1. Introduction
The interior western United States (IWUS)1 is mostly arid, but is also home to the headwaters of several major river systems, for example, the Colorado, Missouri, and Snake Rivers (Woodhouse 2004). There is much interest in quantitative precipitation estimation (QPE) in this region, in a range of disciplines including meteorology, hydrology, agriculture, forestry, and ecology (Mote et al. 2005; Bales et al. 2006; Ebert et al. 2007; Barnett et al. 2008; Rasmussen et al. 2011). Most of the precipitation over the IWUS falls as snow over its mountains (Daly et al. 1994). QPE is especially challenging in complex terrain (e.g., Liu et al. 2011), and gauge-based snowfall rate estimation is more uncertain than rain rate estimation (Rasmussen et al. 2012). In the mountainous IWUS, the Snowpack Telemetry (SNOTEL) network, operated by the Natural Resources Conservation Service (NRCS), has been used as a reference in many studies to evaluate model output (e.g., Liu et al. 2011; Gutmann et al. 2012) and serves as the basis for gridded precipitation estimates over mountains. However, SNOTEL only provides point measurements of precipitation, and the gauge density is low compared to that in highly populated or agricultural regions. Several different techniques have been developed to provide more complete precipitation distribution maps, including “terrain aware” interpolation techniques using gauge measurements as forcing data (e.g., Daly et al. 1994), space-based and ground-based remote sensing retrievals (e.g., Lin and Mitchell 2005), and numerical model simulations (Liu et al. 2017). But the relative performances of different techniques in QPE are not well understood, so a cross validation of different precipitation datasets is necessary.
The terrain-aware interpolation techniques have been widely used to study the precipitation distribution over IWUS (e.g., Daly et al. 1994; Thornton et al. 1997; Xia et al. 2012; Newman et al. 2015a). Daly et al. (1994) developed the Parameter-Elevation Regressions on Independent Slopes Model (PRISM) to produce a terrain-sensitive gridded precipitation dataset, which remains widely used. This model estimates the precipitation in areas without gauges using physically informed statistical relations between terrain and gauge precipitation. Several other gauge-driven datasets have been developed using different statistical downscaling techniques, including Daymet (Thornton et al. 1997), North American Land Data Assimilation System Stage II (NLDAS II; Xia et al. 2012), and the continental United States ensemble gridded datasets (CUSEG; Newman et al. 2015a).
The uncertainties of the gauge-driven gridded datasets have been discussed in several previous studies (e.g., Daly et al. 2008; Gutmann et al. 2012). For example, Daly et al. (2008) tried to estimate the error of PRISM using a data denial cross-validation method. In this method, the PRISM regression with a single gauge removed is compared to that with all gauges in the vicinity of the gauge site. They show that the resulting difference in annual precipitation estimates is 20%–30% over mountains. Daly et al. (2008) also shows PRISM precipitation has a larger uncertainty in winter than in summer. Gutmann et al. (2012) compared the winter precipitation estimates from PRISM against a SNOTEL site in the eastern San Juan Mountains at Moon Pass. The SNOTEL gauge was installed in October 2008 and was not applied in the PRISM used in their study. The comparison indicates PRISM significantly overestimates the winter precipitation (600 mm) compared to SNOTEL (232 mm) at that point. Recently, Henn et al. (2017) compared the precipitation trend estimates from gauge-driven gridded datasets against the streamflow observations in Sierra Nevada and suggests the gauge-driven gridded datasets may have substantial uncertainty at high elevations.
Other than gauge-driven gridded datasets, space-based and ground-based remote sensing techniques have been developed to study the precipitation climatology (e.g., Lin and Mitchell 2005; Hou et al. 2014). Datasets developed using ground-based scanning weather radars, such as the National Centers for Environmental Prediction hourly multisensor precipitation analysis Stage IV (NCEP IV) dataset (Lin and Mitchell 2005), are quite suitable to study the precipitation distribution at high spatial resolution (4 km) over relatively flat terrain, such as in the central and eastern United States. However, the operational weather radar network is challenged over the complex terrain environment of the western United States because of blockage by the first range of mountains and the inability to capture the low-level orographic precipitation growth zone if the lowest unblocked beam is high above the surface, which is common (e.g., Fulton et al. 1998; Maddox et al. 2002; Lin and Mitchell 2005; Lin and Hou 2012; Smalley et al. 2014). A radar network much denser than that currently in operation in the IWUS would be needed to achieve NCEP IV precipitation accuracies comparable to those over the central/eastern United States. Space-based remote sensing has the advantage of vertical incidence and thus no blockage issues. Precipitation estimates using passive remote sensing techniques cannot reveal the variable precipitation distribution over complex terrain (Ebert et al. 2007). Active space-based radar measurements, such as Global Precipitation Measurement (GPM, from 2014 to present; Hou et al. 2014), are inadequate to study the finescale quantitative precipitation climatology now because of lack of overpasses, but they could be useful in the future.
Recently, numerical weather prediction (NWP) models with high resolution (<6 km) have been used to study the precipitation climatology over complex terrain. Ikeda et al. (2010) showed that the Weather Research and Forecasting (WRF) Model at a grid spacing smaller than 6 km well captures the seasonal snowfall in the Colorado Rockies; the difference of cold-season precipitation between the model output and SNOTEL is within 20% for 71% of the SNOTEL sites. Liu et al. (2011) pointed out that this performance is highly sensitive to the choice of cloud microphysics parameterization. Rasmussen et al. (2011, 2014) further confirmed that WRF, at a resolution of 4 km, captures the cold-season precipitation distribution and amount over the Colorado headwaters region well, with a bias of 10%–15% compared to SNOTEL measurements. Because of the good performance in simulating orographic precipitation over complex terrain, high-resolution WRF simulations have been used to assess changes of orographic precipitation in a changing global climate. For instance, Rasmussen et al. (2011, 2014) analyzed the hydrological cycle in the Colorado headwaters region using high-resolution WRF simulations and explored its sensitivity to climate change using a pseudo–global warming approach. Liu et al. (2017) recently extended the work to the continental United States (hereafter CONUS WRF).
Gridded precipitation datasets developed using different techniques have been widely used for various purposes because of their completeness (Lundquist et al. 2015), but their accuracy is not well known. Biases in the gridded precipitation datasets have impacts on the evaluation of hydrologic and climate model simulations and would influence the accuracy of model simulations if they are used as forcing data. Some previous studies have compared different precipitation datasets (e.g., Ebert et al. 2007), but few have specially focused on the mountainous IWUS, and even fewer used high-resolution datasets (Henn et al. 2017). This study aims to cross validate high-resolution precipitation datasets currently available in the IWUS and a 10-yr continuous convection-permitting WRF simulation at 4-km horizontal resolution (hereafter IWUS WRF). The focus is on orographic precipitation across seasons. The model output is compared with gauge data (SNOTEL), a radar-based dataset (NCEP IV), and four gauge-driven gridded datasets (PRISM, Daymet, NLDAS II, CUSEG). IWUS WRF is also compared with CONUS WRF to examine the impact of different driven datasets and boundaries on simulations. The paper is organized as follows. Section 2 describes the WRF Model configuration and introduces the observational datasets. The model output and datasets are compared and statistically analyzed in section 3. Discussion and conclusions are given in sections 4 and 5, respectively.
2. Description of the WRF configuration and observational datasets
a. WRF configuration
A 10-yr continuous simulation is conducted using the WRF Model version 3.7.1 to study the precipitation and snowpack in the IWUS. The Climate Forecast System Reanalysis (CFSR; Saha et al. 2010) is used to provide the initial and lateral boundary conditions. The non-nested model domain, shown in Fig. 1a, has 420 × 410 grid points at 4-km horizontal resolution and 51 vertical levels, with a high layer density close to the ground. The simulation runs from October 2001 to February 2012, and the output from March 2002 to February 2012 is used to study the precipitation patterns in the IWUS. This is different from CONUS WRF, which covers the continental United States and runs from October 2000 to September 2013 using ERA-Interim reanalysis data as input. Comparison between IWUS WRF and CONUS WRF will be made to examine the impact of different input datasets and boundaries. The selection of physical schemes follows CONUS WRF: the Rapid Radiative Transfer Model for General Circulation Models (RRTMG) shortwave and longwave radiation scheme (Iacono et al. 2008), the revised Monin–Obukhov surface layer scheme (Jimenez et al. 2012), the Noah-MP land surface scheme (Niu et al. 2011 and Yang et al. 2011), the Yonsei University (YSU) planetary boundary layer (PBL) scheme (Hong and Pan 1996), and the Thompson cloud microphysics scheme (Thompson et al. 2004). Convection is not parameterized. A 30-yr (1981–2011) retrospective simulation is being conducted based on the same configuration, as a basis to study changes in orographic precipitation in a changing climate (Wang et al. 2017, manuscript submitted to Int. J. Climatol.).
Topography of (a) the WRF Model domain and (b) the study domain. In (b), the three red boxes delineate subdomains referred to as the Greater Yellowstone (upper), Utah (lower left), and Colorado (lower right). The dark gray circles indicate the locations of SNOTEL sites and the gray line is the 2-km height MSL contour. Several mountain ranges are marked in (b). GV, WRR, and SRR are abbreviations for Gros Ventre, Wind River Range and Salt River Range, respectively.
Citation: Journal of Hydrometeorology 18, 9; 10.1175/JHM-D-17-0056.1
This study focuses on complex terrain in a subregion of the model domain, here called the “study domain,” shown in Fig. 1a. We will distinguish three main mountainous regions within this study domain (Fig. 1b): the “Greater Yellowstone” region, the central Utah range (called “Utah”), and the central Rocky Mountain region (called “Colorado”). Most of the precipitation in IWUS falls in areas 2 km above mean sea level (MSL), contoured in Fig. 1b. The SNOTEL sites shown in Fig. 1b are those in operation between March 2002 and February 2012. Data from SNOTEL sites installed after March 2002 are not used in this study.
b. Datasets description
Other than the WRF simulation, we compare six precipitation datasets in this study: SNOTEL, PRISM, Daymet, NLDAS II, CUSEG, and NCEP IV. Basic information for the six datasets is summarized in Table 1.
Information about WRF and the precipitation datasets used in this study.
SNOTEL (NRCS 1977) is an important precipitation network across the IWUS mountains (Doesken and Schaefer 1987; Serreze et al. 1999). A standard sensor configuration includes a pressure-sensing snow pillow (to estimate snow water equivalent), a storage precipitation gauge, and an air temperature sensor. The daily difference in snow water equivalent can be used as a measure of daily snowfall accumulation. We do not use those data in this study because we are interested in all precipitation (not just snow) and because of uncertainties related to snow drift and sublimation. Here we use daily precipitation estimates from the storage gauges. These gauges have a wide orifice and a wind shield to maximize catch efficiency. The SNOTEL sites typically are in small openings in forests to minimize snow drifts and thus to optimize the estimation of the seasonal evolution of snow water equivalent. As a result, the SNOTEL site density is not uniform with elevation across a mountain range (Fig. 1b): there are almost no SNOTEL sites above the tree line, which represents large mountain areas in the Wind River Range or San Juan Mountains, for instance. On the other hand, SNOTEL sites are found only near mountain tops in drier, lower-elevation regions, for example, in Utah west of the Wasatch Range (Fig. 1b). The SNOTEL sensors record data every 15 min; hourly and daily data are archived. SNOTEL has been widely used as the “truth” to evaluate other precipitation datasets and model simulations (e.g., Daly et al. 2008; Rasmussen et al. 2011; Liu et al. 2011). In this study, we also use daily SNOTEL gauge measurements as the reference to evaluate the WRF simulation and the various gridded datasets.
The PRISM dataset (PRISM Climate Group 2016) is based on a statistical model developed to interpolate climate elements to finescale complex terrain. We use the publically available monthly data at 4-km resolution (Table 1), although higher-resolution PRISM data are available commercially. Basically, PRISM uses a precipitation-elevation regression and takes into account other factors such as terrain slope, coastal proximity, and topographic facet orientation to estimate the precipitation at each digital elevation model grid point (Daly et al. 1994, 2002, 2008). Precipitation data are based on various gauge networks. The National Weather Service Cooperative Observer Program (COOP) gauge network (Daly et al. 2007) and SNOTEL are the two main sources of precipitation data used in PRISM. The density of the COOP station network relates to population density and farming intensity. Towns and agricultural areas occupy the lower-elevation plains and valleys in the IWUS. The SNOTEL data largely control PRISM precipitation in the mountains. The PRISM precipitation climatology in the United States has been used as a reference in many studies (e.g., Liu et al. 2017). Uncertainty of PRISM precipitation can be attributed to the quality of gauge data, especially for snow, the precipitation-elevation regression, the density of gauge stations, the complexity of terrain, and assumptions made in the model (Daly et al. 2008).
Daymet (Thornton et al. 2014), NLDAS II (NCEP 2014), and CUSEG (Newman et al. 2015b) are all gauge-driven gridded datasets, similar to PRISM, but they use different statistical interpolation techniques. In Daymet, a terrain-dependent spatial convolution of a truncated Gaussian weighting filter is used to interpolate gauge data to 1-km resolution grids (Thornton et al. 1997). In NLDAS II, the Climate Prediction Center (CPC) gauge-based daily precipitation dataset at 0.125° resolution is used as the initial input. PRISM is used to adjust the magnitude, and NCEP Stage II data are then used to temporally disaggregate the dataset to hourly scale (Xia et al. 2012). CUSEG is a 100-member ensemble daily precipitation dataset. A probabilistic interpolation is used to interpolate gauge data to 0.125° resolution grids. Terrain impacts (e.g., elevation and slope) are included (Newman et al. 2015a). The ensemble approach accounts for the probability density function (PDF) of uncertainties due to spatial undersampling and subsequent extrapolation, projecting irregular measurements onto a grid, and also implicitly accounts for random measurement errors based on the mean monthly precipitation and PDF of daily precipitation estimates from gauges (Clark and Slater 2006).
The NCEP IV dataset (NCEP 2001) uses merged surface radar and rain gauge products to produce hourly precipitation at 4-km resolution (Lin and Mitchell 2005). The finescale distribution of precipitation is controlled by the hourly radar rainfall products from about 140 Weather Surveillance Radar-1988 Doppler (WSR-88D) radars, without direct adjustment for terrain (Lin and Mitchell 2005). The radar-estimated precipitation is then merged with about 3000 hourly gauge reports using the methodology developed by Seo (1998). This dataset has been widely used to evaluate other precipitation products using remote sensing techniques (e.g., Gourley et al. 2010; Tesfagiorgis et al. 2011; Lin and Hou 2012).
3. Results
a. Spatial precipitation distribution
The annual mean precipitation maps over the IWUS, as estimated by SNOTEL, WRF, PRISM, Daymet, NLDAS II, CUSEG, and NCEP IV over the decade from March 2002 to February 2012, are shown in Fig. 2. The terrain is gray-shaded in Fig. 2 to highlight the mountain ranges and to emphasize the orographic nature of precipitation in the IWUS. Not surprisingly, PRISM, Daymet, NLDAS II, and CUSEG appear to be quite consistent with SNOTEL (Figs. 2c–f). Relatively high-resolution datasets (PRISM and Daymet) show finer texture of terrain-relative precipitation distribution than coarser-resolution datasets (NLDAS II and CUSEG) and can resolve orographic precipitation over relatively small mountains (e.g., the Salt River, Wyoming, and La Sal Ranges; see Fig. 1b for the location of these mountain ranges).
Mean annual precipitation maps from March 2002 to February 2012 estimated by (a) SNOTEL, (b) WRF, (c) PRISM, (d) Daymet, (e) NLDAS II, (f) CUSEG, and (g) NCEP IV. The hatched areas depict regions where the absolute difference between WRF and PRISM exceeds the absolute mean difference between PRISM and any other gauge-driven dataset (Daymet, NLDAS II, and the 100 CUSEG datasets; 103 in total). The stars in (g) show the location of WSR-88D radars, and the circles show the 100-km range ring.
Citation: Journal of Hydrometeorology 18, 9; 10.1175/JHM-D-17-0056.1
The WRF precipitation distribution compares well against SNOTEL (Figs. 2a,b) and shows similar patterns against PRISM, Daymet, NLDAS II, and CUSEG. But WRF appears to underestimate the annual precipitation over the Wasatch, which is relatively narrow and steep. WRF estimates more precipitation than the other datasets over the Wind River Range and some plains areas, especially over the high plains of eastern Wyoming; in fact, the |WRF − SNOTEL| difference in this region tends to exceed the mean absolute difference between gauge-driven gridded datasets (as shown by the hatches in Fig. 2b). This bias is most pronounced in spring and summer. Overall, WRF estimates 55% of the annual precipitation falls on mountain areas or high plains above 2000 m, and 27% falls on mountain ranges above 2500 m.
NCEP IV underestimates the annual precipitation in most of the areas, especially over the mountains, although it does capture the general pattern of orographic precipitation (Fig. 2g). NCEP IV orographic precipitation comes closest to other datasets within ~100 km range of WSR-88D radars, for example, in the Wasatch Range, but is a gross underestimation over remote plains and radar-blocked mountains, for example, the Bighorn or the La Sal Ranges.
The annual-mean data in Fig. 2 are partitioned in four seasons in Fig. 3, since precipitation processes are highly dependent on the time of year. Among the four gauge-driven gridded datasets, only PRISM is shown in this figure; Daymet, NLDAS II, and CUSEG provide similar patterns compared to PRISM. The mountains west of the continental divide receive most of their precipitation in winter and experience dry summers, while the high plains are wetter in spring and summer than in fall and winter. During all seasons PRISM, Daymet, NLDAS II, and CUSEG are quite consistent with SNOTEL over mountain areas, which is no surprise. In areas without gauges, their accuracy is unknown. The WSR-88D network can hardly detect shallow orographic storms (Smalley et al. 2014), which prevail in winter (Fig. 3d4). Thus, NCEP IV seriously underestimates winter precipitation over mountains remote from any radar sites, such as the Bighorn and Park Ranges (Fig. 3d4). During summer, most precipitation is generated through deep convection, which is detected over a greater range by radars (Fig. 3d2). In the high plains as well as in the adjacent mountains in Wyoming and Colorado, WRF appears to estimate more spring and summer precipitation than SNOTEL, mainly in spring (MAM) over the high plains and in summer (JJA) over the mountains. This may relate to the inability of the model with a resolution of 4 km to accurately capture convective development and upscale growth (Rasmussen et al. 2014). In addition, our WRF simulation uses the same configuration in all seasons, which may not be optimal (e.g., Ebert et al. 2007).
Mean seasonal precipitation maps estimated by (a1)–(a4) SNOTEL, (b1)–(b4) WRF, (c1)–(c4) PRISM, and (d1)–(d4) NCEP IV. Panels from left to right are precipitation maps in spring (MAM), summer (JJA), fall (SON), and winter (DJF). Hatched areas are defined as in Fig. 2.
Citation: Journal of Hydrometeorology 18, 9; 10.1175/JHM-D-17-0056.1
b. Evaluation of WRF simulation
The good performance of high-resolution WRF in simulating orographic precipitation, especially wintertime precipitation, is evident in previous studies (section 1). In this section, we provide an additional evaluation of the WRF simulation to discuss the terrain-related differences between WRF and SNOTEL. In addition, a comparison between IWUS WRF and CONUS WRF examines the impact of different driven datasets. Moreover, a comparison between simulations at 4- and 1.33-km resolutions is made to examine the impact of boundaries (i.e., domain size) and resolutions.
Scatterplots of seasonal precipitation estimated by WRF against SNOTEL in the entire study domain are shown in Fig. 4. The data are distributed well on both sides of the 1:1 lines during spring and winter, whereas in summer WRF overestimates the precipitation by 30 mm and in fall WRF underestimates the precipitation by 32 mm on average. The WRF root-mean-square bias (RMSB) is larger than 55 mm in all the seasons. The small mean biases and large RMSBs in spring and winter reflect both uncertainties in gauge measurement and small-scale terrain effects that are unresolved in WRF.
Scatterplots of seasonal precipitation estimated by WRF against SNOTEL at all SNOTEL sites in MAM, JJA, SON, and DJF. The black dotted lines are the 1:1 lines. The correlation coefficient, RMSB, and mean bias are shown in each panel.
Citation: Journal of Hydrometeorology 18, 9; 10.1175/JHM-D-17-0056.1
To better understand the difference of winter precipitation between SNOTEL and WRF, we map out the mean bias and correlation coefficient between WRF and SNOTEL for the winter months (Figs. 5a,b). The background color field in Fig. 5a represents the WRF-resolved terrain slope in the direction of the prevailing wind. This is a weighted-average surface wind, the weighting being precipitation rate, using hourly data over 10 winters. The slope for each grid point is calculated as an interpolated height difference over an along-wind distance corresponding to twice the grid spacing. A positive (negative) slope is referred to as the windward (lee) side. For 96% of the SNOTEL sites, the mean monthly winter precipitation bias is within 40 mm month−1, and the correlation coefficient is greater than 0.8, but there are some outliers. Many negative biases are observed on the windward side while positive biases are often observed on the lee side. The negative biases on the windward side may be because a 4-km WRF resolution is not fine enough to resolve some steep windward slopes, for example, in the Wasatch Range, while the positive biases in the lee may be due to SNOTEL snowfall underestimates in the presence of strong surface winds (Rasmussen et al. 2012), which are common in the lee foothills. Figure 5f shows the WRF − SNOTEL bias of monthly precipitation in winter as a function of wind speed; this is the WRF-derived prevailing surface wind during precipitation events. The figure suggests the mean difference between WRF and SNOTEL is very small. For strong winds (>20 m s−1), WRF estimates slightly more snowfall than SNOTEL, and strong winds are more common at lee gauges than at upwind gauges. This may be because SNOTEL underestimates the snowfall due to strong surface wind (Rasmussen et al. 2012), but more studies are needed in the future to better understand the uncertainty of snow gauges and WRF.
(a) Mean monthly precipitation difference and (b) correlation coefficients in DJF between WRF and SNOTEL. The background in (a) is the terrain slope in the direction of prevailing wind during precipitation events. (c) Monthly winter precipitation difference (WRF − SNOTEL) and (d) correlation coefficients between WRF and SNOTEL as a function of elevation. Gray dots in (c) represent the monthly precipitation difference (WRF − SNOTEL) at all SNOTEL sites in DJF from March 2002 to February 2012, and gray dots in (d) represent the correlation coefficients of monthly precipitation in winter from March 2002 to February 2012 between WRF and SNOTEL. The black dots are the mean values at different elevation intervals, and the black bars indicate ±1 standard deviation. (e),(f) As in (c), but for the WRF − SNOTEL difference as functions of elevation bias and prevailing wind speed. Brown and green dots in (f) represent the upwind and lee SNOTEL gauges, respectively.
Citation: Journal of Hydrometeorology 18, 9; 10.1175/JHM-D-17-0056.1
There is no clear pattern either in the bias or in the correlation coefficient; neither parameter relates with terrain elevation (Figs. 5c,d). The mean bias is close to 0, and the standard deviation is smaller than 40 mm, irrespective of elevation, but the precipitation bias (WRF − SNOTEL) is correlated to the height difference between WRF (model terrain height) and SNOTEL (actual height) (Fig. 5e). This reflects a resolution-related weakness in the model: at a resolution of 4 km, WRF underestimates (overestimates) precipitation at SNOTEL sites located on a local mountain (in a local valley or terrain concavity). The monthly precipitation bias between WRF and SNOTEL (Figs. 5c,e) indicates WRF may have larger uncertainty than the mean bias at an individual SNOTEL site; this is probably because the same model configuration is used for the 10-yr simulation, which is not optimal.
A simulation with a resolution finer than 4 km may better resolve the precipitation over complex terrain, but the areal-mean difference across a mountain range is small. This has been shown before over the Colorado mountains (Ikeda et al. 2010) and is confirmed by a comparison of winter precipitation over Greater Yellowstone between simulations with 4- and 1.33-km resolutions (Fig. 6). The 1.33-km resolution simulation is driven by hourly 4-km resolution WRF outputs and is run from December 2007 to February 2008. At higher resolution (Fig. 6b), WRF resolves a finer texture of terrain-relative precipitation distribution than the 4-km simulation (Fig. 6a) and is slightly more consistent with SNOTEL in terms of correlation coefficient and RMSB (Table 2). The 1.33-km simulation estimates 0.77 mm (1.3%) more winter precipitation than the 4-km simulation. The correlation coefficient, RMSB, and root-mean-square percent bias (RMSPB) between the two runs are 0.99, 14.4 mm, and 8.7%, respectively. Over some mountains (e.g., Gros Ventre, Salt River Range), the difference locally is as large as 50 mm, which is still small compared to the difference between simulation and gauge-driven gridded datasets and the uncertainties of gauge-driven gridded datasets (shown later).
Maps of wintertime precipitation in Greater Yellowstone from December 2007 to February 2008 estimated by the IWUS WRF simulation at (a) 4- and (b) 1.33-km resolution, as well as (c) their difference and (d) percent difference.
Citation: Journal of Hydrometeorology 18, 9; 10.1175/JHM-D-17-0056.1
Correlation coefficient, mean bias, and RMSB between WRF simulations and SNOTEL in Greater Yellowstone.
The winter precipitation difference between IWUS WRF and PRISM (IWUS WRF − PRISM) is shown in Fig. 7a. Especially within the Greater Yellowstone area, WRF estimates more winter precipitation over some mountains (Wind River Range, Gros Ventre, Big Horn, and Absaroka) than PRISM, while for others, such as the Teton Range, WRF estimates less precipitation. In the Utah and Colorado subdomains, the agreement generally is better. WRF performance differences between nearby mountain ranges may be due to three types of uncertainties: 1) model related (uncertainties related to WRF physics and boundary conditions), 2) measurement related (often related to local vegetation and finescale terrain factors around the SNOTEL sites), and 3) PRISM algorithm related (uncertainties due to extrapolation in areas with low gauge density). The sensitivity study by Liu et al. (2011) shows that the microphysics scheme has the most significant impact on the simulations. The microphysics scheme used in this study, the Thompson scheme, captures the characteristics of winter orographic precipitation in IWUS best (Liu et al. 2011). Other physical schemes (e.g., land surface, PBL, radiation) have minor impacts on the model results (Liu et al. 2011). However, it is not known whether different driven datasets (boundary conditions) impact the results. Therefore, we compare the CONUS WRF against PRISM (Fig. 7b). CONUS WRF uses a different driver (ERA-Interim reanalysis data), a very different domain (exceeding the continental United States), and a different series of years (October 2000–September 2013) (Liu et al. 2017). Yet, the comparison with PRISM reveals a remarkably similar pattern of winter snowfall biases, suggesting that uncertainties of the second and third type (i.e., uncertainties in the gauge-driven gridded dataset) matter more, which points to errors in the measurements or the gridding of these measurements.
Difference of winter precipitation between (a) IWUS WRF and PRISM and (b) CONUS WRF and PRISM.
Citation: Journal of Hydrometeorology 18, 9; 10.1175/JHM-D-17-0056.1
c. Intercomparison of gauge-driven gridded precipitation datasets
Analyses in section 3a suggest in gauge-deprived areas, such as the Gros Ventre Range, the Elkhead Range, the eastern San Juan Mountains (all labeled in Fig. 1b), the Yellowstone National Park area (northwest corner of Wyoming), and the Great Salt Lake, the accuracy of the precipitation estimated by different gauge-driven gridded datasets is questionable given the lack of data. In this section, we examine the differences between different gauge-driven datasets and try to better understand their uncertainties.
The maximum absolute bias between any two of PRISM, Daymet, NLDAS II, and the 100 CUSEG datasets (103 datasets in total, giving a total of 5253 pairs) at any grid point in the study domain is shown in Fig. 8.2 Bilinear interpolation is used to project different datasets onto a common 4-km grid. The difference between different datasets generally is larger over the mountains than the plains in all seasons, not in a percent sense (not shown), but certainly in an absolute sense (Fig. 8). This likely is due to different statistical relations between precipitation and terrain used for the different datasets. In areas with lots of SNOTEL sites (e.g., the Front Range in Colorado), the difference between the datasets is relatively small. The differences may also be related to the different resolutions of gauge-driven gridded datasets, as shown by the relatively large uncertainty over the Wasatch Range, which is narrow and steep (Fig. 8d). The differences between gridded datasets are relatively small in summer (0–150 mm) and relatively large in winter and spring (0–250 mm; DJF and MAM), which is the wettest period over the mountains.
Seasonal precipitation estimation uncertainty evident from gauge-driven gridded datasets in MAM, JJA, SON, and DJF. The uncertainty is defined as the maximum absolute difference between any two of PRISM, Daymet, NLDAS II, and the 100 CUSEG datasets (103 in total) for each grid box. The blue (black) dots indicate the locations of SNOTEL (other) gauges.
Citation: Journal of Hydrometeorology 18, 9; 10.1175/JHM-D-17-0056.1
The relative performances of the four gauge-driven gridded datasets are assessed in Fig. 9, which compares gauge-driven gridded datasets (PRISM, Daymet, NLDAS II, or CUSEG) against their mean. The upper and lower panels show the summer and winter precipitation difference maps, respectively. All four datasets have a correlation coefficient higher than 0.9 with the mean, because the development of the datasets is strongly terrain related. PRISM is the most consistent with the mean for summer precipitation, and Daymet is the most consistent with the mean for winter precipitation. NLDAS II estimates the most summer precipitation and the least winter precipitation, while CUSEG estimates the least summer precipitation and the most winter precipitation. All the four gauge-driven gridded datasets have larger differences against the mean for winter precipitation than summer precipitation.
Difference of (a)–(d) summer and (e)–(h) winter precipitation between a gauge-driven dataset (PRISM, Daymet, NLDAS II, or CUSEG) and the mean of the four datasets. The correlation coefficient, mean bias, and RMSB (mm) of precipitation for grid points 2 km MSL are shown.
Citation: Journal of Hydrometeorology 18, 9; 10.1175/JHM-D-17-0056.1
In short, gauge-driven gridded datasets using statistical interpolation techniques perform best at SNOTEL sites, since they are anchored to these measurements, but their agreement deteriorates in areas where gauge sites are sparse. The accuracy of the gridded datasets at higher elevations is less certain, not just due to relative gauge sparsity in the mountains, but also due to SNOTEL data uncertainty related to the higher uncertainty of the measurement of snow versus rain, the higher fraction of snow versus rain at higher elevation, the higher uncertainty of the snowfall measurement under stronger wind, and the typically higher wind speed at higher elevations.
d. Comparison of winter precipitation between gridded datasets and WRF
Given these possible data-related uncertainties, we now assume WRF to be the reference to evaluate gridded observational datasets. Maps of the mean wintertime precipitation absolute and relative difference between gauge-driven gridded datasets and WRF are shown in the upper and lower panels of Fig. 10, respectively. The correlation coefficient r, RMSB, and RMSPB of precipitation at locations 2 km MSL are shown. Large differences are evident in some mountains, especially where SNOTEL sites are scarce (e.g., in the Wind River Range, above the tree line; Figs. 10a–d). Of all gauge-driven gridded datasets, PRISM and Daymet compare better against WRF in the study domain than NLDAS II and CUSEG. PRISM and Daymet significantly underestimate the precipitation over the Gros Ventre and Wind River Ranges, NLDAS II has the largest RMSB and has relatively larger uncertainties over Greater Yellowstone than Utah and Colorado, and CUSEG overestimates the precipitation over many of the mountains but underestimates the precipitation over some mountains (e.g., Gros Ventre and Wind River Ranges). Since the four datasets are all SNOTEL based, the pattern of the bias maps is generally consistent with the difference between WRF and SNOTEL (Fig. 5a).
DJF precipitation (a)–(d) bias and (e)–(h) percent bias between gridded observational datasets and WRF. Here WRF is considered to be the truth. The correlation coefficient and RMSB (mm) of precipitation for grid points 2 km MSL are shown in (a)–(d) and RMSPB (%) of precipitation at 2 km MSL is shown in (e)–(h).
Citation: Journal of Hydrometeorology 18, 9; 10.1175/JHM-D-17-0056.1
To better understand the difference between gauge-driven gridded datasets and WRF, we analyze the standard deviations of wintertime precipitation bias (again assuming WRF to be the truth), estimated by all gauge-based gridded datasets, as shown in Fig. 11a. The circles represent the bias standard deviation for different elevation intervals, and the blue histogram represents the amount of grids in the study domain. As seen from the figure, the bias standard deviations increase with terrain height (Fig. 11a), suggesting that these observational datasets have larger uncertainties at higher elevations, consistent with Fig. 8. Indeed, the correlation coefficient between any of the gauge-driven gridded datasets and WRF decreases with decreasing gauge density (Fig. 11b), as in some high-elevation places, indicating uncertainties in the statistical interpolation techniques. Here the gauge density is defined as the number of gauges (including all gauges plotted in Fig. 8, not just SNOTEL gauges) within a range of 30 km for each gauge site.
(a) Standard deviation of winter precipitation bias between gridded datasets and WRF as a function of elevation. (b) Correlation coefficient between observational datasets and WRF as a function of gauge density, which is defined as the number of gauges in a range of 30 km for each gauge. (c) Winter precipitation difference between observational datasets and WRF as a function of wind speed, for grid points 3 km MSL.
Citation: Journal of Hydrometeorology 18, 9; 10.1175/JHM-D-17-0056.1
Since the algorithms behind the gauge-driven gridded datasets lack detailed physical processes, we examine the precipitation bias at 3 km MSL between the gridded datasets against WRF as a function of wind speed (Fig. 11c). As defined in section 3b, this is the WRF-derived prevailing surface wind during precipitation events. The gauge-driven gridded datasets slightly overestimate precipitation when the surface wind is weaker than 16 m s−1 and underestimate precipitation in winds exceeding 24 m s−1. So in mountain areas where strong winds are common (e.g., Wind River Range), all gauge-driven gridded datasets may underestimate the wintertime precipitation.
4. Discussion
Several high-resolution gridded historical precipitation datasets have been developed, using different interpolation techniques, and they are widely used for a range of purposes, for example, hydrometeorology and ecology analyses, forecasting and agricultural operations, and hydrology and land surface modeling (e.g., Bales et al. 2006; Lundquist et al. 2015; Henn et al. 2017). Our analysis shows that gauge-driven gridded datasets have relatively large uncertainties over mountains inadequately covered with gauge sites, especially in winter. Some previous studies using different methods have also suggested that gauge-driven gridded datasets may have large uncertainties in wintertime precipitation estimation over mountains in the IWUS (e.g., Daly et al. 2008; Gutmann et al. 2012; Henn et al. 2017), but it is not known over which particular mountains the largest uncertainties are found and which dataset performs the best. In this study we argue that high-resolution model simulations have become a useful technique to evaluate and improve gauge-driven gridded precipitation estimates. Such simulations are useful also to identify areas with large discrepancies between model and gridded observations, areas where either the SNOTEL measurements are of questionable quality or where gauges are lacking. The 4-km convection-permitting IWUS WRF regional climate simulation nicely captures cold-season orographic precipitation in the IWUS, consistent with previous simulations (Ikeda et al. 2010; Rasmussen et al. 2011). Comparison between WRF and the four gauge-driven datasets suggest PRISM and Daymet are more consistent with WRF for winter precipitation than NLDAS II and CUSEG. There are remarkable similarities in the cold-season precipitation biases over specific mountain ranges (comparing against PRISM) between the IWUS WRF and the CONUS WRF (Liu et al. 2017) simulations (Fig. 7), notwithstanding the fact that these simulations use different domains, different time periods, and different model physics. This indicates that these biases are at least partly due to uncertainties in the measurement of snowfall and/or to uncertainties in the statistical gridding techniques over complex terrain. In the IWUS domain, we find particularly large discrepancies in the Wind River, the Gros Ventre, and Bighorn Ranges in Wyoming, where most gauges are in the foothills, not in the high country. This may motivate the reevaluation of some gauge records over these mountains, and the installation of additional SNOTEL gauges, especially in regions marked by large discrepancies between modeled and gauge-driven precipitation estimates. In Colorado and Utah, more gauges exist closer to mountain tops, and the discrepancies between WRF and observed winter precipitation generally are smaller. The differences between WRF and the four gauge-driven gridded datasets are subtly related to the wind speed, suggesting measurements of ambient conditions may be helpful to improve the gauge-driven gridded datasets.
The 4-km convection-permitting WRF simulation overestimates summertime precipitation in the high plains of Wyoming and Colorado. Changing land surface, PBL, and radiation parameterizations have no large impacts on the results (Rasmussen et al. 2014; Liu et al. 2017). Increasing the model resolution may improve the results because orographic convection generally is rather small and isolated. A previous study shows the summer precipitation over Colorado mountains is significantly overestimated in simulations with relatively coarse resolution (12 km); the results are significantly improved after increasing the resolution to 4 km (Rasmussen et al. 2014). The model representativeness of the summertime precipitation is affected also by its cloud microphysics scheme, which controls the precipitation efficiency and cloud life cycle (e.g., Khain et al. 2015). Further improvements in the resolution of convection-permitting models and in microphysics schemes may improve the simulation of summertime precipitation and its diurnal cycle in the IWUS.
5. Conclusions
A 10-yr, 4-km-resolution, convection-permitting WRF simulation is used to study seasonal precipitation in the IWUS, in particular over the mountains, and to compare orographic precipitation with SNOTEL gauge data, four gauge-driven gridded datasets (PRISM, Daymet, NLDAS II, and CUSEG), and a radar-based dataset (NCEP IV). The main conclusions are as follows:
The IWUS WRF simulation captures wintertime orographic precipitation well, and biases over specific mountain ranges are identical to those in another WRF simulation (Liu et al. 2017), suggesting that these biases are at least partly due to uncertainties in the measurement of snowfall and/or uncertainties in the gridding of these measurements over complex terrain.
Cross validation of the WRF simulation against the gauge-driven gridded dataset suggests that uncertainties of the gridded datasets are larger over mountains in winter. Among the four gauge-driven gridded datasets, NLDAS II estimates the least winter precipitation and the most summer precipitation in the IWUS, while CUSEG estimates the most winter precipitation and the least summer precipitation. PRISM and Daymet are more consistent with WRF than NLDAS II and CUSEG. NCEP IV compares well against other datasets in the vicinity of radars, but grossly underestimates winter precipitation over the mountains.
The discrepancy between WRF and gauge-driven gridded datasets is larger at higher elevations and larger in areas with fewer gauges. Compared to WRF, gauge-driven datasets tend to overestimate the precipitation in areas with prevailing weak surface winds and tend to underestimate the precipitation in areas with strong winds.
In general, regional climate simulations can be used to improve the algorithms of gauge-driven gridded datasets in complex terrain, to inform where gauges are too sparsely distributed, and to indicate which SNOTEL sites may be questionable.
In a separate study in progress, we are using the IWUS WRF simulation and gridded datasets to evaluate summertime precipitation over the IWUS, in particular its diurnal cycle.
Acknowledgments
This work was funded by the Wyoming Water Development Commission and the United States Geological Survey, under the auspices of the University of Wyoming Water Research Program. The lead author was funded by a Wyoming Engineering Initiative Doctoral Fellowship. The research benefitted from comments by Roy Rasmussen.
REFERENCES
Bales, R. C., N. P. Molotch, T. H. Painter, M. D. Dettinger, R. Rice, and J. Dozier, 2006: Mountain hydrology of the western United States. Water Resour. Res., 42, W08432, doi:10.1029/2005WR004387.
Barnett, T. P., and Coauthors, 2008: Human-induced changes in the hydrology of the western United States. Science, 319, 1080–1083, doi:10.1126/science.1152538.
Clark, M. P., and A. G. Slater, 2006: Probabilistic quantitative precipitation estimation in complex terrain. J. Hydrometeor., 7, 3–22, doi:10.1175/JHM474.1.
Daly, C., R. P. Neilson, and D. L. Phillips, 1994: A statistical–topographic model for mapping climatological precipitation over mountainous terrain. J. Appl. Meteor., 33, 140–158, doi:10.1175/1520-0450(1994)033<0140:ASTMFM>2.0.CO;2.
Daly, C., W. P. Gibson, G. H. Taylor, G. L. Johnson, and P. Pasteris, 2002: A knowledge-based approach to the statistical mapping of climate. Climate Res., 22, 99–113, doi:10.3354/cr022099.
Daly, C., W. P. Gibson, G. H. Taylor, M. K. Doggett, and J. I. Smith, 2007: Observer bias in daily precipitation measurements at United States cooperative network stations. Bull. Amer. Meteor. Soc., 88, 899–912, doi:10.1175/BAMS-88-6-899.
Daly, C., M. Halbleib, J. I. Smith, W. P. Gibson, M. K. Doggett, G. H. Taylor, J. Curtis, and P. P. Pasteris, 2008: Physiographically sensitive mapping of climatological temperature and precipitation across the conterminous United States. Int. J. Climatol., 28, 2031–2064, doi:10.1002/joc.1688.
Doesken, N. J., and G. L. Schaefer, 1987: The contribution of SNOTEL precipitation measurements to climate analysis, monitoring and research in Colorado. Proc. 55th Annual Western Snow Conf., Vancouver, BC, Canada, Western Snow Conference, 20–30, https://westernsnowconference.org/node/675.
Ebert, E. E., J. E. Janowiak, and C. Kidd, 2007: Comparison of near-real-time precipitation estimates from satellite observations and numerical models. Bull. Amer. Meteor. Soc., 88, 47–64, doi:10.1175/BAMS-88-1-47.
Fulton, R. A., J. P. Breidenbach, D. J. Seo, D. A. Miller, and T. O’Bannon, 1998: The WSR-88D rainfall algorithm. Wea. Forecasting, 13, 377–395, doi:10.1175/1520-0434(1998)013<0377:TWRA>2.0.CO;2.
Gourley, J. J., Y. Hong, Z. L. Flamig, L. Li, and J. Wang, 2010: Intercomparison of rainfall estimates from radar, satellite, gauge, and combinations for a season of record rainfall. J. Appl. Meteor. Climatol., 49, 437–452, doi:10.1175/2009JAMC2302.1.
Gutmann, E. D., R. M. Rasmussen, C. Liu, K. Ikeda, D. J. Gochis, M. P. Clark, J. Dudhia, and G. Thompson, 2012: A comparison of statistical and dynamical downscaling of winter precipitation over complex terrain. J. Climate, 25, 262–281, doi:10.1175/2011JCLI4109.1.
Henn, B., A. J. Newman, B. Livneh, C. Daly, and J. D. Lundquist, 2017: An assessment of differences in gridded precipitation datasets in complex terrain. J. Hydrol., doi:10.1016/j.jhydrol.2017.03.008, in press.
Hong, S. Y., and H. L. Pan, 1996: Nonlocal boundary layer vertical diffusion in a medium-range forecast model. Mon. Wea. Rev., 124, 2322–2339, doi:10.1175/1520-0493(1996)124<2322:NBLVDI>2.0.CO;2.
Hou, A. Y., and Coauthors, 2014: The Global Precipitation Measurement Mission. Bull. Amer. Meteor. Soc., 95, 701–722, doi:10.1175/BAMS-D-13-00164.1.
Iacono, M. J., J. S. Delamere, E. J. Mlawer, M. W. Shephard, S. A. Clough, and W. D. Collins, 2008: Radiative forcing by long-lived greenhouse gases: Calculations with the AER radiative transfer models. J. Geophys. Res., 113, D13103, doi:10.1029/2008JD009944.
Ikeda, K., and Coauthors, 2010: Simulation of seasonal snowfall over Colorado. Atmos. Res., 97, 462–477, doi:10.1016/j.atmosres.2010.04.010.
Jimenez, P. A., J. J. Dudhia, F. Gonzalez-Rouco, J. Navarro, J. P. Montavez, and E. Garcia-Bustamante, 2012: A revised scheme for the WRF surface layer formulation. Mon. Wea. Rev., 140, 898–918, doi:10.1175/MWR-D-11-00056.1.
Khain, A. P., and Coauthors, 2015: Representation of microphysical processes in cloud‐resolving models: Spectral (bin) microphysics versus bulk parameterization. Rev. Geophys., 53, 247–322, doi:10.1002/2014RG000468.
Lin, X., and A. Y. Hou, 2012: Estimation of rain intensity spectra over the continental United States using ground radar–gauge measurements. J. Climate, 25, 1901–1915, doi:10.1175/JCLI-D-11-00151.1.
Lin, Y., and K. E. Mitchell, 2005: The NCEP stage II/IV hourly precipitation analyses: Development and applications. 19th Conf. on Hydrology, San Diego, CA, Amer. Meteor. Soc., 1.2, https://ams.confex.com/ams/pdfpapers/83847.pdf.
Liu, C., K. Ikeda, G. Thompson, R. Rasmussen, and J. Dudhia, 2011: High-resolution simulations of wintertime precipitation in the Colorado Headwaters region: Sensitivity to physics parameterizations. Mon. Wea. Rev., 139, 3533–3553, doi:10.1175/MWR-D-11-00009.1.
Liu, C., and Coauthors, 2017: Continental-scale convection-permitting modeling of the current and future climate of North America. Climate Dyn., 49, 71–95, doi:10.1007/s00382-016-3327-9.
Lundquist, J. D., M. Hughes, B. Henn, E. D. Gutmann, B. Livneh, J. Dozier, and P. Neiman, 2015: High-elevation precipitation patterns: Using snow measurements to assess daily gridded datasets across the Sierra Nevada, California. J. Hydrometeor., 16, 1773–1792, doi:10.1175/JHM-D-15-0019.1.
Maddox, R. A., J. Zhang, J. J. Gourley, and K. W. Howard, 2002: Weather radar coverage over the contiguous United States. Wea. Forecasting, 17, 927–934, doi:10.1175/1520-0434(2002)017<0927:WRCOTC>2.0.CO;2.
Mote, P. W., A. F. Hamlet, M. P. Clark, and D. P. Lettenmaier, 2005: Declining mountain snowpack in western North America. Bull. Amer. Meteor. Soc., 86, 39–49, doi:10.1175/BAMS-86-1-39.
NCEP, 2001: National Centers for Environmental Prediction Hourly Multisensor Precipitation Analysis Stage IV. Subset: March 2003–February 2012, NOAA NCEP Environmental Modeling Center, accessed 5 August 2016, http://www.emc.ncep.noaa.gov/mmb/ylin/pcpanl/stage4/.
NCEP, 2014: North American Land Data Assimilation System project phase 2. NOAA NCEP Environmental Modeling Center, accessed 11 September 2016, http://www.emc.ncep.noaa.gov/mmb/nldas/.
Newman, A. J., and Coauthors, 2015a: Gridded ensemble precipitation and temperature estimates for the contiguous United States. J. Hydrometeor., 16, 2481–2500, doi:10.1175/JHM-D-15-0026.1.
Newman, A. J., and Coauthors, 2015b: Gridded ensemble precipitation and temperature estimates over the contiguous United States. Earth System Grid at NCAR, accessed 7 November 2016, https://www.earthsystemgrid.org/dataset/gridded_precip_and_temp.html.
Niu, G.-Y., and Coauthors, 2011: The community Noah land surface model with multiparameterization options (Noah-MP): 1. Model description and evaluation with local-scale measurements. J. Geophys. Res., 116, D12109, doi:10.1029/2010JD015139.
NRCS, 1977: Snow Telemetry gauge network. Subset: March 2003–February 2012, NRCS, Department of Agriculture, accessed 31 March 2016, http://www.wcc.nrcs.usda.gov/.
PRISM Climate Group, 2016: Precipitation-elevation Regressions on Independent Slopes Model. Northwest Alliance for Computational Science and Engineering, Oregon State University, accessed 16 February 2017, http://www.prism.oregonstate.edu/.
Rasmussen, R., and Coauthors, 2011: High-resolution coupled climate runoff simulations of seasonal snowfall over Colorado: A process study of current and warmer climate. J. Climate, 24, 3015–3048, doi:10.1175/2010JCLI3985.1.
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, doi:10.1175/BAMS-D-11-00052.1.
Rasmussen, R., and Coauthors, 2014: Climate change impacts on the water balance of the Colorado headwaters: High-resolution regional climate model simulations. J. Hydrometeor., 15, 1091–1116, doi:10.1175/JHM-D-13-0118.1.
Saha, S., and Coauthors, 2010: The NCEP Climate Forecast System Reanalysis. Bull. Amer. Meteor. Soc., 91, 1015–1057, doi:10.1175/2010BAMS3001.1.
Seo, D. J., 1998: Real-time estimation of rainfall fields using radar rainfall and rain gage data. J. Hydrol., 208, 37–52, doi:10.1016/S0022-1694(98)00141-3.
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, doi:10.1029/1999WR900090.
Smalley, M., T. L’Ecuyer, M. Lebsock, and J. Haynes, 2014: A comparison of precipitation occurrence from the NCEP Stage IV QPE product and the CloudSat Cloud Profiling Radar. J. Hydrometeor., 15, 444–458, doi:10.1175/JHM-D-13-048.1.
Tesfagiorgis, K., S. E. Mahani, N. Y. Krakauer, and R. Khanbilvardi, 2011: Bias correction of satellite rainfall estimates using a radar-gauge product – A case study in Oklahoma (USA). Hydrol. Earth Syst. Sci., 15, 2631–2647, doi:10.5194/hess-15-2631-2011.
Thompson, G., R. M. Rasmussen, and K. Manning, 2004: Explicit forecasts of winter precipitation using an improved bulk microphysics scheme. Part I: Description and sensitivity analysis. Mon. Wea. Rev., 132, 519–542, doi:10.1175/1520-0493(2004)132<0519:EFOWPU>2.0.CO;2.
Thornton, P. E., S. W. Running, and M. A. White, 1997: Generating surfaces of daily meteorological variables over large regions of complex terrain. J. Hydrol., 190, 214–251, doi:10.1016/S0022-1694(96)03128-9.
Thornton, P. E., M. M. Thornton, B. W. Mayer, N. Wilhelmi, Y. Wei, R. Devarakonda, and R. B. Cook, 2014: Daymet: Daily Surface Weather Data on a 1-km Grid for North America, Version 2. ORNL DAAC, accessed 11 September 2016, https://doi.org/10.3334/ORNLDAAC/1328.
Woodhouse, C. A., 2004: A paleo perspective on hydroclimatic variability in the western United States. Aquat. Sci., 66, 346–356, doi:10.1007/s00027-004-0723-8.
Xia, Y., and Coauthors, 2012: Continental‐scale water and energy flux analysis and validation for the North American Land Data Assimilation System project phase 2 (NLDAS‐2): 1. Intercomparison and application of model products. J. Geophys. Res., 117, D03109, doi:10.1029/2011JD016048.
Yang, Z.-L., and Coauthors, 2011: The community Noah land surface model with multiparameterization options (Noah-MP): 2. Evaluation over global river basins. J. Geophys. Res., 116, D12110, doi:10.1029/2010JD015140.
