1. Introduction
Water is essential to the economic development of the western United States, from basic needs such as drinking water to irrigation and hydropower generation. An adequate supply of water in the future is critical to maintaining many of these important functions. With the increasing recognition of global and regional climate change patterns, water managers are rightly concerned about their potential impacts on water in the western United States.
In the headwaters region of Colorado (Upper Colorado River basin), approximately 85% of the streamflow comes from snowmelt. Thus, it is paramount that water managers be provided with as accurate an estimate as possible of the likely future changes expected with this resource. Previous climate studies (e.g., Duffy et al. 2006; Coquard et al. 2004; Leung et al. 2003) have shown a wide variety of possible impacts, from no impact (Duffy et al. 2006), to significantly reduced snowpack (Leung et al. 2003), especially in the Sierra Nevada and the Cascades along the Pacific coast. The Colorado Headwaters region, however, seems to be a particularly difficult area for current climate models to properly characterize, with inconsistent snowpack trends in this region from both the Third and Fourth Intergovernmental Panel on Climate Change (IPCC) Assessment Reports [AR3 and AR4, respectively; see Houghton et al. (2001) and Solomon et al. (2007)], despite consistent predictions of temperature increases. Observations over the past 50 yr of temperature and snowpack in the upper Colorado also reflect this predicted trend (Edwards and Redmond 2005), showing increasing temperatures and no identified trend in snowpack in this region.
Analysis by Hoerling and Eischeid (2006) of the AR4 global models indicates that the combination of increased temperature and weak to no trend in snowpack will produce unprecedented drought conditions over the next 50 yr in the southwest United States due to a strong increase in evapotranspiration in this region. Seager et al. (2007), Hoerling and Eischeid (2006), Hoerling et al. (2009), Milly et al. (2005), and Christensen and Lettenmaier (2007) estimate significant reductions in Colorado River streamflow ranging from 5% to 45% due to climate change.
While the above predictions of the global model runs from the AR4 indicate dire consequences for the southwest United States, it also must be noted that the report indicated that global models typically perform poorly in regions of complex terrain due to smoother terrain represented in the model. The hydrology of the upper Colorado region is, in particular, driven by high-altitude snowmelt, so climate assessments in this region using global models are uncertain. Therefore, it is critical to examine climate impacts in this region using higher-resolution models in order to simulate orographic precipitation, snowpack accumulation, ablation, evaporation, and runoff processes more realistically.
In an attempt to achieve this desired improvement, regional climate models have been used to simulate precipitation and runoff in the western United States using global models for lateral boundary conditions. Typical horizontal grid spacings in current regional climate models range from 40 to 60 km, with topography smoothed appropriately. Leung et al. (2003) show that the current regional climate simulations using 40–60-km horizontal grid spacing typically underestimate the snowpack compared to surface snowfall observations and other data by 25%–30%, and identify spatial resolution as one of the most important factors that needs to be improved in order to more accurately simulate snowpack. Leung and Qian (2003) show that decreasing grid spacing from 40 km down to 13 km increased the amount of precipitation to values closer to that observed and also improved the spatial distribution of the snowpack.
Improved topographic representation will result in taller topographic features, which potentially increases orographic precipitation due to enhanced lifting. The taller topography also provides a cooler environment for the snowpack as compared to smoothed topography, which puts the snow at a lower, warmer elevation. Both effects will likely lead to increased snowpack.
The above discussion suggests that regional models of orographic precipitation can be improved if higher horizontal grid resolutions are used. A question that naturally arises is how fine does the horizontal grid spacing have to be? A recent study by Garvert et al. (2007) can help answer this question. They analyzed a winter storm impacting the Oregon Cascades during the second Improvement of Microphysical Parameterization through Observational Verification Experiment (IMPROVE II) field program using aircraft and model simulations and found that to appropriately simulate the snowfall over the complex topography of the Oregon Cascades, model horizontal grid spacings needed to be 4 km or less. The reason for such high resolution was the finding from in situ measurements that local ridges and valleys could produce updraft–downdrafts on the order of 1–2 m s−1, which had a significant impact on the snowfall patterns produced. Their Fig. 3 shows a strong correlation of updrafts observed from the aircraft with model-simulated updrafts when the model horizontal resolution was 1.33 km. At horizontal resolutions coarser than 4 km, the observed strength of the vertical motions was not simulated correctly. The clouds associated with this updraft pattern showed a strong correspondence to the underlying topography. Similar results were obtained for the Sierra Nevada in California (Rasmussen et al. 1988; Grubišić et al. 2005).
The Garvert et al. (2007) study also showed that snowpack accumulation over 3 h (their Fig. 5) in the upwind portion of the Oregon Cascades with a 1-km grid spacing was 4%–14% larger than the snowpack from the 12-km smoothed topography simulation. This result suggests that finer grid spacing and improved representation of topography may improve the snowfall underestimation problem reported by Leung and Qian (2003). Because expected climate change impacts on snowpack are of similar magnitude to these reported model sensitivities, it is important that these types of local effects be properly simulated.
The precipitation-generating mechanisms contributing to this model sensitivity are complex and often scale dependent in relation to properly resolved terrain features. For instance, the case simulated by Garvert et al. (2007) was a high Froude number case (high winds and low stability), leading to flow predominantly going over the ridges. In other cases, weaker winds and/or stronger stability can lead to blocking of the flow. Because the barrier height is critical to whether flow is blocked or not, a properly simulated barrier height is important to whether air goes over a barrier and produces clouds and precipitation or goes around it and interacts with local ridges and valleys.
Another complex process related to terrain specification is the potential suppression (downward) of the freezing level during intense snowfall events. Due to the stronger updrafts in the high-resolution simulations, higher snowfall rates will occur as compared with lower-resolution simulations. As this snow falls through the freezing level, it will cool the surrounding air due to melting, and potentially drive the 0°C level to lower altitudes that may potentially allow the snow to fall to lower levels than it would under lighter snowfall conditions associated with smoother terrain. The more intense the snowfall rate, the stronger this effect will be. Thus, an important process to be assessed in the higher-resolution simulations is how the lowering of the freezing level may influence the accumulation and distribution of snow.
Alternatively, a recent study by Ikeda et al. (2007) showed that embedded convection was very important in the formation of efficient precipitation in the relatively warm winter climate of the Oregon Cascades. As the climate warms, it is unclear how the Colorado Headwaters region will be impacted due to changing amounts of embedded convection during wintertime storms. Proper depiction of relatively small-scale convective processes may also be important in determining precipitation efficiency and amounts.
Finally, the evapotranspiration rate simulated in the models is driven by local winds and subsaturation, as well as net radiation and water availability. These factors are strongly impacted by the shape and orientation of local slopes, ridges, and valleys through variations on incoming solar radiation, and thus higher-resolution simulations should provide a more accurate estimate of evaporative forcing. From an overall water balance perspective, it is critical to examine whether the current estimates of evapotranspiration in current global and regional models using relatively coarse representations of topography are correct. Assessing and reducing such errors should be viewed as a key first step in developing more reliable estimates of runoff and water resources in regional climate scenarios.
The current paper will explore several of these issues by attempting to address the following questions:
Can a properly configured high-resolution coupled atmospheric–land surface modeling system correctly simulate snowfall, snowpack, and runoff over the Colorado Headwaters region?
What is the role of highly resolved vertical motions in achieving improved estimates of snowfall?
What are the first-order climate impacts due to increased temperature and water vapor content on snowfall, snowpack, and runoff in this region?
Section 2 presents the model setup and description of the pseudo–global warming (PGW) approach. Comparisons of retrospective model runs to SNOTEL observations will be presented in section 3. A discussion of the role of improved vertical velocity simulation on snowfall with different model resolutions is given in section 4. Section 5 will provide results from the pseudo–global warming simulations. Results and discussion are given in section 6, and a summary and conclusions are presented in section 7.
2. WRF model setup for historical and pseudo–global warming simulations
The Weather Research and Forecasting (WRF) regional weather and climate model version 3.0 (Skamarock et al. 2005) was used for both the historical verification runs and the PGW simulations. The relevant model parameterizations included the
Noah land surface model (Chen and Dudhia 2001; Ek et al. 2003);
Mellor–Yamada–Janjić (MYJ) planetary boundary layer scheme (Skamarock et al. 2005);
Community Atmosphere Model’s (CAM) longwave and shortwave schemes (Collins et al. 2006); and
Thompson et al. (2008) cloud microphysics scheme.
Surface processes in WRF are simulated using the Noah land surface model (LSM; Chen et al. 1996; Chen and Dudhia 2001; Ek et al. 2003), which is based on a diurnally dependent Penman potential evaporation approach, a multilayer soil model, a modestly complex canopy resistance scheme, and frozen-ground physics. For snow-covered surfaces, Noah considers a blended snow–vegetation–soil layer and simulates the snow accumulation, sublimation, melting, and heat exchange at snow–atmosphere and snow–soil interfaces based on the simple snow parameterization of Koren et al. (1999). The version of the Noah LSM used in this study does not have independent snow layers, nor does Noah have a canopy snow interception component.
a. Retrospective simulations with the high-resolution WRF model
To have confidence in the application of the high-resolution model for future climate simulations, extensive retrospective testing was performed. The model was configured for a single domain of 1200 km × 1000 km with 45 vertical grid levels (Fig. 1a). Horizontal grid spacing in the domain was varied between 2, 6, 18, and 36 km depending on the run. A list of simulations conducted is given in Table 1. Simulations over a 6-month period from 1 November through 1 May at 2-km horizontal resolution were conducted for 2001–02, 2003–04, 2005–06, and 2007–08. These four years were chosen to represent one low, two average, and one high wintertime precipitation years, respectively. In addition, a full-year simulation was conducted for 2007–08 in order to span the snowmelt and summer convective seasons. A large domain size was chosen in order to properly account for the influence of upstream mountains (namely the Wasatch Range in central Utah and Wind Rivers, Medicine Bows, and Sierra Madres in Wyoming) on the flow and moisture depletion. Model initial conditions and lateral boundary forcing were taken from the North American Regional Reanalysis (NARR; Mesinger et al. 2006). This dataset is available every 3 h over North America with 32-km horizontal grid spacing.
Retrospective model domain and location of SNOTEL sites (black dots) for (a) the full model domain and (b) a subdomain focused over the Colorado Headwaters region. Locations of some cities are indicated by stars.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Retrospective model domain and location of SNOTEL sites (black dots) for (a) the full model domain and (b) a subdomain focused over the Colorado Headwaters region. Locations of some cities are indicated by stars.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Retrospective model domain and location of SNOTEL sites (black dots) for (a) the full model domain and (b) a subdomain focused over the Colorado Headwaters region. Locations of some cities are indicated by stars.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
List of simulations performed in the current study. Simulations using a PGW scenario were also examined from 1 November through 1 May during the four water years at 2-km grid spacing. The rightmost column indicates the number of SNOTEL sites that were operational during the respective years in the subdomain (Fig. 1b). The SNOTEL data were used for model verification.
b. Pseudo–global warming procedures and model setup
The PGW approach is based on the procedures developed by Schär et al. (1996), Hara et al. (2008), and Kawase et al. (2009). Hara et al. (2008) verified the approach using historical data, and Kawase et al. (2009) applied the approach to future snowfall in Japan. Lynn et al. (2009) independently developed the technique and applied it to Hurricane Katrina of the 2005 Atlantic hurricane season in a set of future climate perturbations. The procedure consists of adding a mean climate perturbation to the NARR initial and boundary conditions based on expected changes predicted by global climate model (GCM) runs.
A currently popular approach is to bias correct the AR4 global simulations and use statistical downscaling to distribute the precipitation according to present-day observations (Wilby et al. 2000; Wood et al. 2004). However, the current resolution of topography in global models is so coarse (in most cases only one large mound of topography covering the entire western United States significantly shorter than the actual topography), that it is unclear if the resulting downscaled climate change signal is representative of what would actually occur, especially considering that the large mound does not properly modulate the storm tracks and jet stream pattern and that nonlinear local effects are disregarded by a statistical downscaling approach (McAfee and Russell 2010). The PGW approach used in this study, while simple, provides a first-order estimate of the potential climate warming impacts on snowfall, snowpack, and runoff that we feel is consistent with the accuracy of the large-scale climate signal produced by the current generation of global models.
The PGW methodology consists of adding a climate perturbation signal to a contemporary high-resolution analysis of the atmosphere during the future period of interest. The climate perturbation’s primary impact is on the large-scale planetary waves and associated thermodynamics, while the weather patterns entering the domain boundary remain structurally identical in both simulations in terms of frequency and intensity. The weather events, however, can evolve within the regional model domain due to the altered planetary flow and thermodynamics.
To develop a climate perturbation signal field, we subtracted the current (1995–2005) monthly 10-yr climatology from a future (2045–55) monthly 10-yr climatology, ΔCCSMmonthly mean, both from the Community Climate System Model (CCSM) IPCC simulations for the A2 emissions scenario performed by the National Center for Atmospheric Research’s (NCAR) Climate and Global Dynamics Division (Collins et al. 2006). Note that the A2 and A1B scenarios are nearly identical out to 2050.
In applying this method, it is necessary to use the difference between climatological fields that have as little spatial variation as possible due to weather perturbations, so the decadal climatology is used to average out the weather. The significant benefit of this is that the climate perturbation signal can be added to the NARR analysis while maintaining hydrostatic and geostrophic balances because the difference of two balanced mean fields retains the linear relationships between the virtual temperature, geopotential height, and horizontal wind perturbation fields. Therefore, adding this difference retains the consistency in the large-scale fields, while also preserving the small-scale weather patterns and imbalances of the period of interest.
A comparison of the CCSM ensemble mean to the AR4 multimodel ensemble mean of temperature, humidity, and precipitation shows reasonable comparison (Solomon et al. 2007). In particular, the mean increase in temperature of 2.0°C and of boundary layer water vapor mixing ratio of 10%–15% is similar in both. The mean precipitation increase over the western United States in the AR4 multimodel simulations is around 7% (Solomon et al. 2007) while the CCSM increase over the full domain of the model is 4%, which is about half of the ensemble mean.
It is important to emphasize that the PGW methodology is used within the context of a well-constrained experiment and, as such, the method has a few caveats to note. First, we rely on the CCSM climate change projection, which has its own uncertainties, to perturb the weather. Taking the ensemble mean of many climate model simulations could reduce this, but we expect the general features of the climate signal to be the same. Another simplification here is the assumption that the same frequency and intensity of weather perturbations enter the regional simulation domain in the future climate’s mean state, while in reality there may be a corresponding shift in intensity and frequency due to a change in storm track and baroclinicity as a result of warming (Yin 2005). It is interesting to note, however, that Bengtsson et al. (2006, 2007) suggest little, if any, change in extratropical storm intensity and frequency in the Colorado region due to climate warming using the ECHAM5 climate model, although they do find northward storm track displacements in the Northern Hemisphere Atlantic and Pacific Ocean storm corridors. In this paper we assume no change in cyclone track, frequency, or intensity over Colorado from the twentieth to the twenty-first centuries. As will be shown later, the observed CCSM change in precipitation accumulation in the A2 emission scenario from the twentieth to the mid-twenty-first centuries shows only a 4% increase and no change in the center of mass of precipitation (located in southern Idaho), consistent with little, if any, change in storm tracks or storm frequency. Simulating no change in storm tracks or storm frequency likely constitutes the largest source of uncertainty in the PGW simulations. This fact emphasizes that the model experiments presented here should be interpreted as an “imposed” warming experiment and not a robust scenario of future climate.
c. Features of the CCSM mean climate for 2045–55
For 2045–55, the 700-hPa mean temperature from the 10-yr average shows a northward shift of warmer air by about 200 km as compared to the current mean temperature (Fig. 2). This shift results in a ~2.0°C temperature perturbation over Colorado (as indicated by Fig. 2c). The mean future relative humidity at 700 hPa (Fig. 3) over Colorado remains similar to the average relative humidity over the contemporary period (~45%–50%), indicating an increase in mixing ratio on the order of 14%, consistent with a 2.0°C temperature increase and the Clausius–Clapeyron equation. Individual months, however, are observed to have relative humidity difference patterns that indicate higher–lower patterns of humidity in the future (Fig. 4). This can lead to water vapor mixing ratio increases of up to 25%, depending on the temperature.
(a) Current, (b) future, and (c) difference between future and current mean annual 700-mb ensemble temperature. The ensembles consist of CCSM3 current climate runs from 1995 to 2005, and CCSM3 future climate model runs from 2045 to 55.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
(a) Current, (b) future, and (c) difference between future and current mean annual 700-mb ensemble temperature. The ensembles consist of CCSM3 current climate runs from 1995 to 2005, and CCSM3 future climate model runs from 2045 to 55.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
(a) Current, (b) future, and (c) difference between future and current mean annual 700-mb ensemble temperature. The ensembles consist of CCSM3 current climate runs from 1995 to 2005, and CCSM3 future climate model runs from 2045 to 55.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
(a) Current, (b) future, and (c) difference between future and current mean annual 700-mb ensemble relative humidities. See Fig. 2 for details.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
(a) Current, (b) future, and (c) difference between future and current mean annual 700-mb ensemble relative humidities. See Fig. 2 for details.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
(a) Current, (b) future, and (c) difference between future and current mean annual 700-mb ensemble relative humidities. See Fig. 2 for details.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Monthly mean difference in 700-mb relative humidity (%) between current and future climates. See Fig. 2 for details.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Monthly mean difference in 700-mb relative humidity (%) between current and future climates. See Fig. 2 for details.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Monthly mean difference in 700-mb relative humidity (%) between current and future climates. See Fig. 2 for details.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
3. Verification of the high-resolution model using SNOTEL data
SNOTEL observations (see Serreze et al. 1999 for a recent description) provide a long-term record of precipitation from precipitation gauges and snowpack liquid equivalent measured with snow pillows at over 100 sites throughout the Colorado Headwaters region (Fig. 1b, Table 1). These sites are owned and operated by the Department of Agriculture’s Natural Resource Conservation Service (NRCS) and are located at elevations typically above 2400 m mean sea level, which are the locations of the highest snowfall.
Daily SNOTEL data are available on the NRCS Web site and have 0.1-in. (2.5 mm) resolution. A variety of authors have described the use of these data, including known deficiencies (e.g., Serreze et al. 1999; Serreze et al. 2001; Johnson and Marks 2004). The main issue with weighing-type gauges for snowfall estimation is the undercatch due to wind (Serreze et al. 2001; Yang et al. 1998; Rasmussen et al. 2001). Based on the location of SNOTEL gauges in a forest clearing, wind speed is typically less than 2 m s−1, for which an undercatch of approximately 10%–15% is expected (Yang et al. 1998).
a. Comparison of the 2-km WRF model to SNOTEL observations for four water years
The precipitation accumulations at SNOTEL sites from the four retrospective 6-month period simulations with 2-km resolution are shown in Fig. 5. The average SNOTEL precipitation gauge accumulation from SNOTEL observation sites in the subdomain (Fig. 1b, Table 1) is used to compare to the model runs. The arithmetic average of model precipitation from the four nearest grid points to each SNOTEL site is used for comparisons with the SNOTEL observations. Comparisons were also made to the nearest grid-point value and model precipitation computed from a bilinear interpolation and inverse-distance weighted mean. In all cases, at 2-km grid spacing, the results did not significantly change. We also present Parameter-elevation Regressions on Independent Slopes Model (PRISM) monthly averaged snowfall estimates for each case.
Comparison of 2-km WRF to SNOTEL site average accumulative precipitation (mm) for a 6-month simulation period during the (a) 2001–02 (dry year), (b) 2003–04 (average year), (c) 2005–06 (average year), and (d) 2007–08 (wet year) water years.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Comparison of 2-km WRF to SNOTEL site average accumulative precipitation (mm) for a 6-month simulation period during the (a) 2001–02 (dry year), (b) 2003–04 (average year), (c) 2005–06 (average year), and (d) 2007–08 (wet year) water years.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Comparison of 2-km WRF to SNOTEL site average accumulative precipitation (mm) for a 6-month simulation period during the (a) 2001–02 (dry year), (b) 2003–04 (average year), (c) 2005–06 (average year), and (d) 2007–08 (wet year) water years.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
The results show that the WRF model agrees to within 10% of the SNOTEL data for the 2003–04 and 2007–08 simulations, and 15% for the 2001–02 and 2005–06 cases. The model-simulated precipitation is higher than the SNOTEL average precipitation for all four years. Considering the various uncertainties in SNOTEL precipitation measurements (e.g., undercatch up to 15% due to wind effects), the agreement is remarkable.
Figure 6 shows the monthly accumulation time series for the four simulation years. The simulated precipitation curve follows the average SNOTEL observation well in each month indicating that individual weather events are well represented by the 2-km resolution model. The monthly accumulation agrees within 5%–20% for all four years except for March and April 2006 in which the WRF model produces 40% more precipitation than was measured at the SNOTEL sites.
Monthly time series of accumulative precipitation (mm) for (a) 2001–02, (b) 2003–04, (c) 2005–06, and (d) 2007–08. Gray shades represents one standard deviation from the average daily precipitation totals at SNOTEL sites.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Monthly time series of accumulative precipitation (mm) for (a) 2001–02, (b) 2003–04, (c) 2005–06, and (d) 2007–08. Gray shades represents one standard deviation from the average daily precipitation totals at SNOTEL sites.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Monthly time series of accumulative precipitation (mm) for (a) 2001–02, (b) 2003–04, (c) 2005–06, and (d) 2007–08. Gray shades represents one standard deviation from the average daily precipitation totals at SNOTEL sites.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
The spatial patterns of monthly total precipitation from December 2007 to April 2008 from model runs and SNOTEL data are shown in Fig. 7. The spatial patterns of accumulated precipitation in the model compares well to the observed SNOTEL accumulation. Further results regarding the model comparison to observations are given in Ikeda et al. (2010).
Spatial pattern comparison of monthly total precipitation between the (left) 2-km WRF simulation, (middle) SNOTEL observations, and (right) model bias at SNOTEL sites from November 2007 to April 2008. (left) Circles in the WRF simulation results are SNOTEL locations. (middle) Color-filled circles indicate precipitation amounts at SNOTEL sites. (right) Positive (negative) model biases are shown with red (blue) circles.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Spatial pattern comparison of monthly total precipitation between the (left) 2-km WRF simulation, (middle) SNOTEL observations, and (right) model bias at SNOTEL sites from November 2007 to April 2008. (left) Circles in the WRF simulation results are SNOTEL locations. (middle) Color-filled circles indicate precipitation amounts at SNOTEL sites. (right) Positive (negative) model biases are shown with red (blue) circles.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Spatial pattern comparison of monthly total precipitation between the (left) 2-km WRF simulation, (middle) SNOTEL observations, and (right) model bias at SNOTEL sites from November 2007 to April 2008. (left) Circles in the WRF simulation results are SNOTEL locations. (middle) Color-filled circles indicate precipitation amounts at SNOTEL sites. (right) Positive (negative) model biases are shown with red (blue) circles.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
b. Examination of the impacts of horizontal resolution on WRF model snowfall accumulation during the 2007–08 water year
One of the goals of this study is to determine the model resolution required to make accurate snowfall predictions for both weather forecast and climate models. In particular, what grid spacing do regional climate and weather models need to accurately depict the snowfall amounts and spatial distribution such that hydrologists can use the data to predict streamflow and other hydrological quantities? To investigate this question, we performed WRF model simulations with 6-, 18-, and 36-km grid spacings over the same domain shown in Fig. 1 for the 2007–08 case. At large grid spacings, a cumulus parameterization scheme is typically necessary to account for unresolved convection. Ikeda et al. (2010) compared results from various cumulus parameterization schemes for the 18- and 36-km simulations and found no significant difference for cold-season precipitation simulations. Results shown in this paper used the Betts–Miller–Janjić cumulus parameterization scheme.
Figure 5d provides a 6-month accumulation time series of precipitation for the SNOTEL locations during the 2007–08 water year for the four horizontal resolutions listed. The accumulation for the 2- and 6-km simulations are nearly identical, and within 3% of the SNOTEL accumulation. The 18- and 36-km simulations, however, are 14% and 28% lower than the SNOTEL accumulation, respectively. The decreasing snowfall as the model grid spacing increases is also evident at specific SNOTEL locations, as shown in Fig. 8. Further comparison of model results to SNOTEL in Ikeda et al. (2010) shows that the 2-km grid-spacing experiment is typically within 20% of the SNOTEL accumulation 71% of time, while the 18- and 36-km resolution runs are within 20% of the SNOTEL values only 48% and 39% of the time, respectively, clearly indicating that at least 6 km or less resolution is required to properly simulate snowfall in the Colorado Headwaters region.
Accumulative precipitation comparison between measurements at four SNOTEL sites [(a) Willow Park, (b) Grizzly Peak, (c) Ripple Creek, and (d) Columbine Pass], WRF simulations at 36-, 18-, 6-, and 2-km horizontal resolutions, and PRISM data. Values from the WRF simulations and PRISM data at the four SNOTEL sites were determined using bilinear interpolation.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Accumulative precipitation comparison between measurements at four SNOTEL sites [(a) Willow Park, (b) Grizzly Peak, (c) Ripple Creek, and (d) Columbine Pass], WRF simulations at 36-, 18-, 6-, and 2-km horizontal resolutions, and PRISM data. Values from the WRF simulations and PRISM data at the four SNOTEL sites were determined using bilinear interpolation.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Accumulative precipitation comparison between measurements at four SNOTEL sites [(a) Willow Park, (b) Grizzly Peak, (c) Ripple Creek, and (d) Columbine Pass], WRF simulations at 36-, 18-, 6-, and 2-km horizontal resolutions, and PRISM data. Values from the WRF simulations and PRISM data at the four SNOTEL sites were determined using bilinear interpolation.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
4. Examination of the impacts of higher resolution on local vertical motions and resulting snowfall
The previous section showed that accurate yearly snowfall simulations were possible with horizontal grid spacing of 6 km or less. In this section we examine the sensitivity of higher model resolution on local vertical motions and the resulting snowfall and snowpack.
Figure 9 presents an example of the magnitude of updrafts–downdrafts at 600 hPa from the 2-km simulation at 0500 UTC 1 December 2007 during a snowstorm. The 2-km grid simulation produced vertical motions with updraft–downdraft strengths on the order of 1–2 m s−1. These motions are gravity waves forming over terrain features, such as isolated peaks, complexes of high terrain, ridges, and valleys. These updrafts create locally large condensate supply rates, leading to higher liquid water contents and snow mixing ratios than would occur in the typically weak updrafts associated with coarser-resolution (18 and 36 km) simulations [on the order of cm s−1; Ikeda et al. (2010)]. As a result, snow formation is focused close to the peaks of the terrain where updrafts are typically strongest, resulting in snowfall contours that closely resemble topography contours in the 2- and 6-km simulations (e.g., Fig. 7; cf. Grubišić et al. 2005). Note, however, that snowfall on the ground is usually greatest slightly downwind of the peaks where the corresponding downdrafts are the strongest, as in Garvert et al. (2007) and Medina et al. (2005).
Vertical (color) and horizontal (arrows) wind speeds at 600 hPa at 0500 UTC 1 Dec 2007 from the 2-km simulation. Thin gray contours show the underlying topography (see Fig. 1b for the elevations).
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Vertical (color) and horizontal (arrows) wind speeds at 600 hPa at 0500 UTC 1 Dec 2007 from the 2-km simulation. Thin gray contours show the underlying topography (see Fig. 1b for the elevations).
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Vertical (color) and horizontal (arrows) wind speeds at 600 hPa at 0500 UTC 1 Dec 2007 from the 2-km simulation. Thin gray contours show the underlying topography (see Fig. 1b for the elevations).
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Even though the snowfall distribution is distinctly different, the domain total precipitation values in the 2- and 36-km simulations are similar (Ikeda et al. 2010). The reason for this is related to the fact that stronger vertical motion in the 2-km simulation occurs over a smaller spatial area than the broader and weaker updraft in the 36-km simulation. This also holds true for downdrafts. Figure 10, which shows an example of the 3-h total precipitation for a particular cross section of the model for two different resolutions, illustrates this effect. Airflow is parallel to the cross section from left to right. Figure 10a shows that the 6-km simulation has four distinct peaks of precipitation associated with strong updrafts (Fig. 10c) up to 0.5 m s−1 correlated with distinct ridges (Fig. 10d). In contrast, the 36-km simulation only has two distinct precipitation peaks that are out of phase with the 6-km peaks. The 36-km simulation precipitation peaks are associated with significantly weaker updrafts (Fig. 10c) that are produced by the weaker topographic gradients at 36 km as compared to 6 km (Fig. 10d). The first precipitation peak from the 36-km simulation is further upstream and broader than the leftmost precipitation peak in the 6-km simulation. The second precipitation peak in the 36-km simulation is actually a combination of the three separate peaks from the 6-km peak. The peak is broader and has a lower amplitude than the 6-km peaks. The broadening of the precipitation maxima results in the accumulation of precipitation in valleys of the 6-km simulation, and thus the precipitation is not located on the topographic peaks as observed.
Cross section of (a) 3-h total precipitation, (b) 3-h rain and snow accumulation, (c) instantaneous vertical velocity (m s−1), (d) elevation, and (e) instantaneous surface temperature at 0000 UTC 1 Dec 2007 from A to B shown in (e). The cross section is parallel to the mean upper-wind direction.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Cross section of (a) 3-h total precipitation, (b) 3-h rain and snow accumulation, (c) instantaneous vertical velocity (m s−1), (d) elevation, and (e) instantaneous surface temperature at 0000 UTC 1 Dec 2007 from A to B shown in (e). The cross section is parallel to the mean upper-wind direction.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Cross section of (a) 3-h total precipitation, (b) 3-h rain and snow accumulation, (c) instantaneous vertical velocity (m s−1), (d) elevation, and (e) instantaneous surface temperature at 0000 UTC 1 Dec 2007 from A to B shown in (e). The cross section is parallel to the mean upper-wind direction.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
The precipitation peak near 290 km in both the 36- and 6-km simulation is of particular interest. The precipitation type in the 6-km simulation is snow (Fig. 10b) due to the cold temperature at that location (−2.5°C; Fig. 10e). In contrast, the 36-km simulation peak consists of rain due to the significantly warmer temperature (+2.5°C; Fig. 10e) associated with the lower height produced by the terrain smoothing (Fig. 10d). Clearly, these types of differences can have a significant impact on the hydrological processes in a particular river basin such as the snowpack evolution and runoff. Analysis of the grid-spacing impacts on snowpack evolution and runoff production is the subject of ongoing work that will be presented in a future paper.
5. Pseudo–global warming simulation results
A feature of the future atmosphere consistently simulated by climate models out to 50 and 100 yr (such as the models used in the IPCC report) is an increase in temperature of about 2.0°C and an increase in moisture (mixing ratio) of about 7% K−1 of warming, consistent with the Clausius–Clapeyron relation (Trenberth et al. 2003). Precipitation in these models, however, is typically found to increase only by 1%–2% K−1. In this section, we present an evaluation of climate warming and humidification by performing four PGW simulations (method described in section 2b) for the same four water years described in section 3.
a. Results of the pseudo–global warming simulations
Table 2 summarizes the percent change in precipitation from the PGW simulations compared with the WRF runs with retrospective NARR data. Overall, snowfall increases by ~10% at high elevations where SNOTEL sites are located for the high- and low-precipitation years (2007–08 and 2001–02, respectively), and ~7% for the two average precipitation years (2003–04 and 2005–06, respectively). Over the subdomain, an increase in winter precipitation of up to 16% occurs for the high- and low-precipitation years, with only 10% during the average precipitation years (Table 2). Over the full domain, a precipitation increase of ~26% is simulated for the four years in this study (analysis performed ignoring the grid cells nearest the boundary).
Current climate accumulation (mm), PGW accumulation (mm), PGW − current current (mm), and percent change in monthly and 6-month total precipitation from the PGW simulations compared to WRF simulations with retrospective NARR data. Values in this table represent the model average at SNOTEL sites, subdomain average, and domain average for (a) 2001–02, (b) 2003–04, (c) 2005–06, and (d) 2007–08.
As evidenced by the values in Table 3 (discussed below), a significant fraction of the precipitation in the PGW simulation is rain in both the subdomain and domain totals compared with the rain fraction from the retrospective runs. Two factors affect the distribution of rain versus snow in addition to the general warming of the atmosphere. The first factor is the timing of the storm during the season. Early and late storms tend to be warmer than midwinter storms. The second factor is the storm track. Storms arriving from the north tend to be colder with a higher snow fraction than storms coming from the south. In the following, the PGW simulations for the four years are described with respect to the storm timing and track.
Total precipitation (mm) for PGW and current climate simulations and the fraction of total PGW precipitation associated with snow, rain, and graupel for each simulation year in the subdomain.
Figures 11–14 show time series of the total amount of precipitation, snow, rain, and graupel from the retrospective and PGW simulations (bottom panels), and the spatial distribution of total precipitation from each of the two simulations and their differences (top panels), respectively, for the model subdomain during the four simulation years (2001–02, 2003–04, 2005–06, and 2007–08). Table 3 summarizes the fraction of the precipitation for each year in the subdomain distributed between snow, graupel, and rain. Note that the spatial patterns of total precipitation show that the precipitation in the PGW run is enhanced both over the higher terrain (i.e., enhancement at many of the SNOTEL sites) and over the lower elevations. The enhancement in the lower elevations is largely due to the presence of mesoscale weather features, an example of which will be shown in the next section.
(top) Spatial distribution of the 6-month total precipitation from simulations using (a) 2001–02 retrospective NARR data (current climate), and (b) perturbed NARR data (PGW). (c) Difference in the total precipitation between the PGW and retrospective (current climate) simulations. (bottom) Time series of subdomain average (d) precipitation, (e) snow, (f) rain, and (g) graupel accumulations from the WRF model simulation using 2001–02 NARR data (dashed line) and perturbed NARR data for a PGW scenario (solid line). See text for details on the process used to create the perturbed NARR data.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
(top) Spatial distribution of the 6-month total precipitation from simulations using (a) 2001–02 retrospective NARR data (current climate), and (b) perturbed NARR data (PGW). (c) Difference in the total precipitation between the PGW and retrospective (current climate) simulations. (bottom) Time series of subdomain average (d) precipitation, (e) snow, (f) rain, and (g) graupel accumulations from the WRF model simulation using 2001–02 NARR data (dashed line) and perturbed NARR data for a PGW scenario (solid line). See text for details on the process used to create the perturbed NARR data.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
(top) Spatial distribution of the 6-month total precipitation from simulations using (a) 2001–02 retrospective NARR data (current climate), and (b) perturbed NARR data (PGW). (c) Difference in the total precipitation between the PGW and retrospective (current climate) simulations. (bottom) Time series of subdomain average (d) precipitation, (e) snow, (f) rain, and (g) graupel accumulations from the WRF model simulation using 2001–02 NARR data (dashed line) and perturbed NARR data for a PGW scenario (solid line). See text for details on the process used to create the perturbed NARR data.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
As in Fig. 11, but for the 2003–04 simulation.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
As in Fig. 11, but for the 2003–04 simulation.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
As in Fig. 11, but for the 2003–04 simulation.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
As in Fig. 11, but for the 2005–06 simulation.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
As in Fig. 11, but for the 2005–06 simulation.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
As in Fig. 11, but for the 2005–06 simulation.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
The change in precipitation type for all four years is similar (Table 3), with the fraction of snow reduced by 6%–9% of the total grid cell precipitation, and rain increased by an equivalent amount for the given year. Graupel increases are less than 0.4% in all four years. The maximum fraction of snow occurs in 2001–02 and 2007–08. The spatial pattern of precipitation change is distinct from year to year (Figs. 11–14). The spatial pattern of precipitation increase for 2001–02 reveals that most of the increased precipitation under climate change occurs in the northern part of the domain; thus, the higher fraction of snow is not surprising in this year (Table 3). The precipitation increase for 2007–08 occurs in the western part of the domain, and occurs throughout the year. Storms in this case may have been colder than normal, leading to a higher snow fraction than normal.
The two average precipitation years (2003–04 and 2005–06) had the lowest snow fraction (0.87 and 0.90, respectively) and highest rain fraction (0.12 and 0.09, respectively) in the retrospective simulation. Climate warming decreased the snow fraction to 0.78 and 0.82 in 2003/2004 and 2005–06, respectively, and increased the rain fraction to 0.21 and 0.17, respectively. During 2003–04, most of the additional precipitation occurred in the southern part of the domain (Fig. 12) and in April, both of which are consistent with warmer storms. In 2005–06, a significant amount of the precipitation occurred in April, even though most of the precipitation increase was in the northern part of the domain (Fig. 13). Thus, the snow fraction was slightly higher than during the 2003–04 water season. These results reveal some of the year-to-year variability in the change signal in precipitation amount and type. In all four cases, climate warming decreases the fraction of snow by 6%–9% and increases the fraction of rain by an equivalent amount. Graupel is increased by a less than 1% in all four years.
The overriding reason for this pattern of behavior is the increased height of the melting level in the future climate simulations, which is on the order of 200 m as shown in Table 4. During 2003–04 and 2005–06, the amount of total snow increase in the PGW simulation was close to zero (Figs. 12 and 13). This occurs because the decrease in snowfall area that results from the increase in the melting level of ~200 m in the PGW run is compensated by an increase in the average snowfall rate over the domain. The reduction in area is estimated to be 11% (Table 4), and the snowfall rate increased by a similar amount in order for the total snow amount to be the same as the current run.
Subdomain-average freezing-level heights (m AGL) and average percent of subdomain area below 0°C from the retrospective and PGW simulations for (a) 2001–02, (b) 2003–04, (c) 2005–06, and (d) 2007–08.
b. Mesoscale influences
The PGW simulations reveal that much of the precipitation increases predicted by PGW simulations are associated with mesoscale features in the flow. This is illustrated by considering the first significant storm in the model domain during the 2007–08 simulation, which occurred between 30 November 2007 and 3 December 2007. Figure 15 shows a time history of precipitation accumulation and precipitation rates from this storm. The storm total subdomain average precipitation was 46.7 and 39.1 mm in the PGW and current climate runs, respectively. The precipitation rates were higher in the PGW simulation compared with the current climate simulation throughout the storm event.
Time history of precipitation rate (mm h−1, solid and dashed lines in black) and precipitation accumulation (mm, solid and dashed lines in gray) from the current climate and PGW simulations from 30 Nov to 3 Dec 2007.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Time history of precipitation rate (mm h−1, solid and dashed lines in black) and precipitation accumulation (mm, solid and dashed lines in gray) from the current climate and PGW simulations from 30 Nov to 3 Dec 2007.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Time history of precipitation rate (mm h−1, solid and dashed lines in black) and precipitation accumulation (mm, solid and dashed lines in gray) from the current climate and PGW simulations from 30 Nov to 3 Dec 2007.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
The difference plot of the precipitation rates (PGW minus current climate) reveals mesoscale features throughout the domain (Fig. 16). Note the linear alternating regions where precipitation rates were higher–lower in the PGW simulation compared with the current climate simulation. In addition, note that the precipitation enhancement in the PGW simulation is not confined solely to the mountain peaks, but also occurs in the upstream valleys, which is suggestive of mesoscale influences not associated with topographic forcing. Wind speeds are impacted by ±2 m s (Fig. 16d), depending on the location. These aspects will be further investigated in future publications.
Spatial distribution of precipitation rate from (a) the current climate and (b) PGW simulations at 2200 UTC 1 Dec 2007. (c) The precipitation rate difference between the two simulations. (d) The difference in wind speed at 600 mb between the two simulations. Blue (red) arrows indicate wind directions in the current climate (PGW) simulation.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Spatial distribution of precipitation rate from (a) the current climate and (b) PGW simulations at 2200 UTC 1 Dec 2007. (c) The precipitation rate difference between the two simulations. (d) The difference in wind speed at 600 mb between the two simulations. Blue (red) arrows indicate wind directions in the current climate (PGW) simulation.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Spatial distribution of precipitation rate from (a) the current climate and (b) PGW simulations at 2200 UTC 1 Dec 2007. (c) The precipitation rate difference between the two simulations. (d) The difference in wind speed at 600 mb between the two simulations. Blue (red) arrows indicate wind directions in the current climate (PGW) simulation.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
c. Microphysical influences
To diagnose the microphysical factors leading to the enhanced precipitation over a barrier, two-dimensional idealized orographic precipitation simulations were conducted with a sounding taken from the 1 December 2007 storm described earlier for both current and future conditions. These simulations had a grid spacing of 1 km and a width of 1000 km. The barrier was 2 km high and 50 km wide at half height. This barrier is comparable to the San Juan Mountains located in the southwest portion of the subdomain. The vertical cross section is from SW to NW along the flow direction.
Figure 17 depicts the microphysical processes along this cross section from current and future 2D idealized simulations. The cloud water over the barrier is shown to increase in magnitude and area in the future climate simulation. The increase is especially evident in the region above 0°C at the midmountain level in Fig. 17b.
Idealized 2D simulations of orographic precipitation formation over a bell-shaped mountain. (a),(b) Current and future climate cloud liquid water content, (c),(d) current and future climate snow mixing ratio, (e) difference in snow mixing ratio between current and future climate, (f),(g) current and future climate rain mixing ratio, (h),(i) current and future climate graupel mixing ratio, and (j) accumulated precipitation after 6 h of simulation. Arrows in (j) indicate locations where future climate simulation produced more precipitation compared with the current climate simulation. The X axis shows the model grid points in the horizontal direction.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Idealized 2D simulations of orographic precipitation formation over a bell-shaped mountain. (a),(b) Current and future climate cloud liquid water content, (c),(d) current and future climate snow mixing ratio, (e) difference in snow mixing ratio between current and future climate, (f),(g) current and future climate rain mixing ratio, (h),(i) current and future climate graupel mixing ratio, and (j) accumulated precipitation after 6 h of simulation. Arrows in (j) indicate locations where future climate simulation produced more precipitation compared with the current climate simulation. The X axis shows the model grid points in the horizontal direction.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Idealized 2D simulations of orographic precipitation formation over a bell-shaped mountain. (a),(b) Current and future climate cloud liquid water content, (c),(d) current and future climate snow mixing ratio, (e) difference in snow mixing ratio between current and future climate, (f),(g) current and future climate rain mixing ratio, (h),(i) current and future climate graupel mixing ratio, and (j) accumulated precipitation after 6 h of simulation. Arrows in (j) indicate locations where future climate simulation produced more precipitation compared with the current climate simulation. The X axis shows the model grid points in the horizontal direction.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
The snow mixing ratio is shown to increase in the future simulation by 5% (Fig. 17d) as a result of enhanced riming growth. The enhanced snow region is located just over the peak of the mountain (Fig. 17e).
The rain mixing ratio is shown to increase in depth (as a result of the increase of the melting layer by 200 m) and magnitude over the high cloud water region on the upstream region of the barrier. The increase in rain in this region is clearly due to the accretion of cloud water by the raindrops produced by the melting of snow. Also note that the rain in the lee of the mountain is increased due to the rise in the melting level, allowing snow to be converted to rain and to fall out before being advected past the barrier.
Graupel is shown to have a very low mixing ratio in both the current and future simulations. The increase in the graupel mixing ratio due to climate change is less than 1%, consistent with the full 3D model simulations (Table 3). The reason for the low amounts of graupel in both cases is the low value of the cloud water mixing ratio at temperatures below 0°C of only 0.1–0.15 g kg−1. At these low values snow will be the predominant precipitation type, suggesting that the main climate change signal on precipitation in this orographic situation is enhanced snow growth and increased precipitation efficiency due to the higher fall speed of rain as compared to snow in the lee of the barrier. Figure 17j shows three main regions of precipitation enhancement over the barrier (indicated by the arrows): 1) the upstream region, due to rain enhancement by increased cloud droplet accretion; 2) the peak region enhancement, due to increased snow growth by riming; and 3) the downstream region, due to the more rapid fall velocity of raindrops compared to snow.
d. PGW influence on river basin fluxes
Next, we examine the influence of the PGW-imposed climate sensitivity on river-basin-averaged hydrologic variables of total accumulated precipitation (rain plus snow), snow–water equivalent, total accumulated evapotranspiration, and total accumulated runoff, where total runoff is the sum of surface runoff and deep soil drainage from the Noah LSM (Fig. 18). In the version of the Noah LSM used in this study, no horizontal routing of surface overland flow, subsurface flow, or channel flow is performed. Instead, all values provided are area-averaged values from the 1D (vertical) gridded configuration of the Noah LSM. Of the seven basins analyzed in Fig. 18, four drain to the west, into the Colorado River basin, and three drain to the east. While Fig. 18 only shows results from the 2007–08 simulations, other years (not shown) exhibit generally similar patterns of behavior through the ends of their simulation periods (1 May).
Accumulated difference (PGW − current climate) in basin-average values of precipitation (Precip), ET, SWE, and total runoff from the 2007–08 PGW and current climate simulations for the (a) Upper Yampa River, (b) Upper Colorado River, (c) Gunnison River, (d) San Miguel River, (e) Upper Arkansas River, (f) Upper South Platte River, and (g) Boulder Creek basins.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Accumulated difference (PGW − current climate) in basin-average values of precipitation (Precip), ET, SWE, and total runoff from the 2007–08 PGW and current climate simulations for the (a) Upper Yampa River, (b) Upper Colorado River, (c) Gunnison River, (d) San Miguel River, (e) Upper Arkansas River, (f) Upper South Platte River, and (g) Boulder Creek basins.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Accumulated difference (PGW − current climate) in basin-average values of precipitation (Precip), ET, SWE, and total runoff from the 2007–08 PGW and current climate simulations for the (a) Upper Yampa River, (b) Upper Colorado River, (c) Gunnison River, (d) San Miguel River, (e) Upper Arkansas River, (f) Upper South Platte River, and (g) Boulder Creek basins.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Figure 18 shows a series of plots of the accumulated differences (PGW minus current climate run) of precipitation, evapotranspiration (ET, including sublimation), snow–water equivalent, and total runoff from seven headwater basins (Upper Yampa River, Upper Colorado River, Gunnison River, San Miguel River, Upper Arkansas River, Upper South Platte River, and Boulder Creek) of various sizes, locations, elevations, and aspects. Despite differences in basin physiographic attributes, the hydrological sensitivities to the PGW forcing compared with the current climate run are generally similar. All basins exhibit greater precipitation during the cool season (November–April) in the PGW simulation. Beginning in May, several basins begin to exhibit a slight to moderate deficit in precipitation in the PGW run compared with the current climate run. However, in nearly all basins, the deficit in summer rainfall in the PGW runs is not enough to compensate for the increased wintertime precipitation. The one basin that exhibits approximately the same amount of annual precipitation between the PGW and current climate runs is the Upper South Platte basin. Similar differences are found in the plots of accumulated ET, where, beginning in late winter, all basins exhibit an increase in ET in the PGW simulations. In most basins, the difference in ET in the PGW run levels off during the summer and in some basins actually decreases slightly during August, presumably due to increased moisture limitation by the end of the PGW summer.
Changes in basin-averaged snow water equivalent (SWE) initially show increases in the PGW versus current climate run through early winter, which then change to strong deficits by February–April. The differences in SWE extend to May but decay out to late June when nearly all snow has melted from both simulations. A plausible explanation for this signal is that increased precipitation in the PGW run contributes to an earlier development of the snowpack in November–January, but by early spring, warmer temperatures and an increase in rain versus snow in the PGW run work to ablate (melt and sublimate) the snowpack several weeks to a month earlier than occurs in the current climate simulation. It is interesting to note that peak SWE values (Table 5) remain relatively similar (2%–10% difference) but the dates of peak SWE in the PGW simulation are shifted earlier in the year compared with the current climate simulation. As a result of the changes in precipitation, ET, and SWE, the total runoff also exhibits marked differences between the PGW and control simulations. In all basins, the total annual runoff is greater in the PGW simulation with increases being most pronounced during the springtime, presumably in response to earlier snowmelt. Differences in runoff diminish somewhat throughout the summer in response to later snowmelt and greater summer precipitation in the current climate run. The runoff responses achieved in the PGW runs suggest that increases in total accumulated precipitation, as opposed to changes in snowpack conditions, more than compensate for increases in evaporative demand imposed by greater temperatures in the PGW climate.
Values of the fractional difference [(PGW − current climate)/current climate] in basin-averaged total accumulated precipitation, accumulated total runoff, peak SWE, and accumulated evapotranspiration between current climate and PGW runs averaged over all simulation years (2001–02, 2003–04, 2005–06, and 2007–08). Also included are the changes in dates of maximum SWE for each of the seven basins where the values represent control minus PGW sensitivity dates of maximum SWE in units of Julian date (DOY). Time period is from 1 November to 1 May for all years.
Total precipitation, Date of maximum snow water equivalent (DOY), maximum SWE (mm), evapotranspiration (mm), and total runoff (mm) for each of the basins for all four years for current (first set of years) and PGW (second set of years). The average values in this table were used to develop the summary table 5a.
We note that the Upper Colorado, Gunnison, San Miguel, and Upper Arkansas basins possess a spurious increase in runoff in the PGW simulation versus the control simulation at initialization lasting for 1–3 weeks. Soil moisture and temperature fields were not spun up in an offline mode prior to coupled model initialization, and soil moisture fields were initialized with the same values in the PGW simulation as they were in the current climate simulation while the soil temperature field was reinitialized according to the PGW methodology described above. The spurious runoff is caused by the “unfreezing” and subsequent drainage of initialized soil water in the PGW simulation versus the control simulation and should be recognized in comparing current climate and PGW run results. This spurious drainage signal impacts locations and basins where soil temperature differences between the current climate and PGW simulation cross the freezing point. Additionally, this “release” of frozen soil water in the PGW simulation will also impact simulated ET values because frozen soil water is not available for plants to use for transpiration.
Values of the fractional difference [current climate–PGW)/current climate] in basin-averaged total accumulated precipitation, accumulated total runoff, peak SWE, and accumulated ET between current climate and PGW runs averaged over all simulation years through May 1 are tabulated in Table 5. Also included in Table 5 are the changes in dates of maximum SWE for each of the seven basins. Despite the small sample sizes of 4 yr, the PGW climate sensitivity signal is generally consistent in the averaged values. Note that the change in the accumulated wintertime precipitation (+8% to 20%) and the change in accumulated wintertime runoff (+11% to 34%) through the end of the respective simulations are positive while the date of peak SWE is 2–17 days earlier in the PGW run than in the control run. [We note that the two basins with large mean changes in dates of maximum SWE values (Boulder Creek and Upper Yampa River) are heavily influenced by changes during one or two years and may not be reflective of a long-term mean.] Accumulated wintertime ET is also greater in the PGW simulation compared with the control run by 9%–12%, depending on the basin, though this may be partially compensated for by the decrease in late summer ET reported above.
While six of the seven basins do exhibit a decrease in peak SWE values in the PGW simulation, the lack of a uniform signal across all basins and the relatively small magnitude of the changes likely reflect the competition between increased precipitation and stronger melting in the PGW run compared with the control run. These two forcings appear to potentially offset one another with regard to peak SWE values in these basins. However, additional years and additional basins should be evaluated to assess the robustness of these results.
e. Simulated snowpack issues
Although comparisons between WRF-simulated precipitation and SNOTEL-observed precipitation were favorable, verification of WRF–Noah SWE against SNOTEL snow pillow observations reveals that the snowpack simulated by Noah started to melt too soon, resulting in the maximum seasonal SWE being too low (Fig. 19). Past studies have found a negative bias in the Noah representation of SWE, snow cover extent, and in the timing of spring snow depletion (Sheffield et al. 2003; Livneh et al. 2009). Difficulty in simulating snowpack evolution is a common weakness in many land surface models (e.g., Pan et al. 2003).
Time series of SWE for the 9-month simulation starting 1 Nov 2007 averaged over all SNOTEL locations. The control simulation is the black solid line. The other lines are the accumulated addition of model changes: Livneh albedo formulation (red solid), surface terrain effect (black dashed), WRF stability formulation (red dashed), snow albedo set to 0.85 (blue solid), and roughness length formulation (blue dashed). Black dots are the SNOTEL SWE observations.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Time series of SWE for the 9-month simulation starting 1 Nov 2007 averaged over all SNOTEL locations. The control simulation is the black solid line. The other lines are the accumulated addition of model changes: Livneh albedo formulation (red solid), surface terrain effect (black dashed), WRF stability formulation (red dashed), snow albedo set to 0.85 (blue solid), and roughness length formulation (blue dashed). Black dots are the SNOTEL SWE observations.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Time series of SWE for the 9-month simulation starting 1 Nov 2007 averaged over all SNOTEL locations. The control simulation is the black solid line. The other lines are the accumulated addition of model changes: Livneh albedo formulation (red solid), surface terrain effect (black dashed), WRF stability formulation (red dashed), snow albedo set to 0.85 (blue solid), and roughness length formulation (blue dashed). Black dots are the SNOTEL SWE observations.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
The Noah LSM offers a relatively simple treatment of snow, and a community effort to include a multilayer snowpack model is under way. As a short-term solution to the snow modeling deficiency, Barlage et al. (2010) added five modifications to the Noah model: 1) a time-varying snow albedo, 2) surface solar radiation adjustment for terrain slope and orientation, 3) surface exchange coefficient reduction for stable boundary layers, 4) fresh snow albedo increase, and 5) roughness length decrease when snow is present. The Noah LSM was executed in an offline mode from 1 November 2007 to 1 August 2008 over the 2-km domain, and its results were evaluated against individual SNOTEL sites. As shown in Fig. 19, the most effective way to improve the magnitude and timing of the seasonal maximum SWE is to use a time-varying albedo formulation and to increase fresh snow albedo. Minor improvements are obtained by the remaining three modifications. The terrain slope and orientation addition does have an effect on local surface energy components especially for north- and south-facing cells. The net effect of all changes is to improve the magnitude and timing of seasonal maximum SWE, but, as shown in Fig. 19, the snow period end is now too late.
The PGW warming results for SWE indicate that the average snowpack depth on 1 April will be reduced by ~25% due to climate warming. This result, however, must be considered as being preliminary due to the factors mentioned above. Since modeling snowpack for forested areas is a challenging issue, even for very complex snow models (Rutter et al. 2009), work is under way to improve the Noah model in offline mode for the Colorado Headwaters region.
6. Discussion
This study used SNOTEL observations of snowfall to evaluate the ability of a high-resolution regional climate model to accurately simulate snowfall over the Colorado Headwaters region. The WRF model simulations at 2-km grid spacing compared well to both the spatial and temporal distributions of SNOTEL snowfall observations over four water years. These results provide a high degree of confidence that the currently chosen configuration of the WRF model is capable of reliably simulating snowfall over a full winter season for a variety of conditions.
a. Model resolution impacts
The comparison of precipitation amounts as a function of model resolution for the current climate using the NARR boundary conditions indicates very good agreement with SNOTEL-observed precipitation if a model grid spacing of 6 km or less is used (Table 6). At this resolution, snowfall is predicted to fall mostly at the higher elevations, consistent with observations, with significantly less in the valleys. At lower resolutions, less snowfall occurs at the SNOTEL sites (14%–28% less at 18- and 36-km resolutions, respectively), while comparatively more precipitation is distributed in the valleys. While the total accumulations are similar over the subdomain for all resolutions examined (Table 6), the spatial precipitation pattern is significantly different due to the weaker and broader vertical motions in the coarse-resolution simulations. Domain total precipitation, however, is 18% less at the 36-km grid-spacing simulation and 7% less in the 18-km grid-spacing simulation with respect to the 2-km grid-spacing simulation (Table 6). These results suggest that the precipitation underestimate observed by Leung et al. (2003) for regional climate models run at 40–60-km grid spacing may be at least partially attributed to the lack of adequate model resolution. The agreement in accumulation in the subdomain for all resolutions (Table 6) is likely due to the overestimate of precipitation simulated in the valleys of the 36-km simulation compensating for the underestimate of precipitation at the peaks in this smaller region dominated by a nearly continuous sequence of ridges and valleys.
Monthly and 6-month total precipitation amounts (mm) simulated with 2-, 6-, 18-, and 36-km grid resolutions. Shown are the average values at SNOTEL sites (observations are listed in the top row, bold face), over the subdomain, and for the full domain. Model values at SNOTEL sites were determined from a bilinear interpolation method (as oppose to a simple four-point average shown in Table 2).
b. Pseudo–global warming impacts on the precipitation process
The PGW experiments indicate a consistent increase in precipitation on the order of 10%–15% in the future over the Colorado Headwaters region. The increase over the full domain was ~26%, which is significantly more. These amounts are significantly greater than both the CCSM and IPCC AR4 multimodel estimates of precipitation change for this region [~4% for CCSM and 7% for the multimodel ensemble; Solomon et al. (2007)]. An analysis of the vertical velocity differences between the current and PGW simulations (not shown) shows very little difference. Because the condensate supply rate is driven by vertical motion and moisture, these results suggest that the primary cause of the increased precipitation is the higher water vapor content in the lower troposphere. However, whether the condensed water falls out during passage over a mountain barrier will depend on the dominant microphysical processes.
The microphysical analysis conducted in section 5c shows that the increased precipitation in the PGW simulations can be attributed to two main factors:
increased cloud water mixing ratio leading to more riming growth of snow over the mountain and enhanced accretional growth of rain on the upstream slopes of the mountain and
increased precipitation efficiency due to the conversion of snow to rain, causing more of the condensed water to fall out before passing by the barrier.
Because there are multiple peaks in the Colorado Headwaters region, the first peaks encountered by a storm will produce more precipitation assuming all other factors are equal. The precipitation efficiency of a mountain will depend on a variety of factors including updraft velocity, width of the barrier, horizontal wind speed, and cloud depth. If a given barrier is on the order of 30% efficient (typical) in converting condensed moisture to precipitation, then the passage of an air parcel over multiple peaks will deplete 90% of the available moisture within three to four cycles. Vertical cross sections of precipitation differences from the current and future simulations verify this pattern of behavior (Fig. 20) and also show that the climate change precipitation enhancement is smaller farther downstream as a result of this depletion effect.
(a) Cross section of the 3-h total precipitation from the PGW and current climate simulations (left axis) and the percent difference between PGW and current climate simulations (right axis). (b) Cross section of the topography and the wind direction.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
(a) Cross section of the 3-h total precipitation from the PGW and current climate simulations (left axis) and the percent difference between PGW and current climate simulations (right axis). (b) Cross section of the topography and the wind direction.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
(a) Cross section of the 3-h total precipitation from the PGW and current climate simulations (left axis) and the percent difference between PGW and current climate simulations (right axis). (b) Cross section of the topography and the wind direction.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
c. Full-domain precipitation type, amounts, and patterns in the PGW simulations
Total precipitation in the full domain from the current climate and PGW simulations for the 2007–08 case is shown in Fig. 21. An interesting pattern emerges upon examination of the total accumulated precipitation difference from the climate perturbation (Fig. 22). It was noted that the PGW climate perturbation was on average 10%–15% for the subdomain, but ~26% in the full domain. The reason for this is evident in Fig. 22 as enhanced snow accumulation on the mountains surrounding the subdomain region. The year-to-year examination of this pattern reveals that the region enhanced during a particular year depends on the dominant storm track during that year, with the first mountains encountered by the storm experiencing the enhancement. This result is consistent with the discussion in the previous section noting that the first mountain in a sequence will produce the most precipitation as a result of being the first condensation–precipitation cycle. Because the subdomain region (except for the San Juan Mountains in the southwest) is surrounded by upstream mountains, the Colorado Headwaters region actually experiences a “rain shadow” effect from the surrounding topography. Thus, any climate change enhancement of precipitation will be less pronounced in the Colorado Headwaters regions due to this effect. The current simulations suggest that the effect will be about a factor-of-2 less precipitation over the 6-month period than for the full domain.
Domain precipitation accumulation from 1 Nov 2007 to 1 May 2008 from the (a) current climate and (b) PGW simulations.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Domain precipitation accumulation from 1 Nov 2007 to 1 May 2008 from the (a) current climate and (b) PGW simulations.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Domain precipitation accumulation from 1 Nov 2007 to 1 May 2008 from the (a) current climate and (b) PGW simulations.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Difference in total precipitation between the PGW and current climate simulations (PGW − current in mm) in the model domain for (a) 2001–02, (b) 2003–04, (c) 2005–06, and (d) 2007–08. Red box indicates the subdomain (Fig. 1b).
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Difference in total precipitation between the PGW and current climate simulations (PGW − current in mm) in the model domain for (a) 2001–02, (b) 2003–04, (c) 2005–06, and (d) 2007–08. Red box indicates the subdomain (Fig. 1b).
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Difference in total precipitation between the PGW and current climate simulations (PGW − current in mm) in the model domain for (a) 2001–02, (b) 2003–04, (c) 2005–06, and (d) 2007–08. Red box indicates the subdomain (Fig. 1b).
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
Comparison of the PGW precipitation increase in the full domain to the CCSM precipitation increase in the same domain (for the simulations used to derive the PGW simulation) shows that the CCSM-projected increase in precipitation is only 4% compared to the PGW-predicted increase of 26%. The increase in water vapor in the CCSM is of similar magnitude to that in the WRF runs; thus, the high-resolution simulations are significantly more efficient in producing precipitation than the coarse-resolution CCSM simulation. As discussed earlier, the multiple condensation–precipitation process associated with the passage of storms over multiple peaks is likely the cause for this higher efficiency.
The impact is a factor of 6 greater precipitation in the high-resolution simulation than is predicted by CCSM. Statistical downscaling techniques use either direct GCM estimates of precipitation (Brekke et al. 2009) or regional climate model (RCM) output to create local estimates of precipitation not resolved by the models. Because GCM boundary conditions are also used to run RCMs, the current results suggest that current downscaled estimates of future precipitation are low in the Colorado Headwaters region as well as potentially other regions of complex terrain. The key point is that the large-scale flow pattern in the GCM is not correct over complex terrain such as the Rocky Mountains (McAfee and Russell 2010) due to the poor resolution of the topography (simulated as a shallow mound for the entire Rockies). The resulting CCSM precipitation from the current and future A2 simulations (Figs. 23a and 23b) reflects the position of the peak of the mound over southern Idaho. The resulting change in precipitation (Fig. 23c) suggests a minimum over Idaho and a maximum in New Mexico, and has a significantly different pattern and magnitude than the current results (Fig. 22). Any regional model, or downscaling technique driven by GCMs, is only as good as the driving boundary conditions, which have known difficulties over complex terrain (AR4; S. A. McAfee and J. L. Russell 2010, unpublished manuscript). In our case, the flow around the mountains is driven by a 32-km NARR reanalysis, which provides a reasonable estimation of the flow pattern past the Rockies. As a result, the precipitation and 3D flow from the headwaters simulations are much more realistic, as shown by the SNOTEL comparison.
The 10-yr average of 6-month total precipitation (1 November–1 May) over the model domain used in this study from a CCSM A2 simulation. CCSM data were interpolated to the 36-km CO Headwaters model domain. Shown are the results for (a) 1995–2005, (b) 2045–55, and (c) the difference in precipitation between the current and future climates.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
The 10-yr average of 6-month total precipitation (1 November–1 May) over the model domain used in this study from a CCSM A2 simulation. CCSM data were interpolated to the 36-km CO Headwaters model domain. Shown are the results for (a) 1995–2005, (b) 2045–55, and (c) the difference in precipitation between the current and future climates.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
The 10-yr average of 6-month total precipitation (1 November–1 May) over the model domain used in this study from a CCSM A2 simulation. CCSM data were interpolated to the 36-km CO Headwaters model domain. Shown are the results for (a) 1995–2005, (b) 2045–55, and (c) the difference in precipitation between the current and future climates.
Citation: Journal of Climate 24, 12; 10.1175/2010JCLI3985.1
7. Summary
The following conclusions summarize this paper:
Comparison of WRF high-resolution simulations of annual snowfall to SNOTEL observations over the Colorado Headwaters regions shows very good agreement if horizontal grid spacings at or below 6 km are used.
WRF model water-year accumulated snowfall at SNOTEL sites agrees within 20% of the observations for 71% of the 112 Colorado Headwaters SNOTEL sites.
High-resolution simulations of annual snowfall suggest that current global and regional model estimates of snowfall at the ground (18-km grid spacing and higher) underestimate high-elevation snow by 20%–40% and overestimate low-elevation snowfall by a similar amount.
The pseudo–global warming simulations indicate enhanced precipitation on the order of 10%–25% over the Colorado Headwaters region over four winter seasons representing low, average, and high snowfall years. In all four years the enhancement is less over the core headwaters region due to the topographic reduction of precipitation upstream of the region (rain-shadow effect).
The average melting level increased by ~200 m in the PGW simulations.
The main impacts of climate change are enhanced rain and melting at the lateral boundaries of the snowpack and increased snowfall in the center of the snowpack. The resulting change in snow mass is near zero in the headwaters region due to these two compensating effects being nearly equal (10% increase in snowfall rate compensated for by a 10% reduction in area). In regions upstream of the Colorado Headwaters area, total snowpack is increased due to increases in snowfall amounts that exceed the reduction in snowpack area (25% increase in snowfall associated with a 10% reduction in area). These results are similar to the observed change in the Greenland ice sheet, with increasing snow depth in the interior (2–5 cm yr−1) and thinning along its periphery (e.g., Krabill et al. 1999, 2000, 2004; Johannessen et al. 2005), although the mechanism for thinning is likely not due to air temperature increases but to the warm waters in the fjords melting the leading edges of the glaciers. Thus, the acceleration of the hydrological cycle through enhanced snowfall and the increased melting of the snowpack are both confirmed in these simulations, with interior regions having either unchanged or increased snowpack depending on the amount of increased snowfall and peripheral regions having enhanced melting.
The PGW simulations indicate a significantly higher percentage snowfall increase in the future than AR4 models are predicting (10%–25% in the current high-resolution simulations versus 4% in CCSM model and 7% in AR4 ensemble results).
The PGW future simulations indicate that the dynamics of storms are not significantly impacted as reflected by only small changes in the vertical velocity distributions between current and future climate simulations.
The increased precipitation in the future simulations is attributed to two main factors:
increased cloud water mixing ratio leading to more riming growth of snow over the mountain and enhanced accretional growth of rain on the upstream slopes of the mountain and
increased precipitation efficiency due to the conversion of snow to rain, causing more of the condensed water to fall out before passing by the barrier.
Multiple precipitation cycles over the ridges enhanced the precipitation efficiency over a region significantly.
Preliminary results on SWE indicate little to only modest reductions in the maximum SWE in contrast to previous studies. This is true in general and also on a basin-to-basin basis. This is due to the enhanced basin-averaged precipitation (+7%–13%) compensating for enhanced melting due to a warmer climate. The 1 April SWE, however, is reduced by 25% in the warmer climate, and the date of maximum SWE occurs 2–17 days prior to current climate results, consistent with previous studies.
The 6-month simulation period (1 November–1 May) used in this work did not allow for a thorough assessment of potential changes in runoff behavior under our imposed warming experiment. In all simulation years studied, accumulated wintertime (i.e., through 1 May) total runoff was greater (9%–22%) in the PGW scenario. Results from the year-long 2007–08 simulation showed that while nearly all basins analyzed exhibited less summertime streamflow in the PGW scenario than in the current climate, the reduction in summer runoff was not sufficient to negate the increases in wintertime and springtime runoff, resulting in net increases in annual runoff. However, this result needs to be interpreted with caution given issues related to the spinup and initialization of soil temperature and soil moisture in the PGW case. These issues are being addressed in a forthcoming paper.
Acknowledgments
The authors acknowledge the support of the NCAR Computational and Information Systems Laboratory’s (CISL) Accelerating Science Discovery program for providing the computing time necessary to perform these high-resolution simulations. We would also like to recognize CISL’s outstanding support of the data storage and processing requirements needed to conduct this study. We also acknowledge the strong encouragement and support of Rit Carbone and Brant Foote during the conduct on this study. Karen Griggs provided excellent editorial support. This work was supported by the National Science Foundation under the NCAR Water System Program.
REFERENCES
Barlage, M., and Coauthors, 2010: Noah land surface model modifications to improve snowpack prediction in the Colorado Rockies. J. Geophys. Res., 115, D22101, doi:10.1029/2009JD013470.
Bengtsson, L., K. I. Hodges, and E. Roeckner, 2006: Storm tracks and climate change. J. Climate, 19, 3518–3543.
Bengtsson, L., K. I. Hodges, and M. Esch, 2007: Tropical cyclones in a T159 resolution global climate model: Comparison with observations and re-analyses. Tellus, 59, 396–416.
Brekke, L. D., E. Maurer, J. Anderson, M. Dettinger, E. Townsley, A. Harrison, and T. Pruitt, 2009: Assessing reservoir operations risk under climate change. Water Resour. Res., 45, W04411, doi:10.1029/2008WR006941.
Chen, F., and J. Dudhia, 2001: Coupling an advanced land surface–hydrology model with the Penn State–NCAR MM5 modeling system. Part I: Model implementation and sensitivity. Mon. Wea. Rev., 129, 569–585.
Chen, F., and Coauthors, 1996: Modeling of land-surface evaporation by four schemes and comparison with FIFE observations. J. Geophys. Res., 101, 7251–7268.
Christensen, N. S., and D. P. Lettenmaier, 2007: A multimodel ensemble approach to assessment of climate change impacts on the hydrology and water resources of the Colorado River Basin. Hydrol. Earth Syst. Sci., 11, 1417–1434.
Collins, W. D., and Coauthors, 2006: The Community Climate System Model version 3 (CCSM3). J. Climate, 19, 2122–2143.
Coquard, J., P. B. Duffy, K. E. Taylor, and J. P. Iorio, 2004: Present and future surface climate in the western USA as simulated by 15 global climate models. Climate Dyn., 23, 455–472.
Duffy, P. B., and Coauthors, 2006: Simulations of present and future climates in the western United States with four nested regional climate models. J. Climate, 19, 873–895.
Edwards, L. M., and K. T. Redmond, 2005: Climate factors on Colorado River Basin water supply. Colorado River Basin Climate: Paleo, Present, Future. California Department of Water Resources, 14–22.
Ek, M. B., K. E. Mitchell, Y. Lin, E. Rogers, P. Grunman, V. Koren, G. Gayno, and J. D. Tarpley, 2003: Implementation of Noah land surface model advances in the National Centers for Environmental Prediction operational mesoscale Eta Model. J. Geophys. Res., 108, 8851, doi:10.1029/2002JD003296.
Garvert, M. F., B. Smull, and C. Mass, 2007: Multiscale mountain waves influencing a major orographic precipitation event. J. Atmos. Sci., 64, 711–737.
Grubišić, V., R. K. Vellore, and A. W. Huggins, 2005: Quantitative precipitation forecasting of wintertime storms in the Sierra Nevada: Sensitivity to the microphysical parameterization and horizontal resolution. Mon. Wea. Rev., 133, 2834–2859.
Hara, M., T. Yoshikane, H. Kawase, and F. Kimura, 2008: Estimation of the impact of global warming on snow depth in Japan by the pseudo-global warming method. Hydrol. Res. Lett., 2, 61–64.
Hoerling, M., and J. Eischeid, 2006: Past peak water in the Southwest. Southwest Hydrol., 6 (1), 18–19.
Hoerling, M., D. Lettenmaier, D. Cayan, and B. Udall, 2009: Reconciling projections of Colorado River streamflow. Southwest Hydrol., 8 (3), 20–21, 31.
Houghton, J. T., Y. Ding, D. J. Griggs, M. Noguer, P. J. van der Linden, X. Dai, K. Maskell, and C. A. Johnson, Eds., 2001: Climate Change 2001: The Scientific Basis. Cambridge University Press, 881 pp.
Ikeda, K., R. Rasmussen, W. D. Hall, and G. Thompson, 2007: Observations of freezing drizzle in extratropical cyclonic storms during IMPROVE-2. J. Atmos. Sci., 64, 3016–3043.
Ikeda, K., and Coauthors, 2010: Simulation of seasonal snowfall over Colorado. Atmos. Res., 97, 462–477.
Johannessen, O. M., K. Khvorostovsky, M. W. Miles, and L. P. Bobylev, 2005: Recent ice-sheet growth in the interior of Greenland. Sci. Express, 310, 1013–1016, doi:10.1126/science.1115356.
Johnson, J. B., and D. Marks, 2004: The detection and correction of snow-water equivalent pressure sensor errors. Hydrol. Processes, 18, 3513–3525.
Kawase, H., T. Yoshikane, M. Hara, F. Kimura, T. Yasunari, B. Ailikun, H. Ueda, and T. Inoue, 2009: Intermodel variability of future changes in the Baiu rainband estimated by the pseudo global warming downscaling method. J. Geophys. Res., 114, D24110, doi:10.1029/2009JD011803.
Koren, V., J. Schaake, K. Mitchell, Q.-Y. Duran, F. Chen, and J. M. Baker, 1999: A parameterization of snowpack and frozen ground intended for NCEP weather and climate models. J. Geophys. Res., 104, 19 569–19 585.
Krabill, W., and Coauthors, 1999: Rapid thinning of parts of the southern Greenland ice sheet. Science, 283, 1522–1524, doi:10.1126/science.283.5407.1522.
Krabill, W., and Coauthors, 2000: Greenland ice sheet: High-elevation balance and peripheral thinning. Science, 289, 428–430, doi:10.1126/science.289.5478.428.
Krabill, W., and Coauthors, 2004: Greenland ice sheet: Increased coastal thinning. Geophys. Res. Lett., 31, L24402, doi:10.1029/2004GL021533.
Leung, L. R., and Y. Qian, 2003: The sensitivity of precipitation and snowpack simulations to model resolution via nesting in regions of complex terrain. J. Hydrometeor., 4, 1025–1043.
Leung, L. R., Y. Qian, J. Han, and J. Roads, 2003: Intercomparison of global reanalyses and regional simulations of cold season water budgets in the western United States. J. Hydrometeor., 4, 1067–1087.
Livneh, B., Y. Xia, K. Mitchell, M. Ek, and D. P. Lettenmaier, 2009: Noah LSM snow model diagnostics and enhancements. J. Hydrometeor., 11, 721–738.
Lynn, B. H., R. Healy, and L. M. Druyan, 2009: Investigation of Hurricane Katrina characteristics for future warmer climates. Climate Res., 39, 75–86.
Medina, S., B. F. Smull, R. A. Houze Jr., and M. Steiner, 2005: Cross-barrier flow during orographic precipitation events: Results from MAP and IMPROVE. J. Atmos. Sci., 62, 3580–3598.
Mesinger, F., and Coauthors, 2006: North American Regional Reanalysis. Bull. Amer. Meteor. Soc., 87, 343–360.
Milly, P. C. D., K. A. Dunne, and A. V. Vecchia, 2005: Global pattern of trends in streamflow and water availability in a changing climate. Nature, 438, 347–350.
Pan, M., and Coauthors, 2003: Snow process modeling in the North American Land Data Assimilation System (NLDAS): 2. Evaluation of model-simulated snow-water equivalent. J. Geophys. Res., 108, 8850, doi:10.1029/2003JD003994.
Rasmussen, R. M., and Coauthors, 1988: Comparison of orographic cloud and precipitation development of three different geographic locations using a 3-D nested-grid cloud model. Ann. Meteor., 1, 353–355.
Rasmussen, R. M., and Coauthors, 2001: Weather Support to Deicing Decision Making (WSDDM): A winter weather nowcasting system. Bull. Amer. Meteor. Soc., 82, 579–595.
Rutter, N., and Coauthors, 2009: Evaluation of forest snow processes models (SnowMIP2). J. Geophys. Res., 114, D06111, doi:10.1029/2008JD011063.
Schär, C., C. Frie, D. Lüthi, and H. C. Davies, 1996: Surrogate climate-change scenarios for regional climate models. Geophys. Res. Lett., 23, 669–672.
Seager, R., and Coauthors, 2007: Model projections of an imminent transition to a more arid climate in southwestern North America. Science, 316, 1181–1184.
Serreze, M. C., M. P. Clark, R. L. Armstrong, D. A. McGinnis, and R. S. Pulwarty, 1999: Characteristics of the western United States snowpack from snowpack telemetry (SNOTEL) data. Water Resour. Res., 35, 2145–2160.
Serreze, M. C., M. P. Clark, and A. Frei, 2001: Characteristics of large snowfall events in the montane western United States as examined using snowpack telemetry (SNOTEL) data. Water Resour. Res., 37, 675–688.
Sheffield, J., and Coauthors, 2003: Snow process modeling in the North American Land Data Assimilation System (NLDAS): 1. Evaluation of model-simulated snow cover extent. J. Geophys. Res., 108, 8849, doi:10.1029/2002JD003274.
Skamarock, W. C., J. B. Klemp, J. Dudhia, D. O. Gill, D. M. Barker, W. Wang, and J. G. Powers, 2005: A description of the Advanced Research WRF Version 2. NCAR Tech. Note TN-468 STR, 88 pp.
Solomon, S., D. Qin, M. Manning, M. Marquis, K. Averyt, M. M. B. Tignor, H. L. Miller Jr., and Z. Chen, Eds., 2007: Climate Change 2007: The Physical Science Basis. Cambridge University Press, 996 pp.
Thompson, G., P. R. Field, R. M. Rasmussen, and W. D. Hall, 2008: Explicit forecasts of winter precipitation using an improved bulk microphysics scheme. Part II: Implementation of a new snow parameterization. Mon. Wea. Rev., 136, 5095–5115.
Trenberth, K. E., A. Dai, R. M. Rasumussen, and D. B. Parsons, 2003: The changing character of precipitation. Bull. Amer. Meteor. Soc., 84, 1205–1217.
Wilby, R. L., L. E. Hay, W. J. Gutowski Jr., R. W. Arritt, E. S. Takle, Z. Pan, G. H. Leavesley, and M. P. Clark, 2000: Hydrologic responses to dynamically and statistically downscaled general circulation model output. Geophys. Res. Lett., 27, 1199–1202.
Wood, A. W., L. R. Leung, V. Sridhar, and D. P. Lettenmaier, 2004: Hydrologic implications of dynamical and statistical approaches to downscaling climate model outputs. Climatic Change, 62, 189–216.
Yang, D., B. E. Goodison, J. R. Metcalfe, V. S. Golubev, R. Bates, T. Pangburn, and C. L. Hanson, 1998: Accuracy of NWS 8” standard nonrecording precipitation gauge: Results and application of WMO intercomparison. J. Atmos. Oceanic Technol., 15, 54–68.
Yin, J. H., 2005: A consistent poleward shift of the storm tracks in simulations of 21st century climate. Geophys. Res. Lett., 32, L18701, doi:10.1029/2005GL023684.