Where and When Does Streamflow Regulation Significantly Affect Climate Change Outcomes in the Columbia River Basin?

The Columbia River basin is a large transboundary basin located in the Pacific Northwest. The basin spans seven US states and one Canadian province, encompassing a diverse range of hydroclimates. Strong seasonality and complex topography are projected to give rise to spatially heterogeneous climate effects on unregulated streamflow. The basin's water resources are economically critical, and regulation across the domain is extensive. Many sensitivity studies have investigated climate impacts on the basin's naturalized hydrology; however, few have considered the large role of regulation. This study investigates where and when regulation affects projected changes in streamflow by comparing climate outcomes across 80‐member ensembles of unregulated and regulated streamflow projections at 75 sites across the basin. Unregulated streamflow projections are taken from an existing data set of climate projections derived from Coupled Model Intercomparison Project version 5 Global Climate Models. Regulated streamflow projections were modeled by the US Army Corps of Engineers and the US Bureau of Reclamation by using these unregulated flows as input to hydro‐regulation models that simulate operations based on current and historical water demands. Regulation dampens shifts in winter and summer streamflow volumes. Regulation generally attenuates changes in cool‐season high flow extremes but amplifies shifts in warm‐season and annual high flow extremes at historically snow‐dominant headwater reservoirs. Regulation reduces dry‐season low flow changes in headwater tributaries where regulation is large but elsewhere has little effect on changes in low flows. Results highlight the importance of accounting for water management in climate sensitivity analysis particularly in snow‐dominant basins.

and transient rain-snow watersheds (Chegwidden et al., 2019;Fritze et al., 2011;Hamlet et al., 2010;Payne et al., 2004;Stewart et al., 2005). Peak timing shifts will likely be more pronounced in transient watersheds where winter temperatures are at or near freezing and therefore more sensitive to warming (Bureau of Reclamation, 2016;Vano et al., 2015). Extreme precipitation events are increasing (IPCC, 2014;May et al., 2018;Warner & Mass, 2017), and are projected to lead to substantially more severe flood events (Salathé et al., 2014). Queen et al. (2021) projected pervasive increases in Columbia River basin flood magnitudes based on unregulated streamflow projections.
Throughout the past century, expansive water resource infrastructure has changed the streamflow regime by creating man-made reservoirs and altering flows. While climate is a primary driver of natural basin hydrology, extensive regulation modulates this natural hydrology and thus streamflow (Figure 1; Arheimer et al., 2017). The Columbia River basin is heavily regulated by federal and state agencies and private utilities for a range of system objectives including flood risk management, hydropower, irrigation, navigation, fish passage, and recreation. More than 250 reservoirs exist across the system and streamflow regulation sustains an economically critical food-water-energy nexus. Columbia River system operations follow transnational guidelines defined by the Columbia River Treaty, an international agreement between the US and Canada on how water is allocated across the US-Canadian border. Ratified in 1964, the treaty informs the joint management of three upper Columbia Canadian storage dams to coordinate transboundary flood control and is currently being renegotiated for modernization post-2024 (Stern, 2020).
Climate change impacts on Columbia River basin naturalized or unregulated streamflow have been extensively studied; however, only a limited number of large-scale studies have considered the large role of regulation. To test the reliability and vulnerability of Columbia River system operations under changing historical conditions, Jones and Hammond (2020) investigated observed intra-annual timing of reservoir inflows and outflows. Between 1950 and 2012, May through October inflows declined but outflows increased due to low flow augmentation. Zhou et al. (2018) investigated the effect of regulation on the timing of hydrologic regime shifts for large basins across the western US. Their study used climate projections from three Coupled Model Intercomparison Project version 5 (CMIP5) global climate models (GCMs) for Representative Concentration Pathway (RCP) 4.5 and 8.5 emissions scenarios. GCM meteorology was statistically downscaled to the 1/8-degree grid resolution. Regulated flows were simulated by the Model for Scale Adaptive River Transport  Water Management (Voisin et al., 2013) model (MOSART-WM); a simplified hydro-regulation model that uses operational rules based on historical monthly mean inflows and water demands. Zhou et al. (2018) found that for the Columbia River basin, regulation delayed the timing of regime shifts for all seasons except autumn. Studies in subbasins west of the Cascade Mountain range have also shown large differences between projected changes in unregulated and regulated streamflow extremes where, in some cases, regulation amplifies the climate signal (Lee et al., 2016(Lee et al., , 2018. Singh and Basu (2022) analyzed historical seasonal streamflow trends in neighboring natural  (1976-2005), 2030s (2020-2049), and 2070s (2060-2089). Monthly averages are taken from the median of each ensemble. For each location, the left panel shows the unregulated hydrograph, and the right panel shows the regulated hydrograph.
10.1029/2022WR031950 3 of 18 and regulated watersheds across North America. They showed that regulation effects on seasonal streamflow trends vary widely across space and time ranging from amplification to no effect to dampening.
As climate change poses potential challenges for managed freshwater systems, concerns regarding where and when climate impacts will manifest and what they mean for the future of water resources are ever-growing. Large-scale climate sensitivity studies that do not account for regulation effects on streamflow may lead to inaccurate characterizations of projected outcomes. To investigate where and when extensive regulation modifies climate impacts on Columbia River basin streamflow, this study uses two 80-member ensembles of unregulated and regulated streamflow projections developed from 10 CMIP5 GCM projections for the RCP 8.5 emissions scenario (RMJOC, 2018(RMJOC, , 2020 to compare climate outcomes under unregulated and regulated conditions. Regulated flow projections were modeled by the US Army Corps of Engineers (USACE) and the US Bureau of Reclamation (USBR) using hydro-regulation models that are used by USACE and USBR to support long-and short-term federal water management planning. These models use operational rule-curves based on current and historical water management objectives that vary temporally and spatially and account for local and system-wide flood risk management, hydropower, irrigation, navigation, and ecological constraints (RMJOC, 2020). Operational rule-curves are based on current and historical water demands and do not change to account for future changing conditions. This study seeks to answer two key questions: (a) How does regulation modify projections of streamflow volumes and extreme streamflow events under climate change? (b) How do the signatures of climate change and hydro-regulation vary seasonally and across the domain? We address these questions by comparing projected climate impacts on seasonal volumes and high and low flow extremes for unregulated conditions and regulated conditions. We investigate changes in extremes for 51 diverse sites across the basin where hydro-regulation was modeled at a daily time step. Seasonal volume changes are examined for 75 sites using output from hydro-regulation models at both a daily and monthly time step. Outcomes for the 2030s (2020-2049) and the 2070s (2060-2089) are compared to the control period , and we test relationships to river network location and the level of regulation by grouping locations by region and the degree of upstream regulation, respectively. The control period is selected to represent the most recent 30-year period in the historical streamflow used to validate simulated flows (RMJOC, 2018). The 2030s and 2070s are selected to represent the near future and far future, respectively. Analysis is performed for water years rather than calendar years. A water year is a 12-month period used by hydrologists to represent temporal precipitation patterns that influence the water cycle (e.g., wet-season winter snow accumulation and dry-season summer snow melt) and is defined as October 1 of the previous calendar year through September 30 of the given year.

Study Area
The Columbia River basin is a transnational river system covering 673 thousand km 2 of the Pacific Northwest. The basin encompasses a diverse range of hydroclimates from arid lowlands to glaciated mountain regions. The Cascade and Rocky Mountain ranges pass through the western and eastern edges of the basin, respectively, and high elevation snowpack supplies much of the basin's freshwater through the spring freshet. Three hydrologic regimes exist across the domain (Hamlet & Lettenmaier, 2007): the rain-dominant regime where streamflow peaks in the cool-season primarily driven by rainfall; the transient regime where two annual peak streamflow pulses result from cool-season rainfall and warm-season snowmelt; and the snow-dominant regime where streamflow peaks in the warm-season primarily driven by snowmelt . These three regimes can be distinguished by the ratio of peak snow water equivalent (SWE) to cool-season precipitation (Barnett et al., 2005). Pacific Northwest peak SWE typically occurs around April 1. Following the work of Mantua et al. (2010), we classify hydrologic regimes for each 30-year period by the ensemble median ratio of drainage area-averaged April 1 SWE to drainage-area averaged October through March precipitation (SWE/P) where SWE/P less than 0.1 indicates a rain-dominant regime; SWE/P between 0.1 and 0.4 indicates a transient regime; and SWE/P greater than 0.4 indicates a snow-dominant regime ( Figure 2). 10.1029/2022WR031950 4 of 18

Regions
The hydroclimate across the domain is diverse, and system operations vary widely depending on the authorized purposes of water management infrastructure and regional water demands. To test the relationship between regulation effects and river network location, sites are grouped into 10 regions defined by location on a tributary or the mainstem (Figure 3a).

Degree of Upstream Regulation
Regulation effects on streamflow can be attributed to temporal reservoir storage and delayed releases that alter streamflow timing (Grill et al., 2019). The degree of upstream regulation (DOR) is a measure of annual storage effects on unregulated streamflow and is defined as total upstream storage capacity normalized by annual streamflow volume (Dynesius & Nilsson, 1994;Grill et al., 2019;Lehner et al., 2011). Higher DOR indicates greater capacity to store water throughout the water year and as a result, larger regulation effects on the streamflow Figure 2. Map of the Columbia River basin and hydrologic regime ratios for the 75 sites and three periods investigated in this study. The basin is located in the Pacific Northwest region of the US straddling the US-Canadian border. Regime classification color scheme adapted from Mantua et al. (2010). Base map provided by Esri (2009). For location details including drainage area see Table S1 in Supporting Information S1. regime. We group locations by their DOR to test the relationship between annual storage effects and regulated climate outcomes (Figure 3b), with DOR calculated as where DOR is the DOR at site , SV is the storage volume of any reservoir upstream of site , is the total number of reservoirs upstream of site and AV is the unregulated annual streamflow volume at site (Grill et al., 2019).

Seasonal Volumes
Reservoir releases vary widely throughout the year and operational constraints are highly seasonally dependent. As a result, changes in seasonal streamflow volumes have been identified by stakeholders as indicators of system vulnerability for a wide range of water management objectives (RMJOC, 2020). The hydro-regulation models used to generate the regulated flow ensemble examined in this study use rule-curves and other operational targets that vary by season based on historical hydroclimate (RMJOC, 2020); however, the seasonality of unregulated hydrographs is projected to shift under climate change ( Figure 1) and large shifts in seasonal volumes may drive large regulation effects. We compare climate change effects on seasonal volumes by examining the relative seasonal volume change, defined as the ratio of future seasonal volumes to control period volumes, under both unregulated conditions and regulated conditions for September, October, November (SON; autumn); December, January, February (DJF; winter); March, April, May (MAM; spring); and June, July, August (JJA; summer).

Level of Seasonal Volume Regulation
The level of seasonal volume regulation (LRsv) is defined as the ratio of regulated seasonal volume to unregulated seasonal volume, where the seasonal volume is the total amount of flow observed at one of the 75 sites identified in Figure 2. LRsv values greater than 1 indicate the regulated seasonal volume exceeds the unregulated seasonal volume while

of 18
LRsv values less than 1 indicate the regulated seasonal volume is less than the unregulated seasonal volume. We compare control period LRsv to future LRsv to examine how the relationship between regulated and unregulated volumes is changing in the future. It is important to keep in mind that both the numerator and denominator change when applying Equation 2 to different periods.

Extremes
An analysis of changes in extremes can provide critical information for adaptation planning given extensive flood risk management practices and competing demands for water. We investigate regulation effects on high flow extremes by comparing relative changes in annual peak flows with a 50-year return period (Q50RP) under regulated conditions to those under unregulated conditions. The Log-Pearson 3 (LP3) distribution curve is fit to regulated and unregulated annual maximum flow time series for each 30-year period (Text S1 in Supporting Information S1). By using a 30-year sample size rather than a larger (e.g., 50 or 75-year) sample size, we limit the effects of non-stationarity in sample statistics used to generate the LP3 curve. We examine 50-year return period (2% annual exceedance probability) maxima rather than 100-year return period maxima due to lower confidence in the 1% annual exceedance probability that results from using a 30-year sample size. From the LP3 distributions, 50-year return period peak flow changes are investigated by calculating the ratio of future to control period Q50RP. The LP3 fit for unregulated high flow extremes is recommended by United States Geological Survey (USGS) Bulletin 17C (England et al., 2018) which established federal guidelines for flood frequency analysis. Regulated high flow frequency curves are typically generated using graphical fitting methods; however, we use the LP3 distribution to fit both unregulated and regulated high flow frequency curves in order to apply a consistent method across a large number of sites. Warming temperatures will shift streamflow maxima toward winter where they historically occurred in spring (indicated by widespread regime shifts across the domain as shown in Figure 2), and seasonally varying operations could explain large changes in regulated Q50RP. To identify seasonal climatic changes and operations that drive annual Q50RP changes, we also examine changes in cool-season (October-March) Q50RP and warm-season (April-September) Q50RP.
We investigate regulation effects on low flow extremes similarly, by comparing relative changes in 7-day minimum flows with a 10-year return period (7Q10) under regulated conditions to those under unregulated conditions. The LP3 distribution curve is fit to regulated and unregulated 7-day minimum flow time series for each 30-year period (Text S2 in Supporting Information S1). From the LP3 distributions, we examine changes in the 10-year return period 7-day minimum by calculating the ratio of future to control period 7Q10. High snow dominance can lead to annual minimums occurring during cool-season snowpack accumulation (Tohver et al., 2014; see Figure 1). Shifts in dry-season low flow extremes can indicate ecosystem vulnerability and motivate changes in late summer and early autumn ecological operations. Rather than taking the 7Q10 from the annual time series, we limit our analysis to the dry-season (July-October) when low flow operational constraints occur across the domain (RMJOC, 2020).
Model evaluation for high and low flows is performed by comparing the simulated high and low flow statistics to observed using the No Regulation No Irrigation (NRNI) data set (Figures S18 and S19 in Supporting Information S1). NRNI data represents flows that would have been observed without the effects of regulation or irrigation (RMJOC, 2018). Except for the Pend Oreille, which exhibits an approximate 20%-30% high bias for high flows (Q50RP), the ensemble median is within 10% of the NRNI Q50RP across nearly all sites. For low flows (7Q10), an overall high bias exists ranging from approximately 10%-30% on the mainstem of the Columbia relative to the NRNI flows.
The Kolmogorov-Smirnov (KS) and probability plot correlation coefficient (PPCC) goodness of fit tests (Stedinger et al., 1993) are performed to test the suitability of the LP3 analytical curves for fitting the unregulated and regulated low and high flow extreme time series (Figures S3-S14 in Supporting Information S1). Across periods and locations, approximately 83% (89%) of regulated and 91% (100%) of unregulated high flow (low flow) LP3 curves are accepted by the KS test. Approximately 92% (89%) of regulated and 99% (100%) of unregulated high flow (low flow) LP3 curves are accepted by the PPCC test.

Level of Q50RP Regulation
Similar to the level of seasonal volume regulation defined in Section 2.4, we define the level of Q50RP regulation (LR Q50RP ) as the ratio of regulated Q50RP to unregulated Q50RP, LR Q50RP greater than 1 indicates that the regulated Q50RP exceeds the unregulated Q50RP while LR Q50RP less than 1 indicates the regulated Q50RP is less than the unregulated Q50RP. Because the Q50RP amounts are determined independently from the regulated and unregulated flow time series, they do not necessarily denote the same event.

Unregulated Streamflow Projections
Unregulated streamflow projections are taken from Chegwidden et al. (2017). This data set consists of Columbia River basin simulated streamflow at a daily time step from the Precipitation Runoff Modeling System (PRMS; Leavesley et al., 1983) and Variable Infiltration Capacity model (VIC; Liang et al., 1994). Both PRMS and VIC were forced using statistically downscaled CMIP5 GCM projections for the RCP 4.5 and RCP 8.5 emissions scenarios. For the purposes of this study, we limit our analysis to emissions scenario RCP 8.5 which represents the amount of radiative forcing that is projected to occur if no effort is made to decrease greenhouse gas emissions. GCM forcings were statistically downscaled and bias-corrected at the 1/16th degree grid resolution using two different methods: the multivariate adaptive constructed analogs (MACA) method, and the bias correction, spatial disaggregation (BCSD; Wood et al., 2004) downscaling method. Ten GCMs, two meteorological downscaling methods, two hydrology models, and three model parameter sets for the VIC model resulted in an 80-member ensemble of unregulated streamflow projections for RCP 8.5 at locations across the Pacific Northwest (Chegwidden et al., 2019). From this data set, we analyze streamflow changes at 75 Columbia River basin locations that map to sites where hydro-regulation was modeled by USACE and USBR.

Regulated Streamflow Projections
Regulated streamflow projections were developed by USACE and USBR. With the exception of the Yakima, Upper Snake, and Deschutes, regulation across the basin was simulated at a daily step using the USACE Hydrologic Engineer Center's Reservoir System Simulations model (HEC-ResSim) (USACE, 2013) developed by USACE for Columbia River basin planning studies (RMJOC, 2020). This model was recently updated with operating rules based on a preferred alternative from a National Environmental Policy Act (NEPA) Environmental Impacts Study of system operations. These include an updated set of operational constraints and targets for 14 major storage projects that integrate water management for improved anadromous fish habitat and survival (USACE, 2020).
Regulation in the Yakima River basin was simulated by USBR at a daily time step using the RiverWare model (Zagona et al., 2010). The Yakima regulation model was developed to simulate operations and irrigation under 2010 conditions (Bureau of Reclamation, 2010b; RMJOC, 2020). Regulation in the Upper Snake and Deschutes was modeled at a monthly time step using MODSIM (Labadie, 2006) to simulate operations and irrigation under 2008 conditions (Bureau of Reclamation, 2009Reclamation, , 2010aRMJOC, 2020).
Reservoir storage targets and outflows vary inter-annually, year-to-year, and spatially, specific to in-season forecasts of seasonal runoff. Monthly storage requirements are determined by seasonal runoff volume forecasts (water supply forecasts) that are issued at the beginning of each month. These forecasts are required by the hydro-regulation models as input for flood risk management operations. In-season water supply forecasts for nine primary storage projects were developed for each projection using principal component regression and the projected hydrologic states. These forecasts were then used as input to the hydro-regulation models to set local and system operational storage requirements across the domain (RMJOC, 2020). While reservoir operational patterns can vary annually and seasonally based on forecasted basin hydrology, the underlying logic for forecast-based storage requirements is defined based on current hydrologic patterns and operational constraints and was not 10.1029/2022WR031950 8 of 18 modified to account for shifts in hydrology under climate change. For example, the April-August runoff period, for which forecasts are currently made, may be less relevant in the future when runoff occurs earlier.
The unregulated streamflow projections described in Section 3.1 were input into the hydro-regulation models after adjustments were made to account for the effects of irrigation and reservoir evaporation. Irrigation and evaporation extractions were based on historical patterns and depletions adjusted to the level of irrigation of the year 2010 (Bonneville Power Administration, 2011), and were not modified to reflect changes in diversion patterns under climate change. In the Upper Snake, Deschutes, and Yakima subbasins, irrigation withdrawals varied based on historically observed demand patterns for the projected water year type (i.e., irrigation demand patterns during drought years were different than those during extremely wet years) (Bureau of Reclamation, 2011; RMJOC, 2020). Outside of these subbasins, where irrigation depletions represent a much smaller amount of the annual flow volume, irrigation withdrawals were fixed based on historical depletion levels (RMJOC, 2020). The resulting model output is an 80-member ensemble of regulated Columbia River basin streamflow projections for the RCP 8.5 emissions scenario.

Regulation Dampens Seasonal Volume Changes in Winter (DJF) and Summer (JJA)
We investigate projected seasonal volume changes by taking the ratio of future volumes to control period volumes and compare ratios under unregulated and regulated conditions ( Figure 4). Results show changes across all seasons for both conditions by the 2030s and 2070s with the largest shifts occurring by the 2070s. We limit discussion of seasonal volume results to the 2070s, when the greatest changes and differences between unregulated and regulated outcomes occur.
Autumn (SON) unregulated volumes experience the least change out of all seasons (generally less than 25% change across locations) and the direction of change varies spatially. The greatest unregulated volume changes occur in winter (DJF) due to increases in precipitation and more cool-season precipitation falling as rain rather than snow.  year. In the summer (JJA), unregulated volumes are projected to decrease across locations. The summer months are historically water limited. Shifts in snowmelt timing coupled with warmer and drier summers drive large reductions in snowmelt-driven streamflow. The greatest summer volume reductions occur at locations in the Yakima, Spokane, Lower Snake, and Pend Oreille where snow-dominant regimes shift to transient by the 2070s (greater than 50% percent change).
The effects of regulation vary spatially in autumn and spring. Autumn unregulated volumes at upstream sites of the Upper Columbia (Mica; MCD and Revelstoke; RVC) decrease by 8%-14% but augmentation effects under regulation result in increases of 17%-20%. These strong regulation effects diminish downstream (Table S2 in Supporting Information S1). The opposite effects occur in the Yakima and Upper Snake. Except for a single location in the Yakima, autumn unregulated volumes increase by 2%-13% while regulated volumes decrease by 4%-38%. In the spring, sites in the Upper Snake that transition from snow-dominant to transient exhibit the greatest differences between unregulated and regulated volume changes where regulation amplifies change (greater relative change under regulation).
Regulation generally dampens change (less relative change under regulation) in winter and summer. Winter unregulated volumes are projected to increase by over 200% at some locations; however, regulation significantly reduces these changes where upstream regulation (DOR) is greater than 30%. Many locations show large winter unregulated volume increases but no projected change or decreases under regulation. These effects predominantly occur downstream of headwater reservoirs in the Upper Columbia, Kootenai, and Pend Oreille that are snow-dominant well into the future or remain snow-dominant through the 2030s and have large DOR. For example, at Hungry Horse (HGH) in the Pend Oreille subbasin, winter unregulated volumes increase by 140%, but regulation results in a 23% decrease. As in winter, summer regulation results in dampening of the climate signal downstream of headwater reservoirs where DOR is large. For some locations in the Yakima and Upper Columbia, summer low flow augmentation results in future summer volume increases where unregulated volumes decrease.

Level of Seasonal Volume Regulation Explains Large Regulation Effects in Winter (DJF) and Summer (JJA)
In Section 4.1.1, we showed that seasonal volumes are projected to change for both unregulated and regulated conditions; however, regulation effects on the magnitude and direction of change vary widely across seasons. Much of this can be explained by seasonally varying operations that alter flow timing and the seasonality of the annual hydrograph. Figure 5 shows the relationship between regulated and unregulated flow volumes and how these effects are projected to change in the future.
Spring (MAM) straddles the period of the strong snowmelt pulse which typically occurs late spring/early summer. Streamflow across the basin is primarily snowmelt driven and control period peak flows typically occur during the spring freshet (see Figure 1). In the spring, reservoirs begin refilling (storing large volumes of water) for spring flood risk management and LRsv (Equation 2) is less than 1 because regulated flow volumes are less than unregulated flow volumes. Stored spring volumes are later used to augment dry season low flows, and winter drafting of reservoirs increases flood storage space in preparation for the next spring freshet. As a result, control period LRsv is greater than 1 across autumn (SON), winter (DJF), and summer (JJA) for locations where DOR is greater than 60%.
By the 2070s, large changes in the LRsv occur in winter and summer when regulation effects on volume changes exhibit the strongest patterns (widespread dampening effects in winter and summer). Unregulated winter volumes are projected to increase significantly; however, as the system is operated to maintain flows and reservoir storage for the management of flood risk, regulated volume changes are relatively smaller and LRsv-values approaches unity. Unregulated summer volumes are projected to decrease. Summer system operations maintain low flow conditions and flow augmentation results in less change under regulation and LRsv-values that exceed 1 where DOR is large.

Large Regulation Effects on Q50RP Flow Changes Occur at Headwater Tributary Sites Where DOR Is Large
Unregulated annual Q50RP flows are projected to increase by the 2070s across the domain (Figure 6a). The largest unregulated increases occur in the Yakima, Pend Oreille, Spokane, and Upper Snake subbasins (ordered from greatest increase to least). The effects of regulation vary spatially. Regulation significantly dampens changes in the Yakima and Spokane, tributaries where hydrologic regimes shift to rain-dominant by the 2070s. Regulation amplifies Q50RP changes for a number of locations across the domain. The greatest amplification effects occur downstream of headwater reservoirs in the Pend Oreille (Hungry Horse; HGH), Kootenai (Libby; LIB, Duncan; DCD), and Lower Snake (Dworshak; DWR). Strong regulation effects also occur at Arrow Lakes (ARD), a reservoir in the Upper Columbia. Amplification effects from these reservoirs diminish further downstream where dampening effects occur, particularly in the Kootenai and Upper Columbia ( Figure S15 in Supporting Information S1).
We take a closer look at seasonal Q50RP changes to determine whether differences between unregulated and regulated conditions are driven by changes in the cool or warm season. During the cool season, unregulated Q50RP flows are projected to increase across the domain (Figure 6b) as a result of enhanced winter precipitation. Regulated Q50RP flows are also projected to increase; however, operations result in significantly less change where DOR is greater than 30%. Some of the largest dampening effects occur at Hungry Horse, Libby, and Dworshak. By the 2070s, regulation at Libby results in no cool-season change. At Arrow Lakes and Duncan, 2070s cool-season unregulated Q50RP flows exhibit increases of 63% and 74%, respectively; however, regulation amplifies these changes to 112% at both sites.
The warm season is a period when peak flows are driven by the spring freshet and climate change effects during this period are spatially variable (Figure 6c). Warm-season unregulated Q50RP flows are projected to decrease by the 2070s for regions that exhibit regime shifts to rain-dominance (Willamette, Spokane, and Yakima). Increases are projected for all other locations. Regulated changes generally follow unregulated changes. Exceptions occur at Hungry Horse, Libby, and Dworshak, where warm-season regulation results in greater relative change.
Arrow Lakes and Duncan remain snow dominant through the 2070s, yet regulation dampens the warm-season signal and amplifies the cool-season signal indicating that annual amplification effects are driven by cool-season increases in regulated flows. Hungry Horse, Libby, and Dworshak see amplification effects in the warm season, indicating that annual amplification effects are driven by warm-season increases in regulated flows. Figure 7 shows the level of Q50RP regulation (LR Q50RP ) (Equation 3) averaged by region for the cool season and the warm season. Across seasons and regions, LR Q50RP remains less than 1 in the future indicating that as the unregulated Q50RP increases the system still reduces unregulated high flow extremes in the future (also see Figure  S17 in Supporting Information S1); however, regulated Q50RP will generally increase ( Figure 6). Cool-season LR Q50RP -values decrease in the future (regulated Q50RP is significantly less than unregulated Q50RP) indicating the system is largely reducing cool-season unregulated floods. In contrast, warm-season LR Q50RP -values show little change or increases in the future (as unregulated Q50RP flows increase, regulated Q50RP flows also increase), indicating that regulation has little effect on the warm-season Q50RP signal and may be less effective at reducing unregulated high flow extremes in the future.

Low Flow Extremes (7Q10)
Unregulated 7Q10 flows are projected to decrease by both the 2030s and the 2070s across most sites (Figure 8)

Discussion
Seasonally, regulation dampens winter and summer flow volume changes where DOR is greater than 30%. Unregulated seasonal volume changes are largest in winter, a period when precipitation is projected to increase the most and warmer temperatures result in more cool-season precipitation falling as rain rather than snow. All locations exhibit future increases in unregulated winter volumes and there is strong agreement across the ensemble in the direction of change (Table S3 in Supporting Information S1). Regulated winter volumes also increase across most sites, but water management operations result in significantly smaller relative changes. In the summer, warmer and drier conditions drive decreases in unregulated volumes, and, like winter, models agree on the direction of summer change (Table S5 in Supporting Information S1). Regulation generally reduces summer volume changes and sites with very large DOR exhibit increasing volumes due to large summer flow augmentation.
These results align with other studies that have investigated regulation effects on changing conditions in the Columbia River basin. Jones and Hammond (2020) investigated historical trends in inflows and outflows for  large reservoirs across the basin and found that during the dry season, inflows to reservoirs decreased while outflows increased due to the effects of low flow augmentation. Zhou et al. (2018) examined regulation effects on the timing of climate signal emergence (defined by the timing of hydrologic regime shifts) across the western US. Regulation effects in the Columbia River basin showed high seasonal dependence and delayed the timing of climate signal emergence during winter and summer.
The seasonal dependence of regulation effects can be explained by seasonal water management operations and projected future hydroclimate. Seasonal reservoir storage and the delayed release of inflows alter streamflow timing. In the Columbia River basin, reservoirs store large snowmelt driven spring volumes that are later used to augment late summer/early autumn flows and are then released (drafted) throughout winter in preparation for the next spring freshet. This delayed release results in summer, autumn, and winter reservoir outflows that exceed unregulated flows ( Figure 5). As unregulated summer volumes decrease in the future, reservoirs release more water to augment lower summer volumes resulting in a dampened summer climate signal and, also, reservoirs that are less full by winter. As unregulated winter volumes increase, less full reservoirs release less inflow to meet spring flood risk management objectives resulting in smaller relative change and a dampened winter signal.
The results for autumn and spring vary spatially. Unlike projections for the winter and summer, flow projections for the autumn and spring exhibit uncertainty in the direction of change across the ensemble (Tables S2 and S4 in Supporting Information S1) driven by large uncertainty in precipitation patterns (RMJOC, 2018). Nevertheless, results for autumn and spring can be explained by the seasonality of operations. For snow/transient subbasins that shift to transient/rain by the 2070s, regulated autumn volumes decrease where little to no unregulated change occurs. As snowpack decreases and summers become drier, large summer augmentation effects could result in less water stored in reservoirs by autumn and consequently, decreased autumn outflows. Spring regulation effects vary depending on hydroclimate and site-specific operational constraints. March through May straddles the onset of the spring refill period. As warming shifts snowmelt timing earlier, reservoirs that historically empty through late spring could experience an amplified spring signal if large volumes that occurred during reservoir refill shift earlier to periods of drafting.
Regulation results in dampening and amplification of high flow extreme changes and these effects exhibit high seasonal and spatial dependence. For most locations, winter regulation significantly reduces relative increases in cool-season extremes when unregulated high flow extremes are projected to increase the most as a result of enhanced precipitation; however, warm-season and annual changes are spatially variable. Unregulated annual Q50RP values are projected to increase across the domain (Figure 6a). This is in agreement with other studies that used the same unregulated flow data set to investigate changes in future extremes (Chegwidden et al., 2020;Queen et al., 2021). Outflow from historically snow dominant headwater reservoirs where DOR is large exhibit amplification of annual Q50RP changes, but these effects generally diminish downstream where, in many cases, dampening occurs.
The phenomenon of regulated flows exhibiting greater sensitivity to climate change has been discussed before, although, not in the context of extreme flows. Zhou et al. (2018) found that some regulated basins in the western US are projected to be more sensitive to the climate signal, experiencing earlier shifts in the hydrologic regime relative to unregulated conditions. They explain this phenomenon as the result of less variation in the seasonality of a "flattened-out" hydrograph under regulation. Small seasonal shifts in outflow from reservoirs during periods when streamflow is historically regulated can lead to greater relative change under regulation. For example, if a reservoir releases less water after a high flow extreme event in the past (historically low reservoir outflow after an unregulated high flow event) but releases more water after these events in the future, the changes under regulation can be large. At Hungry Horse (HGH), Libby (LIB), and Dworshak (DWR), high elevation headwater reservoirs in the Pend Oreille, Kootenai, and Lower Snake subbasins, respectively, hydrologic regimes shift from snow-dominant to transient or near-transient by the 2070s (Figure 2). Regulation for these locations reduces (flattens) the seasonality of the annual hydrograph ( Figure 1). Operations at these sites result in significant dampening of cool-season extreme changes ( Figure 6b) and amplification of warm-season changes (Figure 6c). Each of these reservoirs is operated for winter and spring flood risk management (RMJOC, 2020). In a transitional climate, large increases in the magnitude of unregulated winter flood events could result in difficulty in meeting spring draft and refill requirements and lead to higher reservoir outflows in the future and an amplified signal.
In contrast, sensitivity studies for rain-dominant basins in East Asia found that reservoir operations resulted in widespread dampening of high flow extreme changes (Dong et al., 2019;Wang et al., 2017;Yun et al., 2021). Wang et al. (2017) studied regulation effects in the Lancang-Mekong River basin and found that the largest attenuation effects occur in headwater basins where DOR is large and weaken downstream. They argue that stronger regulation effects occur at upstream reservoirs due to relatively smaller annual discharges. We show the largest amplification effects occur at historically snow-dominant headwater reservoirs with large DOR and generally diminish downstream where, in most cases, dampening occurs ( Figure S15 in Supporting Information S1). These contrasting effects are likely due to historical regime patterns. The regulated flows used in this study result from seasonal operations based on current and historical water demands that do not change to account for large regime shifts from snow dominant. As streamflow timing shifts in the future, historically based patterns of reservoir draft and refill result in greater outflow during periods when it was historically regulated. Amplification upstream and dampening downstream can be explained by the effects of operations. Headwater reservoirs will store more water during larger unregulated flood events thereby reducing the signal downstream. Water stored during these events is released after the events when downstream flood risk is reduced, which could locally lead to future increases in high flows but have less effect further downstream.
Although large changes in high flow extremes occur for both regulated and unregulated conditions, flood risk management operations continue to reduce unregulated floods into the future (Figure 7 and Figure S17 in Supporting Information S1). Increases in regulated high flow extremes do not necessarily indicate increased flooding downstream. This study identified these increases and linked them to higher reservoir outflows; however, did not examine the likelihood of high flows reaching levels where flood damages occur.
Unregulated July through October low flows are generally projected to decrease and the effects of regulation vary spatially. Significant dampening occurs in tributaries where DOR is large. High DOR locations have the highest augmentation effects in the future ( Figure 5) and consequently, large regulation effects on the low flow extreme signal. Regulated flows on the mainstem are susceptible to the natural climate signal exhibiting little to no regulation effect on low flow changes.
Spatial patterns in the regulated 7Q10 response may be the result of spatially variable operational objectives. Dudley et al. (2020) compared 7-day low flow historical trends between unregulated and regulated gages across the US. In the western US, unregulated low flows exhibited a downward trend. Regulated low flows generally exhibited an upward trend indicating that reservoir operations mitigated climate change impacts on low flows; however, regulated trend results were mixed which they described as a likely consequence of diversity in regional climate and operational purpose. Large DOR occurs downstream of tributary and headwater reservoirs where annual discharge volumes are smaller relative to the mainstem. These regions could be more susceptible to low flow-associated risks (e.g., ecosystem vulnerability) and large DOR could indicate more operational flexibility to dampen large shifts in low flow extremes. DOR is smaller on the mainstem, and results could indicate less operational flexibility in the future to meet low flow storage targets during extreme events. Jones and Hammond (2020) showed that while water management in the Columbia River basin has historically adjusted to long-term downward trends in low flows, operational risks associated with the inability to meet seasonal flow targets increased during periods with lower than average flows.
Columbia River basin reservoir operations vary seasonally and serve multiple and sometimes competing system objectives. In the autumn, winter, and spring, reservoirs are primarily operated to provide storage for flood risk, hydropower, and ecological health. Summer flows and volumes are important for meeting hydropower, navigation, recreation, and irrigation demands (RMJOC, 2020). Climate change is projected to shift the hydrologic regime from snow-dominant to transient or rain-dominant, with large increases in unregulated winter volumes and decreases in summer volumes. Reservoir operations generally reduce these changes by storing larger inflows in the winter and augmenting more in the summer. High and low flow unregulated extreme events will be exacerbated under climate change. The patterns of seasonal reservoir operations result in dampened effects for cool-season high flow increases, but the regulated system exhibits sensitivity to high flow increases during the warm-season. Decreases in low flow extremes are dampened by regulation in tributaries but sensitivity on the mainstem could result from operations that are less flexible given historically based objectives. The reservoir system's ability to meet competing operational objectives in the future may be strained by large shifts in seasonal hydrology; however, as streamflow timing shifts earlier in the water year, the operational patterns of seasonal storage (drafting to create flood space for the spring freshet and refill) could also shift, potentially reducing sensitivities. For example, adjustment of refill operations to accommodate shifts in snowmelt timing and volume could increase the likelihood of refill and the effectiveness of the system to store water to augment lower dry-season flows. The system may also encounter new competing objectives. An example of competing objectives is the increased frequency of bi-modal flood seasons. Storing inflows to reduce winter flooding could interfere with drafting reservoirs to create flood storage for the spring freshet.
This study examined the effects of regulation on climate change outcomes in the Columbia River basin; however, we attribute outcomes to broad basin characteristics across a large domain and results may be generalized for other regulated snow and transient basins. Large regime shifts in hydrologically diverse basins combined with seasonal and multi-objective reservoir operations result in regulated outcomes that vary widely across time and space. Considering the role of regulation in climate sensitivity analysis, particularly for basins that exhibit strong seasonality, is the first step in identifying adaptive management needs.

Limitations
The regulated flows examined in this study result from operating criteria based on the current operational constraints of the system. It is unlikely that seasonal operational patterns will remain static in the future with large shifts in seasonal hydrology. As climate and water demands shift, water management will need to adapt to changing conditions to continue meeting the multi-objectives of the system and operational patterns will need to change. Here we identified the effects of climate changed hydrology under current operating criteria and highlight the importance of considering reservoir operations in climate sensitivity analysis. While modeling the effects of modified operations is not within this scope, these findings will inform future work on adaptative management design for climate change.
We characterize outcomes by the medians of large ensembles of seasonal volume and extreme flow changes. The ensembles were driven by a diverse set of climate scenarios, statistical downscaling methods, and hydrologic model configurations that capture a wide range of uncertainty not examined in this paper for the sake of concisely describing key differences for several metrics important for water management. Furthermore, while rigorous model validation and bias-correction techniques were applied at each element of the modeling chain during data set development (Chegwidden et al., 2019;RMJOC, 2018RMJOC, , 2020, there are uncertainties associated with each step in the process not addressed in this paper (Chegwidden et al., 2019(Chegwidden et al., , 2020Queen et al., 2021;RMJOC, 2018RMJOC, , 2020Rupp et al., 2021).
The unregulated, modeled low flows (7Q10) exhibit a high bias relative to the NRNI reference data set ( Figure  S19 in Supporting Information S1); however, our findings regarding the effects of regulation are largely insensitive to the magnitude of this bias. For example, in the upper, middle, and lower Columbia, where unregulated, modeled flows exhibit a high bias of approximately 10%, 30%, and 20%, respectively, the effects of regulation are similar independent of the bias or of the relative change in unregulated 7Q10. For tributaries, the lack of sensitivity to the low flow bias reflects the seasonality of operations. July through October low flow extremes are small relative to the amount of water stored during the winter and spring and low flow shifts are dampened by regulation regardless of the magnitude of model bias or shift in the unregulated low flow. That is, low flow changes are dampened whether the unregulated future to control period ratio is 0.8 or 0.4. On the mainstem, where DOR is smaller and annual discharge is higher, operations may be less constrained by shifts in low flows and regulation has little effect regardless of the relative shift in unregulated low flow.
Additional uncertainty was introduced by extrapolating extreme events from analytical distributions (LP3) fit to the unregulated and regulated empirical time series. As we report in Supporting Information S1, two goodness of fit tests were performed on these analytical distributions and, generally, the LP3 curve provided robust estimates for both high and low flow extremes for most locations across the ensemble (Figures S3-S14 in Supporting Information S1). Some exceptions occurred for estimates at highly regulated headwater locations; however, despite these results there was strong correlation between the analytical and empirical distributions for a majority of ensemble members and the LP3 analytical fits provided a consistent framework for evaluating all locations.

Conclusions
Regulation modulates the seasonality of the annual hydrograph. The signature of regulation on streamflow patterns varies across time and across the basin. Reservoir operations result in significantly less change for winter and summer streamflow volumes at locations where DOR is large, but results for autumn and spring vary widely depending on local operational constraints and hydroclimate. Regulation effects on high flow extreme changes are also variable. Winter operations reduce changes in cool-season high flow extremes for locations where DOR is large. Annually and in the warm-season, regulation at historically snow-dominated headwater reservoirs amplifies the climate signal on high flows. These increases in flow reflect changes in reservoir release patterns as the system attempts to meet operational objectives under different hydrological conditions. In some cases, the operations developed for historical hydrological conditions are less effective in meeting these objectives as the hydrology changes. In many cases, not adjusting operations for streamflow timing and regime shifts results in greater relative high flow changes under regulation. Dry-season low flow extreme outcomes are dependent on location in the river network. On the mainstem, the regulated system exhibits sensitivity to low flow extremes following changes in unregulated low flows; however, for tributaries where upstream DOR is large, regulation significantly dampens low flow changes.
The reality of freshwater systems world-wide is that the majority are heavily fragmented by reservoirs (Grill et al., 2019). This study has shown that water resource infrastructure and reservoir operations are a constraint that can have large effects on climate outcomes, particularly in snow-dominant watersheds where large regime shifts challenge historically based assumptions. These effects will have implications for managed freshwater systems and the future of water resources in regulated systems. By accounting for the role of regulation in climate sensitivity analysis, a more accurate characterization of climate outcomes will help inform where and how to adapt water management systems for a future climate.