentally relevant levels of concern (LOC), including drinking water standards and ecoregional nutrient criteria.For sites with water quality problems related to total nitrogen and phosphorus (2002 concentration was above the LOC), improvement of water quality conditions (2012 concentration moved below LOC) was uncommon.In contrast, for inorganic nutrients (ammonia and nitrate) and major ions (sulfate, chloride, and total dissolved solids (TDS)), concentrations remained below the LOC, even for sites with positive trends.Overall, water quality trend results, which have focused on sediment, nutrients, and major ions, indicate that improvements in water quality have been highly variable based on the constituent examined in the study along with the region and dominant land-use type of the study watersheds.
The summary of groundwater quality trends research includes studies published before 2013, because monitoring of groundwater wells generally occurs less frequently (once per decade) as compared to surface water quality monitoring sites (multiple times per year).These studies reported decadal changes that occurred between 1990 to 2000 and 2001 to 2010 and focused on constituents that pose potential risks to human health.Groundwater quality data were derived from networks which included 20-30 wells representing water-quality conditions in a given aquifer and represent nearly 80% of the estimated groundwater withdrawals used for drinking-water supply in the nation (Lindsey & Rupert, 2012).Lindsey and Rupert (2012) found that nitrate, chloride, and salinity concentrations increased in half of the well networks that were sampled, and the proportion of wells with nitrate and TDS concentrations exceeding human health thresholds also increased.The detection frequency of pesticides in groundwater samples increased across all land-use types, including agriculture, urban, and mixed (Toccalino et al., 2014).However, the occurrence of pesticides exceeding human health benchmarks was rare, and there was no change in the number of pesticide compounds exceeding human health benchmarks between the two time periods.

| Examples of methods used to evaluate climate effects on water resources trends
Here, we provide some examples of the methods used to evaluate the impact of climate on the four water resources domains.These studies address the complex behavior of streamflow change and its relationships with climate (Serago & Vogel, 2018), and climatic variables including precipitation patterns, evaporation rates, and drought frequency are often non-stationary, so historical climate conditions cannot be assumed These recent national-scale studies focused on evaluating change using datasets from river and aquifer systems and were published within the last decade (2013-onward).The time period requirement was expanded for groundwater quality trend studies.
TA B L E 1 (Continued) in driver attribution analysis (Cheng et al., 2014;Kilgore et al., 2019).Several of the studies highlighted in the Recent Advances, Section 2.3 (Dudley et al., 2017(Dudley et al., , 2020;;Hammond et al., 2022), derived trends from watersheds that were minimally affected by human disturbance to focus on the role of climate and examined various metrics of streamflow (low flow, drought) and climate indices (precipitation, air temperature, and climate indices like the El Niño/Southern Oscillation (ENSO)) using correlations.Simple monthly water-balance models can also be used to resolve the transient response of streamflow to varying climate at a continental scale (McCabe et al., 2017;McCabe & Wolock, 2011).
Changes in precipitation, evaporation, and drought frequency can also have a significant impact on groundwater recharge and levels (Jiménez Cisneros et al., 2014).Empirical approaches typically rely on statistical methods, such as the Mann-Kendall trend test (Russo & Lall, 2017), to establish a trend in groundwater levels indicative of changes in aquifer storage, and Spearman correlation coefficients have been used to establish a link between precipitation and baseflow trends (Ficklin et al., 2016).Singular spectrum analysis (SSA), wavelet coherence Climate can affect water quality as a result of water temperature increases, and more frequent high-intensity rainfall events mobilize pollutants (U.S. Global Change Research Program, 2018).Attribution of nitrate trends has been the focus of many coupled climate/surface water quality regional trend studies.In the Mississippi River (MR) Basin, correlation methods like Kendall's tau were used to explore the role of antecedent moisture conditions on nitrogen export (Murphy et al., 2014).The effect of large-scale climate phenomena, like the ENSO, on interannual variation in MR nitrate loads was examined using multivariate autoregressive state space modeling (Smits et al., 2019), and a stepwise regression approach was used to show the effects of current and historical (1 year lag) precipitation on monthly and annual MR nitrogen loads (Baeumler & Gupta, 2020).Spatially referenced regression approaches have shown that terrestrial nitrogen losses associated with changes in climatic forcing, specifically annual temperature and precipitation, may offset the effects of increased nitrogen loading to the Chesapeake Bay on decadal time scales (Ator et al., 2022;Chanat & Yang, 2018).National-scale studies have used empirical regression models to examine the effects of precipitation and temperature impacts on increasing nitrogen loads (Ballard et al., 2019;Sinha et al., 2017).The Mann-Kendall test was also used to examine the relationship between 27 different climate indices and nutrients, sediment, major ions, and carbon at a national scale (Ryberg & Chanat, 2022).
Climate change impacts on groundwater quality have received less attention than climate change impacts on surface water quality.Singular spectrum analysis was used to examine the role of climate cycles, such as PDO and the ENSO, on groundwater chloride and nitrate concentrations (Gurdak et al., 2007) and Levy et al. (2021) assessed the percent area of the Central Valley with year-to-year groundwater level and nitrate changes to better understand the extent to which improving or degrading groundwater quality co-occurred with rising or declining water levels, which can be influenced by climate.

| Examples of methods used to evaluate land-use effects on water resources trends
Here, we provide some examples of the methods used to evaluate the impact of land use on the four water resources domains.Stationarity in historical land-use conditions cannot be assumed in driver attribution analysis, as humans have been altering the natural landscape for centuries (Klein Goldewijk et al., 2011).In the CONUS, half or more of the historical forest extent in the east has been cleared in the last three centuries, with forest area rebounding during the twentieth century as agricultural activities shifted westward (Sohl et al., 2016).Expansion of agriculture in the Midwest and Great Plains reduced native prairie and wetland vegetation (Davidson, 2014;Sohl et al., 2016).Urban growth has also altered the landscape (Sohl et al., 2016), and more recently (from 1992 onward), exurbanization, the process by which urban residents move into rural areas, has increased, particularly in the South, Southwest, and Mid-Atlantic (Falcone et al., 2018).
The effect of land use on surface water quantity trends is often examined using methods that also account for climate and water use.
For example, the National Hydrologic Model Precipitation Runoff Modeling System has been use to evaluate the effects of precipitation, air temperature, and land use on changes in streamflow (Hodgkins et al., 2020), and Kemter et al. (2023) developed machine-learning models to better understand the relative influences of climate and land use on streamflow trends at a national scale.A time-varying Budyko framework and the Mann-Kendall test have been used to investigate the influence of both climate and land use on hydrology at continental and regional scales (Ayers et al., 2019;Li & Quiring, 2021).Land use and water use are tightly linked drivers of groundwater quantity trends, and changes in groundwater levels or storage have been determined by examining the variation of trends as a function of dominant land use.For example, areas of the glacial aquifer system, which has relatively low human influence tended to have higher percentages of significant increases for mean ground-water levels than wells that had a high human influence (Hodgkins et al., 2017;Scanlon et al., 2005).developed landscapes, on nutrient trends.Nonlinear models were used to evaluate the variable impacts of agriculture on nitrate loading trends in Iowa streams (Green et al., 2014), and WRTDS has been used to partition and assign water quality trends to changes in streamflow as compared to watershed management, which often varies depending on land use (Murphy & Sprague, 2019).Finally, structural equation modeling has been used to link climate and land use to pesticide trends (Ryberg, Stone, & Baker, 2020).
Groundwater quality trends can be attributed to land-use change by establishing well networks that control for the dominant land use surrounding each well, and networks located in agricultural areas have been compared to well networks located in urban or undeveloped areas (Lindsey et al., 2018).Associating a land-use change with a concurrent change in groundwater quality is complicated by the lag time between when a change occurs on the land surface and when the impact on water quality is observed at a well.Tracer-based groundwater ages have been used to estimate groundwater lag times and link land-use and groundwater quality changes (Tesoriero et al., 2007).In sum, these driver attribution studies demonstrate that climate and land-use drivers rarely occur in isolation, and changes in one driver may amplify or lessen the effect of the other.

| G APS AND P OTENTIAL SOLUTI ON S
Continued development of datasets and methodologies to assess changes in surface and groundwater quality and quantity remains important.However, to better address current demographics and potential future shifts based on climate change and other pressures (Villarini & Wasko, 2021;Vörösmarty et al., 2000;Wada & Bierkens, 2014), improvements in water resources trend assessments are needed.The concept of water availability is well-suited to address the aforementioned pressures (Liu, Yang, et al., 2017;Pastor et al., 2014), as the term water availability describes the spatial and temporal distribution of water resources that are required to meet human and ecosystem needs (Evenson et al., 2018).The four water resources domains discussed at length in the previous section can be used to document changes in some components of water supply but linking these to a water availability framework requires the development and integration of additional capabilities.
Here, we identify some of the challenges or gaps related to water resources trend assessments and present potential solutions to improve the outcomes of future trend assessments to better inform critical changes in water availability.

| Produce trend results more efficiently
Management of water resources relies on timely assessments of trends.However, some of the surface-and ground-water quality trend assessments highlighted in Recent Advances Section 2.3 (Murphy, 2020;Shoda et al., 2019;Stets et al., 2020) were published in 2019 and 2020, reporting on trend results with a period of record ending in 2012.The lengthy time lag between the end of the trend period of record and the publication of the trend results limits the use of this information for management, which requires more immediate insights into the magnitude and direction of changes in water resources.Using automated or semi-automated coding workflows for data harmonization and trend analysis, leveraging increased capacity to store harmonized datasets and trend results, and improved access to high-performance computing to complete trend analyses, will help ensure shorter time lags between data collection and regularly updated trend results.

| Improve spatial coverage of trend results
It is important to increase the spatial representativeness of trend results because information captured by monitoring data is often at a different scale than what is desired for water resource decision-making.For example, streamflow trends at the mouth of the Mississippi River provide spatially and temporally integrated information for approximately 40% of the CONUS.However, insights on sub-watershed trends and drivers cannot be inferred without additional information at finer spatial scales.
Of the four water resources domains, empirical streamflow data have the richest spatial coverage, and this section will focus on methods to provide improved spatial coverage of streamflow trends.Despite impressive data collection and synthesis efforts, there is a relatively low spatial density of monitoring stations relative to the large and varied environment.For example, the 8500 streamgage sites (Eberts et al., 2019) only represent 0.3% of the 2.7 million stream reaches in the National Hydrography Dataset Plus Version 2 (Moore & Dewald, 2016).In addition, relatively few monitoring stations have the requisite record length and data completeness for robust trend analysis.Because of these data limitations, many of the studies cited in Recent Advances Section 2.3 used fewer sites, and the number of streamgages ranged from 345 to 2482 (Archfield et al., 2016;Dudley et al., 2020;Rice et al., 2015).
Generalizing hydrologic trends observed at monitoring locations to unmonitored areas to increase the spatial coverage of trend results typically involves geostatistical/regional (McCabe & Wolock, 2014), process-based (Hodgkins et al., 2020), or artificial intelligence/machinelearning (Miller et al., 2018;Rice et al., 2015) models, which first predict daily or monthly streamflow and then derive long-term trends from these predictions.While process-based models hold the greatest potential for understanding physical drivers, reproducing observed longterm trends is often not a focus of model calibration, which may result in poor predictive performance of trends.Hodgkins and others (2020), for example, found that the process-based National Hydrologic Model performed more poorly than geostatistical methods for estimating longterm trends in undisturbed catchments.An intercomparison of 14 process-based models with CONUS-scale streamflow predictions found general agreement among models in trend direction but disagreement in trend significance at the regional scale (Saxe et al., 2021).Hybrid approaches to correct process-based models using machine-learning approaches may yield further improvements in the ability to generalize trend observations to unmonitored areas (Konapala et al., 2020).In addition, the methods that are developed to improve the spatial coverage of trend analyses can also be used to better understand drivers of change, as climate and land use are commonly used input datasets into the machine-learning models.

| Account for groundwater influences on surface water trends
Surface and groundwater are often managed and regulated separately (Scanlon et al., 2023), and results for groundwater and surface water trends are also often reported separately (Recent Advances, Section 2.3).This separation is in part due to response lags in groundwater quantity and quality which typically develop at decadal to century scales, whereas surface water systems respond much more rapidly, reflecting short-term (seasonal and interannual) dynamics that are substantially muted in groundwater systems.In addition, the accurate estimation of the perturbation lag time between a typical surface and its connected groundwater system is challenging (Sanford & Pope, 2013;Van Meter & Basu, 2015).Furthermore, the structural boundaries for surface and groundwater are not congruent.Surface water is typically organized by topographically defined catchments or watersheds, while groundwater is organized by subsurface-defined aquifer systems.
Despite the separations that appear in groundwater and surface water management and trend evaluations, these two resources are inherently connected; changes in groundwater quantity or quality often propagate into connected surface waters.For example, streams that are hydraulically connected to an aquifer are subject to volume reductions because of decreasing groundwater discharge into the stream and/or by inducing seepage from the stream into the aquifer (Konikow & Bredehoeft, 2020;Konikow & Kendy, 2005;Perkin et al., 2017).In addition, surface water quality issues related to elevated salinity and nitrate concentrations can be traced back to watersheds where streams are influenced by groundwater.Runoff or leaching processes transport excess salt and nitrate used in deicing of urban areas or fertilization of agricultural fields into groundwater reservoirs, which are then released as a legacy sources to streams with lag times on annual to decadal time scales (Corsi et al., 2015;Rumsey et al., 2023;Tesoriero et al., 2013).
Indicators that account for the link between groundwater and surface water systems can provide essential information to better understand processes and drivers governing water resources change to inform management.Baseflow index (BFI) integrates information from surface water and groundwater; in the absence of any precipitation in the basin, baseflow should match recharge, thus providing an estimate of groundwater contributions to streamflow.Lin et al. (2021) found that it was necessary to account for both land use and baseflow to explain the variability in stream nitrate concentrations at a national scale.In agricultural areas, the BFI was positively related to stream nitrate, but the opposite was true in areas with lower agricultural inputs; there baseflow was not a nitrate source.
New datasets and modeling approaches could be used to better link surface and groundwater in future trend assessments.Modeling approaches to estimate BFI time series at streamgage locations have been implemented at a national scale (Foks et al., 2019).National-scale maps of recharge and groundwater discharge as baseflow are also now available (Reitz et al., 2017;Zell & Sanford, 2020), and paired air and water temperature analysis can also provide a similar characterization of baseflow contributions to streamflow, with the added benefit of assessing the depth of groundwater contributions (Hare et al., 2021).Datasets of age-tracers in well samples and advancements in modeling capabilities now make it possible to provide information on groundwater residence times at regional scales, even in locations without numerous empirical well samples (Green et al., 2021;Tesoriero et al., 2019).The BFI estimates and maps, paired water air and water temperature, and age-tracers would have to be adapted for use in a trend context, but they show promise as datasets to identify some important processes and drivers of changes in water resources.

| Account for water quality influences on water availability
Global-scale models that are built to assess scarcity and availability issues are multifaceted, incorporating factors such as energy, water, land, economy, and climate, but they typically only account for the influence of water quantity, not quality (Cui et al., 2018;Evenson et al., 2018;Schewe et al., 2014).New approaches have been developed to place water quality into a water scarcity context by accounting for the extra water withdrawal required to dilute contaminated water to obtain water of acceptable quality for each water-use category (Ma et al., 2020;van Vliet et al., 2021).The inclusion of surface and groundwater quality trends would also benefit management, as trends in a particular contaminant concentration could also indicate that the amount of water required for dilution has changed.In the CONUS, increasing nitrate and salinity concentrations have been linked to reduced water available for drinking, irrigation, and thermoelectric power generation uses (Pan et al., 2018;Pennino et al., 2017;Rumsey et al., 2021).Therefore, the development of a modeling framework that accounts for both quality and quantity changes is needed for water availability assessments conducted at a national scale.
In addition to the modeling framework described above, another potential way to integrate water quality changes into water availability assessments is to expand on some established approaches described in Recent Advances, Section 2.3 (Lindsey & Rupert, 2012;Shoda et al., 2019).These studies indicated how water availability for human use changed by comparing water quality trends to drinking water standards.Future trend assessments that continue to compare the magnitude and direction of surface and groundwater water quality trends will require updated information on potential risk of water quality contaminants in drinking water, and this information is available in the U.S.
Environmental Protection Agency ECOTOX database (Olker et al., 2022).However, there are additional important human use categories to also consider if the application of water resources trends is extended into an evaluation of changes in water availability.Poor water quality can influence the amount of irrigation water available in agricultural systems (Fipps, 2003) or feed water available for cooling systems at thermoelectric power generation plants (Pan et al., 2018).Thresholds for these additional use categories would need to be accounted for if water quality trends were to be used in this expanded water availability context.

| Balance human water use with ecosystem needs
The tension between human water use and ecological needs has been well documented (Poff & Zimmerman, 2010;Richter, 2014).Globally, the inclusion of national-scale water resources to support ecological communities has been recognized as an important priority, and many nations identify the protection of water resources to support ecological needs in their water policies (Le Quesne et al., 2010).Documenting the effects of surface and groundwater quantity and quality trends on water availability to support healthy aquatic environments, otherwise referred to as ecological use (Evenson et al., 2018;Kennen et al., 2018), can be used to inform management.Shoda et al. (2019) compared changes in river nutrient concentrations to thresholds related to the trophic status of rivers (Recent Advances, Section 2.3).This type of trend evaluation benefits ecosystem needs because the maintenance of nutrient concentrations below these thresholds reduces the likelihood of eutrophication; anoxic or hypoxic conditions, as a result of eutrophication, can pose a significant risk to aquatic life.Unlike drinking water standards, which are consistent at a national scale, nutrient criteria vary by ecoregion, and many rivers and streams cross ecoregional boundaries.The spatial mismatch between trend metrics capturing changes in rivers and the thresholds based on ecoregional boundaries will need to be accounted for if water quality trend results are going to be integrated into evaluations of changes in water availability at a national scale.
Depleted high flows, homogenization of flows, and erratic flows are useful metrics to evaluate how streamflow changes may affect aquatic ecosystem fish and invertebrate communities (Carlisle et al., 2017), but the establishment of related thresholds to protect ecological communities can be contentious (Capon & Capon, 2017;Kennen et al., 2018).A presumptive standard calls for the maintenance of altered streamflow within a percentage-based range of natural flows (Meador & Carlisle, 2012;Richter et al., 2012), and narrative thresholds have been adopted by some states (Novak et al., 2015).However, the lack of a standardized set of thresholds complicates the evaluation of streamflow trends in the context of ecological use at a national scale.
Issues related to thresholds used to evaluate the changes in water availability for ecological use remain, but changes to the types of metrics included in trend studies could improve water resources management for ecological use.Many water quality trend studies evaluated in the Recent Advances, Section 2.3 (Kaushal et al., 2018;Murphy, 2020;Shoda et al., 2019;Stets et al., 2020) reported annual mean concentrations trends.However, annual changes in water quality concentrations may not be at an adequate temporal resolution to detect when water quality changes imperil supplies for ecological use (Moore et al., 2020;Qian, 2015).Additional metrics including seasonal trends (Kelly et al., 2019;Molot et al., 2022;Rumsey et al., 2023) or changes in the likelihood of a given constituent exceeding a threshold (Corsi et al., 2015;Worrall et al., 2020) may be more useful.

| Continue improvement of driver attribution studies
Driver attribution is critical to understanding the factors influencing changes in water resources, and one challenge highlighted in Recent Advances, Sections 2.4 and 2.5, was that many drivers are interrelated, so it may be difficult to tease apart the effects of one driver from another to determine which one has the largest influence on a trend.Improvement of existing datasets may help address this challenge.
For example, lower precipitation rates may reduce streamflow and lead to increased groundwater withdrawal as water purveyors switch from using surface water to groundwater.Understanding the relative importance of drivers of streamflow trends in this scenario requires information about surface and groundwater quality trends, the contributions of ground to surface water, and information related to climate and water use.Methods to account for climate change and groundwater and surface water connections and water use have been developed (Russo et al., 2014;Russo & Lall, 2017), and solutions to better account for groundwater influences on surface water were outlined in Gaps and Potential Solutions, Section 3.3.However, water use has not yet been explicitly addressed.Withdrawal estimates for various use types (domestic, irrigation, livestock, aquaculture, industrial, mining, and thermoelectric power) have been available at 5-year intervals and the state level (Dieter et al., 2018), but improved spatial and temporal resolution of water use information (Evenson et al., 2018;Marston et al., 2022) could be used to help discern the impacts of interrelated climate and human water use drivers on surface and groundwater quantity trends.
Another challenge related to trend attribution is that many drivers influence water resources trends, beyond climate and land use, but datasets to account for these influences in a modeling framework are not available (Irvine et al., 2015;Murphy, 2020).For example, land-use management can influence river salinity and alkalinity trends through the use of agricultural lime to improve crop yields or salt to deice roads (Kaushal et al., 2018;Stets et al., 2014); agricultural best management practices (BMP), like adding riparian buffer strips or using cover crops, can reduce the transport of surplus nutrients from fields to streams (Liu, Engel, et al., 2017).Data for national scale road salt applications are available (Bock et al., 2018), but comprehensive national-scale data on liming practices or BMP implementation are not (Bock et al., 2018;Stackpoole et al., 2021;Stets et al., 2014).Development of these additional datasets is needed to better tease apart the effects of multiple drivers on changes in water resources trends.
In conclusion, continued development of datasets and methodologies to assess changes in surface and groundwater quality and quantity remains important.However, to advance future water availability trend assessments and better inform the management of water resources, the development and integration of additional capabilities is needed.Among these are timely and efficient delivery of trend results; the capability to estimate trends in ungaged areas; increased understanding of groundwater-surface water interactions; better quantification of water use for human needs (Evenson et al., 2018); and, explicit representation of ecological flow requirements (Freeman et al., 2022; Figure 2).In addition, methods to account for how changes in quality affect total estimates of water supply and the development of improved driver attribution datasets to distinguish the interrelated impacts of land use, human water use, and climatic influences are also needed.
analysis, and lag correlation were used to quantify the effects of various climate indices like ENSO, North Atlantic Oscillation (NAO), Pacific Decadal Oscillation (PDO), and Atlantic Multidecadal Oscillation (AMO) on precipitation and groundwater levels across principal aquifers of the CONUS(Kuss & Gurdak, 2014;Russo & Lall, 2017), andRusso and Lall (2017) employed a wavelet analysis to show declining groundwater levels in deep aquifers are likely due to increased groundwater abstraction to offset shallow water supplies impacted by climate change.
Land use can have large impacts on water quality.Stets et al. (2020) assessed changes in the estimated marginal means of nutrient and major ion concentrations based on -dominant land use.Multiple linear regression(McIsaac et al., 2016) and spatially referenced regression models like SPARROW(Chanat & Yang, 2018) have been used to assess the influence of point sources, which are often associated with more Conceptualization; visualization; writing -original draft; writing -review and editing.Gretchen P. Oelsner: Conceptualization; project administration; writing -original draft; writing -review and editing.Edward G. Stets: Conceptualization; writing -original draft; writing -review and editing.Jory S. Hecht: Conceptualization; writing -original draft; writing -review and editing.Zachary C. Johnson: Conceptualization; writing -original draft; writing -review and editing.Anthony J. Tesoriero: Conceptualization; writing -original draft; writing -review and editing.Michelle A. Walvoord: Conceptualization; writing -original draft; writing -review and editing.Jeff G. Chanat: Writing -original draft.Krista A. Dunne: Writing -original draft.Phillip J. Goodling: Writing -original draft; writing -review andF I G U R E 2Components of future integrated national-scale water resources trend assessments.Information from the four water resources domains remain at the core of trend assessments (from Figure1), but the development and integration of additional capabilities are needed and include increased understanding of groundwater-surface water interactions (blue arrows); better quantification of water use for human needs, and the refinement of trend metrics to account for the competing needs of society and ecological integrity (middle blue circle).Changes in water resources will need to be evaluated within the context of large-scale driving forces like climate, land use, and human water use (outer orange circle).

TA B L E 1
Examples of studies reporting on trend results across the four water resources domains.

a Water resources trend domain Time period Metric used in trend analysis Trend result Reference
Year-long surface area, including lake, reservoirs, rivers, streams, and ponds Year-long surface area decreased in the Southwest and Northwest; increased in the Southeast, Northern Great Plains, and Southern Midwest Zou et al., 2018 Surface water quantity 1980-Streamflow intermittency measured as no-flow duration, dry-down period, and no-flow timing Increased intermittency in southern U.S., decreased intermittency in northern U.S. Zipper et al., 2021 Groundwater quantity 1900-Groundwater depletion rate Decreases in groundwater volume; largest decreases were in the parts of the Gulf Coastal Plain, High Plains aquifer, Arizona Alluvial Basin, and the Central Valley, CA.