Role of surface-water and groundwater interactions on projected summertime streamflow in snow dominated regions: An integrated modeling approach

Authors


Corresponding author: J. L. Huntington, Division of Hydrologic Sciences, Desert Research Institute, 2215 Raggio Pkwy., Reno, NV 89512, USA. (justinh@dri.edu)

Abstract

[1] Previous studies indicate predominantly increasing trends in precipitation across the Western United States, while at the same time, historical streamflow records indicate decreasing summertime streamflow and 25th percentile annual flows. These opposing trends could be viewed as paradoxical, given that several studies suggest that increased annual precipitation will equate to increased annual groundwater recharge, and therefore increased summertime flow. To gain insight on mechanisms behind these potential changes, we rely on a calibrated, integrated surface and groundwater model to simulate climate impacts on surface water/groundwater interactions using 12 general circulation model projections of temperature and precipitation from 2010 to 2100, and evaluate the interplay between snowmelt timing and other hydrologic variables, including streamflow, groundwater recharge, storage, groundwater discharge, and evapotranspiration. Hydrologic simulations show that the timing of peak groundwater discharge to the stream is inversely correlated to snowmelt runoff and groundwater recharge due to the bank storage effect and reversal of hydraulic gradients between the stream and underlying groundwater. That is, groundwater flow to streams peaks following the decrease in stream depth caused by snowmelt recession, and the shift in snowmelt causes a corresponding shift in groundwater discharge to streams. Our results show that groundwater discharge to streams is depleted during the summer due to earlier drainage of shallow aquifers adjacent to streams even if projected annual precipitation and groundwater recharge increases. These projected changes in surface water/groundwater interactions result in more than a 30% decrease in the projected ensemble summertime streamflow. Our findings clarify causality of observed decreasing summertime flow, highlight important aspects of potential climate change impacts on groundwater resources, and underscore the need for integrated hydrologic models in climate change studies.

1. Introduction

[2] There is growing consensus that increased greenhouse gas (GHG) concentrations in the global atmosphere are causing long-term changes to the Earth's climate [Christensen et al., 2007]. The combination of rising GHG forcings, ongoing natural-climate variability, and uncertainty in climate model projections make future climates more uncertain for water resource managers [Brekke et al., 2008]. Additionally, the fact that hydrologic processes, such as runoff, recharge, and evapotranspiration (ET), all covary in time and space, and are correlated to each other, makes it difficult to analyze cause and effects for any one hydrologic process without an integrated framework to model all these processes simultaneously. In environments where summertime streamflow and groundwater discharge is critical for water resources and biological demands, an accurate understanding of the causality of historical and future hydrologic change during these periods is especially important.

[3] The mechanisms causing observed historical and projected hydrologic change in high-elevation catchments is poorly understood, especially regarding surface water/groundwater interactions (SW/GW). For example, streamflow records across the Western United States indicate predominantly decreasing summertime flow [Kim and Jain, 2010], and 25th percentile annual flows [Luce and Holden, 2009] where groundwater discharge is a major component of the total streamflow. These opposing trends could be viewed as paradoxical, given that several studies suggest that increased annual precipitation will equate to increased annual groundwater recharge, and therefore high summertime flow [Jyrkama and Sykes, 2007; Allen et al., 2010]. Many hydrologic modeling studies support observed decreases in summertime flow, asserting that earlier snowmelt and runoff is the primary cause [Hamlet and Lettenmaier, 1999; Wilby and Dettinger, 2000; Dettinger et al., 2004; Scibek et al., 2007; Mantua et al., 2010; Maurer et al., 2010]. Although these modeling studies provide an explanation of decreasing summertime flow, shifts in snowmelt and runoff timing alone are not complete explanations. Additional clarification on the causality of decreasing summertime flow, and ties to changes in hydrologic timing are needed to assess historical and future trends [Luce and Holden, 2009]. A thorough understanding of the linkage between changes in snowmelt timing and SW/GW interactions will help address an important question in hydroclimate research, that is, how do changes in snowmelt and streamflow timing impact groundwater resources and groundwater-derived surface water resources?

[4] Recent findings show significant shifts in the timing of snowmelt and observed streamflow in several watersheds in the Sierra Nevada [Coats, 2010], and vulnerability of groundwater to changing climate in the region [Singleton and Moran, 2010]. The purpose of this work is to develop a process-based explanation for decreasing summertime flows that have been reported by previous investigators by using an integrated modeling framework to analyze changing SW/GW interactions. We show that decreased summertime flow is likely part of a broader hydrologic change that is occurring due to earlier onset of the snowmelt pulse and the resulting earlier seasonal drainage in these watersheds. Six different climate model projections are used to force the hydrologic model and demonstrate that projections of earlier snowmelt recession results in decreased summertime flow over a wide range in projected precipitation amounts, including both decreasing and increasing long-term precipitation trends. The use of multiple climate projections are important for providing greater evidence for our explanation of why summertime flows are decreasing because the period of record for these watersheds is short, and thus the climate projections provide greater credence to the statistical significance of decadal or longer trends in the historical streamflow data.

[5] To simulate the effects of earlier snowmelt runoff on watershed drainage and SW/GW interactions, we rely on the integrated SW/GW interactions model, GSFLOW. Both observed historical data, as well as climate model projections for the 21st century are used to evaluate the significance and implications of decreased summertime flow in the Sierra Nevada. Projections of future hydrologic conditions complement the historical simulations by allowing for a longer simulation period to discern persistent shifts in hydrologic conditions. Models are constructed for three snow-dominated watersheds of the eastern Sierra Nevada tributary to Lake Tahoe and Truckee Meadows hydrographic areas of California and Nevada (Figure 1). The study area is of special interest with regard to water resources because it is representative of many low-permeability bedrock snow-dominated mountainous regions of the Western United States that provide primary water supplies to nearby developed watersheds. The study area is representative of the greater Sierra Nevada because topography, geology, climate, and hydrology are similar over much of the upland regions, where precipitation is the greatest. Important characteristics that are shared among the upland (i.e., >2000 m) watersheds of the Sierra Nevada are the large topographic relief and relatively impermeable shallow bedrock that accentuate the dominance of shallow groundwater-flow paths in the regional system. Because the alluvial aquifers are small and have limited storage, the alluvial aquifers are likely to be more sensitive to climate fluctuations than large valley aquifers. There is additional interest in the drainage processes within the Incline and Third Creek watersheds because these watersheds transport sediment and nutrients to Lake Tahoe, which is nationally recognized for its clarity and recreational value.

Figure 1.

Study area illustrating Incline Creek, Third Creek, and Galena Creek watersheds and model domain (thick black line indicating watershed boundaries).

1.1. Modeling Background

[6] Due to model limitations and computing constraints, simulating climate change effects on groundwater hydrology typically has been done with compartmentalized models, in which SW/GW interactions are decoupled or neglected [Vaccaro, 1992; Middelkoop et al., 2001; Scibek et al., 2007; Jyrkama and Sykes, 2007; Tague and Grant, 2009; Allen et al., 2010]. In these studies, if the unsaturated zone is explicitly considered, it is represented as a soil column through which water flows independently of the underlying water table. These models calculate recharge independently of dynamic groundwater levels and SW/GW interactions. Furthermore, the important interplay between snowmelt-derived streamflow and SW/GW interactions are not simulated in a coupled manner, which we will show is a key process that must be considered to evaluate climate-change impacts on summertime flow in snow-dominated regions. In short, the effects of climate on the interactions between SW/GW and resulting summertime flow are not fully understood due to various compartmental model limitations and assumptions [Scibek et al., 2007].

[7] Recently, with the development of sophisticated computer codes, several studies have applied integrated models to simulate climate change effects on water resources [Maxwell and Kollet, 2008; Ferguson and Maxwell, 2010; Sulis et al., 2011]. These models have provided greater insight into climate change effects on watershed hydrologic processes due to their ability to more realistically simulate feedback between hydrologic processes that occur above and below land surface. Here, we add to these past works by calibrating over a longer period to evaluate the model's ability to simulate low-frequency variations in summertime flow that are associated with groundwater storage, considering climate projections from six climate models and two GHG scenarios, and projecting hydrologic conditions over the next century to assess the combined effects of low-frequency weather cycles and future climate change. Natural climate variability will be an important component of future climate conditions, and a good representation of these historical cycles allows for more realistic projections of water availability and the severity of climate extremes. Researchers have observed both interdecadal and intradecadal periodicities in precipitation and streamflow [Hanson et al., 2006; Perry, 2006], and groundwater levels [Hanson et al., 2006; Laque-Espinar et al., 2007]. These low-frequency signals have been linked to Quasi-Biannual Oscillation (QBO), El Niño Southern Oscillation (ENSO), Pacific Decadal Oscillation (PDO), tidal, and solar cycles [Barco et al., 2010; Burroughs, 2003]. Accurately predicting historical low-frequency responses is central to predicting future low-frequency responses in groundwater storage, discharge to streams and springs, and water-dependent biota. Integrated models that are calibrated to historical interactions of SW/GW over wet and dry periods, and are forced with future climate data over many decades, are better suited to assess how climate change might affect water resources, and in particular, groundwater resources.

1.2. Model Description

[8] GSFLOW was used to simulate all near-surface and groundwater hydrologic processes within three watersheds of the eastern Sierra Nevada (Figures 1 and 2). GSFLOW simultaneously accounts for climatic conditions, runoff across the land surface, variably saturated subsurface flow and storage, plus connections among terrestrial systems, streams, lakes, wetlands, and groundwater. Runoff and interflow (shallow subsurface flow) cascade to receiving streams or lakes, while including effects of saturation-excess runoff caused by shallow water table conditions. GSFLOW and its precursors have been applied in several basins across the United States to simulate SW/GW interactions [e.g.,Hunt et al., 2008; Markstrom et al., 2008; Niswonger et al., 2008; Doherty and Hunt, 2009; Koch et al., 2011].

Figure 2.

Three dimensional and cross section representation of the hydrogeologic framework model illustrating vertical and horizontal model discretization and hydrogeologic units.

[9] GSFLOW is the integration of the Precipitation Runoff Modeling System (PRMS) and the Modular Groundwater Flow model (MODFLOW). Integration of PRMS and MODFLOW was facilitated by an implicit iterative coupling approach using the Newton linearization method [Niswonger et al., 2011]. Markstrom et al. [2008] and Niswonger et al. [2011]provide a complete description of GSFLOW and its theory, and only a broad description is provided herein. PRMS is a modular deterministic, distributed-parameter, physical-process watershed model used to simulate precipitation, climate, and land use on watershed response [Leavesley et al., 1983]. PRMS simulates snowpack processes using a distributed two-layered system that is maintained and modified on both a water equivalent basis and as a dynamic heat reservoir. PRMS simulates snowmelt- and rain-generated runoff in a fully distributed sense, where runoff can cascade among four neighboring surface grid cells, reinfiltrate, or flow to a stream. The soil zone is represented by coupled continuity equations with storages that represent different components of soil porosity (i.e., dead-end verses kinematic and macropore porosity), conceptualized in PRMS as the preferential, gravity, and capillary reservoirs. Water in the soil zone can percolate into the deeper unsaturated zone (MODFLOW), flow horizontally to a receiving grid cell or stream, or evapotranspire to the atmosphere. In areas where the water table is above the base of the soil zone, groundwater can seep into the soil zone. Additionally, groundwater discharge occurs to the surface in areas where groundwater heads are above land surface.

[10] ET is derived from the vegetation canopy and land surface (sublimation from the snowpack and evaporation off of land surface), within the soil zone, and the deeper unsaturated and saturated zones. Evaporation also can be simulated from surface water, such as from the surfaces of lakes and streams. ET is simulated as a function of the potential (PET), water storage in the vegetation canopy and in the soil zone. Beneath the soil zone, ET is a function of the PET that is not satisfied from the soil zone, root available water content in the deeper unsaturated zone, and water table elevation in the deeper saturated zone. If the water table elevation is above the root depth (i.e., extinction depth) and the PET is not met by the soil and unsaturated zones, then ET is removed directly from groundwater using the formulation developed in the MODFLOW ET Package [McDonald and Harbaugh, 1988]. There are three options in GSFLOW for calculating PET. These formulas are empirical and rely on climate data including, air temperature, solar radiation, and elevation. For this work, the Jensen and Haise [1963]solar radiation-temperature empirical formulation for calculating PET was used.Markstrom et al. [2008]provide further details, including the distribution of climate data on the landscape and calculations of energy-budget components.

[11] Flow beneath the base of the soil zone is simulated by MODFLOW, including vertical unsaturated flow, groundwater flow, and with a wide variety of boundary conditions that represent streams, lakes, groundwater development, and many other hydrologic processes. Vertical unsaturated flow is simulated by MODFLOW using the Unsaturated-Zone Flow (UZF1) Package [Niswonger et al., 2006], in which unsaturated flow is simulated using the kinematic-wave equation. The relation between the unsaturated hydraulic conductivity and water content in the unsaturated zone is defined on the basis of the Brooks-Corey function [Brooks and Corey, 1966]. The version of MODFLOW used in this application of GSFLOW is called MODFLOW-NWT, which is a Newton formulation of MODFLOW-2005 that provides capabilities to simulate drying and wetting of groundwater cells [Harbaugh, 2005; Bedekar et al., 2011; Niswonger et al., 2011]. MODFLOW simulates three-dimensional (3-D) confined and unconfined groundwater flow using the conservative form of the continuity equation that is discretized using block-centered finite differences; groundwater head is calculated at the cell center, and flows are calculated at the interface between cells [Harbaugh, 2005]. Following the approach of MODFLOW for solving the 3-D unconfined groundwater-flow equation, the water table is resolved at the subgrid scale that allows a coarse vertical discretization of the subsurface without degradation of the unconfined solution. Similarly, unsaturated flow is simulated using the method of characteristics solution of the kinematic-wave equation that is not dependent on grid-cell thickness [Smith, 1983; Niswonger and Prudic, 2004; Niswonger et al., 2006]. Thus, vertical discretization of GSFLOW models is guided by geologic information rather than constraints associated with numerical stability and accuracy. However, the equations used in GSFLOW are more approximate than full 3-D Richards' equation, which results in some error that must be balanced against errors in parameterization. All surface water in GSFLOW, other than overland runoff, is simulated by MODFLOW packages, including the modified lake (LAK7) and streamflow routing (SFR2) packages [Merritt and Konikow, 2000; Niswonger and Prudic, 2005]. Readers are referred to Markstrom et al. [2008] for details regarding SW/GW interactions, including groundwater interactions with overland flow and lakes.

2. Methods

2.1. Model Setup

[12] Gridded datasets of elevation, geology, vegetation, soils, and land use were used to discretize and parameterize GSFLOW. Model cells were set to a 60 × 60 m spatial resolution over the 54-km2model domain. Climate was distributed spatially across the model (1,900–3,000 m above Mean Sea Level AMSL) based on the Parameter-elevation Regression on Independent Slopes Model (PRISM) mean monthly precipitation patterns [Daly et al., 1994], and daily temperature and precipitation recorded at the Natural Resource Conservation Service (NRCS) Mt. Rose SNOTEL station located at 2700 m elevation, and the Tahoe City National Oceanic and Atmospheric Administration (NOAA) cooperative-observer weather station, located within 20 km of the model domain at 1900 m elevation (AMSL). Mean annual precipitation within the model domain ranges from 380 to 1650 mm, with 90% of the precipitation occurring between November and March. Monthly average extreme temperatures range from 30°C in August to −10°C in January. Vegetation consists of subalpine and conifer forest, with some deciduous riparian and meadows association.

[13] Mountain block geology is composed of granodiorite and andesite, overlain with glacial moraines and stream deposits in low-elevation areas making up the alluvial aquifers, while soils generally are shallow and derived from parent rock consisting of mostly sand.Plume et al. [2009] recently compiled and evaluated geologic, geophysical, and hydrogeologic data for the study area for examining the extent and characteristics of the hydrogeologic units that comprise the aquifers. Spatial hydrogeologic and stratigraphic data reported by Plume et al. [2009] were used to develop the conceptual hydrogeologic framework model (HFM) and vertical and horizontal model discretization (Figure 2). Alluvium in these watersheds consists primarily of decomposed granite, glacial outwash, and stream deposits. Accordingly, the alluvial layers increase in thickness around the streams and toward Lake Tahoe [Plume et al., 2009]. Figure 2c illustrates a cross section of the HFM for the Incline Village area, starting at the mountain block and ending near the Lake Tahoe shore line. The thickness of the layers representing the alluvium follows values provided by well logs and geophysical data, and was linearly interpolated on the basis of distance from a stream channel and valley bottom in order to define areas between data locations. Based on hydrogeologic and stratigraphic data reported by Plume et al. [2009], Incline, Third, and Galena Creek watersheds were discretized vertically into five layers, and horizontally into approximately 16,500 grid cells per layer, for a total of 83,000 active cells. The model was divided into four basic geologic units, including topsoil, alluvium, weathered bedrock, and less-weathered bedrock. Coarse (three layers) and fine (eight layers) vertical layering models also were developed to test the effects of vertical discretization on hydrologic response, while keeping geologic units the same for all models. The five-layer model produced similar results as the eight-layer model, and thus, the five-layer model was adopted for this work.

[14] Drainage in these watersheds occurs rapidly due to the great topographic relief and relatively shallow, permeable aquifers that sit on the low-permeable bedrock. The main stem and tributary stream channels drain the shallow soils and alluvial aquifers such that nearly all recharge within the Third and Incline Creek watersheds discharges to streams before entering Lake Tahoe, as indicated by observed shallow groundwater gradients at the base of the watershed. This process also was supported by simulation results that showed that subsurface groundwater flowing to the lake was negligible over a broad range in model parameters. Recharge in Galena Creek watershed partially drains as groundwater beneath the stream valley and alluvial fans that extend east to the valley bottom. Based on the steep topography near the watershed divides, no-flow boundary conditions were assigned along the edges of the model domain that coincide with watershed divides. Head-dependent flux boundary conditions were set where a portion of the model extended into Lake Tahoe, and where the streams crossed the model boundary at the outlets of the model domain. Lake Tahoe water surface elevation was used to represent the head-dependent boundary condition for the portion of the lake in the model. Land surface slopes were used to define groundwater gradients at the boundary conditions beneath where the stream crossed the model boundary. The upper soil and alluvial layers (layers 1–3) were assigned a zero thickness where there were no soils or where bedrock outcropped at land surface.

[15] The stream network was divided into 861 stream reaches, where a stream reach is the length of a stream that is contained within a single model grid cell. Streams were delineated using a geographic information system according to the contributing area method, where a minimum threshold was used to define streams that correlated with field observations and stream delineations from 1:24,000 topographic maps. Streams were delineated in order to define their subgrid geometries; however, defining a stream reach does not require that water flows in the reach. Stream reaches naturally flow and dry depending on whether there is runoff or subsurface flow entering the reach. Generally, all streams are perennial in the study area and serve as drains for shallow aquifers, except in the upper reaches where flow is intermittent. Stream cross-sectional geometries, slopes, and lengths for each reach were estimated from surveys using a differential global positioning system and a 10 m digital elevation model. Runoff that occurs on a grid cell that does not contain a stream reach is assumed to flow as surface flow or interflow according to the overland-flow routing equations in PRMS.

[16] In many surface water model parameterizations, shrub and tree root depths, which affect plant available soil water and ET, generally are assumed to be between 0.4 and 2 m, but limited to the depth of the soil zone [Leavesley et al., 1983; Liang et al., 1994; Flerchinger et al., 1996]. However, it has been documented that roots extend beneath the soil zone and into weathered bedrock and bedrock fractures [Stone and Kalisz, 1991; Canadell et al., 1996; Hubbert et al., 2001]. In the Sierra Nevada, at least 70% of the water used by the forest during the growing season is extracted from weathered bedrock and bedrock fractures from at least 3.5 m [Witty et al., 2003], as this unsaturated zone stores more than twice as much plant-available water by virtue of its greater thickness as compared to the soil layer [Hubbert et al., 2001]. For these reasons, and given the ability to model ET derived from the deeper weathered bedrock and bedrock unsaturated and saturated zones using the Unsaturated Zone Flow Package (UZF1) in GSFLOW [McDonald and Harbaugh, 1988; Niswonger et al., 2006], and on the basis of calibration, roots were assumed to extend to a maximum of 4 m below land surface.

[17] Calibration of the integrated model followed two different conceptual models (CM1 and CM2) in order to determine the most accurate conceptualization of drainage from these watersheds on the basis of the analytically derived water balance, observed streamflows, and groundwater heads. CM1 proposed that the major watershed drainage mechanism consists of snowmelt recharging shallow alluvial aquifers that drain to streams. Because alluvium is shallow and overlays bedrock, saturation excess runoff is likely to occur in response to the water table rising to land surface, particularly near streams and wetlands. CM1 relies on the unconfined groundwater-flow equation solved by MODFLOW to simulate most of the lateral subsurface flow, whereas the soil zone has very low storage and smaller capability to conduct water to streams. CM2 assumes that most of the lateral subsurface flow occurs through macropores in the soil zone. In this case, the soil zone has significant storage and conductance, and is represented by the kinematic-wave formulation to simulate lateral subsurface stormflow, as calculated by PRMS [Beven, 1981; Markstrom et al., 2008]. Saturation excess runoff is assumed to play a lesser role in CM2 due to the ability of the soil zone to conduct water laterally, resulting in faster drainage of shallow groundwater. During snowmelt periods, macropore flow was observed around eroded boulders and holes within the shallow soils and decomposed granite. Furthermore, overland runoff outside of the channels and wetlands was mostly nonexistent.

2.2. Calibration

[18] For calibration purposes, the model was forced with historical temperature and precipitation observations from Mt. Rose SNOTEL and Tahoe City NOAA weather stations, in which streamflow was simulated during an 18 year historical period (1992–2008). The model was calibrated using a 3 step process. For the first step of the calibration process, PRMS was calibrated independent of MODFLOW for the 18 year period by matching observed streamflows. PRMS was manually calibrated and a separate calibration was performed for each conceptual model (CM1 and CM2). This was done by running PRMS for 1 year as a “spin-up” period to establish initial storages in the soil zone. The calibration procedure consisted of a multiobjective, stepwise procedure where PRMS is calibrated first by adjusting parameters that affect the distribution of solar radiation and potential ET in order to match the average flow of water through the watershed and observed annual water balance. Simulated snow covered area (SCA) was then compared to SCA estimated from satellite remote sensing data derived from the MODIS Terra instrument to verify the simulated timing and spatial distribution snow pack development and melt. Some adjustment of the PRMS snowpack module parameters was required to better simulate the timing of snowmelt, specifically the parameters that determine the shape of the snowpack areal depletion curve for each grid cell. Following calibration of the snowpack module, parameters that affect the timing and magnitude of runoff and shallow subsurface flow were then adjusted until the model provided a good fit between the simulated and observed daily streamflow. Goodness of fit between the simulated and observed daily streamflow was assessed using the Nash-Sutcliffe statistic [Nash and Sutcliffe, 1970].

[19] For the second step of the calibration process, MODFLOW was run independent of PRMS (MODFLOW-only) using a steady state stress period (i.e., storage terms in the groundwater-flow equations were set to 0). Long-term average recharge rates estimated by the PRMS-only simulations were used for the steady state recharge distribution. The steady state recharge rates were scaled until there was a good correspondence between the simulated steady state flows in streams and the 18 year average of the observed summertime flow in the streams. The steady state groundwater simulation was calibrated by adjusting aquifer hydraulic conductivity values and matching groundwater levels, summertime flow, and the locations of springs in the watersheds. Mapped wetland and spring areas were used to calibrate the steady state groundwater model by comparing the surface elevations of wetland spring areas to the spatial distribution of simulated heads that were within 1 m of the land surface. For the third step of the calibration process, GSFLOW was run in integrated mode, and the aquifer storage parameters were adjusted to match observed variations in observed low flows and to match dominant frequencies in climate signals exhibited in the streamflow data (i.e., 6 month, 1, 2, 3, and 11 year periods). Additionally, further refinement was done by adjusting parameters that affect the timing and rates of runoff and subsurface flows to the stream. The integrated model calibration was assessed on the basis of the goodness of fit between simulated and observed streamflow on the basis of the Nash-Sutcliffe statistic and root-mean-square error of groundwater heads and wetland areas.

2.3. Future Climate Forcing

[20] To assess future hydrologic change and to extend the simulation period, bias-corrected and spatially disaggregated general circulation model (GCM) projections of daily temperature and precipitation were used as direct input to GSFLOW. The projections came from six different GCMs that contributed to the Intergovernmental Panel on Climate Change (IPCC) Fourth Assessment [Christensen et al., 2007], considering the Special Report on Emissions Scenarios A2 and B1 scenarios. We used data from six climate models and for two greenhouse gas (GHG) emission scenarios to consider uncertainty in future hydrologic conditions [Hay and Clark, 2003; Prudhomme et al., 2003; Wilby and Harris, 2006]. Downscaled projections of temperature and precipitation from 2010 to 2100 at 12 km resolution were developed from the bias-corrected spatial disaggregation (BCSD) method [Maurer and Hidalgo, 2008; Cayan et al., 2009] for GCMs of CNRM CM3.0, GFDL CM2.1, MIROC3.2 (med), MPI ECHAM5, NCAR CCSM3, NCAR PCM1, responding to B1 and A2 GHG scenarios. Specifically, climate model projections were taken from two 12 km grid cells that were coincident with the Mt. Rose SNOTEL and Tahoe City NOAA weather stations. These projections were further interpolated to the Mt. Rose SNOTEL and Tahoe City NOAA weather stations using a quantile-quantile mapping approach [Panofsky and Brier, 1968] to account for biases in temperature and precipitation due to elevation differences between 12 km GCM projections and weather-station elevations. Comparison between the 12 km GCM projected climate and the historical climate observed from the climate station clearly indicated the need for this second level of bias correction. The final resolution of climate data after bias correction and spatial distribution using average spatial relations provided by PRISM was equal to the hydrologic model grid cells (60 m). Hydrologic simulations were run using daily time steps on 12 desktop computers, one for each GCM forcing. The steady state, 18 year historical, and 100 year projections required approximately 10 s, 12 h, and 3 days of computational time, respectively.

3. Results

3.1. Calibration

[21] Results of the steady state groundwater-model calibration indicate that the groundwater model was able to simulate the limited amount of observed heads and the locations and extent of small wetlands and spring areas within the watershed, without defining any structural features or heterogeneities in our hydraulic conductivity fields beyond the original HFM (i.e., additional geologic heterogeneities or faults that act as barriers or conduits for flow), indicating that nearly all springs and wetlands in these watersheds are topographically derived (Figure 3). However, the model did not predict groundwater discharge to land surface for two of the mapped springs. These springs are not topographically controlled like the other springs in these watersheds. For these two springs, CFC-estimated apparent ages are more than 15 years older than other springs and near-stream seepage faces that were sampled, and have apparent ages older than samples from wells screened about 200 m below land surface. CFC apparent ages of the two springs that are not considered to be topographically controlled were 38 and 41 years, whereas the apparent ages of all other springs were less than 15 years. Adjustments to the model input were not made to better simulate these two structurally controlled springs because of uncertainties in their origin and because structurally controlled springs that originate from deep groundwater are considered less important for this study relative to shallow groundwater-discharge areas.

Figure 3.

Spatial distribution of groundwater heads within 1 m of land surface shown as red-transparent grid cells where the hydraulic conductivity was varied by (a) a factor of 0.1 lower than calibrated values for all layers, (b) a factor of 10 greater than calibrated values for all layers, and (c) calibrated values for all layers. Black hallow grid cells are specified stream cells, and thick black polygons illustrate mapped springs and wetlands. The optimally calibrated case (Figure 3c) illustrates that the water table intersects the land surface in areas that coincide with mapped streams, springs, and wetlands.

[22] Using spring and wetland locations to constrain the steady state calibration proved very useful. For example, Figure 3 shows a sensitivity analysis that demonstrates the tightly constrained aquifer hydraulic conductivity (K) values. The spatial distribution of heads within 1 m or above land surface was plotted for K distributions that were scaled by factors of 0.1 and 10 of the calibrated K values. For a factor of 0.1 (Figure 3a), it is evident that the model overpredicts heads, as it would only be expected to have heads within 1 m or above land surface around springs, wetlands, and perennial streams. For a factor of 10 (Figure 3b), it is evident that the model underpredicts heads in the upland areas and does not provide shallow groundwater levels around springs, wetlands, and perennial streams. Clearly, the calibrated K distribution provides the most accurate representation of wetland, spring, and perennial stream areas. Figure 4shows simulated versus observed heads in wells, and land-surface elevations for spring and wetland areas. The 1-to-1 plot (Figure 4a) shows that the model simulates the head distribution accurately over a wide range in head values, with RMSE and normalized RMSE values of 3.2 m and 0.4%, respectively. A small normalized RMSE (i.e., RMSE/total head loss) as shown in this work indicates that model errors are only a small part of the overall model response [Anderson and Woessner, 1992]. The errors in simulated heads (Figure 4b) indicate that there is a slight bias in overpredicting the groundwater heads observed in wells, while underpredicting the heads in spring and wetland areas. Adding further complexity to the model to better match heads was not warranted given the model scale and uncertainty in these observation data. Most of the wells in the study are located in steep terrain, making direct comparisons between simulated and measured heads in these wells difficult due to the grid scale. Additionally, errors in simulated wetland heads are acceptable given the subgrid variability in land surface from which the wetland observation heads were derived. Water levels in wetland areas are not always at land surface, but near land surface and within the root zone of identified wetland areas. Thus, a bias toward underpredicting the wetland heads is consistent with our conceptualization of groundwater levels in the wetland areas. Additionally, annual average water balance calculations using observed streamflow and precipitation data were used to further constrain simulated ET values. The calibrated steady state water budget corresponded well with the 18 year annual average water budget; precipitation and streamflow were 350,000 and 158,000 m3 d−1, and 350,000 and 155,000 m3 d−1 for the simulated and observed values, respectively. The ratio of streamflow to precipitation for these watersheds is 45%, indicating that 55% of the precipitation in these watersheds is lost to ET. As previously discussed, groundwater flowing out of these basins through the subsurface is an insignificant component of the water budget. These results compare well with ET derived from recent watershed modeling, chloride mass balance, and Darcian flux estimates of water yield (i.e., runoff + mountain front recharge) in adjacent watersheds with similar geology, vegetation, and precipitation magnitudes [Maurer et al., 1996; Maurer and Berger, 1997; Jeton and Maurer, 2007].

Figure 4.

(a) Observed and simulated groundwater head, and (b) residual head error relative to the observed head.

[23] In accordance with the conceptual model developed for these basins, calibration results favored decreases in K with depth and there is a large contrast in K at the interface between the alluvium and shallow bedrock. Calibrated values for K decrease from 21.5 m d−1 for shallow soils to 1.4 m d−1 for the alluvium, and from 1.4 m d−1 to 0.005 m d−1 at the alluvium/bedrock interface (Table 1). Calibrated anisotropy (Kh/Kv) values for the alluvium are equal to 3.5 and the bedrock was assumed to be isotropic. Calibration results strongly favored CM2, suggesting that most of the lateral subsurface flow occurs through macropores within the soil zone (represented as an equivalent porous media in the model). A good fit to the streamflow hydrograph could not be attained for CM1 because saturation excess runoff occurred over all reasonable ranges in model parameters and this resulted in a hydrograph that had flows that changed much too abruptly relative to the observed streamflows. Thus, parameterizing the soil zone to represent a mixture of macropore and matrix flow provided the best fit to observed streamflows. Table 1lists calibrated saturated and unsaturated zone hydraulic properties. The calibration of hydraulic properties is robust as determined from many simulations that were run, despite the many input parameters that are required in the integrated model. This was mostly due to the large amount of relief in these watersheds and the distribution of groundwater-discharge areas that constrain aquiferK values (Figure 3c). Additionally, the character of the hydrograph and the gross water balance calculations put tight constraints on parameters that control flow and storage in the soil zone. However, despite observation constraints on model input, there is uncertainty around the estimated parameter values, especially K. The effects of uncertainty in K were assessed with regards to simulated streamflow using sensitivity analysis, as shown in climate projection section.

Table 1. Major Hydraulic Properties Used for GSFLOW Modela
 SoilsbShallow AlluviumcDeep AlluviumcShallow Weathered BedrockcDeeper Weathered Bedrockc
  • a

    The specific storage, Brooks-Corey exponent, and air-entry pressure were not required for layer 1.

  • b

    Flow in this layer is calculated using a kinematic wave formulation in PRMS.

  • c

    Flow in this layer is calculated using the groundwater flow equation in MODFLOW.

Horizontal hydraulic conductivity (m d−1)21.52.81.40.0050.001
Vertical hydraulic conductivity (m d−1)11.50.80.40.0050.001
Specific storage (m−1)1 × 10−61 × 10−61 × 10−61 × 10−6
Specific yield (unitless)0.180.180.100.0050.005
Brooks-Corey exponent (unitless)4.04.04.04.0
Saturated water content (unitless)0.23–0.480.350.350.0060.006
Air entry pressure (m)−0.15−0.15−0.15−0.15

[24] Results of the calibrated integrated model (i.e., PRMS + MODFLOW) indicate that historical daily streamflows are well simulated, with an average Nash-Sutcliffe value [Nash and Sutcliffe, 1970] of 0.73 (0.77 for log streamflow; 1.0 indicating a perfect fit), as are 6 month, 1–3 year, and 11 year periodicities exhibited in simulated streamflow for the 18 year period of record, where the 11 year cycle is the most notable of the cycles greater than 1 year (Figures 5a and 5b). The 6 month period in observed and modeled streamflow shown in Figure 5b is the result of October–November rain and snowmelt runoff along with later spring runoff that typically occurs about 6 months later, and is evident by close inspection of Figure 5a. The 11 year period in observed summertime flow is a result of precipitation and groundwater-recharge cycles and associated distributions of residence times of water flowing through the subsurface, effectively resembling the hydraulic memory of the watershed governed by climate, geology, and geomorphology [Smakhtin, 2001]. Periods of wet climate on the order of 2–4 years sustain increases in summertime flow, and seem to contribute to the strong 11 year period in streamflow. It is recognized that the statistical significance of the 11 year cycle in observed and simulated streamflow shown in Figure 5bis low given the limited period of record of only 18 years. However, after analyzing many long-term streamflow records in the region, the 11 year cycle is a common attribute and is statistically significant at the 95% confidence level when tested against red noise [Gilman et al., 1963]. Given that the model produces an 11 year cycle from the input precipitation, suggests that this cycle has significance, even if it is not statistically significant due to the short period of record.

Figure 5.

Simulated and observed (a) daily streamflow and (b) periodograms for Incline Creek.

[25] Spatial distributions of groundwater recharge during the winter (Figure 6a), early spring (Figure 6b), and late spring (Figure 6c) indicate that the greatest groundwater-recharge rates occur near stream channels, mountain fronts, and across the alluvial aquifers, where the alluvium is relatively permeable as compared to the upland bedrock areas. Recharge occurs in the upland bedrock areas; however, deep percolation in these areas is restricted by the relatively low vertical hydraulic conductivity of the weathered bedrock. These results are consistent with recent findings from a noble gas and isotopic tracer study of recharge in a nearby high-elevation catchment with similar geology, which suggests that most groundwater recharge to the alluvial aquifer occurs on the lower slopes of the catchment [Singleton and Moran, 2010].

Figure 6.

Spatial distributions of groundwater recharge (a) during winter, (b) early spring, and (c) late spring of 2005. Red grid cells indicate negligible recharge, where yellow, green, and blue grid cells indicate low, moderate, and high groundwater recharge, respectively.

[26] Recharge in the alluvial areas occurs quickly following the onset of snowmelt because of shallow water tables and high rates of deep percolation. Shallow water tables also result in saturated-excess runoff and subsurface stormflow near streams and groundwater discharge areas due to the lack of storage for deep percolation. Accordingly, during peak runoff, the simulated relative proportion of presnowmelt and postsnowmelt event water is dominated by postsnowmelt event water. This result is supported by observed math formula18O in snowmelt, streamflow, and spring samples taken from Third Creek and springs within the watershed. Figure 7 shows a time series of math formula18O from Third Creek for a 1 year period. The large proportion of snowmelt derived water in the stream is illustrated by the large change in stream math formula18O following the onset of the spring snowmelt period and the resulting peak flow in the stream. Following the hydrograph recession, math formula18O in the stream is representative of groundwater, as represented by spring flow and summertime streamflow math formula18O values (Figure 7).

Figure 7.

Time series of (a) collected snowmelt and rainfall amount and math formula18O, and (b) streamflow volume and math formula18O for Third Creek. Mean summertime flow and spring math formula18O also are shown for reference, which illustrates that the spring hydrograph is mostly composed of snowmelt.

3.2. Climate and Hydrologic Projections

[27] Climate-change simulations run with GSFLOW indicate that projected temperatures and precipitation strongly influence all water budget components. Daily average temperatures are projected for the study watershed from 2°C to 4°C for the B1 and A2 GHG scenarios, respectively, during 2010–2100 relative to the base period of 1950–2010. Long-term changes in projected precipitation also are apparent during the next century. For GHG scenario A2, four GCMs predict a steady decrease in annual precipitation, while the other two predict a steady increase in precipitation (Figure 8a1). For GHG scenario B1, five GCMs predict a steady decrease in precipitation, while one predicts a decrease up to about year 2040 and then an increase in precipitation for the remainder of the century (Figure 8b1). The ensemble 30 year annual average precipitation from 2010–2040 to 2070–2100 changes by −5 and −9% for the A2 and B1 climate scenarios, respectively. Discrepancies among GCMs in their projections of precipitation over the next century suggest a large amount of uncertainty in precipitation for these basins. Variations in long-term precipitation trends presented byCoats [2010] for various watersheds in the Sierra Nevada are consistent with variations in projected precipitation among GCM projections of precipitation. Thus, by using several GCM climate projections, we are able to (1) utilize a long period of record that cannot be developed using historical data alone, and (2) evaluate the mechanisms for decreasing summertime flow that is consistent across many watersheds in the Sierra Nevada that are experiencing disparate trends in precipitation.

[28] Similar to recent PRMS simulations using future climate for many watersheds across the U.S. [Hay et al., 2011; Markstrom et al., 2012], our results indicate that annual snow water content is projected to decrease for all GCMs and GHG scenarios due to increased temperature, snowmelt rates, and precipitation falling as rain (Figures 8a2–8b2). Annual streamflow projections mimic precipitation projections, with a majority indicating decreases (Figures 8a3–8b3). In analyzing projected streamflow, attributes of a 6 month streamflow cycle are common among GCMs; however, a 7 year cycle is most clearly evident among the projected streamflow results. Although less pronounced than in observed data, the streamflow simulated on the basis of the CNRM CM3.0 climate projection exhibited an 11 year signal that most closely corresponds to the observed streamflow in the area. GCMs do show some ability to project realistic weather cycles; however, improving the GCMs ability to better simulate these decadal weather cycles would make their projections and subsequent seasonal and decadal hydrologic responses more realistic.

[29] Annual overland runoff (i.e., infiltration excess and saturation excess overland flow) generally increases or remains steady for A2 and B1 scenarios, respectively (Figures 8a4–8b4). The increase in runoff is caused by more precipitation falling as rain, higher frequency of rain-on-snow events, and increased snowmelt rates, and has been well documented in snow dominated regions [Barnett et al., 2005]. Overland runoff typically is the fastest pathway to a stream, and increased overland runoff could result in larger peak streamflow rates and a greater occurrence of flooding, which has been previously pointed out for the region [Hayhoe et al., 2004].

[30] Annual groundwater recharge, groundwater storage, and groundwater discharge to streams exhibit a decreasing trend in four of the six A2 climate scenarios (Figures 8a5–8b7). Annual streambed losses increase for all simulations with decreased precipitation as a result of decreased groundwater heads beneath streams (Figures 8a7–8b7). Annual groundwater discharge and streambed losses generally are steady for the A2 MPI ECHAM5 and NCAR PCM1 GCMs (Figures 8a7–8a8), despite large increases in annual groundwater storage (Figure 8a6), which reflects the earlier drainage of the watersheds and decreased groundwater heads beneath streams following snowmelt recession. In summary, as annual precipitation, streamflow, and groundwater recharge decrease, so does the annual groundwater storage and discharge to streams, while at the same time, streambed losses to the aquifer increase. These results illustrate the important interplay between surface water and groundwater and underscore the need to run long-term simulations within an integrated modeling framework when making inferences about the effects of climate change on surface water and groundwater resources. Long-term simulations are important for these analyses because of the long-term autocorrelation exhibited in hydrologic variables that are related to groundwater storage.

[31] Broadly speaking, results in Figure 8 suggest that changes in annual precipitation drive changes in annual groundwater fluxes; however, seasonal variations in groundwater fluxes are driven by the timing of snowmelt runoff, and more directly by the depth of flow in streams. The effects of increased air temperature on the hydrology of these basins become clear when streamflow components are analyzed on a seasonal basis. To better demonstrate the interplay of seasonal stream gains and losses, Figure 9 illustrates simulations of these variables on a daily basis during a selected 2 year time period (2027–2028) for the CNRM CM3.0 climate model and A2 GHG scenario. The net groundwater discharge to streams is significantly reduced during peak snowmelt runoff due to the bank storage effect [Cooper and Rorabaugh, 1963; Pinder and Sauer, 1971]. The bank storage effect is important in these watersheds due to rapid runoff and interflow that elevates the stream head more abruptly than the rise in groundwater head near streams. Elevated stream head increases streambed losses to the groundwater and suppresses groundwater discharge to streams, effectively reducing the net groundwater discharge to streams (black line in Figure 9). Earlier snowmelt and streamflow increases the period of time during which groundwater drains to streams, where a longer groundwater-drainage period causes a decrease in July–October streamflow. These results indicate that there is an asymmetric shift toward earlier snowmelt recession that is not completely compensated by earlier onset of snowmelt, thus resulting in a longer period of groundwater drainage to streams during each year.

Figure 8.

Time series of simulated yearly average hydrologic variables. Simulated hydrologic variables for different GCMs (colored lines) and greenhouse gas emission scenarios (a) A2 and (b) B1. Time series were smoothed using a 30 year moving average. Hydrologic variables included (1) precipitation, (2) snow water content, (3) streamflow, (4) runoff, (5) groundwater recharge, (6) groundwater storage relative to initial conditions, (7) groundwater discharge to streams, and (8) streambed loss.

Figure 9.

Selected 2 year time series from 2027–2028 of projected streamflow, streamflow loss to groundwater, groundwater discharge to streams, and net groundwater discharge to streams (i.e., groundwater discharge to streams minus streamflow loss to groundwater), illustrating the seasonality of surface and groundwater interactions.

[32] Figure 10 was developed on the basis of the simulated results, and illustrates our conceptualization of the seasonal drainage of these watersheds. During winter, the snowpack builds, and cold conditions result in negligible recharge, groundwater storage is at minima from previous summer and autumn drainage, and streamflow (i.e., groundwater discharge) is at a minima (Figure 10a). During the spring snowmelt period, runoff and interflow fill the stream channels and elevate the stream head, suppressing groundwater discharge to the stream that causes the stream to lose water to the streambanks and deeper subsurface (Figure 10b). Higher stream head during the snowmelt period increases horizontal flow parallel to the stream in the down valley direction. Following the peak snowmelt runoff period, the stream head subsides and bank storage and regional groundwater seeps into the stream, resulting in peak groundwater seepage to the stream (Figure 10c). Shallow aquifers surrounding the stream are then drained and groundwater discharge to the stream decreases and reaches a minimum during summer and early autumn that is exacerbated by riparian ET (Figure 10d). The transition from a gaining to a losing stream during high flows (Figure 10b) is caused by bank storage and groundwater suppression and is a well-documented process [Cooper and Rorabaugh, 1963; Pinder and Sauer, 1971]. This process is clearly evident in our simulations of streambed losses and groundwater discharge to streams. Due to similar climactic, geologic, and geomorphologic characteristic among other basins, the drainage process illustrated in Figure 10likely represent drainage processes in many snow-dominated mountain block watersheds.

Figure 10.

Conceptualization of the seasonal drainage of a snowmelt dominated stream – aquifer system for, (a) early winter with negligible recharge, groundwater storage, and groundwater discharge, (b) spring snowmelt with elevated the stream head, seepage losses to bank storage and shallow aquifers, and suppressed shallow aquifer heads, (c) summer stream recession with peak shallow and regional groundwater discharge to the stream, and (d) late autumn recession of groundwater discharge to the stream.

[33] When evaluating projected hydrologic change on a seasonal basis, as expected, model results clearly show that increased temperatures projected for these watersheds result in significant timing shifts. Although the GCM ensemble of precipitation (Figures 11a1–11b1) shows little change among the 30 year time periods (i.e., 2010–2040, 2040–2070, and 2070–2100), increased temperatures result in an overall decrease in the snowpack, expressed as snow-water equivalent (Figures 11a2–11b2). The earlier snowmelt (Figures 11a3–11b3) cascades through the hydrologic system and impacts the timing of all other important hydrologic processes, including streamflow, groundwater recharge, and groundwater discharge to streams (Figures 11a4–11b6). As our simulated hydrograph separation and conceptual illustrations suggest (Figures 9 and 10), groundwater discharge to streams is inversely correlated to streamflow (Figures 11a4–11b4 and Figures 11a6–11b6). Additionally, groundwater discharge peaks approximately 1 month later than recharge, further indicating that the timing of peak groundwater discharge to streams follows the timing of streamflow recession rather than the timing of recharge (Figures 11a5–11b6).

Figure 11.

(1) Mean monthly precipitation, (2) snow water equivalent, (3) snow melt, (4) streamflow, (5) groundwater recharge, and (6) groundwater discharge to streams for different time periods and greenhouse gas emission scenarios (a) A2 and (b) B1.

[34] Increased air temperatures and earlier snowmelt also greatly affect soil moisture. Increased air temperatures reduce soil moisture in two ways, directly by providing more energy to drive the ET process, and indirectly by causing earlier snowmelt and drainage from the soil zone. The 30 year annual average ensemble soil moisture from 2010–2040 to 2070–2100 is decreased by 13 and 7% as compared to precipitation being decreased by 5 and 9% for the A2 and B1 climate scenarios, respectively. Seasonal variations in soil moisture show that the decrease in soil moisture occurs during April–November. Soil moisture is largely unchanged during winter and early spring, while the soil zone is at its water-holding capacity. Drier conditions during April–November significantly reduce ET from early to late century, and more than compensate for late century increased ET during December–May that is associated with high air temperatures. Accordingly, the 30 year annual average ensemble ET from 2010–2040 to 2070–2100 is reduced by 5 and 4% for the A2 and B1 scenarios, respectively, due to reduced growing-season soil moisture. Saturated zone ET that occurs near springs and wetlands increases over the next century indicating that groundwater levels near springs and wetlands do not change appreciably in these simulations.

[35] With regards to projected changes in ET, the two main competing processes affecting ET are high temperatures, increasing PET, and reduced soil moisture, which decreases ET below the PET. The relative impact of these two processes is further revealed by 30 year running means of precipitation, total ET, and saturated zone ET for each GCM climate projection. Trends in annual basin wide total ET correlate well with annual precipitation, which indicates that total ET is limited by water availability. Increases in ET are dampened for GCM models that project increases in precipitation. For example, the NCAR PCM1 A2 scenario projects an increase in precipitation; however, total ET for this projection remains relatively constant over the next century. Similarly, the NCAR CCSM3 A2 scenario projects an increase in precipitation during 2050–2080, whereas ET decreases during this period. These results illustrate the impact of the earlier snowmelt and earlier drainage of soil water, causing drier conditions in the summer that limit and/or decrease annual ET.

[36] Figure 12 illustrates average July–October soil moisture, total ET, net groundwater discharge to streams, and streamflow shown as a 30 year running average. By focusing on the warm season period, the dramatic decreases in soil moisture, total ET, net groundwater discharge, and streamflow are clearly evident. Decreased net groundwater discharge to streams, along with increases in saturated zone ET surrounding riparian areas and stream zones, decrease streamflow during the hottest months of the year, despite projected increases in annual precipitation, groundwater recharge, and groundwater storage by some GCMs (Figure 8).

Figure 12.

Time series of 30 year moving average July–October (late summer and early autumn) (1) soil moisture, (2) total ET, (3) net groundwater discharge, and (4) streamflow for different GCMs (colored lines) and greenhouse gas emission scenarios (a) A2 and (b) B1. Note the July–October net groundwater discharge and streamflow decreases even if annual precipitation and groundwater recharge increases for GCMs NCAR PCM1 and MPI ECHAM5 (Figures 8).

[37] To test the sensitivity of the model with respect to climate change, a simple sensitivity analysis was conducted, in which hydraulic conductivity (K) was perturbed by ±50% from calibrated values, and the model was run with the GCM NCAR PCM1, A2 climate scenario, in which annual precipitation is projected to increase. This sensitivity analysis was used to analyze the July–October streamflow sensitivity to scaled hydraulic conductivity distributions. Figure 13 illustrates the sensitivity of July–October streamflows to changes in K, where streamflow decrease even for the perturbed K simulations. Thus, the major results of this paper that July–October streamflows are projected to decrease is a robust result, despite uncertainty in K. It should be noted that under prediction of shallow groundwater heads could result in incorrect seepage rates (and directions) under some sections of the streams. However, because the streams act as drains within these shallow aquifers, greater aquifer drainage would occur if simulated groundwater heads were greater. Thus the effect of earlier drainage on summertime flow would still occur as shown here.

Figure 13.

Time series of 30 year moving average July–October streamflow for GCM NCAR PCM1, A2 climate scenario, for different values of K, illustrating the sensitivity of summertime flows to changes in K.

4. Discussion

[38] An important discussion point that is highlighted in many hydrologic studies, and is the driver of this work, is the principle cause of historical and projected changes in summertime flow in small mountainous watersheds of the Sierra Nevada. Several studies point out that increased conceptual understanding, derived from better observations and increased model structure, is needed to better understand observed and projected decreases in summertime flow and groundwater dominated flows [Scibek and Allen, 2006; Luce and Holden, 2009; Kim and Jain, 2010; Maurer et al., 2010]. As indicated by these studies, primary deficiencies in observations are limited headwater precipitation and groundwater-monitoring networks. The primary model structural deficiency in snow-dominated basins is the simulation of transient SW/GW interactions, starting with snowpack development and melt, groundwater recharge and storage, and linking these states and fluxes with in stream SW/GW interactions, as done in the simulations presented herein. While considering all of these coupled processes, our results indicate that summertime streamflows decrease in the model over all reasonable ranges in precipitation and recharge values, indicating that decreased summertime flow is independent of precipitation and recharge, and is a result of temperature changes and the resulting shift in the snowmelt recession. Furthermore, summertime streamflows decrease in simulations with perturbed hydraulic conductivity distributions indicating that these results are robust given uncertainties in hydraulic conductivity. Future work should focus on making frequent and spatially distributed head measurements in streams and adjacent shallow aquifers to provide verification of the strong relationship between the timing of snowmelt recession and peak groundwater discharge to streams that is illustrated by simulations presented herein.

[39] Our results demonstrate the important inverse relation between streamflow and groundwater discharge to streams that is caused by the effects of elevated stream depths during snowmelt runoff that suppresses groundwater discharge to streams, often referred to as the bank storage effect. Furthermore, the timing of peak groundwater discharge is not correlated to the timing of recharge, indicating that snowmelt recession is the dominant mechanisms controlling summertime flow as compared to the timing of groundwater recharge. This distinction is important because if summertime flow were correlated to the timing recharge then the effects of earlier snowmelt recession would be compensated by earlier snowmelt onset and earlier recharge. Earlier snowmelt recession decreases stream depths, which allows the shallow aquifers to drain to streams earlier in the season, thereby decreasing the amount of groundwater discharge to streams during the summer months (Figures 14a3–14b3). This explanation is in contrast with previous explanations that attributed earlier drainage of aquifers to earlier aquifer recharge, and that summertime flows are expected to increase with increased groundwater recharge [Jyrkama and Sykes, 2007; Allen et al., 2010]. Our results suggest the opposite, where simulated groundwater discharge to streams and summertime flow decrease for all GCM climate projections, even those with increased annual precipitation and groundwater recharge. These results are consistent with observations of decreasing summertime flow and dry year annual flows [Luce and Holden, 2009; Kim and Jain, 2010] coincident with increasing trends of annual and winter time precipitation [Groisman and Easterling, 1994; Karl and Knight, 1998; Mote et al., 2005; Coats, 2010].

[40] The mechanisms for reduced summertime flow are indeed linked to increasing atmospheric temperatures and the resulting changes in the timing of snowmelt, as previously suggested [Eckhardt and Ulbrich, 2003; Scibek and Allen, 2006; Maurer et al., 2010]; however, we highlight strong evidence for a direct mechanism, the shift in the timing of groundwater discharge, which is dependent on the stream stage and timing of streamflow recession (i.e., the timing of hydraulic gradient reversals between the stream and underlying groundwater). Due to numerous similarities in physical characteristics of other watersheds to those studied in this research (i.e., climate, geology, topography, vegetation), we propose that the results from this study would most likely extend to other mountainous, snow-dominated basins. This generalization is confirmed by our results being congruent with observations, that is, summertime flow that decreases even during periods when annual precipitation and groundwater recharge increases.

[41] The model informs us of a plausible process-based explanation of what may be occurring in these watersheds in response to earlier snowmelt. More field measurements are required to verify our explanation of decreased summertime flow, including distributed measurements of streamflow and streambed hydraulic gradients observed during a snowmelt cycle. However, it is difficult to compare field-scale measurements to watershed-scale response that is exhibited by the model. The model indicates that SW/GW interactions are highly variable in time and space and that a large proportion of seepage through the streambed consists of water that is recycled in and out of the stream. Complex topography causes streams to gain and lose over short reaches and large variations in streamflow causes time variable hydraulic gradients. Capturing this variability with field measurements remains a challenge, especially with regards to upscaling field measurements to infer broader scale SW/GW interactions. The model, however, could be used to guide field measurements to better understand the transient and spatial nature of SW/GW interactions and how these interactions affect or control watershed drainage.

5. Summary and Conclusions

[42] Results suggest that high temperatures have an indirect and compounding effect on groundwater storage, discharge, and streambed losses, due to interactions between surface water and groundwater. Hydraulic gradients between the stream and underlying groundwater become neutral or reversed from earlier snowmelt, streamflow, and groundwater discharge to streams. Accordingly, groundwater is depleted during the summer and there is less water to flow to streams, resulting in low summertime flow. Simulations show that the timing of peak groundwater discharge to the stream is inversely correlated to snowmelt runoff and groundwater recharge due to the bank storage effect and reversal of hydraulic gradients between the stream and underlying groundwater. That is, groundwater flow to streams peak following the decrease in stream depth caused by snowmelt recession. These changes in SW/GW interactions result in more than a 30% decrease in summertime flow when averaged across all GCM projections. Based on these results, similar snow-dominated watersheds may become more arid during the hottest part of the year, and dry season water stresses will likely become more severe even if annual precipitation increases.

[43] Groundwater will be pivotal for future water supplies, yet our current understanding of climate change impacts on groundwater is extremely limited. These findings clarify causality of decreasing summertime flow and dry year annual flows, and highlight important aspects of potential climate change impacts on groundwater resources, while explicitly considering interactions between groundwater and surface water in an integrated modeling framework.

Acknowledgments

[44] Support for the first author was provided by the NV State Engineer's Office and the Bureau of Reclamation NV Water Resources Evaluation Program, funded by a grant under Public Law 109-103, section 2.8(a), Cooperative Agreement 06FC204044. Support for the second author was provided by the U.S. Geological Survey's (USGS) Groundwater Resources Program through the Office of Groundwater. The authors would like to thank Mary Tiree, Scripps Institution of Oceanography, and Mike Dettinger, USGS, for their help in obtaining GCM data, and Dave Prudic, USGS emeritus, for his help with conceptual model development, Steve Regan, Steve Markstrom, and Paul Barlow, USGS, for their help with model development, and Toby Welborn for help with figure illustration. The authors also would like to thank Greg Pohll, John Mejia, and Darko Koracin, Desert Research Institute, for their ideas and input, Lisa Wable and Anna Knust, Desert Research Institute for their illustration and editing efforts, and Jim Thomas and Simon Poulson from the Desert Research Institute and University of NV Reno for financial assistance and isotopic analysis.

Ancillary