1. Introduction
Snowpack provides drinking water to over 40% of the world’s population (Meehl et al. 2007) and generates between 50% and 80% of total runoff within the western United States (Dettinger 2005), making it an important resource. Simulations of snow and hydrological processes have been found to be critically dependent on the meteorological forcing data (Reed et al. 2004; Mote et al. 2005; Tobin et al. 2011). The accuracy of precipitation and air temperature data, which ultimately determine the quantity and phase (rain or snow) of modeled precipitation, directly affects the accuracy of simulated runoff and snow accumulation, particularly at relatively small spatial scales and shorter time scales. Other meteorological forcing variables are also important. For the western slopes of the Sierra Nevada in California, for example, over 80% of the snow surface energy balance is controlled by net irradiance (Marks and Dozier 1992); thus, errors in downward shortwave (SW) and longwave (LW) irradiance data strongly influence simulated snowpack melt rates. In addition, errors in the relative humidity (RH) forcing degrade simulations of snow sublimation as well as evapotranspiration. Although not exhaustive, Table 1 represents some common sources of forcing data used for hydrological studies within mountainous terrain. Evaluating these different sources of forcing data, however, is not a trivial task, especially in high-elevation mountain basins where very limited observations of these forcing variables are typically available (Weingartner and Pearson 2001; Lundquist et al. 2003).
Typical availability, estimation, and distribution techniques of hydrological modeling variables.
In this paper we examine various combinations of 1) in situ observations, 2) empirical models (see section 3), and 3) physically based simulations from the Weather and Research Forecasting Model (WRF, described in section 3) (Skamarock and Klemp 2008) in a well-instrumented maritime mountain basin, the North Fork of the American River basin (hereafter NF American basin; Fig. 1) in the northern Sierra Nevada, California. Because of the previously referenced dominance of the irradiance terms in the surface energy balance in this river basin (Marks and Dozier 1992), we do not examine turbulent fluxes and wind speed sources (W), but instead focus on the sources of precipitation, temperature, relative humidity, and downward SW and LW irradiance data. Common empirical methods for estimating the downward irradiance variables are critically dependent on the diurnal temperature range and the relative humidity used. Therefore, we also investigate these effects. Below we describe specific experiments based on the three cases (see Table 2), for which specific forcing sets over the NF American basin were created and evaluated.
(a) Map of the NF American basin located on the western slope of the Sierra Nevada. Shown are the meteorological stations (HMT, CDWR, and ibuttons) and WRF grid cell centers used in this study. Labeled stations show stations measuring net irradiance (*) and downward solar irradiance (^). (b) Location of basin (arrow) and extent of the WRF 6-km domain (white outline). (c) Stars show the locations of the eight snow pillows used for model analysis. Red outlines show the boundaries of the Onion Creek and East Fork subbasins in which stream stage was recorded. The black contour running north to south shows the climatological average snow line at 1500 m.
Citation: Journal of Hydrometeorology 14, 3; 10.1175/JHM-D-12-0102.1
Common cases of meteorological forcing data sources (more fully described in section 3) and examples of prior studies using each approach. Obs stands for observations; DTR stands for diurnal temperature range.
Case 1 represents a study basin in which there are only one or two stations that measure precipitation, temperature, and relative humidity. These point observations are then extrapolated to the entire basin using spatiotemporal weights based on historical observations from regional studies. This methodology has been employed in studies including Waichler and Wigmosta (2003), Surfleet et al. (2011), Kuraś et al. (2011), and Whitaker et al. (2003). Table 1 highlights the typical availability of meteorological observations and common methods of estimating and distributing the unmeasured variables required for hydrological modeling. In comparison with temperature and precipitation, dewpoint temperature and SW and LW irradiance are sparsely measured variables, which has led previous studies to estimate dewpoint temperature and irradiance using empirical models based on measured predictors such as the daily minimum temperature (dewpoint) and diurnal temperature range (SW irradiance) (e.g., Thornton and Running 1999). Such empirical models are often fit to observed data from a particular region or climate, and their transferability to other regions has been noted as a significant limitation (Flerchinger et al. 2009).
Case 2 represents a study where only atmospheric model output is used to generate surface forcing data (e.g., as applied in Zhao et al. 2009; Westrick 2001; Westrick and Storck 2002; Leung et al. 1996). An important advantage in using an atmospheric model is that the model can provide all meteorological variables at grid points with a greater spatial density than most observational networks. Increased spatial resolution of each grid cell has allowed numerical weather prediction models to resolve the topography that drives orographic precipitation gradients (Barros and Lettenmaier 1994; Anders et al. 2007), giving greater confidence in the underlying physical basis of the models. However, the accuracy of modeled precipitation has been shown to be sensitive to the microphysics scheme chosen (Chin et al. 2010; Dettinger et al. 2012; Minder et al. 2011). Also, a more physically based estimation of downward SW and LW irradiance is achieved through the model’s vertical representation of atmospheric moisture content, vertical temperature profile, and cloud cover.
Finally, in case 3, an atmospheric model generates surface temperature and precipitation at grid points within a basin, but irradiance is empirically estimated at each point based on the diurnal temperature range and relative humidity of the atmospheric model (e.g., as in Dettinger et al. 2012; Das et al. 2011). Using an empirical model as the source of irradiance input may be preferable because of biases in atmospheric models stemming from grid size restrictions (Ruiz-Arias et al. 2011) and representation of cloud cover (Edwards and Slingo 1996). However, it is not clear whether empirical models that have been fitted to observed diurnal temperature range and relative humidity will perform as well with atmospheric model input that may be characteristically different from in situ observations.
For each case of meteorological forcing data, we ask the following questions:
How do simulated meteorological variables compare with in situ observations, particularly in high-elevation areas where stations are typically unavailable?
How and where does the source of forcing data impact simulations of snowpack?
How does the choice of forcing data impact streamflow simulations in basins draining different elevation ranges?
To answer these questions, we selected the NF American basin (Fig. 1) as a case study. The basin is particularly well suited for exploring the hydrological sensitivity to the choice of meteorological forcing for several reasons. First, it is a relatively simple basin in terms of subsurface contributions. Shallow soils, steep topography, and negligible groundwater contribution during the cool season mean that accurate simulation of streamflow is largely dependent on the meteorological forcing. Second, the National Oceanic and Atmospheric Administration’s (NOAA) Hydrometeorology Testbed (HMT) program has maintained a dense network of meteorological stations that cover the basin (Ralph et al. 2005). Third, the upper portions of the basin are largely unaffected by water management (such as dams or diversions for water supply), and observations of snowpack and streamflow are also available for the basin.
We use the HMT network and other observational datasets (described in section 2) to evaluate different cases of empirical and mesoscale input data. In section 3 we outline specific cases of meteorological sources and use them to determine where they have the greatest impact on simulated snowpack and streamflow using the Distributed Hydrology Soil and Vegetation Model (DHSVM; Wigmosta et al. 1994). Results are reported in section 4, a brief discussion is provided in section 5, and conclusions are summarized in section 6.
2. Observed data
Our first question was addressed by comparing each forcing dataset to all suitable meteorological observations within the NF American basin. Locations of all meteorological stations used in this study are shown in Fig. 1a and listed in appendix A. Because 98% of the NF American basin precipitation occurred from October through June (National Climatic Data Center, http://www.ncdc.noaa.gov/oa/ncdc.html), we restricted our analysis to these months. Hourly observations of precipitation and air temperature for water years 2001–10 (October – September) were obtained from eight weather stations operated by the California Department of Water Resources (CDWR; data available through the California Data Exchange Center at http://cdec.water.ca.gov/). Measurements of precipitation, temperature, and relative humidity taken every two minutes were obtained at 13 HMT stations in or near the study basin (data available at http://hmt.noaa.gov/). Three stations (Desert Research Institute; HMT) provided downward SW irradiance measurements, and three HMT stations observed net irradiance from 2006 to 2010 (Fig. 1a). In addition, 52 self-recording sensors (iButtons) were distributed in trees across the basin (following methods in Lundquist and Huggett 2008) to provide temperature and relative humidity data from 2008 to 2010.
Continuous daily streamflow measurements at the basin outlet were acquired from the U.S. Geological Survey (USGS; station #11427000, North Fork Dam). To allow verification of the simulated internal basin streamflow timing, additional stream stage data (described in Lundquist et al. 2009) for water years 2007–10 were measured in two subbasins, shown in Fig. 1c.
Snow water equivalent (SWE) data were obtained from six snow pillows from the CDWR and two snow pillows from the U.S. Natural Resources Conservation Service’s Snowpack Telemetry (SNOTEL) network (data available at http://www.wcc.nrcs.usda.gov/snow/; Fig. 1c). Four of the snow pillows are located near the basin’s climatological snow line at 1500 m (Mizukami and Smith 2012; Shamir and Georgakakos 2006), and four are located above 2000 m.
All data were closely quality controlled for unrealistic outliers, constant values, and extreme jumps, following Meek and Hatfield (1994). The 2-min precipitation, temperature, and vapor pressure measurements from HMT were aggregated to hourly values if at least 75% of that hour was available; otherwise, the hour was considered missing. The temperature data from the self-recording sensors required minor filling to prevent a cold bias when sensors became briefly snow covered (see appendix B for details). Because of the limited measurement of wind speed at all stations, no correction for precipitation gauge undercatch was attempted (Sevruk 1983).
3. Methods
We created three cases of forcing data sources, which are the NF American basin–specific counterparts to the general cases described in Table 2. The source of the variables in each case is shown in Table 3, and the methods used to generate them are described below. For each source, we compared wet season (defined as October–June) averages of distributed temperature, relative humidity, and accumulated precipitation to observations. Comparing elevational gradients of estimated relative humidity is problematic as it may include temperature errors; therefore, all observed relative humidity values and WRF modeled mixing ratio values were converted to vapor pressure using paired temperature and pressure data (sections 3a and 3b). The empirical irradiance equations examined here parameterize the effect of clouds using surface relative humidity; therefore, to compare this method separately from clear day irradiance, estimated SW and LW irradiance data were separated into clear and cloudy days using a clear sky index based on the observed daily SW irradiance as follows. Daily averaged values greater than half of the averaged potential clear sky irradiance for each day of the year were defined as clear. Modeled streamflow from DHSVM was evaluated using the percent bias of flow, the determinate of correlation, and the Nash–Sutcliffe efficiency (NSE).
Cases of meteorological variable sources used to force DHSVM. For all sources, wind speed was taken from the Secret Town station. Empirical estimates are based on the diurnal temperature range (DTR) and the relative humidity (RH).
a. Case 1: Limited observations
A common constraint for hydrological studies within complex terrain is to have only one available observational station measuring maximum and minimum temperature, daily accumulated precipitation, and relative humidity at low to middle elevations. The SW and LW irradiance inputs required by distributed hydrological models must then be estimated at the station, and all variables distributed to the rest of the basin. We simulated this type of observationally based forcing set by selecting the Secret Town station (Fig. 1a) as our base station. This station was selected because it is located below the snow line at 829 m, which allowed easier maintenance and resulted in minimal missing data over the 2001–10 period of interest (0.5% missing precipitation data, 1% missing temperature data, and 1% missing relative humidity data).
Following the methods of Thornton and Running (1999), the potential daily SW irradiance and its transmissivity through the atmosphere were calculated at the Secret Town station based on the observed diurnal temperature range and daily relative humidity averaged from hourly instantaneous measurements. Daily values of SW irradiance were then disaggregated to a 3-hourly time step according to the local solar zenith angle. Following the recommendations of Flerchinger et al. (2009), we used the algorithm of Dilley and O’Brien (1998) to estimate mean daily downward clear sky LW irradiance from the diurnal temperature range and daily relative humidity. Adjustments for cloudy sky were made following Marks et al. (1998) using estimates of precipitable water from Prata (1996). We note that there exists variability among empirical irradiance models, although we focus only on the above selections for brevity.
b. Case 2: Output from the WRF
We selected the WRF (Skamarock and Klemp 2008) as a mesocale model to generate 6-km surface meteorological data for 10 partial water years (October–June of water years 2001–10). Boundary conditions were provided by the North American Regional Reanalysis (NARR), a 32-km gridded product created by ingesting surface and upper-air observations over the continental United States (Mesinger et al. 2006). To allow the surface output of the WRF to be comparable to the historic observations, we reinitialized the lateral and internal boundary conditions every 5 days with 3 h of spinup, while continuously updating the lateral boundary conditions every 3 h, resulting in a temporally continuous simulation. This technique takes advantage of the accurate forecasting range of the WRF (~5 days) and has been shown to capture the variability in meteorological conditions over an 11-yr period in Southern California when generated with the fifth-generation Pennsylvania State University–National Center for Atmospheric Research (NCAR) Mesoscale Model (MM5), WRF’s predecessor (Abel and Hall 2009).
The downscaling simulation contained two domains: an 18-km horizontal resolution domain that covered California and extends west over the Pacific Ocean and a 6-km domain over California (Fig. 1b). Two-directional nesting was employed between the 18- and 6-km domains. Each domain contained 27 vertical levels, with the vertical grid stretched to place the highest resolution in the lower troposphere. In the 18-km domain, the Kain–Fritsch cumulus parameterization was used (Kain 2004); in the 6-km domain only explicitly resolved convection could occur. Both domains used the Yonsei University (YSU) boundary layer scheme (Hong et al. 2006), the Morrison two-moment microphysics scheme (Morrison et al. 2009), the rapid radiative transfer model LW radiation scheme (Mlawer et al. 1997), the Dudhia SW radiation scheme (Dudhia 1989), and the Noah land surface model with four ground layers (Chen and Dudhia 2001).
c. Case 3: Combination of the WRF and empirical models
This case of meteorological data emulates those studies that used temperature, precipitation, and relative humidity from mesoscale model output and then estimated SW and LW irradiance using empirical models (case 3). Temperature and precipitation were used as in case 2 from the WRF, and SW and LW irradiance were estimated at these points using the empirical methods described in case 1. In addition, we created variants on case 3 in order to independently test the impact of different relative humidity sources on empirical estimation of SW and LW irradiance. Case 3A empirically estimated SW irradiance using WRF relative humidity based on the dewpoint assumption and WRF LW irradiance. We compared case 3A and case 2 to illustrate the different characteristics of downward SW irradiance estimation between an empirical model and the WRF model.
In cases 3B and 3C, we empirically estimated LW irradiance based on relative humidity data from the WRF and from the dewpoint assumption, respectively. Both case 3B and case 3C used the WRF SW irradiance. The comparison between cases 3B and 3C allowed us to access the importance of accurate relative humidity data for empirical estimation of LW forcing data. These last two variants of case 3 were then compared to the WRF (case 2) and observations in order to identify how, and at which locations, errors from different sources of LW irradiance most impact simulated snowpack and streamflow.
d. Hydrological simulations of the NF American basin
To evaluate the impact of simulated hydrologic variables, each meteorological forcing set in Table 3 was then used to drive the DHSVM, which is a fully distributed, physically based hydrological model that simulates the surface energy and water balance and the transport of water as a function of meteorological forcing, topography, soil drainage characteristics, and vegetation cover (Wigmosta et al. 1994; Waichler and Wigmosta 2003). It has commonly been applied at grid scales ranging from 30 to 180 m over basins that include complex terrain (Leung et al. 1996; Westrick 2001; Westrick and Storck 2002; Zhao et al. 2009; Nijssen et al. 1997). For each time step at each grid cell, DHSVM uses a full energy balance approach to determine snow accumulation and ablation within the overstory and understory (if present) and at the surface (Andreadis et al. 2009). Surface and subsurface moisture transport are simulated via the topographic gradients between grid cells, with water ultimately propagating to the simulated stream network, where a one-dimensional routing scheme is used to calculate streamflow at the basin outlet (Wigmosta et al. 1994).
DHSVM was set up to run at a 150-m grid resolution and 3-h time step over the NF American basin (Fig. 1a) and in point mode at selected snow pillow stations (Fig. 1c). This grid cell size was chosen to balance the need to accurately represent the NF American basin’s topography with computational expense. All simulations used the same initial conditions of soil moisture and temperature, which were created from a separate DHSVM simulation using the most available observational stations (4) that had continuous records for the study period. Because of the dry natural of soils in the NF American basin during late summer each year, simulated streamflow showed little sensitivity to initial soil conditions. SWE was initialized to zero for all simulations in accordance with snow pillow observations. Soil type and land cover data were obtained from the State Soil Geographic (STATSGO) and USGS databases (Soil Survey Staff 2006; Fry et al. 2011). The basin is predominantly covered by evergreen needle leaf (71%), as well as open shrub (17%) and mixed forest (6%). At every time step and at each grid cell, DHSVM requires the following meteorological forcing data: precipitation, temperature, vapor pressure, wind speed, and downward SW and LW irradiance.
A full model calibration of DHSVM for the NF American basin was beyond the scope of our investigation, and instead, standard parameters were chosen based on previous studies in maritime environments (Wigmosta et al. 1994; Storck 2000). Model performance as it relates to our specific use of the model in the experiments is evaluated in subsequent sections.
4. Results
a. How do different sources of precipitation compare?
Wet season accumulated precipitation values from case 1 (one measurement site extrapolated using PRISM climatology) and case 2 (WRF output) were plotted against elevation and compared to observations in Fig. 2. For most years, the PRISM forcing (case 1) and the WRF (case 2) capture the correct averaged orographic precipitation gradient; yet some years, such as 2002 and 2003, show large biases. For example, WRF has a +234 mm (+26%) and +458 mm (+34%) wet bias during water years 2001 and 2003, while the PRISM-based estimates more closely match observations. In contrast, the PRISM-based estimations had a −237 mm (−18%) dry bias in 2002, while WRF more closely matched observations. Also, during water year 2006, when the study domain received ~180% of the climatological average, the PRISM forcing (case 1) overestimated precipitation observations by +409 mm (+17%), while WRF (case 2) had a −68 mm (−3%) bias. These errors in case 1 are related to the relationship between the PRISM base station (Secret Town, the lowest-elevation green dot in Fig. 2) and the higher-elevation stations. During years when the ratio of accumulated precipitation between the Secret Town station and the surrounding stations was different than climatology, the case 1 errors were large (years 2002 and 2006, Fig. 2). Averaged over the entire study period, WRF had a consistent bias of +196 mm (17%), although some of this difference is likely because of gauge undercatch, which was not accounted for in the observations.
Total wet season accumulated precipitation vs elevation for water years 2001–10. Black asterisks show the WRF precipitation for each of the (61) 6-km grid cells. The gray dots represent the observed precipitation at the Secret Town station extrapolated using weights from PRISM climatology (see section 3 for details). Observations from HMT and CDWR stations are shown as diamonds and circles, respectively. Note that water year 2006 required a different x-axis scaling.
Citation: Journal of Hydrometeorology 14, 3; 10.1175/JHM-D-12-0102.1
b. Impact of precipitation forcings on simulations of streamflow
Over the 2001–10 period, the DHSVM modeled streamflow had adequate performance when compared to daily observations at the basin outlet. For the entire study period, case 1 and case 2 had daily coefficient of determination (R2) values of 0.58 and 0.54, percent biases of 3% and 1%, and NSE values of 0.57 and 0.53, respectively. The poor NSE values reflect errors in precipitation timing (i.e., errors in spatial weighting in case 1 and storm track errors in case 2) and errors as a result of the uncalibrated model. In addition, the period of statistics considered, October–June, includes more variable streamflow than the summer months that were not examined here. Differences in basin-averaged evapotranspiration between cases were small compared to differences in streamflow during the October–June period, but would likely become more significant if summer months were considered. While the average statistics for each forcing case were indistinguishable, the least biased forcing case varied between years depending on the source of precipitation forcing.
Two example years, 2003 (Fig. 3) and 2006 (Fig. 4), highlight how biases in precipitation propagate into simulated streamflow. Simulated streamflows from both cases at the basin outlet during 2003 are shown in Fig. 3a. The largest simulated streamflow errors occurred at the end of December and early January (Fig. 3b), reflecting differences in the precipitation forcing shown in Fig. 3d. Interestingly, case 2 resulted in a lower absolute bias in the total accumulated streamflow (expressed in basin equivalent precipitation and shown in Fig. 3c) of +144 mm (+20%) compared with the case 1 bias of −349 mm (−48%), even though case 2 had a larger absolute accumulated precipitation bias compared to observations (Fig. 2).
(a) DHSVM’s modeled streamflow at the NF American basin outlet. (b) The error compared to observations at the NF USGS gauge for water year 2003. (c) Cumulative streamflow and (d) basin-averaged precipitation.
Citation: Journal of Hydrometeorology 14, 3; 10.1175/JHM-D-12-0102.1
As in Fig. 3, but for water year 2006.
Citation: Journal of Hydrometeorology 14, 3; 10.1175/JHM-D-12-0102.1
In contrast, during water year 2006 (Fig. 4), the biases in precipitation forcings (and resulting simulated streamflow bias) had opposite signs as 2003. Case 1 had a precipitation bias of +409 mm (17%) and resulting total streamflow bias of +403 mm (26%), while case 2 had a precipitation bias of −68 mm (−3%) and a total streamflow bias −252 mm (−16%). These two examples illustrate the importance of selecting an accurate source of precipitation forcing for unbiased simulation of total streamflow. Differences in the timing of simulated streamflow resulting from the two different forcing cases could be due to precipitation errors as well as the timing of simulated snow accumulation and ablation. Because the simulation of snow processes is dependent on other meteorological forcing variables, we examine these impacts below.
c. How do different sources of temperature compare?
The two sources of estimated temperature agreed well with observations of mean temperature and with each other over all water years. Figure 5a shows the mean wet season mean temperature plotted against elevation for water years 2008 to 2010, when additional HMT, CDWR, and iButton records were available for comparison. The WRF output was not statistically different from the temperatures estimated using the simpler case 1 method during the averaged wet season period. However, on subdaily time scales, the two sources of temperature had different averaged diurnal temperature ranges (Fig. 5b). WRF’s diurnal temperature range (case 2) was on average 3.0°C smaller than the average of all observations sites, while case 1 was on average 2.6°C larger. Interestingly, the differences between these two estimated forcings were as large as the variance between observations. In regards to hydrological modeling, errors in the diurnal temperature range are important because they can bias empirical estimates of SW irradiance and simulations of transpiration.
(a) Mean October–June air temperature vs elevation and (b) average diurnal range of temperature from October to June for (left to right) water years 2008–10. Dashed line indicates constant diurnal temperature range resulting from lapsing temperature from the Secret Town station.
Citation: Journal of Hydrometeorology 14, 3; 10.1175/JHM-D-12-0102.1
d. How do different estimates of vapor pressure compare?
Empirical models of LW irradiance are dependent on the surface vapor pressure to provide an estimate of cloudiness and of atmospheric transmissivity during clear sky conditions (Kimball et al. 1982); therefore, three sources of estimated surface vapor pressure were compared to observations (Fig. 6). For case 1, the assumed −1.25°C km−1 dewpoint lapse rate [proposed by Franklin (1983) and used in Running et al. (1987)] from the Secret Town station did not match the decreasing gradient of observed vapor pressure over the NF American basin. WRF (case 2) generally captured the observed gradient of vapor pressure, yet had a mean bias overall of −190 Pa across all elevations. When we used WRF’s minimum air temperature to estimate the dewpoint, the resulting average vapor pressure (case 3C) had a mean bias overall of +245 Pa. For comparison, the averaged difference between case 2 and case 3C in terms of relative humidity is 32%.
Average October–June vapor pressure vs elevation for (left to right) water years 2008–10 within the region of the NF American basin. All relative humidity estimates and observations were converted to vapor pressure as described in text.
Citation: Journal of Hydrometeorology 14, 3; 10.1175/JHM-D-12-0102.1
e. Comparison of estimated irradiance forcing sources
Three stations measured downward SW irradiance from 2006 to 2010 (Fig. 1a). We examined the mean daily fluxes between March and June to highlight differences in the estimated forcings during the melt season. Comparisons between daily observed fluxes and the estimated forcing sources are shown in Fig. 7a for clear days and in Fig. 8a for cloudy days (defined in section 3). On clear days, differences between case 1 and case 2 at all stations were not significant; however, the empirical estimation of SW irradiance using the output from the WRF temperature range (case 3A) had average median values 37 W m−2 less than WRF (case 2) and 24 W m−2 less than observations. The latter result implies that the empirical SW irradiance equations used here are not appropriate to use with temperature from WRF (i.e., case 3A), and the irradiance from WRF itself should be used. On cloudy days (Fig. 8a), both of the empirical estimates (cases 1 and 3A) were consistently positive biased, while WRF (case 2) was not statistically different from observations at Big Bend and Foresthill and was the least biased option at the Sierra Snow Laboratory. In general, the WRF SW irradiance performed better than the empirical models tested.
Averaged daily statistics from each radiation forcing over clear days between March and June of (a) downward SW irradiance, (b) downward LW irradiance, and (c) simulated net irradiance compared to available observational stations. The x axis identifies the case number for each model run (see legend), and the columns identify the site location sorted by elevation. We define clear days when the mean observed SW irradiance is >½ of the daily averaged potential SW irradiance. Plus signs show outliers.
Citation: Journal of Hydrometeorology 14, 3; 10.1175/JHM-D-12-0102.1
As in Fig. 7, but for cloudy days. We define cloudy days when the mean observed SW irradiance is less than half of the daily averaged potential SW irradiance.
Citation: Journal of Hydrometeorology 14, 3; 10.1175/JHM-D-12-0102.1
No direct observations of downward LW irradiance were available to test the accuracy of the estimated LW forcings. However, substantial differences between sources of estimated LW irradiance were found during both clear and cloudy periods (Figs. 7b, 8b). The LW irradiance from case 1 was empirically estimated at the Secret Town station and uniformly distributed across the basin, resulting in a median value of 311 W m−2 on cloudy days. In contrast, the LW irradiance calculated by WRF (case 2) decreased with elevation, with median values of 329, 295, and 281 W m−2 at Foresthill (1040 m), Blue Canyon (1609 m), and Big Bend (1739 m), respectively. This behavior is expected based on the observed decrease of surface temperature and vapor pressure in the atmospheric column at higher elevations, resulting in lower LW radiance emission (Marty et al. 2002). The median values of cases 3B and 3C also decreased with increasing elevation, but within a smaller range.
Each case of irradiance forcing was used to force the DHSVM over the NF American basin, and simulated net irradiance values at the surface were compared to point observations at the sites shown in Fig. 1a. For both clear and cloudy periods, the differences between each source of SW and LW forcing propagated into the simulated net irradiance (Figs. 7c, 8c). However, the evaluation of net irradiance was complicated by differences in the snow surface albedo, for which no direct observations were available. The simulated albedo decay was parameterized in DHSVM as a function of days since last snowfall (Laramie and Schaake 1972), which was the same for cases 2 and 3A–C, which all had the same precipitation input. During clear days (Fig. 7c), WRF had larger median biases in net radiation at the higher-elevation station, Big Bend (−39 W m−2), than at the lower station, Blue Canyon (−9 W m−2). The empirical estimations of LW (cases 3B and 3C) resulted in slightly less biased median values of −32 and −28 W m−2 at Big Bend and +3 and +9 W m−2 at Blue Canyon, respectively.
f. Impact of LW irradiance forcings on simulations of snowpack
Because the differences between estimated LW forcings showed the greatest variation during the melt season (Fig. 7a), we examined the impact of this variable on simulating snowmelt. Two snow observing stations were selected to illustrate the impacts of estimated forcings at high (Sierra Snow Laboratory, 2100 m) and middle (Blue Canyon, 1609 m) elevation sites. Figure 9a shows observed and simulated SWE at the Sierra Snow Laboratory (2100 m) for each different LW forcing case over water year 2009. All other forcing variables were taken from WRF and were constant between simulations. SWE simulated at the high-elevation site (Sierra Snow Laboratory) only differed among cases 2, 3B, and 3C after March when the simulated snowpack temperature increased to the melting point (Fig. 9c). During this period, the differences in simulated net irradiance (Fig. 9i) between WRF (case 2) and the empirical forcings (case 3B and 3C) resulted in different simulated melt rates. Case 2 had the lowest simulated melt rats and the lowest net irradiance, which was consistent with the negative bias found in the WRF net irradiance (Fig. 7c). The difference in LW forcings between case 3C and case 2 resulted in the simulated snow disappearance date at the Sierra Snow Laboratory station to shift later by 12 days, which was consistent with a shift toward later snow disappearance during other years (not shown). Empirical estimates of LW (cases 3B and 3C) were relatively insensitive to the source of relative humidity during this spring melt period (Figs. 9a,g,i).
Simulated SWE from DHSVM compared to observations at the (a) Sierra Snow Laboratory (2100 m) and (b) Blue Canyon (1609 m) during 2008–09. (left) Sierra Snow Laboratory and (right) Blue Canyon modeled (c),(d) internal pack temperature, (e),(f) downward SW irradiance, (g),(h) downward LW irradiance, and (i),(j) net irradiance. Note that (e),(g),(i) focus on the melt season (March–May), and (f),(h),(j) focus on the erroneous midwinter melt event during January. The time periods are marked with vertical dashed lines in (a) and (b).
Citation: Journal of Hydrometeorology 14, 3; 10.1175/JHM-D-12-0102.1
In contrast, the simulated snowpack at the Blue Canyon station (Fig. 9b), at a lower elevation of 1609 m, had a warmer snowpack temperature throughout the winter (Fig. 9d) and thus was more sensitive to winter biases in the surface energy balance. During a 10-day period in January, the empirically estimated net irradiance forcings (Fig. 9j) were biased high by ~50 W m−2, resulting in high simulated melt rates that were not observed at the Blue Canyon snow pillow (Fig. 9b). This bias in midwinter melt was largest for case 3C (empirical LW with relative humidity estimated from Tmin from WRF), and resulted in an offset in simulated SWE for the rest of the season. These January LW irradiance estimates (Fig. 9h) were very sensitive to the source of relative humidity, whereas the spring irradiance estimates (Fig. 9g) were not.
This example illustrates how simulated snowpacks at middle elevation sites are more sensitive to biased irradiance forcings during the winter, whereas higher and colder snowpacks are only sensitive during the spring melt period. To test the robustness of this observation, we compared modeled with observed SWE at eight snow pillows in the area, four above 2000 m and four below (Fig. 10). For snow in clearings, the January melt bias in case 3C (empirical LW) was present at all of the lower-elevation snow pillows (black lines, Figs. 10a,b,c).
(a) Observed SWE during the 2008–09 period at the eight snow pillows in or surrounding the NF American basin, separated into stations above and below 2000 m elevation. Point simulations of SWE with forcings from Case 2 and 3C are respectively shown with (b),(c) no vegetation (bare soil resembling a snow pillow) and with (d),(e) a forest overstory with 0.75 fractional coverage (representing the vegetation type covering 77% of the NF American basin).
Citation: Journal of Hydrometeorology 14, 3; 10.1175/JHM-D-12-0102.1
While the above biases in LW irradiance forcings were shown to have a critical effect on the simulated melt rates at open sites typical of snow pillows, their impact under a forest canopy would be expected to be significantly less, as a higher fraction of LW emission comes from the canopy instead of the atmosphere (Pomeroy et al. 2009). Within DHSVM, this process is parameterized based on the fractional coverage of the vegetation (see appendix C for a full description), which was set at 0.75 for the 77% of the NF American basin covered by an overstory canopy (Fry et al. 2011). Therefore, we evaluate how the presence of an overstory modifies the impact of different LW sources on simulated melt rates. As discussed above, simulated SWE within clearings (Figs. 10b,c) at lower-elevation sites (black lines) showed strong sensitivity to the lower LW forcing of case 2. However, the same forcings applied under a canopy with 0.75 fractional coverage reduced the difference between the simulations in case 2 and case 3C. At lower elevations, the simulated snow under the forest canopy melted in January regardless of how atmospheric LW irradiance was prescribed. Thus, within forested regions, the source of LW irradiance appears to be less important than the correct modeling of the canopy temperature. Because we do not have SWE measurements under the canopy, this remains an area for future investigation.
g. Impact of LW irradiance forcings on timing of simulations of streamflow
Figure 11a shows modeled streamflow at the NF American basin outlet compared to observations for 2008, which is representative of the spring pattern of simulated streamflow during 8 out of the 10 years examined here. The empirical LW forcing (case 3C) resulted in faster spring melt, which is illustrated by more flow in May and less flow in June than the all-WRF forcing scenario (case 2). The impact on basin-aggregated streamflow from different LW irradiance forcings was not as significant as one might expect from the snow pillow simulations due to the high forest density over the NF American basin. Both simulations of streamflow had the largest errors during individual storms and during the melt season (Fig. 11b). At each subbasin (Figs. 11c,d), the observed stage height featured diurnal snowmelt cycles through mid-May and then steadily decayed throughout June. Both simulations of streamflow captured this pattern at the lower East Fork basin, but had prolonged snowmelt and higher streamflow throughout June at the higher Onion Creek basin. These simulated streamflow errors at the higher subbasin propagated into the basin outlet (Fig. 11b) and are therefore important. The source of the higher flow error during June is likely a combination of early spring melt rates biased low and biased high wet season precipitation (+138 mm, 14%; Fig. 2), leading to prolonged simulated snowpack at high elevations and simulated runoff from snowmelt at Onion Creek during June when no was observed.
DHSVM’s modeled daily streamflow at (a) the NF American basin outlet compared to daily observations at the NF USGS gauge for water year 2008. Vertical dashed lines in (a) identify the period shown with a zoomed view in (b),(c),(d). Simulated 3-hourly streamflow at (b) the NF American basin, (c) Onion Creek, and (d) East Fork subbasins compared to 3-hourly observed stream stage (right axis). Note that (c),(d) only compare timing and relative magnitude of streamflow.
Citation: Journal of Hydrometeorology 14, 3; 10.1175/JHM-D-12-0102.1
5. Discussion
The evaluation of which source of forcing data is most accurate and consistent for hydrological simulation was hampered by the lack of measurements required to fully diagnose errors in the simulated snowpack energy balance, despite the higher than average station network around the NF American River basin. However, the meteorological forcings from the 6 km WRF compared to in situ observations were found to perform as well as, or better than, other sources examined in this study. This is a particularly important finding because such models can be produced in areas with limited station coverage, or even in remote areas that do not have any access to any other sources of data. That said, it is important to note that WRF was configured to produce a “best case” forcing data scenario, using updates from a reanalysis product every five days to prevent the large-scale weather systems from diverging from the historical observations. Data assimilation incorporated in the large-scale forcing used to drive the WRF simulations (and, in particular, accurate vertical soundings from nearby meteorological stations) would not be present in remote areas where inaccuracies in the large-scale forcing would be expected to degrade performance. For similar reasons, degradation of the accuracy of the WRF output is likely to occur when a free-running global circulation model is used instead of a reanalysis product (e.g., in climate change assessments). In addition, the transferability of these results to other regions may be dependent on the configuration of WRF. For example, the choice of the microphysics scheme and representation of cloud cover can affect the SW and LW irradiance at the surface (Edwards and Slingo 1996). However, for this configuration of WRF, we found equal or superior performance compared to other sources of driving data examined over the NF American basin.
Biases in the precipitation forcings from the Secret Town station and PRISM weights (case 1) and WRF (case 2) were found to vary in magnitude and sign between years. Averaged over the entire study period, case 2 had a consistent positive bias of +196 mm, although water year 2006 did have a negative bias (−68 mm). In contrast, case 1 had an overall lower average bias of +91 mm but did not show a consistent accumulated precipitation bias, underpredicting observed values on average by −123 mm during 2001–03 and overpredicting by +183 mm from 2004 to 2010. The variation of the bias in case 1 is related to the relationship between the PRISM base station and the higher-elevation stations, which underwent a change during 2003, and is consistent with the shift in gauge inconsistency found by Mizukami and Smith (2012), who used a larger dataset surrounding the NF American basin. In contrast, the uniform bias in case 2 may be caused by the choice of microphysics scheme in WRF, as Chin et al. (2010) suggests in a study over California, and/or by gauge undercatch. These results suggest that if the only available station measuring precipitation is unrepresentative of the basinwide orographic gradient, then WRF may provide a more consistent precipitation forcing. Given that in most locations we have limited means of assessing whether one existing station is representative, WRF is likely a better choice in areas of limited surface observations.
The biases in precipitation forcing for a given year were found to significantly impact the DSHVM-simulated streamflow at the NF American basin outlet. Two example years were used to illustrate how biased dry and wet precipitation forcings propagated into the total volume of simulated streamflow. DHSVM-simulated streamflow forced by the WRF model (case 2) had smaller total streamflow biases, yet simulated streamflow from case 1 had a higher average determinate of correlation because the precipitation forcing from Secret Town captured storm timing better than WRF. Although DHSVM was uncalibrated, because the NF American basin has steep slopes, shallow soils, and little storage, we expect the relative magnitude of the simulated streamflow response to different precipitation biases to remain unchanged following calibration. In addition, any attempt to calibrate DHSVM to a particular source of forcing data may have biased the evaluation of another forcing set, especially given the large differences between sources examined here.
In general, no single meteorological forcing set considered here was able to capture the observed melt rates at all elevations and during the entire study period. At higher-elevation sites [~(2000–2500) m], the simulated melt rates during early spring were lower than observations at snow pillow locations, contributing to a bias in simulated streamflow timing at the higher subbasin (Onion Creek) and in the aggregated basin output. Simulated snowpacks at midelevation sites [~(1500–2000) m] were found to be more sensitive to short-term positive biases in the radiation forcing than higher elevation sites. This was illustrated during a warm period in January of 2009, when higher values of empirically estimated LW irradiance caused high melt rates that were not observed. The sensitivity of warmer snowpacks in clearings has also been noted by Kuraś et al. (2011, their Fig. 4), who found DSHVM simulated melt prematurely in clearings but not under forest canopy and suggested the cause was the representation of the snow albedo decay. Our results suggest that this bias may, alternatively, be due to errors in estimates of incoming LW irradiance. However, another explanation is that errors in the simulated turbulent energy fluxes, which are on average small or opposite in sign (Marks and Dozier 1992), may have become significant for short periods, yet this remains for future work.
Shortwave irradiance generated directly by WRF performed better than empirical estimations using the WRF temperature, especially during cloudy days. However, the impact on simulated snowpack between different sources was negligible because the majority of melt events occurred on clear days. One possible explanation for the improved performance over other methods is that the physical representation of atmospheric moisture content led to the improved SW irradiance values over the NF American basin, although further work will be needed to test this hypothesis.
The observational network within the NF American basin region (Figs. 1a,c and appendix A) provided a unique opportunity to examine the accuracy and modeling impact of different forcing data sources. Although some unobserved variables (albedo and LW irradiance) limited the diagnosis of biased melt rates, this study was able to highlight the contrasting sensitivity of simulated snowpack at different elevational zones, indicating that the measurement of irradiance at one station alone does not provide representative forcing data. In addition, the choice of the least biased source of wet season accumulated precipitation varied between years and resulted in significant differences in simulated streamflow amount and timing. These results suggest that when selecting a forcing data source for hydrological simulation, it is critical that irradiance and precipitation datasets capture local elevational and seasonal patterns.
6. Summary of conclusions
This study used the Distributed Hydrology Soil and Vegetation Model to examine the impacts of different cases of meteorological forcing data on simulated snowpack and streamflow within and near the North Fork of the American River basin in the Sierra Nevada in California. Three cases of forcing data were created that represent commonly used configurations of forcing data from observations and empirical- and mesocale-model sources. The averaged wet season accumulated precipitation from the Secret Town station and PRISM forcings had the lowest mean bias of +91 mm (6%), but showed an inconsistent relationship with higher-elevation stations before and after the end of water year 2003. WRF had a higher mean bias of +196 mm (17%), but with the exception of 2006, was consistently positive biased between years. When DHSVM was forced with the precipitation forcing from the PRISM dataset and WRF, the percent biases of simulated flow at the NF American basin outlet over the entire study period were 3% and 1%, suggesting that undercatch likely represented a significant portion of the precipitation biases. However, model calibration may also affect the percent bias of a simulation through changes to evapotranspiration.
The elevational gradients of annual mean temperature from WRF agreed well with the climatological −6.5°C km−1 lapse rate, yet failed to capture the daily diurnal temperature range, which was biased low on average by 3.0°C. Surface vapor pressure estimated by assuming that saturation occurs during the minimum WRF temperature performed worse than the output from the WRF itself. Likewise, the empirical estimation of SW irradiance based on the WRF temperature had larger average median errors than the WRF model output, especially during cloudy days. Longwave irradiance from WRF captured the expected decreasing mean values with elevation, but whether a low bias in LW irradiance caused simulated melt rates to be too low over the Sierra Nevada remains for future work with additional observations.
In general, the choice of which case of meteorological forcing was best was not the same for each year, for each time of year (winter or spring), or at each elevation. This highlights the importance of evaluating sources of forcing data over long time periods and large areas. While the empirical estimated LW irradiance at high-elevation sites resulted in melt rates lower than observations, at lower elevations the same forcing caused midwinter melt that was not observed. The higher sensitivity of simulated snowpacks at lower-elevation sites, which observes intermittent snow cover, represents a difficult modeling environment that should be further used to test methods of generating forcing data as well as snow model formulation.
Acknowledgments
Funding was provided by the National Science Foundation (Grant EAR-0838166), by a University of Washington Program on Climate Change Graduate Fellowship, and a University of Washington Valle Fellowship. Support for in situ instrumentation and fieldwork was provided by NOAA through their Hydrometeorology Testbed (HMT) and through the Joint Institute for the Study of the Atmosphere and Ocean (JISAO) under NOAA Cooperative Agreement NA17RJ1232 and NA10OAR4320148.
APPENDIX A
Observational Stations Used in this Study
See Table A1 for observational stations used in this study.
Observational stations used in this study.
APPENDIX B
Filling of Temperature Sensor Gaps
Because the self-recording temperature sensors (iButtons) were placed in trees, they occasionally become covered in snow, resulting in low diurnal temperature ranges that have a mean near 0°C. Any day that exhibited this type of behavior was removed from the dataset. However, because these days are unrepresentatively cold, their removal introduces a warm bias to long-term means. Therefore, we employed empirical orthogonal function (EOF) reconstruction as described by von Storch and Zwiers (1999) and Beckers and Rixen (2003) and selected as the best reconstruction method by Henn et al. (2013), in order to estimate some periods of missing temperature data. Filling of missing data via EOF reconstruction was only applied to those stations that recorded data at least 90% of each October–June period between 2008 and 2010. The resulting quality-controlled and filled temperature data were then aggregated into monthly means if 90% of each month’s hourly values were present. This method expands the available temperature dataset while guaranteeing that wet season (October–June) means do not suffer from warm biases.
APPENDIX C
Fractional Longwave Irradiance Calculation under a Canopy
REFERENCES
Abel, M. R., and Hall A. , 2009: Local and synoptic mechanisms causing Southern California’s Santa Ana winds. Climate Dyn., 34, 847–857, doi:10.1007/s00382-009-0650-4.
Alduchov, O. A., and Eskridge R. E. , 1996: Improved Magnus form approximation of saturation vapor pressure. J. Appl. Meteor., 35, 601–609.
Anders, A. M., Roe G. H. , Durran D. R. , and Minder J. R. , 2007: Small-scale spatial gradients in climatological precipitation on the Olympic Peninsula. J. Hydrometeor., 8, 1068–1081.
Andreadis, K. M., Storck P. , and Lettenmaier D. P. , 2009: Modeling snow accumulation and ablation processes in forested environments. Water Resour. Res., 45, W05429, doi:10.1029/2008WR007042.
Barros, A. P., and Lettenmaier P. , 1994: Dynamic modeling of orographically induced precipitation. Rev. Geophys., 32, 265–284.
Beckers, J. M., and Rixen M. , 2003: EOF calculations and data filling from incomplete oceanographic datasets. J. Atmos. Oceanic Technol., 20, 1839–1856.
Bristow, K. L., and Campbell G. S. , 1984: On the relationship between incoming solar radiation and daily maximum and minimum temperature. Agric. For. Meteor., 31, 159–166, doi:10.1016/0168-1923(84)90017-0.
Brunt, D., 1932: Notes on radiation in the atmosphere. Quart. J. Roy. Meteor. Soc., 58, 389–420, doi:10.1002/qj.49705824704.
Brutsaert, W., 1975: On a derivable formula for long-wave radiation from clear skies. Water Resour. Res., 11, 742–744.
Chen, F., and Dudhia J. , 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.
Chin, H.-N. S., Caldwell P. M. , and Bader D. C. , 2010: Preliminary study of California wintertime model wet bias. Mon. Wea. Rev., 138, 3556–3571.
Daly, C., Neilson R. P. , and Phillips D. L. , 1994: A statistical-topographic model for mapping climatological precipitation over mountainous terrain. J. Appl. Meteor., 33, 140–158.
Daly, C., Gibson W. , Doggett M. , Smith J. , and Taylor G. , 2004: A probabilistic-spatial approach to the quality control of climate observations. Preprints, 14th AMS Conf. on Applied Climatology, Seattle, WA, Amer. Meteor. Soc., 7.3. [Available online at https://ams.confex.com/ams/pdfpapers/71411.pdf.]
Daly, C., Halbleib M. , Smith J. I. , Gibson W. P. , Doggett M. K. , Taylor G. H. , Curtis J. , and Pasteris P. P. , 2008: Physiographically sensitive mapping of climatological temperature and precipitation across the conterminous United States. Int. J. Climatol., 28, 2031–2064.
Das, T., Dettinger M. D. , Cayan D. R. , and Hidalgo H. G. , 2011: Potential increase in floods in California’s Sierra Nevada under future climate projections. Climate Change, 109, 71–94, doi:10.1007/s10584-011-0298-z.
Dettinger, M. D., 2005: Changes in streamflow timing in the western United States in recent decades. U.S. Fact Sheet 2006-3018, 4 pp. [Available online at http://pubs.usgs.gov/fs/2005/3018/pdf/FS2005_3018.pdf.]
Dettinger, M. D., and Coauthors, 2012: Design and quantification of an extreme winter storm scenario for emergency preparedness and planning exercises in California. Nat. Hazards, 60, 1085–1111, doi:10.1007/s11069-011-9894-5.
Dilley, A. C., and O’Brien D. M. , 1998: Estimating downward clear sky long-wave irradiance at the surface from screen temperature and precipitable water. Quart. J. Roy. Meteor. Soc., 124, 1391–1401.
Dozier, J., and Frew J. , 1990: Rapid calculation of terrain parameters for radiation modeling from digital elevation data. IEEE Trans. Geosci. Electron., 28, 963–969.
Dudhia, J., 1989: Numerical study of convection observed during the winter monsoon experiment using a mesoscale two-dimensional model. J. Atmos. Sci., 46, 3077–3107.
Edwards, J. M., and Slingo A. , 1996: Studies with a flexible new radiation code. I: Choosing a configuration for a large-scale model. Quart. J. Roy. Meteor. Soc., 122, 689–719, doi:10.1256/smsqj.53106.
Flerchinger, G. N., Xaio W. , Marks D. , Sauer T. J. , and Yu Q. , 2009: Comparison of algorithms for incoming atmospheric long-wave radiation. Water Resour. Res., 45, W03423, doi:10.1029/2008WR007394.
Franklin, A. I., 1983: Climate of the Priest River Experimental Forest, northern Idaho. General Tech. Rep. INT-159, USDA Intermountain Forest Range and Experimental Station, Ogden, UT, 58 pp. [Available online at http://www.fs.fed.us/rm/pubs_int/int_gtr159.pdf.]
Fry, J., and Coauthors, 2011: Completion of the 2006 National Land Cover Database for the Conterminous United States. Photogramm. Eng. Remote Sens., 77, 858–864.
Hamlet, A. F., and Lettenmaier D. P. , 1999: Effects of climate change on hydrology and water resources in the Columbia River basin. J. Amer. Water Resour. Assoc., 35, 1597–1623.
Henn, B., Lundquist J. D. , and Fisher A. , 2013: An evaluation of the effectiveness of methods of filling gaps in near-surface air temperature data. J. Hydrometeor., in press.
Hong, S.-Y., Noh Y. , and Dudhia J. , 2006: A new vertical diffusion package with an explicit treatment of entrainment processes. Mon. Wea. Rev., 134, 2318–2341.
Hungerford, R., Nemani R. R. , Running S. W. , and Coughlan J. C. , 1989: MTCLIM: A mountain microclimate simulation model. Res. Paper INT-414, USDA Forest Service Intermountain Research Station, Ogden, UT, 56 pp. [Available online at http://www.fs.fed.us/rm/pubs_int/int_rp414.pdf.]
Huwald, H., Higgins C. W. , Boldi M.-O. , Bou-Zeid E. , Lehning M. , and Parlange M. B. , 2009: Albedo effect on radiative errors in air temperature measurements. Water Resour. Res., 45, W08431, doi:10.1029/2008WR007600.
Kain, J. S., 2004: The Kain–Fritsch convective parameterization: An update. J. Appl. Meteor., 43, 170–181.
Kimball, B. A., Idso S. B. , and Aase J. K. , 1982: A model of thermal radiation from partly cloudy and overcast skies. Water Resour. Res., 18, 931–936.
Kimball, B. A., Running S. W. , and Nemani R. , 1997: An improved method for estimating surface humidity from daily minimum temperature. Agric. For. Meteor., 85, 87–98.
Kuraś, P. K., Alila Y. , Weiler M. , Spittlehouse D. , and Winkler R. , 2011: Internal catchment process simulation in a snow-dominated basin: Performance evaluation with spatiotemporally variable runoff generation and groundwater dynamics. Hydrol. Processes, 25, 3187–3203.
Laramie, R. L., and Schaake D. J. , 1972: Simulation of the Continuous Snowmelt Process. Massachusetts Institute of Technology, Cambridge, MA, 167 pp.
Leung, L., Wigmosta M. S. , and Ghan S. , 1996: Application of a subgrid orographic precipitation/surface hydrology scheme to a mountain watershed. J. Geophys. Res., 101, 12 803–812, 817.
Lundquist, J. D., and Huggett B. , 2008: Evergreen trees as inexpensive radiation shields for temperature sensors. Water Resour. Res., 44, W00D04, doi:10.1029/2008WR006979.
Lundquist, J. D., Cayan D. R. , and Dettinger M. D. , 2003: Meteorology and hydrology in Yosemite National Park: A sensor network application. Information Processing in Sensor Networks: Second International Workshop, F. Zhao and L. Guibas, Eds., Lecture Notes in Computer Science, Vol. 2634, Springer-Verlag, 518–528. [Available online at http://tenaya.ucsd.edu/~jessica/Lundquist_sensornetwork.pdf.]
Lundquist, J. D., Huggett B. , Roop H. , and Low N. , 2009: Use of spatially distributed stream stage recorders to augment rain gages by identifying locations of thunderstorm precipitation and distinguishing rain from snow. Water Resour. Res., 45, W00D25, doi:10.1029/2008WR006995.
Male, D. H., and Granger R. J. , 1981: Snow surface energy exchange. Water Resour. Res., 17, 609–627.
Malek, E., 2008: The daily and annual effects of dew, frost, and snow on a non-ventilated net radiometer. J. Atmos. Res., 89, 243–251, doi:10.1016/j.atmosres.2008.02.006.
Marks, D., and Dozier J. , 1992: Climate and energy exchange at the snow surface in the alpine region of the Sierra Nevada: 2. Snow cover energy balance. Water Resour. Res., 28, 3043–3054.
Marks, D., Kimball J. , and Tingey D. , 1998: The sensitivity of snowmelt processes to climate conditions and forest cover during rain-on-snow: A case study of the 1996 Pacific Northwest flood. Hydrol. Processes, 12, 1569–1587.
Marty, C., Philipona R. , Fröhlich C. , and Ohmura A. , 2002: Altitude dependence of surface radiation fluxes and cloud forcing in the alps: results from the alpine surface radiation budget network. Theor. Appl. Climatol., 72, 137–155.
Matsui, N., and Coauthors, 2012: Evaluation of Arctic broadband surface radiation measurements. Atmos. Meas. Tech., 5, 429–438, doi:10.5194/amt-5-429-2012.
Meehl, G., and Coauthors, 2007: Global climate projections. Climate Change 2007: The Physical Science Basis, Cambridge University Press, 747–845.
Meek, D. W., and Hatfield J. L. , 1994: Data quality checking for single station meteorological databases. Agric. For. Meteor., 69, 85–109.
Mesinger, F., DiMego G. , and Kalnay E. , 2006: North American Regional Reanalysis. Bull. Amer. Meteor. Soc., 87, 343–360.
Minder, J. R., Mote P. W. , and Lundquist J. D. , 2010: Surface temperature lapse rates over complex terrain: Lessons from the Cascade Mountains. J. Geophys. Res., 115, D14122, doi:10.1029/2009JD013493.
Minder, J. R., Durran D. R. , and Roe G. H. , 2011: Mesoscale controls on the mountainside snow line. J. Atmos. Sci., 68, 2107–2127, doi:10.1175/JAS-D-10-05006.1.
Mizukami, N., and Smith M. B. , 2012: Analysis of inconsistencies in multi-year gridded quantitative precipitation estimate over complex terrain and its impact on hydrologic modeling. J. Hydrol., 428–429, 129–141, doi:10.1016/j.jhydrol.2012.01.030.
Mlawer, E. J., Taubman S. J. , Brown P. D. , Iacono M. J. , and Clough S. A. , 1997: Radiative transfer for inhomogeneous atmosphere: RRTM, a validated correlated-k model for the long-wave. J. Geophys. Res., 102, 16 663–16 682.
Morrison, H., Thompson G. , and Tatarskii V. , 2009: Impact of cloud microphysics on the development of trailing stratiform precipitation in a simulated squall line: Comparison of one- and two-moment schemes. Mon. Wea. Rev., 137, 991–1007.
Mote, P. W., Hamlet A. F. , Clark M. P. , and Lettenmaier D. P. , 2005: Declining mountain snowpack in western North America. Bull. Amer. Meteor. Soc., 86, 39–49.
Murray, F. W., 1967: On the computation of saturation vapor pressure. J. Appl. Meteor., 6, 203–204.
Nijssen, B., Lettenmaier D. P. , Liang X. , Wetzel S. W. , and Wood E. F. , 1997: Streamflow simulation for continental-scale river basins. Water Resour. Res., 33, 711–724 , doi:10.1029/96WR03517.
Pomeroy, J., Marks D. , Link T. , and Ellis C. , 2009: The impact of coniferous forest temperature on incoming longwave radiation to melting snow. Hydrol. Processes, 23, 2513–2525.
Prata, A. J., 1996: A new long-wave formula for estimating downward clear-sky radiation at the surface. Quart. J. Roy. Meteor. Soc., 122, 1127–1151, doi:10.1002/qj.49712253306.
Ralph, F. M., and Coauthors, 2005: Improving short-term (0–48 h) cool-season quantitative precipitation forecasting: Recommendations from a USWRP workshop. Bull. Amer. Meteor. Soc., 86, 1619–1632.
Rasmussen, R., and Coauthors, 2012: How well are we measuring snow: The NOAA/FAA/NCAR winter precipitation test bed. Bull. Amer. Meteor. Soc., 93, 811–829.
Reed, S., and Coauthors, 2004: Overall distributed model intercomparison project results. J. Hydrol., 298, 27–60, doi:10.1016/j.jhydrol.2004.03.031.
Ruiz-Arias, J. A., Pozo-Vázquez D. , Lara-Fanego V. , Santos-Alamillos F. J. , and Tovar-Pescador J. , 2011: A high-resolution topographic correction method for clear-sky solar irradiance derived with a numerical weather prediction model. J. Appl. Meteor. Climatol., 50, 2460–2472.
Running, S. W., Nemani R. R. , and Hungerford R. D. , 1987: Extrapolation of synoptic meteorological data in mountainous terrain and its use for simulating forest evapotranspiration and photosynthesis. Can. J. For. Res., 17, 472–483, doi:10.1139/x87-081.
Satterlund, D. R., 1979: An improved equation for estimating long-wave radiation from the atmosphere. Water Resour. Res., 15, 1649–1650.
Sevruk, B., 1983: Correction of measured precipitation in the Alps using the water equivalent of new snow. Weather, 14, 49–58.
Shamir, E., and Georgakakos K. P. , 2006: Distributed snow accumulation and ablation modeling in the American River basin. Adv. Water Resour., 29, 558–570, doi:10.1016/j.advwatres.2005.06.010.
Sicart, J., Pomeroy J. , and Essery R. , 2006: Incoming longwave radiation to melting snow: observations, sensitivity and estimation in northern environments. Hydrol. Processes, 20, 3697–3708.
Skamarock, W., and Klemp J. , 2008: A time-split nonhydrostatic atmospheric model for weather research and forecasting applications. J. Comput. Phys., 227, 3465–3485, doi:10.1016/j.jcp.2007.01.037.
Soil Survey Staff, cited 2006: U.S. General Soil Map (STATSGO2) for California. Natural Resources Conservation Service, United States Department of Agriculture. [Available online at http://soils.usda.gov/survey/geography/ssurgo/description_statsgo2.html.]
Storck, P., 2000: Trees, snow and flooding: An investigation of forest canopy effects on snow accumulation and melt at the plot and watershed scales in the Pacific Northwest. Water Resources Series Tech. Rep. 161, University of Washington, Seattle, WA, 198 pp. [Available online at http://ir.library.oregonstate.edu/xmlui/bitstream/handle/1957/5136/PB2001104138.pdf?sequence=1.]
Surfleet, C., Skaugset A. E. III, and Meadows M. W. , 2011: Road runoff and sediment sampling for determining road sediment yield at the watershed scale. Can. J. For. Res., 1980, 1970–1980, doi:10.1139/x11-104.
Susong, D., Marks D. , and Garen D. , 1999: Methods for developing time-series climate surfaces to drive topographically distributed energy- and water-balance models. Hydrol. Processes, 13, 2003–2021.
Tapash, D., Pierce D. W. , Cayan D. R. , Vano J. a. , and Lettenmaier D. P. , 2011: The importance of warm season warming to western U.S. streamflow changes. Geophys. Res. Lett., 38, L23403, doi:10.1029/2011GL049660.
Thornton, P. E., and Running S. W. , 1999: An improved algorithm for estimating incident daily solar radiation from measurements of temperature, humidity, and precipitation. Agric. For. Meteor., 93, 211–228.
Tobin, C., Nicotina L. , Parlange M. B. , Berne A. , and Rinaldo A. , 2011: Improved interpolation of meteorological forcings for hydrologic applications in a Swiss Alpine region. J. Hydrol., 401, 77–89, doi:10.1016/j.jhydrol.2011.02.010.
USACE, 1956: Snow hydrology summary report of the snow investigations. Tech. Rep., U.S. Army Corps of Engineers, North Pacific Division, Portland, OR, 437 pp.
von Storch, H., and Zwiers F. W. , 1999: Statistical Analysis in Climate Research. Cambridge University Press, 496 pp.
Waichler, S. R., and Wigmosta M. S. , 2003: Development of hourly meteorological values from daily data and significance to hydrological modeling at H. J. Andrews Experimental Forest. J. Hydrometeor., 4, 251–263.
Weingartner, R., and Pearson C. , 2001: A comparison of the hydrology of the Swiss Alps and the Southern Alps of New Zealand. Mt. Res. Dev., 21, 370–381.
Westrick, K., 2001: An evaluation of a high-resolution hydrometeorological modeling system for prediction of a cool-season flood event in a coastal mountainous watershed. J. Hydrometeor., 2, 161–180.
Westrick, K., and Storck P. , 2002: Description and evaluation of a hydrometeorological forecast system for mountainous watersheds. Wea. Forecasting, 17, 250–262.
Whitaker, A., Alila Y. , Beckers J. , and Toews D. , 2003: Application of the distributed hydrology soil vegetation model to Redfish Creek, British Columbia: Model evaluation using internal catchment data. Hydrol. Processes, 17, 199–224, doi:10.1002/hyp.1119.
Wigmosta, M. S., Vail L. W. , and Lettenmaier D. P. , 1994: A distributed hydrology-vegetation model for complex terrain. Water Resour. Res., 30, 1665–1679.
Zhao, Q., Liu Z. , Ye B. , Qin Y. , Wei Z. , and Fang S. , 2009: A snowmelt runoff forecasting model coupling WRF and DHSVM. Hydrol. Earth Syst. Sci., 13, 1897–1906, doi:10.5194/hess-13-1897-2009.