Concurrently assessing water supply and demand is critical for evaluating vulnerabilities to climate change

Aligning water supply with demand is a challenge, particularly in areas with large seasonal variation in precipitation and those dominated by winter precipitation. Climate change is expected to exacerbate this challenge, increasing the need for long‐term planning. Long‐term projections of water supply and demand that can aid planning are mostly published as agency reports, which are directly relevant to decision‐making but less likely to inform future research. We present 20‐year water supply and demand projections for the Columbia River, produced in partnership with the Washington State Dept. of Ecology. This effort includes integrated modeling of future surface water supply and agricultural demand by 2040 and analyses of future groundwater trends, residential demand, instream flow deficits, and curtailment. We found that shifting timing in water supply could leave many eastern Washington watersheds unable to meet late‐season out‐of‐stream demands. Increasing agricultural or residential demands in watersheds could exacerbate these late‐season vulnerabilities, and curtailments could become more common for rivers with federal or state instream flow rules. Groundwater trends are mostly declining, leaving watersheds more vulnerable to surface water supply or demand changes. Both our modeling framework and agency partnership can serve as an example for other long‐term efforts that aim to provide insights for water management in a changing climate elsewhere around the world.


| INTRODUC TI ON
The overriding challenge of managing water resources, especially in regions with large seasonal variation in precipitation and those dominated by winter precipitation, is aligning water supply with demand.This challenge will only become more difficult in the future as the climate changes-affecting the amount and timing of water supply and demand (Cai et al., 2015)-and as the population grows-increasing demands (Butler & Memon, 2005).To manage available water resources under these changing conditions, water managers need reliable projections of where and when water is available (water supply) and who and how much water users will require (water demand).However, conducting water supply and demand projections is difficult in large watersheds with complex system dynamics.Tailored models require massive amounts of data to adequately represent the breadth of physical and social factors affecting water supply and demand, must address substantial uncertainty in their projections, and often require significant stakeholder input, which can be difficult to integrate into modeling efforts.
Obtaining data to parameterize and validate process-based models in large watersheds, including discovery, privacy, access, quality control, compatibility, and integration, can pose significant challenges, especially when data are collected at different scales and housed across multiple agencies.Despite efforts to mitigate these challenges, complete, directly accessible, and integrated datasets are still rare.Assembling the necessary datasets sometimes requires significant iterative stakeholder input and coordination across multiple agencies, along with trustbuilding, collaboration, clear data-sharing agreements, transparency, and accountability.Such efforts are time-and resource-intensive, and commonly underappreciated.
Even with access to necessary resources and data, water supply and demand projections remain prone to uncertainty, making it difficult to capture a large watershed's complexity.Projections become less certain over longer future timeframes (Done et al., 2021) and depend on the assumptions made about responses and future actions of individuals and entities operating within the system.For example, water management agencies, who are often the main audience of these types of analyses, may be influenced by the projections themselves, making the actions and projections interdependent.
Finally, while incorporating stakeholder input into water planning has been shown to improve water management (Megdal et al., 2017), stakeholders are rarely integrated into the process of developing supply and demand projections (though see Austin Water, 2018;New Mexico Interstate Stream Commission, 2018; USBR, 2012; Utah Division of Water Resources, 2021 in table S1).There are multiple challenges to the kind of co-production necessary to include stakeholders in efforts to quantify future water availability.These include difficulties in resolving differences in the needs and goals between researchers and stakeholders and the added capacity and time to build relationships and understanding of each other's needs.Teams must also work to address the differences in terminology, experiences and norms, institutional structures, and bureaucracies that do not lend themselves to collaborative efforts or sharing of knowledge and data, all while being hampered by a general lack of training and expertise in carrying out co-production successfully (Djenontin & Meadow, 2018;Howarth et al., 2022;Lemos et al., 2018).
For these reasons, long-range projections that are co-produced with water managers to directly inform water management decisions are relatively uncommon and are rarely reported in the peer-reviewed literature (e.g., Ahmadi et al., 2020;Cross et al., 2017;Haque et al., 2014;Harma et al., 2012;Heidari et al., 2021;Lacroix et al., 2016;Zamani Sabzi et al., 2019).However, a number of such long-term water supply or demand forecasts or projections have been carried out, primarily in the more arid watersheds of the western United States (see examples in Table S1).Many of these analyses are not published in academic journals, but rather exist as agency produced reports made for direct use by decision-makers such as city and state governments, or watershed level decision-making entities (e.g., Austin Water, 2018;New Mexico Interstate Stream Commission, 2018;State of Oregon Water Resources Department, 2015;Utah Division of Water Resources, 2021;Wyoming Water Development Commission, 2007; Table S1).They vary in terms of their temporal and spatial scale, their focus, the sophistication of their methods, and whether they include supply, demand, or both (Table S1).While these reports may be useful for site-specific decision-making, a lack of representation in the academic literature means that valuable information on modeling processes, methods, and results from such studies are unlikely to inform future research efforts (Brown et al., 2015).Meanwhile, peer-reviewed supply and demand analyses (e.g., Christensen & Lettenmaier, 2007;Harma et al., 2012;Heidari et al., 2021;Table S1) tend to not be explicitly tied to management and decision-making, and their results are consequently less likely to be translated into actionable outcomes.
Our work has sought to fill this gap between academic research and decision tools for management by publishing, within the academic literature, projections of water supply and demand that were developed in partnership with a state-level decision-making body, Washington State Department of Ecology's Office of Columbia River (OCR).This partnership has been built over 15 years (see Hall et al., 2016Hall et al., , 2022;;
The Columbia River is the fourth largest river in the US by discharge volume (237 billion m 3 , US Census Bureau, Economics, & Statistics Administration, 2011) and is the main water source in a region that is home to about eight million people (Zhang et al., 2021).More than 20,000 km 2 are irrigated from water originating in the CRB (National Research Council, 2004), and agriculture is an important economic sector, with total agricultural cash receipts totaling $10.1 billion in Washington in 2021 (USDA, 2021) and food processing generated an additional $21.8 billion in revenues (WSDA, 2022).The Columbia River system has more than 400 dams, and hydropower covers an estimated 60%-70% of the electricity needs in the Pacific Northwest (USBR, 2021).The CRB is also home to multiple threatened and endangered species of fish dependent on its cold-water flows and habitats to persist.
Annual average temperatures in the Pacific Northwest have risen by about 0.8°C compared to 1901-1960 averages and are projected to further increase an additional 2.0°C (RCP4.5) to 2.6°C (RCP8.5)by mid-century (2036-2065;Vose et al., 2017).Precipitation changes in the region are less certain, but a shift toward drier summers and more precipitation occurring in other seasons is expected (Easterling et al., 2017;Mote & Salathé, 2010).These climatic changes are expected to have significant impacts on streamflow (Hamlet & Lettenmaier, 1999), agriculture (Rajagopalan et al., 2018;Scarpare et al., 2022), energy production (Northwest Power and Conservation Council, 2018), and habitat quality and suitability for cold-water fish species (Hare & Francis, 1995;Mantua et al., 1997;McClure et al., 2003;McGinnis, 1995).The economic and environmental importance of the CRB and the combined impacts expected from climate change make analyses of changes in water supply and demand valuable for OCR and water users in Washington.Additionally, the national importance of the CRB makes it an informative case study that can provide insights that are transferable to other large, complex river basins.

| ME THODS
At the center of the Columbia River Forecast is the integrated modeling of surface water supply (the dominant source of water to meet outof-stream uses in Washington; Lane & Welch, 2015) and agricultural water demand (the dominant out-of-stream water use in Washington; Fasser, 2018) for a historical  and future period (2040: 2026-2055) across the whole CRB.However, to serve the needs of our partner, OCR, we included more detailed data within the Washington portion of the CRB in this modeling (Figure 1), and we expanded our analysis (using simpler methods) to also (a) quantify residential water demand, (b) evaluate impacts of supply and out-of-stream demand patterns on instream flows, and (c) evaluate groundwater trends (Table 1).These additional lines of evidence supported a more robust evaluation of vulnerabilities in water availability due to changes in water supplies and demands in eastern Washington.We also modeled results for another future time period (2070: 2056-2085), which is not included here because most results did not substantially change from 2040 to 2070 and not all data was available for the 2070 time period.Results from the 2070 analysis are available in Hall et al. (2022).

| Modeling framework
We used an integration of three computer models-Variable Infiltration Capacity (VIC; Liang et al., 1994), CropSyst (Stöckle et al., 1994(Stöckle et al., , 2003(Stöckle et al., , 2010)), and R-ColSim (a variation of ColSim, Hamlet & Lettenmaier, 1999)-to simulate surface water supply, out-of-stream agricultural demands, and large reservoir management.VIC (Liang et al., 1994) is a large-scale, process-based water and energy balance model that has been F I G U R E 1 The Columbia River Basin, Eastern Washington, and the Columbia River Mainstem, which constitute our study area.Due to the partnership with the Office of Columbia River in Washington State, additional focus is provided in the eastern Washington portion of the Basin, and the 2021 Water Supply and Demand Forecast modeling focused on areas upstream of the Bonneville Dam.calibrated and used in other studies in the Pacific Northwest, including parts or the whole CRB (Hamlet et al., 2013;Mantua et al., 2010), in the nearby Colorado River Basin (Christensen & Lettenmaier, 2007), and in the Sierra Nevada of California (Maurer, 2007).CropSyst simulates the growth and phenology of numerous annual and perennial crops under both irrigated and dryland conditions and under varying management regimes (Stöckle et al., 2003), and has been calibrated and applied in Washington State and the CRB (Stöckle et al., 2010).VIC and CropSyst have been coupled into the VIC-CropSyst v3.0 model that simulates the hydrologic cycle, soil water budgets, crop growth, and crop yield to quantify the effects of climate change and crop production scenarios on different outputs (see details in Malek et al., 2017).VIC-CropSyst has been well calibrated in hydrological and crop parameters and used to study climate change impacts on crop yields and irrigation technologies in the CRB and its tributary, the Yakima Basin (Malek et al., 2020(Malek et al., , 2021;;Rajagopalan et al., 2018;Yourek et al., 2023).
Reservoir modeling with R-ColSim (the new version of ColSim coded in the open-source computational software, R: K.M. Malek, M. Yourek, J. Adam, A. Hamlet, K. Rajagopalan, and P. Reed, under review; R Core Team, 2022) captured operations of 34 storage and run-of-river dams along the Columbia River and its major tributaries (Snake, Pend Oreille, Flathead, and Kootenai Rivers).Dam operating rules used within the model (hydropower production, flood control, and flow targets) are based on Hamlet and Lettenmaier (1999), with minimal modification to capture important changes to the rules since 1999 (Alan Hamlet, personal communication).Due to the regional importance of the Yakima River Basin, we used a separate reservoir model, Yakima-RiverWare (Malek, Adam, Stockle, et al., 2018;Vano et al., 2010), to incorporate reservoir operations and water management into streamflow estimation.
We obtained water supply estimates from both the unregulated streamflow outputs of VIC-CropSyst and the regulated streamflow outputs of R-ColSim.We simulated the growth and development of over 100 different crops with VIC-CropSyst to capture the diversity of eastern Washington's crop mix.We obtained agricultural water demand estimates from the crop water requirement outputs of VIC-CropSyst, which we adjusted with estimates of surface water and groundwater withdrawal splits and conveyance losses.
Results within eastern Washington were summarized for 34 Water Resource Inventory Areas (WRIAs; WA ECY, n.d.-a).Many WRIAs correspond directly to watersheds, though some are additionally constrained by jurisdictional lines.However, for simplicity, we refer to them as watersheds throughout.In response to OCR's management focus, there are two aggregations that did not follow these boundaries, because, jurisdictionally, they are managed together.These are the Yakima River (WRIAs 37, 38, and 39) and the WRIAs in Douglas County, Washington (WRIAs 44 and 50).Though these aggregations complicate comparisons across watersheds, the benefits of how they can inform management outweigh these complications.

| Model calibration
To calibrate the VIC model for modeling streamflow, we selected 274 gauges from multiple sources, focusing on those with sufficient data for the appropriate time period and capturing flows from sufficient drainage areas.From these stations, 81% have very good performance (i.e., NSE equal or greater than 0.75) and 13% have good performance (i.e., NSE is between 0.65 and 0.75).More information can be found in Adam et al. (2022) and in Table S2.
We also performed a low-data approach crop parameterization and calibration for yield and evapotranspiration under irrigation conditions using the date of occurrence of the main phenological events, in addition to the yield or biomass information at final harvest.Twenty-five crops TA B L E 1 Summary of the components of the Columbia River Forecast and the geographic scopes for which results are presented and discussed.Detail references for each component are provided in the Modeling framework section, below.

| Input data
VIC-CropSyst is run using daily precipitation, maximum and minimum temperature, maximum and minimum relative humidity, shortwave solar radiation, and wind speed.We applied VIC-CropSyst across the whole CRB, downstream to the Bonneville Dam (Figure 1).For the historical simulations (1986-2015 water years, which start October 1 of the previous year and run through September 30), we used climate data from gridMET (Abatzoglou, 2013) for the U.S. portion of the study area (using all above mentioned observed climate variables) and Livneh et al. (2013) for the Canadian portion (using only daily maximum and minimum temperature, precipitation, and wind speed).
To model future conditions, we ran VIC-CropSyst using projected daily weather data from the 2026 to 2055 water years for the 2040 estimates.We used projected daily weather data from 17 General Circulation Models (GCMs) in CMIP5 (Taylor et al., 2012), and two emissions scenarios, RCP 4.5 (Thomson et al., 2011) andRCP 8.5 (Riahi et al., 2011), which have been downscaled using the Multivariate Adaptive Constructed Analogs method (Abatzoglou & Brown, 2012) et al., 2020) for other U.S. states, and from MODIS data by using the same methodology as Pervez and Brown (2010) for Canada.We used 462 and 495 ppm as 2040 carbon dioxide concentrations for RCP 4.5 and 8.5, respectively.We report on the crop phenology and management inputs used in Adam et al. (2022).
We used water management data to estimate curtailment and instream flow deficit frequencies (see details in the Section 2.3, below).We identified interruptible water rights, their points of diversion, and places of use based on data from the Washington Department of Ecology's Water Rights Tracking System.We combined this information with the 2018 WSDA cropland data layer to determine the area and grid cells where curtailments would occur.In the Yakima River Basin, we used a different approach (Hubble, 2012) due to the differences in how interruptible water rights are managed in this basin: naturalized flows, which we used to calibrate and correct bias in estimates of simulated streamflow, were collected from the Bonneville Power Administration's no regulation, no irrigation streamflow dataset (BPA, 2014).If naturalized flows were not available from the BPA source, we used the naturalized flows from Elsner et al. (2010).Operation rules used for reservoir modeling are those used by Hamlet and Lettenmaier (1999) for the reservoirs represented in R-ColSim and by the U.S. Bureau of Reclamation for the Yakima River Basin (USBR, 2002).Instream flow targets, which dictate curtailment decisions on multiple tributaries and at multiple locations along the Columbia River mainstem, are based on Washington Administrative Codes (RCW, 1998).

| Scenarios
We used the integrated models to explore four scenarios (summarized in Table 2).To explore the impacts of changing planting dates in response to warming we considered a scenario where planting would be a week earlier, on average, by 2040.To project future crop mix, we quantified trends in the area of different crops from 1999 to 2019 by fitting statistical models to survey data from the USDA National Agricultural Statistics Service for Washington State (USDA-NASS, 2019), and then extending these trends to estimate the proportions of different crops at the state level in 2040 (Figure 2).These estimated proportions were then assigned to each modeling grid cell based on the crop types in that grid cell in 2020, determining the projected crop mix for 2040.
We compared scenarios 1 and 2 for the supply results, and we used scenarios 1 through 4 to evaluate the contributions of climate change versus human response factors to changes in agricultural demand.All scenarios included historical water management and existing reservoirs, as we are unaware of any available data to directly inform additional scenarios representing water management and reservoirs 20 years into the future.Similarly, future demand scenarios assumed the extent of irrigated agriculture remained constant.However, OCR estimates that 308.4 million m 3 of additional water will be available by 2040 for out-of-stream uses from planned or in-development water supply projects in Washington (OCR, personal communication).We did not have sufficient detail to use this estimation in modeling scenarios, but it was incorporated into our evaluation of expected changes in out-of-stream demands across eastern Washington.

| Data and historical estimates
We defined residential demand, the second largest out-of-stream use in the state (Fasser, 2018), as water that is used in or around the home, both within municipalities supplied by public or private community water providers, and self-supplied domestic use outside of municipal boundaries.This does not include water used for industrial or commercial purposes.
We only estimated the average monthly residential demand in eastern Washington, using publicly available data obtained from the most recent Comprehensive Water System Plans of large water provider systems (serving >1000 connections; these are updated every 7 years).
In some cases, the data included historical and projected population growth, water use, and water rights information.Forty-five large water provider systems in eastern Washington had sufficient data to be included in the analysis.We assumed all remaining household water users were considered domestic water use, and we estimated their demand using the U.S. Geological Survey 2015 Water Use Report, which provides county-level data on annual per capita water use for domestic and self-supplied categories (Dieter, 2018).To estimate monthly values, we assumed that domestic water use within a county would have similar monthly patterns to mean monthly municipal water use within that same county.To aggregate residential demand results for each watershed we assumed that where a county or city area spanned more than one watershed, the partial demand assigned to each watershed was proportional to the percent of the city's or county's land area in that watershed.
While the integrated modeling accounts for both the consumptive use of irrigation water by crops and drainage that returns to the system downstream, these estimates of residential demand do not make that distinction.We therefore needed to estimate consumptive use in the residential sector.We distinguished between indoor and outdoor water use based on a minimum month model that assumes water use during winter months (in our case December, January, and February) is representative of year-round indoor household water use (Mini et al., 2014).
Outdoor water use was assumed to be the remainder of the total water used.We then followed guidance from the Washington Department of Ecology that assumes household consumptive water use for self-supplied water users is 10% (indoors) or 80% (outdoors) of total water use in a given period (Culhane & Nazy, 2015).Estimates of municipal consumptive use were made using the information on source water withdrawals and wastewater discharge locations (for a more detailed description of how municipal consumptive use was estimated, see Adam et al., 2022).

| Projections of future water use
We estimated monthly residential water use in 2040 solely based on projected population data and historical water use.We calculated municipal population projections using information from the relevant comprehensive system plans (Adam et al., 2022).Given that extrapolation TA B L E 2 Scenarios used for the integrated modeling of surface water supply and agricultural water demand.Historical planting dates were obtained from USDA (1997).See the text for further details.

Planting date Crop mix
(1) Historical baseline We obtained domestic population projections from the Washington State Office of Financial Management (2017).We used their low and high projections at the county-level through 2040 to provide a range of projection uncertainty, which is generally higher for rapidly growing and small counties.We estimated future monthly water use for both municipal and domestic sectors by multiplying the relevant municipal-or county-level population by the relevant mean historical per capita monthly water use.

| Instream flows for fish
We are not aware of any comprehensive dataset that quantifies the instream flows required to support sustainable fish populations throughout the Columbia River and its tributaries.We therefore used Washington State adopted instream flows (WA ISF; RCW, 1998) and the Federal Columbia River Power System Biological Opinion instream flows (FCRPS BiOp; NOAA, 2021) as coarse and incomplete measures of the instream water required to fulfill the needs of fish species.We chose these two schemes because of their role in regulating interruptible water rights holders (in the case of the WA ISF) and managing federal dams and the Quad Cities water permit (in the case of the FCRPS BiOp).
We used these instream flow requirements in different ways for different rivers.For tributaries in Washington subject to WA ISF rules, we modeled the frequency and magnitude of curtailments on a weekly basis for both historical and future (2040) periods by comparing modeled water supply (after accounting for agricultural and residential out-of-stream demands) to state instream flow requirements.If the remaining flows were less than the target instream flow in any week, curtailment was reflected in the integrated modeling process by curtailing demand in interruptible grid cells.
The WA ISF rules on the Columbia River itself are addressed differently than the tributaries, and interruptible water rights have only been curtailed once in 2001.Since one data point is insufficient to verify a curtailment model for the Columbia River, we instead estimated the frequency of instream flow deficit at nine dams in Washington State (Adam et al., 2022).We defined an instream flow deficit as a week when water supply at a particular location is insufficient to meet instream flow requirements once agricultural and residential demands have been accounted for.We summarized the instances of instream flow deficits as the frequency of occurrence across the 30-year time windows for the historical and 2040 periods.
Water entitlements in the Yakima River Basin are also handled differently.1905).Under drought conditions, the non-proratable right holders receive their entitlement in full, the proratable water rights users receive a reduced or prorationed portion of their entitlements, and the junior right holders are curtailed in full and receive no water.We analyzed proration levels under the historical baseline and future (2040) climate using the Yakima RiverWare model (Malek, Adam, Stockle, et al., 2018;Vano et al., 2010;Zagona et al., 2001

| Historical trends by aquifer layer
We compiled well depth and depth to water data available across eastern Washington for over 3000 wells.The wells were grouped and analyzed by aquifer layer, based on the Columbia Plateau Regional Aquifer System's (CPRAS) Overburden, Saddle Mountains Basalt, Wanapum Basalt, and Grande Ronde Basalt layers (Burns et al., 2011).Where we had available data, we analyzed trends at locations outside of the CPRAS without distinguishing different aquifers.To inform OCR's needs and to present results at a comparable scale to the surface water results, we aggregated results by groundwater subareas.We defined a subarea as an area with similar hydrogeologic characteristics and groundwater hydraulic connectivity and identified 21 distinct subareas in eastern Washington (Figure 3).
We conducted the trend analysis using only spring high water level measurements to represent the equilibrium state of the aquifer layers.We limited the dataset we used to wells that had at least eight spring high water level records between 2000 and 2020 (EPA, 2009).We evaluated each well individually, using the non-parametric Sen Slope estimator (Sen, 1968) to compute the magnitude of the trend, and the nonparametric Mann-Kendall test to determine which wells had a significant monotonic upward or downward trend (Mann, 1945;Kendall, 1948).The geometric mean trend was computed for each aquifer layer by subarea, where a minimum of three wells were present that met the selection criteria.

| Future vulnerabilities
Future groundwater vulnerabilities were evaluated based on changes in available saturated thickness (AST).We defined AST as the thickness of water between the spring high depth to water level and 20 feet above the bottom of the well, as a measure of the amount of water accessible to the pump.Changes in AST were determined based on a linear projection of the 2000-2020 significant trends forward in time, through 2040.The degree of vulnerability was determined based on the percent change in AST by 2040 and the number of years to a 25%, 50%, and 75% reduction in AST.These thresholds were used to represent when pumps might need to be lowered for continued water supply reliability (25%), a significant reduction in well yields may be observed (50%), and a significant investment or discontinued use may be necessary (75%).

| Co-production and stakeholder interactions
The Forecast was fully co-produced with OCR staff, who engaged in bi-weekly team meetings throughout the length of the project, helping shape the questions asked, methods used, and outcomes and their interpretation, to ensure the research explicitly targeted their concerns and needs.The iterative nature of the forecast allowed the agency to reflect on the past versions of the forecast and deepen understanding on both sides of OCR's goals and needs, as well as the complexities and uncertainties associated with model projections.OCR also provided in-house data that researchers would have been challenged to obtain in the absence of a long-term co-production environment.
OCR facilitated more infrequent, but recurrent, interactions with two other groups.First, OCR's Policy Advisory Group, comprised of representatives from state, local, federal, and tribal governments, irrigators, business, and environmental groups, gave feedback on the overall direction of this Forecast.Second, over the 15 years that the team has been working interactively on Forecasts with OCR, they have convened a "caucus" of their sister state agencies semi-annually.This process has opened avenues for bi-directional data sharing and allowed our research team greater insight into the needs and capabilities of this suite of agencies.In between caucus meetings, the team continued conversations with smaller groups of agency personnel to delve more deeply into specific issues of interest; for example, to explore opportunities to highlight flow issues relating to salmonid fish species.
The team also incorporated a public review process into the Forecast that allowed for feedback with a range of individuals working on water supply issues throughout Washington.Each Forecast was released in draft form, with three virtual or in-person workshops allowing the research team to present the draft results and hear community comments and questions.Written comments were solicited during and outside of workshops, with the joint research-OCR team responding to these comments in the final Forecast (see Adam et al., 2022).

| Changes in surface supply
Overall, annual surface water supply within the Washington portion of the CRB is not expected to change by 2040 (Table 3).The change in timing of surface water supplies is notable, however, with an expected increase in the wet season (November-May) supply of 14.9% and an expected decrease in the dry season (June-October) supply of 28.5% by 2040.The center of timing of supply (i.e., the day of year, starting from October 1st, that represents the temporal centroid of the hydrograph; Stewart et al., 2005) in median flow years across the CRB is thus expected to shift 12 (±2) days earlier by 2040 (Figure 4), a result that is consistent with previous studies (Byun et al., 2019;Stewart et al., 2005).
The shifts in surface water supply vary considerably by watershed (Figure 5a), reflecting variations in the historical rain-to-snow ratio across eastern Washington.Consistent with previous studies (Elsner et al., 2010;Fritze et al., 2011;Hidalgo et al., 2009), watersheds in the central and southern Cascades, which receive a high proportion of their supply from snowmelt (measured as the snowmelt ratio; Figure 5b), are expected to experience the largest shifts in timing.As temperatures increase and winter precipitation changes from snow to rain (Knowles et al., 2006;Reidmiller et al., 2018), streamflows shift earlier in the year, with centers of timing of water supplies expected to shift as much as 24 days earlier by 2040 (Wenatchee watershed, 45 in Figure 5a, where 67% of precipitation falls as snow).The northern watersheds also have high snowmelt contributions (Figure 5b), yet are expected to experience more moderate shifts in timing of water supply, likely because the mean winter temperature is far enough below 0°C that snowmelt will be less sensitive to expected temperature change by 2040 (Adam et al., 2009).Finally, the low elevation watersheds in the heart of central Washington, which have historically been rain-dominated (Li et al., 2017), are expected to experience the smallest shifts in timing, which can be as little as 2 days (Lower Snake, 33 in Figure 5).

| Groundwater trends
Trends in groundwater levels over the last 20 years were predominantly declining across eastern Washington's groundwater subareas and across the four aquifer layers considered (Grande Ronde, Wanapum, Saddle Mountains, and Overburden; Figure 6).These trends are consistent with previous evaluations of groundwater in Washington (e.g., Vaccaro et al., 2015).The most spatially extensive declines are observed in the deepest Grande Ronde layer with the steepest localized declines occurring in the Wanapum layer.Declines in the Overburden tended to be closer to zero, but shallow AST still led to vulnerabilities in this layer.The Rock Glade subarea exemplifies the importance of conducting the trend and vulnerability analysis for each aquifer layer separately.The average trend in Rock Glade is negative in the deeper Grande Ronde and Wanapum layers with wide variability, positive in the shallowest basalt layer in the Saddle Mountains, and near zero in the Overburden.
Overall, a total of 10 subareas across all the aquifer layers have less than 50 years to a projected decline of 25% in AST.In other words, nearly half will require some level of investment in groundwater infrastructure within the next 50 years.

F I G U R E 3
The groundwater subareas within eastern Washington included in our study.
The inclusion of groundwater trends-and the vulnerabilities identified as these are projected into the future and put in the context of access to groundwater-are rare in forecasting or projection-based efforts.Even though the groundwater trends analysis was not integrated with the modeling of surface water supplies (this was beyond the scope of the project), the joint evaluation of changes in both surface and TA B L E 3 Modeled annual and seasonal water supply in the historical  and future (2040) periods in a median flow year for the Washington portion of the Columbia River Basin.The future value is represented by a 90% confidence interval of the ensemble mean flow (17 General Circulation Models [GCMs] for RCP 4.5 and 8.5 each).2017) and the VIC model (Liang et al., 1994) forced using gridMET data (Abatzoglou, 2013).
groundwater supplies creates a more complete understanding of current and future water availability.The dominance of declining trends in groundwater highlights how vulnerable many locations are to future changes in groundwater, an important water source for municipalities, as well as for irrigators responding to drought, even if their main source is surface water (Pitz, 2016).The expansion of groundwater modeling to improve groundwater projections capabilities and surface-groundwater connectivity was documented as a high priority for the next update of the Forecast.under future (elevated) atmospheric carbon dioxide levels.These responses are crop-specific.While some crops may have increased water demand (such as perennial hay crops that can make use of the longer growing season through multiple cuttings), others see decreased demand, such that overall, the region is expected to see a decrease in agricultural water demand.This is exemplified by the watersheds with the largest decreases in water demand (Figure 7).In the counties that overlap with such watersheds, dominant crops include annuals like potatoes and vegetables and tree fruits such as apples and cherries (USDA-NASS, 2019), which are projected to have reduced demand in the future (Rajagopalan et al., 2018;Vano et al., 2010).Although there is an overall reduction in agricultural water demand across eastern Washington, as many as half the watersheds are expecting increases, in some cases up to as much as 16 million m 3 /year by 2040 (e.g., Upper Yakima and Okanogan, 39 and 49 in Figure 7, respectively).This variation highlights the need to concurrently evaluate supply and demand changes, and to do so at finer spatial scales than state or basin.

| Changes in agricultural and residential demands
In terms of changes in timing, demand during the first half of the irrigation season (March-June) is expected to increase by 9% to 13% by 2040 (Table 4), depending on the agricultural production scenarios considered (Table 2).The demand during the second half of the season (July-October) is expected to decrease by a similar amount (10% to 12% by 2040; Table 4).Overall, the center of timing for agricultural demand across the CRB is only expected to shift earlier by 3 (±0.5)days, though the shift in timing, as well, varies greatly by crop type (M.Yourek, F.V. Scarpare, R. Gustine, M. Liu, J. Boll, M. Barik, and J.C. Adam, in prep).
The expected shift in timing of agricultural demand is much smaller than the expected shift in timing of surface water supply.The difference in timing of when water is needed relative to when it is available will therefore increase, exacerbating the current challenges water managers and decision-makers face as they try to fulfill the competing demands for instream and out-of-stream water uses.Quantifying these changes does not address these challenges.But providing relevant and reliable information on the shifts in timing of both supply and agricultural demand-which is not common in similar agency-driven efforts-can help decision-makers prepare for the expected changes over the next 20 years.
Total residential demand, though it only accounts for approximately 11% of water withdrawals in eastern Washington (Fasser, 2018), is expected to increase by 22% by 2040 (Table 5).Of this, 42% of total annual residential demand occurred during the summer months (June, July, and August), thereby coinciding with the timing of decreased water supplies.If agricultural and residential demands are combined, the resulting out-of-stream demands projected for 2040 are expected to be similar to historical, as agricultural demands decrease and residential demands increase.However, planned and under-development water supply projects in the region are expected to provide an additional 308.4 million m 3 /year for new out-of-stream uses (OCR, personal communication).
When this amount of potential new demand is added to the changes in agricultural and residential water demands, the overall median demand is anticipated to increase by more than 7% by 2040.The inclusion of residential demand and of additional potential demand served by planned water supply projects in estimating future out-of-stream demand were key to understanding the range of possible changes in out-of-stream needs.Such insights are also unusual in other comparable efforts, which do not consistently consider changes in infrastructure and future water supply projects (Table S1).

| Implications of concurrent changes in supply and demand
Vulnerabilities related to future changes in supply and demand across eastern Washington emerged when we assessed supply and demand changes concurrently.Different watersheds are expected to experience different degrees of change in surface and groundwater supplies, coupled with increases in some out-of-stream demands but not others.Summarizing results at the watershed level can support effective management (Derepasko et al., 2021), as significant planning in Washington occurs at the watershed scale (e.g., watershed planning; WA ECY, n.d.-b).
We highlight three key vulnerabilities associated with these changes.First, we discuss concurrent changes in surface water supply and agricultural demand (focusing on low supply years, when issues are greatest), given that surface water accounts for approximately 76% of water withdrawals in eastern Washington (Lane & Welch, 2015).Second, we focus on groundwater trends and changes in residential demand (though with reference to changes in surface water supplies for some locations), as groundwater accounts for around 86% of water withdrawals by municipalities (Lane & Welch, 2015).Third, we discuss how state water regulations can affect how important these co-occurring changes are, and what vulnerabilities need to be addressed.

| Surface supply and agricultural demand changes
Two watersheds are expected to be particularly vulnerable to co-occurring decreases in surface water supply during low supply years with increases in agricultural demand by 2040: the Naches and Upper Yakima (both part of the Yakima watershed; 38 and 39 in Figure 8).These watersheds are expected to see both relatively large increases in agricultural demand from May through September (the core irrigation season; TA B L E 4 Modeled agricultural water demand, excluding conveyance losses (known as "top of crop"), in the historical  and future (2026-2055) periods using the median of 34 climate change scenarios (17 GCMs for RCPs 4.5 and 8.5), for the Washington portion of the Columbia River Basin, distinguishing between early and late in the irrigation season.Each future value is represented by a 90% confidence interval of the ensemble mean water demand.Figure 8a) and relatively large decreases in supply during those months (Figure 8b).The watersheds in south-central Washington (31,32,33,36,41,43 in Figure 8c) as well as the Wenatchee and Methow watersheds (45 and 48, respectively in Figure 8c) are relatively less vulnerable to future changes in surface water supply (Figure 8a) and agricultural demand (Figure 8b).
The vulnerabilities highlighted for the Yakima watershed during the core irrigation season are in part related to the shift in timing of water supply (Figure 9).The center of timing for supply is expected to occur 20 days earlier by 2040 in the Yakima watershed, which is close to the maximum shift expected for any eastern Washington watershed (Figure 5a).In the upper Yakima, the increase in agricultural demand appears to have a similar weight as the changes in supply driven by this shifting timing (Figure 8a,b).The changes in agricultural demand were mainly determined by climatic changes rather than being affected by an earlier planting date or the changes in crop mix projected for 2040 (Figure 10).
Even under historical conditions there are months when the supply in the Yakima does not meet demands (e.g., July and August, Figure 11a), and this unmet demand is expected to increase as the climate changes (Figure 11b).These changes in supply and demand could lead the frequency of prorationing to double or close to triple (depending on the RCP scenario considered) by 2040 (Figure 12).The proration rate (the proportion of the full water entitlement received in a particular year) is expected to decrease under future climates, suggesting that prorationing will also become more severe (Figure 12).

| Water supplies and residential demand changes
Municipal water systems often maintain a varied supply portfolio to ensure water availability for customers throughout the year.However, of the 45 of the municipalities we studied, 43 rely at least partially on groundwater (Table 6).Of these large water systems in eastern Washington, 18 are expecting summer residential water demand to increase by 25% or more by 2040 (Table 6).Of those 18, 12 are completely groundwater dependent, and eight of those are likely accessing at least one aquifer layer that is experiencing a decline (Table 6).Although some data for large municipalities was lacking and we did not have access to (or knowledge of) data for the many medium and small public water systems across eastern Washington, the number of large systems that are likely vulnerable due to co-occurring increases in demand and declining groundwater levels is notable.The extent of these vulnerabilities would not be exposed by the most commonly used approaches for water projections (Table S1).
Municipal water providers rely on their inchoate rights (these are rights to begin withdrawing water and putting it to beneficial use in the future, as new demands arise) to meet future water demands.We found that, by 2040, 14 municipal systems are expected to be using less than 25% of their existing water rights, 21 systems are expected to be using 25%-50% of their water rights, six systems would be using 50%-75% of their water rights, and four systems (Connell, Quincy, Pasco, and Airway Heights) are estimated to be using over 75% of their water rights (Figure 13).
Pasco is unique among the 45 municipalities we studied in that it relies solely on surface water supplies (Table 6), though at least nine other municipalities do so partially (others depend on intertie supplies, for which we were unable to determine the originating source of water).Many of the municipalities that can expect increases in demand by 2040 may also be vulnerable to concurrent decreases in surface water supply (Figure 14).Most of the 10 municipalities that rely at least partially on surface water are within watersheds categorized as high (Pasco, Yakima, Ellensburg, Cle Elum, Leavenworth, and Cashmere) or medium (White Salmon, Kennewick, Richland, and Walla Walla) vulnerability due to concurrent decreases in summer water supply and increases in residential demand expected by 2040 (Figure 14c; Table 6).

| Surface supply changes and instream flow requirements
The Columbia River's surface water supply is expected to be sufficient to meet out-of-stream demands (of which irrigated agriculture accounts for ~96%; National Research Council, 2004), at least as far out as 2040 (Figure 15).However, important instream demands are codified in Washington State Instream Flow Rules (WAC Title 173; RCW, 1998) and federal Biological Opinion flow targets (USACE, 2000(USACE, , 2014)).When these flow targets are included, even under historical conditions, the Columbia River's supply in low flow years is insufficient to meet demands for multiple months (Figure 16).Irrigation water rights that are junior to these instream flows will be vulnerable to being interrupted more frequently in the future, as major dams on the Columbia River).Under future climates, deficit frequencies during the late summer are expected to increase, possibly occurring every year during some weeks by 2040 (Figure 17).It is worth noting, however, that the shift of water supply earlier in the year due to changing snowpack conditions is also expected to lead to a reduction in deficit frequencies during the early summer (Figure 17).
The Wenatchee watershed, which was categorized as having medium vulnerability due mainly to the shifts in timing of water supply (Figure 8c), is an example of a tributary watershed that is expected to see similar changes as the Columbia River mainstem.Under historical conditions, modeled curtailments (in the Wenatchee we had sufficient data from past events to guide curtailment modeling) are expected as often as not from late July through the end of October, reflecting how frequently the supply is expected to be insufficient to meet ISF requirements once out-of-stream demands are met (Figure 18).With expected changes in water supply and demand by 2040, the frequency of curtailment is expected to increase, particularly from mid-June through mid-September (Figure 18).
These results do not provide the full spectrum of possible instream flow deficits or curtailments across watersheds and even within the Columbia and Wenatchee Rivers.Additional curtailments are possible when junior water rights holders are required to modify their water use to protect senior water rights (Adam et al., 2022); instream flow deficits may still occur in these and other rivers once water use is curtailed; and curtailment is possible in the Columbia River itself.However, combining supply and demand by modeling changes in prorationing (in the Yakima River), curtailment (in the Wenatchee River) and instream flow deficits (in the Columbia River) provides feasible scenarios of where and when the water management challenges are likely to become more acute as the climate changes.

| Inclusion of management variables in the forecast
A key feature of this Forecast is the inclusion of management variables in addition to climatic influences in our modeling and calculations.This inclusion helps explore the implications of current rules and investments on future changes in supply and demand and improves understanding of when and where management and investment decisions could make the most difference.We discuss three examples.F I G U R E 1 0 Modeled historical  and future (2040) residential and agricultural water demands within the Yakima watershed.Future agricultural water demand was modeled under four scenarios described in the "The Scenarios" section of the Methods.Each bar represents the median demand condition.
First, the inclusion of estimated out-of-stream water demands that planned and under-development water supply projects would serve shifted our results from an expectation of decreased agricultural demand to a 7% increase in out-of-stream demand that could realistically occur by the 2040s.When combined with the expected decreases in dry season water supply (when demands are also higher), the future vulnerability becomes apparent.How, when, and where that additional water is allocated could lead to different outcomes in the ability of surface supplies to meet out-of-stream demands in the future, as there is significant variability across the region in terms of expected changes in supply and demands.Other analyses that provide demand estimates rarely include these investment decisions.Instead, they rely solely on future per capita demand projections or continuing past trends (e.g., Lacroix et al., 2016;New Mexico Interstate Stream Commission, 2018;Zamani Sabzi et al., 2019).
Second, we placed the expected increases in residential water demand and declining groundwater trends in the context of municipalities' inchoate water rights, which highlighted the risk posed by concurrent changes in supply and demand for some municipalities.Three of the four A proration rate of 100% corresponds to fully satisfied water entitlements.These results correspond to an annual proration rate of 70% or less, which has been shown to have significant impacts on crop production (Vano et al., 2010).
TA B L E 6 Details available for the 45 municipal water systems studied.Water systems shown in bold are expecting increases in summer residential demand of 25% or more.The type of water source is identified for each municipal water system as surface water (SW), groundwater (GW), and intertie (IN).In red is the only municipality that is fully surface water dependent.Airway Heights, has initiated water purchase agreements with a neighboring municipal water supplier (Kolb, 2021).Understanding the context of neighboring municipalities and the extent of similar vulnerabilities across the region could be useful to other municipal water systems that may need to begin augmenting their water supplies.
Third, we explored implications for meeting instream flow requirements.In some rivers, like the Columbia and the Wenatchee, vulnerabilities in water availability in the mid-term future only become clear when the codified instream flow requirements are included in the analysis, whether through instream flow deficit or curtailment modeling.Modeling changes in prorationing frequency and rate in the Yakima River highlighted the extent to which water managers must prepare for prorationing, which could happen close to yearly by 2040.Evaluating supply and demand changes within these legal frameworks gives managers information on how the risk of curtailment may change in the future.Even in the Columbia River, where there is insufficient data to effectively model curtailment decisions (curtailment has only occurred once, in 2001), estimates of instream flow deficits helped quantify the risk of not meeting instream flow requirements.

| Transferable elements
Our approach in this long-range supply and demand Forecast has three elements that we believe could be valuable for other similar efforts intended to aid water management and investment decisions.First, our partnership with a water management agency (OCR) throughout the process ensured that researchers were aware of the agency's key concerns and goals.In addition, OCR staff's regular engagement in the modeling process helped them gain a deeper understanding of the modeling framework, its strengths and limitations, as well as contribute to the interpretation of model results.This partnership between researchers and managers has helped shape this scientifically rigorous integrated modeling framework into a robust and useful tool for management.
Second, our application of an integrated set of models to capture the interactions between water supply and the largest out-of-stream demand provided robust estimates of how supply and demand are likely to change, and the vulnerabilities that those changes will bring.
Third, our assessment of additional supply (groundwater) and demand (residential) changes using simpler methodologies allowed for a more complete evaluation of vulnerabilities.The integrated modeling framework in combination with simpler associated analyses can help assess the interrelated nature of climate, hydrology, and water supply management with irrigation water demand, trends in important crops,  , 1977).Data used to create these maps is accessible from https:// story maps.arcgis.com/ stori es/ 08bc5 5d5ef 5a4b6 f9f96 1d708 4866960.
residential water demand, and groundwater trends.The inclusion of all these aspects of water supply and demand allowed for a more complete understanding of the current and future water availability in eastern Washington.

| Remaining limitations
The 2021 Columbia River Long-Term Water Supply and Demand Forecast (Adam et al., 2022;Hall et al., 2022) is a complex body of work integrating a variety of methodologies to characterize changes in water supplies and demands, yet there are still important aspects of future supply and demand that we were unable to include.For instance, at the regional scale, modifications to the federal Columbia River Treaty between the U.S. and Canada (CRS, 2023) could affect operations of U.S. dams on the Columbia River, and therefore affect water supplies and demands.More locally, our modeling framework did not explore the potential impacts of regulatory-or incentive-driven conservation efforts, which could alter future water demands and supplies in complex ways.For example, research has shown that irrigation efficiency may reduce water availability downstream for other users, instream flows, and groundwater recharge (Malek, Adam, Stöckle, & Peters, 2018).Hydropower production is another important instream demand in the Columbia River, but due to lack of data to reliably quantify flows needed to meet hydropower needs, we were unable to include quantitative estimates of this instream demand (though see Adam et al., 2022;Hall et al., 2022 for expected changes in demand for hydroelectric power).
In addition, there are uncertainties associated with the models and data used that should also be considered.Our results are mainly driven by a consistent signal of increasing temperatures (e.g., snowpack dynamics, growing degree days and impacts on crop phenology), yet uncertainties remain around both the direction and magnitude of precipitation changes in the U.S. Pacific Northwest (Dettinger et al., 2015;Elsner et al., 2010;Vose et al., 2017).We also lack a comprehensive set of observed streamflow measurements across the CRB without regulation influence with which to calibrate our surface water supply modeling results.Bias may therefore be induced into the supply results through the processes implemented to remove the effects of regulation (Adam et al., 2022;Georgakakos et al., 2014).The groundwater trends analysis was also constrained by data availability, and our extrapolation of these trends 20 years into the future is simplistic and carries its own uncertainties.
Demand estimates are also subject to uncertainties.We lacked field observations on the extent of irrigated agriculture outside of Washington State, as well as the comprehensive experimental datasets needed for detailed crop parameterization for all crops across a range of conditions.In addition, uncertainty remains in crop responses to increases in atmospheric carbon dioxide and their resulting impact on transpiration rates (Kirschbaum & McMillan, 2018;Vicente-Serrano et al., 2022).Municipal water use data had some gaps, and county-level data were available annually, needing assumptions to interpolate monthly values.
The State Legislature's mandate requires a long-term supply and demand Forecast every 5 years.If funded appropriately, future iterations could allow our partnership to address some of these limitations.

| CON CLUS ION
Policy-makers, water management agencies, and other water managers are working with a complex system where water availability is influenced by a range of interacting biophysical and human-driven factors.In many cases, they are making decisions that have implications well into the future.The Columbia River Long-term Supply and Demand Forecast concurrently explored expected changes in supplies and out-ofstream demands, as well as the implications of those changes for instream flows in the future.
The vulnerabilities that arose from our analysis highlight the complexity of efforts to inform decisions 20 years or more into the future.In the CRB, changes in annual surface water supply are not the main concern; it is the shifting timing in surface water supply that leaves many watersheds without sufficient supply late in the season to meet out-of-stream demands at times when those demands are highest.Similarly, it is not the current trend in annual out-of-stream water demands that raises concern, but rather the specific locations and times of the year when these demands are expected to increase, combined with potential changes in the water allocated to out-of-stream uses through planned water development projects.In consequence, under future conditions, rivers with federal or state instream flow rules established to protect flows could require more frequent and deeper curtailments of junior out-of-stream water rights.And even with these curtailments, rivers may have insufficient flows to meet their instream requirements.The declining trends in the aquifers we analyzed suggest that, to prepare for and mitigate the impacts of future changes in supply and demand, decision-makers need to explore options other than switching demands from surface to groundwater supplies.These joint surface and groundwater declines are not uncommon and there is a need for greater understanding of how different supplies are changing simultaneously in different regions (Condon & Maxwell, 2019).
This Forecast is not all-encompassing or fully comprehensive, and significant uncertainties remain around the modeled future, and whether the scenarios we modeled are indeed the scenarios that will occur.Yet this analysis delves further into integrated modeling to study both supply and demand than the majority of other analyses meant to inform long-term water management decisions (e.g., Austin Water, 2018;Cross et al., 2017;Wyoming Water Development Commission, 2007; see Table S1).The Forecast also explicitly quantified expected changes in the main supplies and demands, and was developed in partnership with the OCR, providing a direct channel of communications that both focused the analysis on key management priorities, and provided an opportunity for shared learning of the results and their implications.The Forecast's results therefore provide entities like the OCR and other water managers with a robust set of potential future conditions that highlight the vulnerabilities that eastern Washington and the greater CRB can expect as supplies and demands change into the future.

AUTH O R CO NTR I B UTI O N S
Sonia A.
Mixed future Agricultural demand: To characterize the combined impacts of climate change and planting date Future: 2040 only 1 week earlier than historical Historical (4) Climate change and future crops Agricultural demand: To characterize the combined impacts of climate change, planting date, and crop mix Future: 2040 only 1 week earlier than historical Future: 2040 onlyhas been shown to be no less accurate than far more complex population projection models(Armstrong, 1984;Chi, 2009;Smith, 1997), we used three extrapolation-based methods to derive a mean estimate of 2030 and 2040 municipal population change: linear estimation, logistic curve fitting, and ratio extrapolation.

F
I G U R E 2 Left panel: Estimated proportions of different crops in Washington, used as inputs to the integrated modeling of agricultural water demand.Right panel: Irrigated hectares of certain crop groups of particular interest for this analysis.The historical (2020) crop mix was estimated using USDA National Agricultural Statistics Service survey data, and the future crop mix was estimated based on a statistical analysis of trends in different crops between 1999 and 2019 and extending those trends through 2040.
Expected changes in water supply across the entire Columbia River Basin by 2040 (2026-2055) for RCP 4.5 and RCP 8.5 compared to historical.Center of timing was quantified at Bonneville Dam for historical and future supplies followingStewart et al. (2005).

F
I G U R E 5 (a) Changes in the center of timing of water supply expected during high flow years (80th percentile) by 2040 compared to historical.Since the spatial variation is minimally affected by climate scenario the median of 34 climate change scenarios (17 GCMs for RCPs 4.5 and 8.5) is shown.Note that one aggregate value is given for watersheds 37, 38, and 39, and one value is given for watersheds 44 and 50, reflecting in each case their management as a single combined administrative unit.(b)Historical (1976Historical ( -2005)  )   snowmelt ratio (the proportion of the supply received from snowmelt), obtained by using methodology fromLi et al. ( Agricultural water demand across eastern Washington is expected to decrease 1.7% by 2040.Although atmospheric evaporative demand tends to increase with warmer temperatures, decreases in actual crop evapotranspiration resulted from (a) accelerated crop development under warming which reduces the time to crop maturity and the length of time when irrigation is needed, and (b) increased water use efficiencies F I G U R E 6 Groundwater trends from 2000 to 2020 by subarea in the four main aquifer layers in Columbia Plateau Regional Aquifer System (CPRAS) and groundwater subareas outside of CPRAS.Only subareas with three or more wells are plotted.Subarea locations are shown in Figure 3.

F
I G U R E 7 Expected changes in annual agricultural water demand between the historical (1986-2015) and future (2026-2055) periods, summarized by watershed.Since the spatial variation is minimally affected by climate scenario the median of 34 climate change scenarios (17 GCMs for RCPs 4.5 and 8.5) is shown.
the timing of supply shifts earlier in the season.The instream flow deficit calculations confirm this, with most years showing deficits in late July and August at Priest Rapids and McNary Dams (Figure 17; see Hall et al., 2022; Whittemore, 2022 for instream flow deficit frequencies at other

F
I G U R E 8 (a) Expected change in May-September agricultural water demand between the historical (1986-2015) and future (2040) periods, by watershed in eastern Washington.Since the spatial variation is minimally affected by climate scenario the median of 34 climate change scenarios (17 GCMs for RCPs 4.5 and 8.5) is shown.(b) Changes in May-September water supply expected during low flow years (20th percentile) by 2040.Since the spatial variation is minimally affected by climate scenario the median of 34 climate change scenarios (17 GCMs for RCPs 4.5 and 8.5) is shown.Note that one value is given for supply changes in 37, 38, and 39, and one value is given for 44 and 50, reflecting in each case their management as a single combined administrative unit.(c) Vulnerability to future co-occurring changes in low flow year water supply and agricultural demand.The vulnerability level was assigned based on the standard (z-score) concurrent changes in both supply and demand, categorized using Jenks natural breaks clustering (Jenks, 1977).

F
Modeled historical (1986Modeled historical ( -2015) )  and 2040 (RCP 4.5 and RCP 8.5) surface water supply generated within the Yakima watershed for the median (50th percentile) supply conditions.

F
I G U R E 11 Surface water supply and agricultural and residential demands for (a) historical (1986-2015) and (b) future (2040) periods in the Yakima watershed.The median value of the range of climate change scenarios is shown here.The top of the bar for agricultural water demand shows the 50th percentile of total surface water demand, and the error bars show the 20th and 80th percentiles of total surface water demands.These results do not consider water curtailment.F I G U R E 1 2 Modeled historical baseline (1986-2015) and future (2040; RCP 4.5 and RCP 8.5) (a) proration frequency, and (b) median annual proration rate for the Yakima watershed.
municipalities expected to require more than 75% of their inchoate water rights to meet demand by 2040-Connell, Quincy, and Pasco-are already actively looking for additional (pending) water rights (City of Connell, 2007; City of Pasco, 2019; City of Quincy, 2014).The fourth,

F
Percent of available water rights used by 2040 for sampled water provider systems.F I G U R E 1 4 (a) Expected change in summer (June-August) residential water demand between the historical (2020) and future (2040) periods by watershed in eastern Washington.(b) Expected change in summer water supply expected during low flow years (20th percentile) by 2040.Since the spatial variation is minimally affected by climate scenario the median of 34 climate change scenarios (17 GCMs for RCPs 4.5 and 8.5) is shown.Note that one value is given for 37, 38, and 39, and one value is given for 44 and 50, reflecting their common management as a single combined administrative unit in each case.(c) Vulnerability to co-occurring changes in summer surface water supply and residential demand by 2040.The vulnerability level was assigned based on the standard (z-score) changes in supply and demand, categorized using Jenks natural breaks clustering (Jenks

F
Comparison of regulated surface water supply and agricultural water demand for historical and future (2040) periods using the median of 34 climate change scenarios (17 GCMs for RCPs 4.5 and 8.5) for the whole Columbia River Basin.F I G U R E 1 6 Historical (1986-2015; left column) and future (2040; right column) regulated surface water supply at Priest Rapids (top row) and McNary (bottom row) dams for low (20th percentile), median (50th percentile), and high (80th percentile) supply years, averaged across 34 climate change scenarios (17 GCMs for RCPs 4.5 and 8.5).Supplies presented here are prior to accounting for out-of-stream demands.Washington State instream flow (WA ISF) and federal Biological Opinion (BiOp) flow targets (bars) are also shown.F I G U R E 17 Left panels: Modeled historical (1986-2015) frequency of instream flow deficits (quantified as the number of years out of 30 years when supply is insufficient to fulfill ISF Rules once out-of-stream demands are met) at Priest Rapids (top) and McNary (bottom) dams.The frequency of instream flow deficits was calculated on a weekly basis.Right panels: Change in expected frequency of instream flow deficits (quantified as the difference in the number of years out of 30 years) between historical (1986-2015) and 2040 (2026-2055) periods at Priest Rapids (top) and McNary (bottom) dams under two emissions scenarios (RCP 4.5 and RCP 8.5).F I G U R E 1 8 Left panel: Modeled historical (1986-2015) frequency of curtailment (quantified as the number of years out of 30 years when supply is insufficient to fulfill ISF Rules once out-of-stream demands are met) in the Wenatchee River.The frequency of curtailment was calculated on a weekly basis.Right panel: Change in expected frequency of curtailment (quantified as the difference in number of years out of 30 years) between historical (1986-2015) and 2040 (2026-2055) time periods under two emissions scenarios (RCP 4.5 and RCP 8.5).
Adam et al. (2022)ts, root crops, leguminous, forages, and oilseed crops were addressed in this study.The data used came from crop field trials, conducted mostly by University Extension groups.The weather conditions under which these trials were conducted cover a wide geographic area and include many years and a range of management practices and crop varieties that represent the diversity of farmers' practices in the Pacific Northwest.Calibration results can be found inAdam et al. (2022).
Adam et al. (2022)tions were run for each individual GCM/RCP combination to avoid issues relating to co-variability in climate forcings and we calculated statistics on model outputs.We chose to run all available scenarios so that we could best represent overall uncertainty.For more information regarding the climate modeling decisions in this study, seeAdam et al. (2022).Results often do not substantially differ between RCPs.When this is the case, results are shown as the median of the 34 climate change scenarios used in this study, which helps highlight other factors (e.g., temporal and spatial variation) that are of greater influence than

climate, historical planting date, historical crop mix Future (2040) climate, future planting date, historical crop mix Future (2040) climate, future planting date, future crop mix
Historical (2020)and future (2040) residential water demand for the Washington State portion of the Columbia River Basin, separated into total annual demand, summer demand, and demand during the rest of the year.