The Colorado River is a major water source and economic engine for the southwestern United States (US). Nearly 90% of its water originates from the snow-dominated watersheds in Colorado, Utah and Wyoming1 to support 40 million people and $1.4 trillion in economic activity2. The Colorado River is currently grappling with a multi-decadal drought3,4 with legal and political tensions rising to develop voluntary cuts in water use as a result of diminishing supply5. Declining streamflow and reservoir storage, however, cannot be explained solely by lower precipitation in its headwater basins. As an example, in 2021, the Upper Colorado River Basin (UCRB) received 80% normal snowpack but delivered only 30% of average streamflow to its receiving reservoir6. Accelerated warming in mountain basins may be part of the explanation7. Warming reduces snowpack accumulation by shifting more precipitation toward rain, enhancing mid-season melt events and increasing snow vapor losses8. Increased snow drought and associated snowpack declines are observed across the western US since the mid-20th century 9,10 and the UCRB is projected to move toward persistent low-to-no snow conditions over 50% of its area by the end of this century11. Research has indicated that changes in snowpack with warming are expected to decrease mountain streamflow through preferential partitioning of precipitation toward increased vegetation water use in the soil zone12,13.
While effects of warming on snowpack and soil water processes have been explored, there remains considerable uncertainty on the relative importance of groundwater sourcing to mountain streamflow14. The prevailing assumption has been that these systems have limited groundwater storage and steep hydraulic gradients with streamflow more reliant on surface processes that control runoff and interflow, or shallow subsurface water flow through the soil and regolith15,16. However, there is growing evidence that groundwater contributes 10 to 50% the annual streamflow in mountain catchments of the Colorado River17–19. While stored groundwater release has the potential to buffer streamflow to short-term climate extremes20–22, how groundwater storage will respond to multi-decadal warming and how this may affect streamflow generation in mountain basins is largely unknown. Understanding groundwater flow to mountain streams is hindered by the need to quantify the interaction of climate and snow dynamics23–26, soil water storage and loss to vegetation27, and their combined effects on groundwater generation, or recharge28. Modeling these processes in mountain basins at the appropriate resolution (100–250 m29) is numerically expensive, while obtaining sufficient data to constrain is difficult due to challenges in site access and harsh climate30.
To address these challenges, we focus on the East River, Colorado (750 km2, Fig. 1a). The East River is representative of many headwater basins of the UCRB31. The climate is continental subarctic, and it spans nearly 2 km in vertical relief (2440–4300 m). It is home to the largest atmosphere-to-bedrock observational network in the US with investment across multiple federal agencies, national laboratories, university partners, and regional stakeholder groups32. Data includes airborne mapping of snowpack33,34, vegetation35, and bedrock resistivity36, as well as dense arrays that monitor snowpack, groundwater, and streamflow. Data informs an integrated hydrologic model of the basin37. A 100-m grid resolution represents topographic complexity and a daily timestep accounts for energy and water partitioning between the snowpack, vegetation, soil zone and bedrock. Details on the land surface model construction and calibration are provided in previous work38. Here it is dynamically linked to a three-dimensional groundwater flow model39 that allows for streamflow gaining and losing conditions40. The groundwater model accounts for nine stratigraphic units41 and extends 400 m below land surface (Fig. 1b). Inclusion of the deep subsurface is based on the growing evidence that non-trivial groundwater flow to streams can occur at considerably greater depth than previously believed42, even in fractured crystalline43 and metamorphic rock44. Simulated depth to the water table is provided in Fig. 1c with open water indicating groundwater-fed perennial streams. Refer to the Methods and Supplemental Information for details on data sources (Table S1), historical climate and model construction (Fig. S1-S9, Table S2).
Hydrologic sensitivity to temperature is tested by applying + 4°C to historical minimum and maximum daily temperatures for water years 1987 to 2022 with no change in precipitation (water year 1 Oct to 30 Sept). This extrapolates locally observed temperature and (lack of) precipitation trends to roughly the end-of-century (Table S3). Independently, sensitivity to fall, winter, spring and summer warming is tested by applying temperature increases to only to those months defining a particular season. The experimental design allows a direct comparison to the historical baseline condition and a comparison of relative effects of season (see Methods for limitations). Using this data- modeling framework we explore historical water budgets and the relevance of groundwater sourcing to streams. We then ask how sensitive is streamflow generation in a significantly warmer climate and how and where does changing groundwater storage factor in to estimated changes in streamflow?
Conceptual model of mountain hydrology and the importance of lateral flow
To predict groundwater to streams in mountain basins it is necessary to track the entire water budget, including snow dynamics, soil moisture, evapotranspiration (ET), and lateral subsurface flow. Modeled historical conditions indicate annual precipitation (P) is high (947 ± 176 mm) and arrives primarily as snow (Fig. 2). The summer monsoon delivers approximately 20 ± 6% of annual water input and this is important for sustaining vegetation water demand in the late summer45. At the basin-scale, the East River is generally energy-limited such that aridity, or the ratio of potential evapotranspiration (PET) to P is less than 1.0. Despite being a wet and cold climate, simulated ET accounts for more than half the water exported out of the basin with the majority of ET derived from the soil zone. Results align with observed data46 and fit the Budyko framework47 such that the ratio of ET/P is tightly coupled to aridity with soil water increasingly used to support ET under drier conditions (Fig. S10). While ET sourced from groundwater (ETg) is a fairly small portion of the average annual water budget at the basin scale (2 ± 1%), it becomes increasingly important in years when aridity is relatively high, especially when summer monsoon rains fail. Despite ET/P sensitivity to interannual variation in aridity, the total ET volume is consistent year-to-year to suggest the East River is soil storage limited48 with average annual soil moisture relatively insensitive to winter and spring precipitation that represent the bulk of water inputs to the basin. Instead, annual soil moisture is more sensitive to summer temperatures and monsoon rain (Table S4). Warm and dry conditions in the summer and fall promote dry soils that last through the winter months49 to reduce recharge, interflow to streams and streamflow efficiency in the subsequent snowmelt season (Fig. S11).
Groundwater generation is divided into areal recharge and seepage. Areal recharge is the vertical flux of water from the soil zone into the deeper subsurface. Areal recharge is relatively low in the East River as a consequence of steep topography, limited soil storage, and permeability contrasts between the soil and underlying bedrock that preferentially partitions excess soil water toward lateral movement, or interflow. Interflow moves snowmelt downgradient towards topographic convergent zones to form streams. Stream channels that are disconnected from groundwater are non-perennial with a portion of their flow seeping back into the underlying groundwater system as a function of hydraulic head differences between the stream water surface and water table elevation. Similar to previous mountain studies50,51, this topographic focusing of snowmelt and gradient-driven seepage loss is estimated as the dominant form of recharge. Interflow and runoff that remains in the river channel (combined Qs) represents the bulk of streamflow in the East River (74 ± 29%). Qs is highly responsive to interannual variability in aridity and plots nonlinearly as an inverse image to ET/P indicating its tight coupling to vegetation water use (refer to Fig. S10d for stream sources). Groundwater discharge to streams (Qg) is estimated at approximately 12% the total water budget and represents 26 ± 7% of total annual streamflow, though extreme wet and dry water years stretch the fraction of Qg in streamflow between 8% and 48% in response to highly variable Qs contributions. In contrast to Qs, volumetric inputs of Qg are a stable source of streamflow that are relatively insensitive to aridity. Instead, Qg responds as a direct, exponential function of groundwater storage (r = 0.85, p < < 0.01) (Fig. S12).
Elevation is a first-order control in the ER, with codependent relationships between climate and vegetation that influence water storage and movement (Fig. 3). Broadly, the alpine occurs above tree line (> 3500 m), the upper subalpine is a mixture of lower density conifer forests, shrubs, and barren ground (3200–3500 m), the lower subalpine is predominantly conifer forests mixed with aspen and meadows (2800-3200m). The montane (< 2800 m) is dominated by shrubs. The transition between the upper and lower subalpine is distinguished as the transition between energy and water limitation during historically low precipitation water years and coincides with the maximum density of conifer cover38. Simulated snow water equivalent (SWE) is orographically controlled, but this is modified by wind and avalanche26 to push maximum snow accumulation towards tree line51. The combined effect of soil evapotranspiration, sublimation, and canopy interception loss (ETa) is largest where PET = P during historically dry conditions. Under average and wet conditions maximum ETa occurs at lower elevations with the downgradient movement of excess interflow 52. At lower elevations and years with low precipitation, groundwater subsidizes ETg.
While average annual basin-scale ET volume is relatively stationary from one year to the next, the spatial distribution of ET does vary (Fig. S13). ET increases during years with low precipitation at higher elevation as aridity approaches unity and the growing season lengthens through reductions in snow-covered area and warmer seasonal temperatures. At lower elevations, ET decreases during years with low precipitation due to water limitation. The largest decreases are along south-southwest aspects. Exceptions occur within riparian corridors where shallow water tables occur, and groundwater can support vegetation water use even under drought. Annual precipitation in excess of streamflow and ET results in a net gain in subsurface storage (dS = P-ET-Q > 0). This occurs primarily in the alpine and upper subalpine zones. Excess water supports both areal recharge and moves laterally through the soil zone to support downgradient streamflow and evapotranspiration in water-limited portions of the basin (dS < 0). Results agree with previous studies in which the lateral transfer of high elevation snow melt (and nutrients) subsidizes lower elevation carbon stocks53, above-ground productivity54 and vegetation55. Maximum streamflow generation is simulated to occur in the forested subalpine as a mixture of both Qs and Qg with the influence of groundwater on streamflow remaining stable but increasing in its relative contributions at lower elevations and during dry water years as soil water preferentially partitions away from Qs and toward vegetation water use.
Hydrologic response to a warmer climate
We compare the annual changes in hydrologic storage and their associated fluxes due to multi-decadal warming during the driest water year with its historical equivalent across different elevations (Fig. 4, refer to Fig S14-15 for other water year examples). With + 4°C warming across all seasons, water-limited conditions shift to higher elevations, and the East River becomes nearly snow-free. Sustained snowpack is concentrated only in the alpine and upper subalpine. Snowpack reductions and increased atmospheric demand drive soil moisture loss across all elevations but losses are greatest at the interface between the upper and lower subalpine where conifer density is highest. Soil moisture reductions are not substantial above tree line where atmospheric demand remains less than precipitation and thin soils are largely reset even under these extremely dry and warm conditions. Despite large losses in snow accumulation, high elevation remnant snowpack is still able to support increased ET56 but does so at the expense of downgradient interflow-derived streamflow and a net loss to recharge. Groundwater released from storage at lower elevations is also reduced in comparison to the historical condition (less negative, or DSg > 0). Despite this reduction in groundwater release, groundwater discharge maintains support of ET at the historical level at lower elevations (DET ~ 0) but is unable to sustain historical groundwater discharge to streams. Qg is reduced volumetrically by 20% with the largest reductions occurring in the montane. Summing Qs and Qg, total streamflow is reduced by 32% in the low precipitation water year example, with summer and fall temperatures responsible for two-thirds of this loss.
Groundwater storage declines amplify streamflow LOSS.
Groundwater storage aggregates climate over a 4-year period. Historically, this multi-year aggregation helps buffer extremely dry water years and promotes groundwater recovery back to the historical mean condition (Fig. 5a). Recently, however, the onset of low/no monsoon rain from 2016–2020 combined with the shock of an extremely dry water year in 2018, has prompted sustained declines in groundwater storage with losses simulated at greater than 2-standard deviations of the historical mean. This agrees with groundwater observations in the East River that indicate water table declines from 2018 are not recovering57 (Fig. S16) despite above average snow accumulation in 2017, 2019 and 2022. Applying + 4°C warming to historical climate sequencing results in a groundwater system unable to recover to the historical average condition even during very wet periods simulated. The largest drops in storage occur after each exceptionally dry water year and losses exceed the historical low point after the first of these events. At the end of the hypothetical warming simulation, groundwater storage loss exceeds six standard deviations of the historical mean and does not appear to have stabilized. Results indicate summer temperature increase is responsible for the largest relative declines in groundwater storage and is nearly twice as influential on storage loss compared to spring warming with effects most pronounced during exceptionally dry years (Fig. S17).
Water table declines due to multi-decadal warming are nonuniformly distributed across the East River with changes largest in the subalpine (Fig. S18). Multiple linear regression analysis indicates high relief catchments covered by conifer forests are expected to experience the largest declines in groundwater storage in response to long-term warming (refer to Methods). Forests have higher rates of ET compared to non-woody vegetation58 and will potentially out-compete for access of deeper water pools59, while dense canopies are responsible for large interception losses to evaporation that are expected to increase in a warmer climate. These processes will reduce the downgradient transfer of interflow, recharge and result in lower water tables60. Extrapolating across the UCRB distinguishes regions most at risk of groundwater decline in a warmer future (Fig. 5b) and identifies possible areas for future forest management to increase groundwater sustainability61. Specifically, mountainous areas in the Colorado Headwater Basin and the Gunnison Basin are at elevated risk in a warmer future. Normalizing initial subsurface storage conditions (refer to Methods), we find gradient-driven seepage loss results in 10–24% annual streamflow decline and 12–27% reductions in the 7-day minimum flow, with the influence of seepage loss on streamflow declines increasing at higher aridity (Fig. 6). Seepage loss is twice as effective at reducing annual streamflow compared to warming with no change in groundwater storage. Mitigation of temperature effects on annual streamflow declines occurs because sharp decreases in summer streamflow are largely offset by increased winter and early spring streamflow in response to earlier snowmelt that occurs when atmospheric demand and vegetation water use are low (Fig. 7). Seepage loss effects on summer flows are smaller but push the East River toward a dry riverbed during below average precipitation that are not predicted if the groundwater system is ignored.
summary
Mountain snowpack is a critical reservoir for downstream water use worldwide62. Dwindling supply is aggravating increased lowland demand in basins like the Colorado River, promoting political tensions and a need to better understand mechanisms of mountain discharge for better adaptive management. Hydrologic modeling studies have identified increased ET as a primary mechanism of streamflow reduction in a warmer climate, largely through decreased albedo with snow loss12, but with spatial variability in response largely dictated by elevation13. While groundwater contributions are increasingly acknowledged as important to mountain streamflow, limited analysis has been done at the resolution for capturing recharge in complex terrain and linking to both snow and ET dynamics. We use a high-resolution, integrated hydrological model extended into the deep subsurface to determine if groundwater will buffer or amplify streamflow losses in a warmer world. Results for a representative headwater basin in the UCRB indicate that groundwater contributions have historically been a significant and stable contribution to stream water (26 ± 7%) and directly responsive to groundwater storage which aggregates climate over a 4-year period to minimize climate extremes. Groundwater sourcing occurs at higher elevations where precipitation inputs are in excess of atmospheric demand and discharges downgradient in water limited portions of the basin to support ET and stabilize groundwater sourcing to streams. Warming by + 4°C reduces snowpack, with snow accumulation largely confined to the highest elevations in the basin. Increased atmospheric demand and longer growing seasons promote more ET in the alpine and upper subalpine, while subsurface water subsidizes lower elevation ET at (or above) historic levels at the expense of interflow and groundwater discharge to streams. With warming, groundwater storage reductions exceed the historical minimum after the first extremely dry water year and are unable to recover to historical conditions even after multiple wet periods simulated. Given the cold and wet nature of this headwater basin, summer warming appears the most influential on groundwater storage loss through elevated ET and reduced soil moisture that adversely affects recharge in the subsequent snowmelt period. Resulting water table reductions promote increased seepage loss from streams to partially buffer recharge declines, but in doing so, significantly reduce annual streamflow, and forces minimum summer flow toward intermittent conditions that are not predicted if groundwater interactions are not considered. Water table declines are non-uniformly distributed. High relief catchments occupied by conifer forests are at particular risk. Results help identify those regions in the UCRB that may benefit from the joint management of forest and water resources to minimize streamflow reductions in a warmer future. Our research stresses that including groundwater response to multi-decadal warming is necessary to forecast future response of mountain hydrology.