1. Introduction
During the last 50 yr, water resources in the western United States have been developed extensively for flood control, hydropower production, irrigation, navigation, fish protection, and recreation. For example, irrigation supports agricultural production of diverse crops in California and the Columbia River basin (CRB) that would otherwise be impossible to grow from natural rainfall. Although less than 15% of harvested cropland in the United States requires irrigation, irrigated crops (mainly concentrated in the western states) produce almost 40% of the value of U.S. crops (Bajwa et al. 1992). Hydropower accounts for over 70% of electric power supply in the northwestern United States. There are numerous other examples in which water resources have been essential to the economic development of the western United States. It is, therefore, worrisome that recent studies suggest that global warming may exert significant impacts on snowpack and streamflow, which may seriously affect water resources in the western United States during the latter half of this century (e.g., Leung and Wigmosta 1999; Leung and Ghan 1999; Mile et al. 2000; Leung et al. 2003d).
Arguably there is a high degree of uncertainty in projecting future water resources in any part of the world. Even estimating the contemporary water budgets of river basins is a challenge. Roads (2002) showed, using various sources of data for atmospheric moisture, runoff, and precipitation, that the total global annual land water budget could be closed to within 10%. At the scales of large river basins, such as the Mississippi, errors in closing the water budgets are near 20%, and become increasingly larger for smaller basins, especially high-latitude basins. In the western United States, observational data in remote mountain areas have been scarce; for example, radar data are problematic because of beam blockage by terrain and other factors (e.g., Westrick et al. 1999). Yet clearly, precipitation and snowpack in remote mountains dictate the cold season water cycle in regions such as the Columbia River and Sacramento–San Joaquin River basins (SSJ) (Fig. 1).
Besides problems associated with observational data, water budgets of rivers in the West are also not adequately represented in modeling studies. Even in global and regional analyses, the accuracy of estimating water budgets is compromised by the spatial resolution of the models and insufficient observational data available for data assimilation. In this regard, the complex topography and orographic effects of the narrow Cascade Mountain and Sierra Nevada ranges challenge existing modeling efforts of the water cycle in these regions.
Supported by the Department of Energy (DOE) Accelerated Climate Prediction Initiative, climate change impacts on regional water resources have been investigated in the West (Barnett et al. 2003). A necessary step in the end-to-end assessment of climate change impact is downscaling (e.g., Giorgi et al. 2001; Leung et al. 2003a). In dynamical downscaling, regional climate models are used to provide regional climate change scenarios based on large-scale conditions simulated by global climate models. We have, therefore, undertaken several studies to systematically investigate a number of issues related to dynamical downscaling in the complex terrain of the western United States. Leung et al. (2003b,c) performed extensive evaluation of a regional climate simulation driven by the National Centers for Environmental Protection–National Center for Atmospheric Research (NCEP–NCAR) reanalyses for 20 yr. They analyzed many hydroclimate features, including spatial distribution of seasonal mean and extreme precipitation, the relationship between temperature and precipitation, the seasonal cycle and its link to terrain features, orographic precipitation, and the snowpack. Leung and Qian (2003) studied the effects of spatial resolution on simulating precipitation and snowpack in the western United States.
Leung et al. (2003b,c) showed that at 40-km spatial resolution, regional climate models realistically captured many distinct features of the mean and interannual variability of water budgets in the western United States. However, some hydroclimatic aspects were not adequately represented. Examples included the spatial distribution of snowpack and orographic precipitation, and summer monsoon rainfall in the Southwest. It is not clear whether the deficiencies were more related to model parameterizations, spatial resolution, or the large-scale conditions used to drive the regional model.
The goal of this study is to understand differences in estimating water budgets of the western United States by comparing three global reanalysis products and three regional climate simulations driven by the reanalyses. By this intercomparison, we can study the effects of spatial resolution, model configurations/parameterizations, and large-scale boundary conditions on basin-scale water budgets. We focus our efforts on studying cold season hydroclimate conditions of the Columbia River and Sacramento–San Joaquin River because winter precipitation and snowpack dominate the regional water cycle of these major river basins. As we describe in this paper, observational datasets developed using different data sources and procedures also differ significantly from each other.
Section 2 describes the observational data, global reanalyses, and regional climate simulations. Section 3 analyzes the differences among the global reanalyses, and section 4 describes the differences among the regional climate simulations. In the discussion and conclusions (section 5), we summarize the differences among all reanalyses and regional simulations and summarize the differences in estimating water budgets of the Columbia River and Sacramento–San Joaquin River basins.
2. Data and model simulations
a. Observations
For comparison with the global reanalyses and regional simulations, we used the 1/8° gridded daily maximum/minimum temperature and precipitation dataset developed by Maurer et al. (2002) for the continental United States. The dataset used temperature and precipitation measurements from the National Oceanic and Atmospheric Administration (NOAA) Cooperative Observer (COOP) stations. Daily precipitation data were scaled to match the long-term average of the Parameter-elevation Regressions on Independent Slopes Model (PRISM) precipitation climatology (Daly et al. 1994), which was developed based on data from COOP and snow telemetry stations and accounts for orographic effects based on statistical relationships between elevation and precipitation. This dataset provides a more realistic depiction of orographic precipitation in the western United States.
For comparison of precipitation at a resolution similar to the global reanalyses, we used the 2.5° gridded data from the Global Precipitation Climatology Project (GPCP; Huffman et al. 1997). This dataset combines gauge measurements with remote sensing data to achieve global coverage over land as well as the ocean.
For comparison of surface water budgets, we used the observed streamflows for the Columbia River and Sacramento–San Joaquin River basins. Gauged flows at any given measurement point are increased or decreased on a monthly or daily basis to account for upstream diversions, storage, or export/import of water to or from other watersheds to obtain naturalized or unimpaired flows. Data for the Columbia River basin were produced for the Bonneville Power Administration by the A. G. Crook Company. Data for the Sacramento–San Joaquin River basin were obtained from the California Department of Water Resources California Data Exchange Center (http://cdec.water.ca.gov/riv_flows.html).
b. Global reanalyses
Three global reanalyses were intercompared and used to provide large-scale conditions for regional climate simulations: 1) NCEP–NCAR reanalysis I (Kalnay et al. 1996), 2) NCEP–DOE second Atmospheric Model Intercomparison Project (AMIP-II) reanalysis II (Kanamitsu et al. 2002), and 3) European Centre for Medium-Range Weather Forecasts (ECMWF) reanalyses (Gibson et al. 1997), here denoted as NRA1, NRA2, and ERA, respectively. Our intercomparison focused on the western United States between 1986–93 where regional climate simulations were available for the same region and period.
Each global reanalysis was archived 6-hourly at 2.5° horizontal resolution and 17 pressure levels. They were, however, originally generated at different horizontal and vertical resolutions. ERA was produced using the ECMWF model at T106 horizontal resolution and 31 vertical levels. NRA1 and NRA2 used the NCEP Global Spectral Model (GSM) applied at T62 horizontal resolution and 28 vertical levels.
NRA2 and NRA1 differed in several aspects that are described in Kanamitsu et al. (2002). NRA2 included some error fixes and improvements in some of the physics parameterizations, prescribed fields (such as albedo, ozone, and sea ice), and diagnostics (such as clouds and snow/water budgets). In addition, the treatments of soil moisture in NRA1 and NRA2 were drastically different. In NRA1, soil moisture was calculated based on the model precipitation, and then damped to an assumed climatology (Roads and Betts 2000). Because of feedback effects between the land surface and atmosphere, biases in model precipitation and soil moisture are related. In NRA2, this problem was fixed by correcting the model-simulated soil moisture based on the difference between the observed and simulated precipitation.
Although NRA2 should be considered an update of NRA1, rather than a next-generation reanalysis, there are some significant differences between NRA1 and NRA2, particularly over data-sparse regions, such as the oceans, and in second-order variables (e.g., Hnilo et al. 1999). Because cold season precipitation in the western United States depends strongly on large-scale moisture flux from the Pacific Ocean, an intercomparison of the global reanalyses can reveal differences in the hydrologic cycles of the reanalyses, as well as differences in regional climate simulations driven by the reanalyses over major river basins of the west.
c. Regional climate simulations
To understand the role of boundary conditions and model differences on regional climate simulations, an ideal experiment requires a matrix of simulations with each model driven by each global reanalysis to separate the two effects. Pan et al. (2001) reported a comprehensive study that used two regional climate models, each forced by three sets of large-scale conditions (a 2 × 3 suite of 10-yr simulations), to evaluate uncertainties in projecting climate change using the nested modeling approach. In this study, with three global reanalyses and two regional models, six simulations would be needed to fully analyze the impacts of model and boundary forcing differences. However, only three simulations were available so that a strict separation of the effects of boundary conditions and model differences was not possible. Nevertheless, a comparison of the three regional simulations did reveal, to some extent, the level of model sensitivity to the two factors.
Two of the simulations were based on the fifth-generation Pennsylvania State University–NCAR Mesoscale Model (MM5; Grell et al. 1993) driven by NRA1 and ERA. The third simulation was based on the NCEP Regional Spectral Model (RSM; Juang and Kanamitsu 1994; Kanamitsu 2000) driven by NRA2. These simulations are referred as MM5-NRA1, MM5-ERA, and RSM-NRA2. Both MM5 simulations were initialized based on the reanalyzed large-scale conditions on 1 July 1980, with a large domain covering the United States and the surrounding oceans at 120-km resolution and a nested domain covering the western United States at 40-km resolution. The RSM simulation was initialized based on the NRA2 conditions on 1 June 1986 and covered the whole United States at approximately 50-km resolution. Because ERA was only available through 1993, we compared these regional simulations for the common period, September 1986 through August 1993, for the portion of the western United States.
The MM5-NRA1 simulation (1981–2000) was discussed in detail by Leung et al. (2003b,c). The simulation realistically simulated the mean and extreme seasonal climate conditions of the western United States (Leung et al. 2003b). Furthermore, the observed interannual variability and climate anomalies associated with the El Niño–Southern Oscillation (ENSO) were surprisingly well reproduced by the simulation (Leung et al. 2003c). The MM5-ERA simulation used an identical model configuration. Therefore, the only difference between these two simulations was the large-scale conditions used to initialize and update the lower (sea surface temperature) and lateral boundary conditions.
The RSM-NRA2 simulation (1986–2000) was analyzed by Han and Roads (2003), which focused on the whole United States. The simulation provides realistic depiction of regional climate over different parts of the country. Besides physical parameterizations, two major differences between RSM and MM5 were the use of spectral versus finite difference numerical representation and internal spectral nudging in RSM. The latter refers to nudging of the RSM-simulated large-scale variables to the global reanalysis in the spectral domain over the entire RSM domain. We might, therefore, anticipate the global reanalyses to have a larger control over the RSM simulation than the MM5 simulations.
Table 1 summarizes the numerics and physical parameterizations used in MM5 and RSM. Although they were applied at similar resolution over the western United States, Fig. 1 shows large differences in the surface topography prescribed in the models. The RSM topography was smoother compared to that of MM5. For example, the Sierra Nevada ascended from 200 to a maximum of 2400 m in MM5. In the RSM, the terrain gradient was between 400 and 2000 m. The Cascade and Wasatch Mountain ranges, and the Olympic Mountain, as well as the Columbia and Snake River basins, and the Sacramento–San Joaquin River valley were all much better resolved in MM5 than RSM. Such differences were partly related to the finite difference versus spectral representation used by the models. In simulating the regional climate of the western United States, these topographic differences can lead to significant differences because of terrain effects on thermodynamics, mesoscale circulation, clouds, precipitation, and snow processes.
3. Intercomparison of global reanalyses
a. Moisture flux and precipitation
Moisture flux from the Pacific Ocean is an important component of the western U.S. cold season hydrologic cycle. Figure 2 shows the seasonal averaged atmospheric moisture flux and moisture flux divergence at 850 mb during winter [December–January–February (DJF)] from NRA1, NRA2, and ERA. The seasonal mean moisture flux was calculated based on averaging the product of the 6-hourly horizontal wind and specific humidity data. It includes both the mean and transient components of the moisture flux. In general, the mean component (the product of seasonal mean horizontal wind and seasonal mean specific humidity) contributes less than 15% to the seasonal average shown in Fig. 2, which, therefore, reflects mainly differences among the reanalyzed transient moisture flux that is more closely related to midlatitude synoptic-scale systems that produce precipitation. As seen in Fig. 2, differences among the reanalyses were small over the central and eastern United States because high-density rawinsonde data were available for data assimilation. Over the oceans and in regions with strong topography, larger differences were found.
During the cold season, the large-scale flow in the northeast Pacific Ocean was near southwesterly, and moisture flux was the strongest in NRA1. Moisture converged further inland in NRA1 and ERA than in NRA2, where moisture convergence was stronger over the ocean and near the coastline of the Northwest. This moisture flux pattern in NRA2 was consistent with the divergence over the Rockies, which was larger and further west compared to NRA1 and ERA. The latter may partly be related to the broader terrain representation of the Rockies in NRA2.
Indeed, Fig. 3 shows that differences appeared in the terrain features of the reanalyses at the same 2.5° spatial resolution. Even NRA1 and NRA2, which used the same global model applied at the same spatial resolution, showed differences that were not insignificant compared to differences between NRA1 or NRA2 and ERA. NRA2 used smoothed topography to avoid the Gibb's effect of the spectral technique, which creates artificial topographic structures that induced erroneous orographic precipitation found in NRA1. Therefore, while NRA1 suffers from spurious precipitation in areas such as Chile and Brazil, mountain ranges in NRA2 suffer from reduced amplitude that affect the spatial distribution of precipitation in areas such as the western United States. The ERA terrain reflects the higher spatial resolution used by the ECMWF model.
In ERA, divergence over the Rockies was narrower but extended over a larger region in the north–south direction. The Rocky Mountains blocked more moisture flux propagating inland from the ocean, which resulted in less moisture flux and convergence on the east side of the mountain. Near southern California, moisture was transported mainly in a northwesterly direction, and was strongest in NRA2. Differences among the reanalyses were larger over the eastern tropical Pacific Ocean, the Rocky Mountains, and on both sides of Mexico. Spatial resolution and terrain representation probably played an important role in creating such differences.
The differences in spatial distribution of moisture flux between ERA and NRA1 were very similar during both seasons (summer not shown). They were larger near the Rockies and the eastern tropical Pacific Ocean. This suggests that the differences were probably more related to differences in spatial resolutions and terrain features in the NCEP and ECMWF models. However, differences between NRA1 and NRA2 were very different between the warm and cold seasons. During winter, the largest differences were found over the eastern tropical Pacific Ocean and the Rockies. During summer (not shown), the largest differences were located east of the Rockies on the Great Plains. This indicates that differences may be more related to differences in physics parameterizations, large-scale circulation, and factors other than spatial resolution and terrain features.
Figure 4 shows the latitude–height distributions of relative humidity and horizontal wind along 125°W. At the higher latitudes, both NRA1 and NRA2 depicted moister (drier) conditions in the lower (upper) atmosphere compared to ERA during winter. In the midlatitudes, ERA had a more uniform distribution of moisture in the vertical than NRA1 and NRA2. Furthermore, both NRA1 and NRA2 showed elevated moisture in the upper atmosphere near 300 mb that did not show up in ERA.
As a result of discrepancies in large-scale circulation, surface climate conditions were also quite different among the global reanalyses. Figure 5 shows the cold season precipitation based on the reanalyses and GPCP observations at the same spatial resolution bilinearly interpolated to the MM5-nested domain. Although they all show a gradual decrease in precipitation from the Pacific coast inland, there are also clear differences in the regional distributions in the Northwest, California, and Southwest. NRA1 and ERA were more similar in the way precipitation was spread inland from the coast toward the northern Rockies and the Southwest. However both NRA1 and NRA2 had less precipitation along the southern California coast than ERA. The GPCP precipitation lacked much of the spatial structure evident in the 1/8° gridded data (Fig. 7), and there were obvious discrepancies between the two datasets, even on a regional basis. This will be discussed further in the next section.
b. Interannual variability
On an interannual basis, there were substantial variations in the moisture flux from the Pacific Ocean. Figure 6a shows the cold season (DJF) total vertically integrated moisture fluxes across the lateral boundaries of the rectangular box shown in Fig. 1 between 1981–93. Similar to Fig. 2, moisture fluxes were calculated based on 6-hourly reanalyzed data summing over the cold season and the boundaries but using units of kilograms. The largest contribution to moisture change in the region was, of course, moisture flux from the Pacific Ocean in the western boundary. Moisture was mainly transported out of the region from the eastern boundary, but there were also small outflow contributions from the southern boundary. Moisture flux at the northern boundary was not only small, but also much more variable from year to year with changes in sign. Less moisture transported in from the western boundary was usually associated with moisture transported in from the northern boundary, which corresponded to northwesterly flows that created drier conditions in 1985, 1989, 1990, and 1991.
Similarly, more moisture transported in from the western boundary (typically larger than 15 × 1014 kg of moisture) was associated with more moisture transported out of the northern boundary. This corresponded to southwesterly flows that were more typical during El Niño years, such as 1983, 1986, 1987, and 1992. These variations were well represented in all global reanalyses.
Comparing the three reanalyses, the moisture flux from the western boundary was the largest in NRA1. Figure 6b shows the area-averaged vertically integrated moisture flux convergence and precipitation in the rectangular box. Larger moisture convergence in NRA1 in the mid-1980s typically resulted in more precipitation compared to NRA2 and ERA. Differences among the reanalyses precipitation were generally about 15% of the average of all reanalyzed precipitation, which indicates the minimum level of difference among global reanalyses for water budget studies in western river basins. Additional differences among the global reanalyses may arise because of errors in the input data and data assimilation procedures.
However, during 1982 and 1987, differences among the reanalyzed precipitation were as large as 30%–45% of the average of all reanalyzed precipitation. Interestingly, there were not any corresponding larger differences among the reanalyzed moisture flux convergence during 1987. On the contrary, larger differences in moisture flux convergence occurred in 1991 and 1992 when precipitation differences were relatively small. This suggests that in addition to moisture flux convergence, factors such as wind direction and interactions with orography may also play an important role in determining precipitation during the cold season in the western United States. This will be discussed further in section 4d.
In summary, over the western United States, it is hard to judge whether NRA1 and NRA2 (which used a very similar model) were more similar to each other than to ERA. Differences could be attributed to many factors, including model physics and numerics, spatial resolutions, prescribed conditions (such as terrain and vegetation), and data assimilation procedures. On a regional basis, it is clear that differences of 15%–20% exist among the reanalyzed large-scale circulation (moisture flux in particular), which will have important impacts on regional climate simulations. Differences among the reanalyzed precipitation were about 15%–30% of the averaged reanalyzed precipitation.
4. Intercomparison of regional climate simulations
a. Precipitation
Figure 7 shows the regional spatial distributions of precipitation during the cold season. At a spatial resolution much higher than the reanalyses, the regional simulations of precipitation all showed regional details that were more comparable to ⅛° observations than were the reanalyses. The two MM5 simulations were very similar in terms of spatial distributions that strongly reflect the underlying terrain features. They both captured orographic precipitation along the Cascades, Sierra Nevadas, and the Rockies. Consistent with stronger moisture flux from the western boundary in NRA1, the MM5/NRA1 precipitation was higher than the MM5/ERA precipitation almost everywhere. With the smoother topography shown previously in Fig. 1, the RSM precipitation missed many peaks along the Cascades and the Sierra Nevadas that were found in the observations and MM5 simulations. RSM precipitated more along the coast than over mountains. Precipitation was also more spatially extensive throughout the western United States.
Figure 8 shows the basin mean cold season precipitation based on all reanalyses, regional climate simulations, and observed precipitation datasets in the CRB and SSJ. Generally, basin mean cold season precipitation was larger in the regional simulations than in the reanalyses. Among the reanalyses, the NRA2 precipitation was the lowest, and agreed the least with the ⅛° observations in both basins. Among the regional simulations, the MM5 ERA precipitation was the lowest and in closest agreement with the ⅛° observations.
A closer examination of Fig. 8 reveals further differences among the reanalyses and regional simulations. In CRB, ERA and MM5 ERA agreed most closely with the 1/8° data, which was dominated by heavy precipitation between November and January over the maritime coastal mountains, and between March and May over the northern Rockies. The seasonal cycles in the MM5 simulations were similar to the observed, except the cold season precipitation was too strong in MM5/NRA1. Precipitation based on NRA1, NRA2, and RSM/NRA2 showed double peaks—one corresponding to the cold season precipitation, and the other during spring or early summer. This similarity among NRA1, NRA2, and RSM precipitation may be related to similarity in parameterizations and model configurations. Precipitation in the second peak was more convective in nature and located further inland compared to the winter peak.
In SSJ, again, ERA and MM5/ERA were in closest agreement with the ⅛° data. All of the reanalyses and regional simulations realistically captured the observed seasonal cycle with precipitation that peaked between November and March. There was an abrupt transition from March to April where precipitation significantly reduced as the Pacific high weakened and retreated to the north.
For the MM5 simulations, more precipitation in MM5/NRA1 was consistent with the higher moisture flux in NRA1 compared to ERA. However, the RSM/NRA2 simulation typically produced larger precipitation despite the consistently low precipitation found in NRA2. Hence, the difference between the precipitation simulated by the regional model and the global model that provided large-scale conditions was largest in the RSM and NRA2 pair in both basins. When separating precipitation into convective and large-scale precipitation simulated by the regional models (not shown), the RSM large-scale precipitation was found to be similar to that of the MM5. However, the RSM convective precipitation was 3–4 times higher than that of MM5. The ratios of convective to total precipitation were about 10%, 50%, and 70% in RSM, and 2%, 20%, and 45% in MM5 during winter, spring, and summer respectively. More sensitivity experiments are needed to help identify whether the larger convective precipitation was related to the convective parameterizations, large-scale conditions, or effects of internal nudging.
Besides differences among the reanalyses and regional simulations, there were also significant differences between the GPCP and ⅛° precipitation data in both basins. The GPCP cold season precipitation was less than 50% of the ⅛° data in CRB and between 30%–50% in SSJ. The upper bound of the differences among the reanalyses, regional simulations, and observations was approximately 100% of the mean of all precipitation datasets during the cold season. For comparison, analyses in section 3b showed that the difference in large-scale moisture flux was 15%–20%. Apparently, terrain features, spatial resolutions, and model treatments of precipitation, clouds, and possibly even dynamics amplified differences in moisture transport, and resulted in much larger differences among precipitation datasets. This indicates the difficulty in developing accurate water budgets for small western river basins like the CRB and SSJ.
Table 2 summarizes the cold season basin mean precipitation and surface temperature based on the reanalyses, regional simulations, and gridded observational datasets. As discussed above, precipitation is greatly amplified with downscaling in CRB. Temperature biases are generally positive (warm bias) and small in CRB except in MM5/NRA1 and RSM/NRA2 where the warm bias reaches over 2°C. In SSJ, temperature biases are also generally small except for NRA2 where the cold bias is over 4°C. Both warm and cold biases are found among the datasets. Note that the general warm biases found in most regional simulations may partly explain the negative bias in the snowpack simulated by the models. This is discussed further in the next section, and by Leung and Qian (2003) where more detailed analyses of observed and simulated snowpack were presented.
b. Transects of elevation and precipitation across mountain ranges
To further investigate how precipitation is distributed differently across mountains and valleys in the regional simulations, Fig. 9 shows a comparison of precipitation and elevation across two east–west transects in the Northwest and California based on observations and simulations. The first transect (approximately 40 km in width) cuts across the Olympic Mountain, Puget Sound, the Cascades, Columbia River basin, and the northern Rockies along 47.8°N. The observations were based on 1/8° data smoothed to 40-km resolution for comparison with the simulations. Clearly both the MM5 and RSM surface terrains were smoother compared to the actual terrain. For example, the Olympic Mountain Peak was only 600 m high in MM5 and was totally missed in RSM, while in reality it is 1200 m.
Major differences can be seen in the precipitation along this transect. First, in the observation and MM5 simulations, precipitation increased rapidly spatially on the western side of mountain ridges, such as the Olympic Mountain and the Cascades. Observed precipitation reached a maximum approximately 1° to the west of the ridge and decreased eastward because of rain-shadow effect. The MM5 simulations did not have this large difference probably mainly because of the lower peak elevation compared to the observation, but they did correctly describe the westward shift in precipitation maxima along the mountains. However, the drying on the eastern side of the ridge (or rain shadowing) was not as strong as observed. The spatial distributions of precipitation along the transect for the two MM5 simulations were very similar. Precipitation in MM5/ERA was typically 20% lower than that in MM5 NRA1.
RSM simulated a rather different precipitation transect across mountains and valleys. First, precipitation decreased gradually from the coast inland despite the presence of mountain ranges. In the observation and MM5 simulations, precipitation was higher over the mountains than along the coast. Second, unlike the observations and MM5 simulations where there was a westward shift of the precipitation peak from the elevation peak, the peak orographic precipitation in RSM aligned with the peak elevation (except for broad mountains, such as the Rockies). Similar situations were found in plots of other RSM transects not shown.
Third, on encountering the Cascades, precipitation only increased slightly with elevation in the RSM. This was partly because the RSM had a lower peak, and without the Olympic Mountains, the Cascades stretched smoothly from the coast with a much smaller topographic gradient, compared to MM5 and observations. If precipitation and elevation had a quasi-linear relationship, RSM should show a larger orographic precipitation than calculated. This suggests that topographic gradient might be an important control on orographic precipitation. Other differences between the RSM and MM5 simulations might also play a role. An example is the difference between the large-scale conditions used to drive the regional models. The weaker southwesterly moisture transport in ERA and NRA2 probably contributed to the lower orographic precipitation on the west of the Cascades in the MM5/ERA and RSM/NRA2 simulations. Further, separation of the effects of large-scale conditions and topography is not possible without a matrix of numerical experiments with different models and large-scale conditions.
The second transect along 42.7°N cuts across the Sierra Nevadas from the California coast and reaches the Wasatch Mountain range on the east. None of the simulations captured the observed large coastal precipitation, possibly because they all missed the sharp topographic gradient near the coast. On the other hand, all simulations were rather similar in showing small topographic forcing upslope of the Sierra Nevadas and the rain shadow effects east of 122°W. Again, the MM5/NRA1 simulation was about 20% higher than the MM5/ERA simulation, and both MM5 simulations showed slightly stronger orographic effects than the RSM. We might argue that large-scale conditions had less of an effect along this transect because the reanalysis conditions were similar in this region over the ocean (Fig. 2). Differences in orographic precipitation in MM5 and RSM were likely more related to differences in model parameterizations and terrain features.
c. Basin mean surface water budgets
1) Columbia River basin
Figure 10 shows the mean seasonal cycle of surface water budgets in CRB and SSJ based on the three regional simulations. Note that the rate of change in snow and soil moisture, evapotranspiration, and runoff were all calculated by the land surface models used in the regional climate models (refer to Table 1). Soil moisture includes water in all soil layers, and runoff was shown as the total of surface and subsurface runoff without any routing. Comparing the RSM/NRA2 and MM5/NRA1 precipitation in CRB, the RSM simulation was larger during spring (March–June) due to larger convective precipitation in the Rockies. Comparing the two MM5 simulations, precipitation was larger when NRA1 was used to provide boundary conditions.
There were major differences between the MM5 and RSM simulations of snowpack and runoff. Although the basin mean cold season precipitation in all simulations was similar, much less precipitation was stored as snowpack in RSM than MM5. This is because snowpack can only be stored in regions with sufficient precipitation and cold temperatures. In RSM, the smoother and, hence, lower terrain features, which mean less orographic precipitation and warmer temperatures in mountains, were less favorable for snow formation. Simulating the correct spatial distributions of precipitation and temperature is probably more important than just simulating a correct regional mean in surface water budgets of snow-dominated basins.
Comparing the two MM5 simulations of snow in CRB, snow accumulation rate was higher during winter (November–January) in MM5/NRA1 because precipitation was also higher. During spring, snowmelt rates peaked during March, which was earlier than the observed peak in April–May. At spatial resolution of 40 km, snow processes were still not accurately simulated because temperatures were still not cold enough along mountain ridges where snow forms. Leung et al. (2003b) showed that the simulated snowpack was only 25%–30% of the observed snowpack at SNOTEL stations, which are typically located at high elevations. They suggested that in addition to spatial resolution, problems with the representation land surface processes also contributed to the negative bias in the simulated snowpack.
The partitions of precipitation into the various components at the surface were quite different among the regional simulations. Because there was very little snowpack, RSM/NRA2 runoff responded almost directly to precipitation (i.e., runoff and precipitation were in phase). Soil moisture increased during winter and decreased during summer. Evaporation in RSM/NRA2 was the highest among all simulations, and it maximized in May and June.
In MM5/NRA1, runoff responded more to snowmelt than precipitation. During winter, runoff was small because precipitation, which was larger over mountains, was stored as snowpack rather then runoff. During spring, runoff peaked at the same time that snowmelt peaked. Compared to RSM/NRA2, soil moisture increased more during winter, although the precipitation in the two simulations was similar. This may partly be related to the generally lower elevation in RSM, which led to warmer temperatures, less snowpack, higher evaporation, and less soil moisture accumulation during winter.
Because precipitation was higher in MM5/NRA1 than MM5/ERA, snowpack accumulation was greater. As a result of the accumulated difference in precipitation and snowpack, a larger difference was found in the simulated runoff. Precipitation differences did not seem to affect evaporation and soil moisture when the same land surface model was used. This is evident from Fig. 10, which shows only small differences between the MM5/NRA1 and MM5/ERA evaporation and soil moisture. This is probably because the soil was already near saturation during winter so that more precipitation only contributed to more runoff and snow accumulation.
Figure 11 summarizes the runoff based on all regional simulations and observations. Comparison was done based on monthly mean runoff so that the lack of routing of the simulated runoff should have minimal effects. In CRB, the biggest difference between the regional simulations and observations was the timing of streamflow. As explained earlier, the model's inability to resolve terrain features at 40–50-km resolution is the main reason for early snowmelt in the MM5 simulations. In RSM/NRA2, the even lower terrain features compared to MM5 limited snow accumulation during winter and created a runoff pattern that corresponded to the timing of precipitation rather than snowmelt. Annual streamflows in the MM5/ERA and RSM/NRA2 simulations were only 5% lower than the observed. For MM5/NRA1, the annual streamflow was 40% higher than observed. The big difference between the annual mean MM5/NRA1 and RSM/NRA2 runoff was not so much due to precipitation differences, but rather to how the land surface partitioned the water.
2) Sacramento–San Joaquin River basin
In SSJ, precipitation was similar between RSM/NRA2 and MM5/NRA1, and larger than MM5/ERA. Convective precipitation was again much higher in RSM/NRA2 than MM5 throughout the year except for July–September. All simulations produced very little snow, partly because temperatures were much warmer compared to CRB, and partly because of inadequate spatial resolution in resolving the narrow and steep Sierra Nevadas range. In MM5/NRA1, precipitation increased runoff, evaporation, and soil moisture during winter. In RSM/NRA2, similar to the findings in CRB, there was greater evaporation and less runoff than occurred in MM5. In the summer, the main balance was between evaporation and reduced soil moisture. Differences between MM5/NRA1 and RSM/NRA2 were small.
Comparing MM5/NRA1 and MM5/ERA, more cold season precipitation in the former was reflected directly in larger runoff. Again, there was little difference between the simulated evaporation and soil moisture accumulation. This was because soil moisture was already near saturation during winter. During summer, there were similar soil moisture reductions and decreased evaporation in the simulations. Runoff was a little higher in MM5/ERA, in correspondence to smaller evaporation. This was attributed to the slightly cooler summer temperatures in the MM5/ERA simulation.
Even though snow accumulation and melt were not as dominant in the SSJ surface water budget as that in CRB during the cold season, the observation (Fig. 11) does show significant amounts of streamflow between April and June that correspond to snowmelt rather than precipitation (because there was an abrupt drop in precipitation beginning April). All regional simulations missed the streamflow during that time for the same reasons they missed the streamflow at CRB. The MM5/NRA1 simulation captured the timing and magnitude of the observed March peak runoff rather well. On an annual basis, the MM5/NRA1 simulation was about 25% lower than the observed and the other two simulations were about 50%–60% too low. The MM5/ERA simulation was better in terms of simulating the summer runoff in both CRB and SSJ. The slightly cooler temperature in the MM5/ERA simulation compared to MM5/NRA1 resulted in slightly less evapotranspiration and soil moisture depletion, which combined to result in more runoff during summer.
d. ENSO precipitation anomaly
As shown previously in Fig. 6, there were unusually large differences among the reanalyzed precipitation during 1987, and, to a lesser degree, 1988. To investigate this issue further, we focus on the 1987 conditions, which corresponded to a strong El Niño year. While the 850-mb moisture transport pattern in NRA2 during 1987 was similar to the mean conditions between 1981–93, the NRA1 pattern had a large anomaly compared to its mean conditions (not shown). During 1987, much more moisture was transported in a near-southerly direction in NRA1, which resulted in a large amount of moisture converging along the Washington and Canadian coasts. There were also stronger northwesterly flow and moisture divergence near southern California in NRA1. Difference in moisture transport in the northeast Pacific was responsible for the large difference between the NRA1 and NRA2 precipitation in the Northwest.
What effects do differences in moisture transport in the global reanalyses have on the regional simulations? Figure 12 shows the January–February–March (JFM) precipitation difference between 1987 and 1988 for each regional simulation and observation. The El Niño event during 1987 maximized during the cold season (JFM) of 1987. Beginning from the 1987 summer, sea surface temperature anomalies gradually diminished and by winter of 1988, sea surface temperatures returned to near normal, which marked the end of the El Niño event. The difference between 1987 and 1988 was, therefore, similar to the typical El Niño precipitation anomaly discussed by Leung et al. (2003c).
According to observations, more precipitation was found during 1987 over California and along the Washington and Oregon coasts. Precipitation was noticeably less along the Cascades and Rocky Mountains. As discussed by Leung et al. (2003c), the reduction was related to a change in wind direction from westerly toward southwesterly during the El Niño year, which led to reduced orographic precipitation (southwesterly is apparently not the preferred flow direction for the north–south-oriented Cascades range). On both sides of the Cascades, precipitation increased during 1987 because southwesterly flow brought in more moisture. The interesting pattern of positive–negative–positive (PNP) precipitation anomaly in that region is, therefore, associated with the interactions of large-scale circulation changes with the underlying topography.
With a regional model that can better resolve topographic features, at least compared to the global reanalyses, we find in Fig. 12 that all regional simulations generally captured the spatial distributions of precipitation anomalies very well. In particular, all simulations showed the PNP anomaly in the Cascades area. The spatial distribution was particularly well simulated by MM5/NRA1. Consistent with the smaller moisture transport in ERA, the magnitude of precipitation anomaly was less in MM5/ERA than MM5/NRA1.
Comparing the RSM and MM5 anomalies, there was a westward shift in the PNP anomaly in RSM because of the RSM topography. The wet anomalies in southern California were more realistically simulated by RSM than MM5. This may in part be related to the weaker southwesterly flows offshore of California in the NRA2 compared to NRA1 and ERA during 1987, which resulted in less precipitation in RSM/NRA2.
5. Discussion and conclusions
We have analyzed many aspects of the hydrological cycle in the western United States and how they differ among the global reanalyses, regional simulations, and observations. Overall, the regional simulations were superior in terms of simulating the spatial distributions of mean precipitation and precipitation anomalies compared to the global reanalyses. However, basin mean cold season precipitation over small basins, including CRB and SSJ, was generally amplified through downscaling in the regional models. Precipitation based on the regional simulations was typically higher than observed, while the opposite was true for the reanalyses. The amplification was largest in RSM/NRA2, which showed the biggest difference between the large-scale (NRA2) and regional (RSM/NRA2) precipitation. ERA and MM5/ERA provided the best estimate when compared to the 1/8° dataset.
It is useful to provide a cursory comparison of regional simulations of precipitation reported here and those based on previous regional climate modeling of the same region. Similar to results reported by Giorgi and Bates (1989), Giorgi et al. (1993), Pan et al. (2001), and Kim and Lee (2003), for example, all using regional climate models at 30–50-km spatial resolution, precipitation in the western United States is quite well captured spatially, except all models missed the precipitation band over narrow mountains, such as the coastal range. Most studies reported positive bias in the Pacific Northwest and negative bias in California. While we find similar positive bias in the Northwest for all regional simulations, we also find positive bias in California for two (MM5/NRA1 and RSM/NRA2) of the three regional simulations.
Pan et al. (2001) showed that differences among precipitation simulations produced by different regional models are larger along the West Coast during the cold season than in other regions of the United States. This finding is consistent with the large differences displayed in Figs. 7 and 8 among the regional simulations. As Pan et al. (2001) discussed, large differences among the simulated cold season precipitation suggest that the representation of interactions between onshore flow and topography differs among different regional models. These differences are likely related to model representation of topography, cloud and precipitation processes, and model numerics, such as spectral versus finite difference methods and representation of pressure gradient in sigma coordinates over complex terrain.
This study shows that large differences remain in estimating the water budgets of western river basins, such as the CRB and SSJ. In terms of atmospheric moisture transport, there is a 15%–20% difference in the global reanalyses over the western United States. In terms of basin mean precipitation, differences among the reanalyses, regional simulations, and observations were as large as 100% of the mean precipitation based on all estimates. In the SSJ, such differences were reflected only in the magnitude of cold season precipitation. In the CRB, differences were also displayed in the seasonal cycles as well. In terms of the spatial distribution of precipitation, large differences were found between the RSM and MM5 simulations because of terrain representations and other factors.
Runoff and snowpack were among the surface water budget components that showed the most sensitivity to model differences in spatial resolutions, physics parameterizations, and model representations. Better simulations of basin mean precipitation do not imply more superior simulations of runoff or snowpack. We find that factors such as terrain features, spatial distributions of precipitation and temperature, and partitioning of precipitation into the various components of surface water budgets were all important in determining the timing and annual amount of streamflow and snowpack. Leung and Qian (2003) performed a more detailed evaluation of the regional simulation of snowpack by comparing model simulation with observations at snow telemetry stations. They concluded that biases in temperature and precipitation explain at most 50% of the large negative bias in the snow simulation. They discussed possible deficiencies in the land surface model and biases in other simulated variables (such as radiation, wind, and temperature profile) that could be responsible for the remaining 50% or more bias in simulating snow.
To summarize the findings of this study, we plotted in Fig. 13 the differences among the reanalyses, regional simulations, and observations. We used two measures of similarity—spatial correlation and root-mean-square (rms) differences—for the comparison. They were calculated based on the DJF mean fields from each dataset, all bilinearly interpolated to the MM5 grid at 40-km resolution, for the rectangular region shown in Fig. 1. When comparing reanalyses and regional simulations, the latter were spatially smoothed to the resolution of reanalyses (2.5°) to compare the large-scale features alone. When comparing datasets at similar resolution (e.g., between reanalyses or regional simulations), comparison was done at the scale of the datasets.
For precipitation (Fig. 13a), the reanalyses were quite similar to one another although comparison of NRA1 with ERA yielded the highest spatial correlation and smallest rms difference. The spatial correlations were all higher than 0.8 and rms differences were within 40%. Among the regional simulations, the two MM5 simulations resembled each other much more than they did the RSM simulation. Comparing different pairs of regional simulations with the reanalyses used to drive the regional simulations, the ERA and MM5/ERA pair yielded the highest spatial correlation and smallest rms difference. Overall, spatial correlations of all comparisons were higher than 0.5 but rms differences ranged from 20%–80%.
The NRA2 and RSM/NRA2 pair had the lowest spatial correlation and second largest rms difference among all the pairs of reanalyses and regional simulations compared. This indicates that RSM produced precipitation at the large scale that was most different from the precipitation generated by the global model, which was used to provide large-scale conditions to the regional model. Surprisingly the RSM simulation resembled more the reanalyzed precipitations from NRA1 and ERA than NRA2.
What are the relative effects of model parameterizations and large-scale conditions in downscaling? The spatial correlations and rms differences in Fig. 13a indicate that the MM5/NRA1 and MM5/ERA pair correlated almost perfectly spatially at the regional scale (spatial correlation coefficient near 1.0 and normalized rms difference of 0.4). This implies that the downscaled precipitation carried a particular spatial signature at the mesoscale that remained nearly constant as long as the large-scale conditions used to drive the regional models were also quite similar. With different regional models, the mesoscale signatures can be very different, hence, the large differences between the MM5 and RSM simulations (spatial correlation coefficients less than 0.7 and normalized rms differences higher than 0.6). It appears that differences between MM5 and RSM created more differences among the regional simulations than different large-scale conditions.
Interestingly, at the larger scale, all regional simulations resembled the precipitation of ERA more than that of NRA1 or NRA2. This may be related to the fact that ERA was obtained using a model applied at a higher spatial resolution than NRA1 or NRA2. The large-scale features of ERA and the regional simulations, therefore, carry similar effects of mesoscale processes, such as orographic precipitation. This again points to the importance of spatial resolution in modeling climate conditions of the western United States.
Similar analyses can be provided for other variables, such as temperature, snowpack, or runoff. We compared the 850-mb winds of the reanalyses and regional simulations in order to determine whether there were stronger links between the reanalyzed and downscaled large-scale conditions than second-order variables such as precipitation. Figure 13b shows the comparison of U-component (or latitudinal) winds. While the spatial correlations were generally higher for comparing winds than precipitation, the same was true for the normalized rms differences. Among the reanalyses, NRA1 and NRA2 were more similar to one another than they were to ERA. This contrasts with the precipitation comparison where NRA1 and ERA were the most alike among the reanalyses.
Among the regional simulations, the two MM5 simulations remained the most similar, yielding a spatial correlation coefficient near 1.0 and normalized rms differences of less than 0.2. In general, the regional simulations resembled each other more than with respect to the reanalyses, even though the comparisons with reanalyses were performed at the large scales. This shows that mesoscale features were prominent in the western United States for the 850-mb wind field.
With all other comparisons yielding a low spatial correlation and large normalized rms differences, the only remaining observation that should be pointed out is the comparison of the regional simulations with the reanalyses used to drive the regional simulations. In contrast to the precipitation comparison, the NRA2 and RSM/NRA2 pair yielded the largest similarity among the three pairs of comparisons. Internal nudging applied to the RSM simulation did provide for more consistency in large-scale circulation between the regional simulation and global reanalysis. This, however, does not translate to more consistency for second-order variables, such as precipitation, as shown in Fig. 13a.
Analyses performed in this study suggest that spatial resolution is perhaps one of the most important factors that distinguish the precipitation among the reanalyses, regional simulations, and observation datasets. Spatial resolution not only affects the representation of topography, which exerts a strong influence on temperature and precipitation simulation, but other components of the water budget are also affected by spatial resolution through its effects on temperature and precipitation, clouds and radiation, and other feedback mechanisms (e.g., Giorgi and Marinucci 1996; Leung and Qian 2003). Although high-resolution modeling offers a solution to yield more realistic spatially distributed surface water budgets in the mountainous western United States, precipitation still remains problematic because regional mean precipitation tends to increase as spatial resolution increases. This is similar to findings by Colle et al. (2000) and Mass et al. (2002) for real-time weather forecasts, and Leung and Qian (2003) for regional climate simulations for the Pacific Northwest. The mean precipitation bias must be somehow removed for realistic simulations of surface water budgets or streamflow forecasts.
Besides high-resolution modeling, more accurate simulations of water budgets, especially snowpack and runoff, may be possible with the use of subgrid parameterizations that account for terrain variations within climate model grid cells (e.g., Leung and Ghan 1998) or by applying climate simulations to offline land surface models with corrections for terrain effects (e.g., Leung et al. 1996). In these models, the terrain effects on temperatures can be easily accounted for in the subgrid parameterization or offline application. The latter also offers the ability to more realistically simulate streamflow over smaller basins or at shorter timescales, where river routing becomes important. However, in offline coupling of climate and hydrologic models, moisture and energy in the climate system are not strictly conserved and feedback between atmospheric and land surface processes cannot be accounted for.
Runoff represents the amount of precipitation integrated over the basin and modulated over time by snow accumulation, snowmelt, and soil moisture storage. Offline application of land surface or hydrologic models may provide an effective way of assessing the accuracy of observed and simulated precipitation at multiple time- and space scales. In this regard, different sources of “observed” (e.g., GPCP versus ⅛° gridded data) or simulated precipitation data should be evaluated with streamflow data to ensure realistic basin mean annual precipitation and a seasonal cycle.
Lastly, to more fully understand the effects of model parameterization and resolution, boundary forcing, and how they contribute to uncertainties in estimating water budgets of river basins, we stress the importance of using experimental design, such as that employed by Pan et al. (2001), to include a matrix of numerical simulations based on different regional climate models, each driven by the same set of boundary conditions. A coordinated effort on model intercomparison and diagnosis can provide the information needed to quantitatively evaluate different models or modeling approaches and define research needs for the future.
Acknowledgments
Funding for this study was provided by the Department of Energy (DOE) Accelerated Climate Prediction Initiative (ACPI), the National Oceanic and Atmospheric Administration (NOAA) Global Energy and Water Cycle Experiment (GEWEX) Americas Prediction Program (GAPP), and the Joint Institute for the Study of Atmosphere and Oceans (JISAO) Climate Impacts Group (CIG) at the University of Washington. The ACPI pilot effort was supported largely by the DOE Office of Biological and Environmental Research. The CIG is supported by the NOAA Office of Global Programs. All MM5 regional climate simulations reported in this study were performed on 64 processors of an IBM-SP3 at the Center for Computational Sciences (CCS), which is supported by the DOE Office of Science, at the Oak Ridge National Laboratory (ORNL). The RSM simulation was supported by a grant from NOAA, under Grant NA17R1231 and by a grant from the DOE ACPI project DOE DE-FG03-98ER62605. Pacific Northwest National Laboratory is operated for the U.S. Department of Energy by Battelle Memorial Institute under Contract DE-AC06-76RLO 1830.
REFERENCES
Bajwa, R. S., Crosswhite W. M. , Hostetler J. E. , and Wright O. W. , 1992: Agricultural irrigation and water use. Agricultural Information Bulletin 638, ERS/USDA, 128 pp.
Barnett, T. P., Malone R. , Pennell W. T. , Stammer D. , Semtner B. , and Washington W. M. , 2003: Project overview and summary. Climatic Change, in press.
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 , 569–586.
Chou, M-D., 1992: A solar radiation model for use in climate studies. J. Atmos. Sci., 49 , 762–772.
Colle, B. A., Mass C. F. , and Westrick K. J. , 2000: MM5 precipitation verification over the Pacific Northwest during the 1997–99 cool seasons. Wea. Forecasting, 15 , 730–744.
Daly, C., Neilson R. P. , and Philips D. L. , 1994: A statistical–topographic model for mapping climatological precipitation over mountainous terrain. J. Appl. Meteor., 33 , 140–158.
Fels, S. B., and Schwarzkopf M. D. , 1975: The simplified exchange approximation: A new method for radiative transfer calculations. J. Atmos. Sci., 32 , 2265–2297.
Gibson, J. K., Kallberg P. , Uppala S. , Hernandez A. , Nomura A. , and Serrano E. , 1997: ERA Description. ECMWF Reanalysis Project Report Series, Vol. 1, ECMWF, 72 pp.
Giorgi, F., and Bates G. T. , 1989: The climatological skill of a regional model over complex terrain. Mon. Wea. Rev., 117 , 2325–2347.
Giorgi, F., and Marinucci M. R. , 1996: An investigation of the sensitivity of simulated precipitation to model resolution and its implications for climate studies. Mon. Wea. Rev., 124 , 148–166.
Giorgi, F., Bates G. T. , and Nieman S. J. , 1993: The multiyear surface climatology of a regional atmospheric model over the western United States. J. Climate, 6 , 75–95.
Giorgi, F., Ding Y. , Griggs D. J. , Nogue M. , van der Linden P. J. , Dai X. , Maskell K. , and Johnson C. A. , 2001: Regional climate information—Evaluation and projections. Climate Change 2001: The Scientific Basis, J. T. Houghton et al., Eds., Cambridge University Press, 583–638.
Grell, G., Dudhia J. , and Stauffer D. R. , 1993: A description of the fifth generation Penn State/NCAR Mesoscale Model (MM5). NCAR Tech. Note NCAR/TN-398 + 1A, 107 pp.
Han, J., and Roads J. O. , 2003: U.S. climate sensitivity simulated with the NCEP Regional Spectral Model. Climatic Change, in press.
Hnilo, J. J., Santer B. D. , Boyle J. , Taylor K. E. , and Doutriaux C. , 1999: Research activities at the Program for Climate Model Diagnosis and Intercomparison. Preprints, Second Int. Conf. on Reanalyses, Reading, United Kingdom WCRP.
Hong, S-Y., and Pan H-L. , 1996: Nonlocal boundary layer vertical diffusion in a medium-range forecast model. Mon. Wea. Rev., 124 , 2322–2339.
Huffman, G. J., and Coauthors, 1997: The Global Precipitation Climatology Project (GPCP) combined precipitation dataset. Bull. Amer. Meteor. Soc., 78 , 5–20.
Juang, H-M., and Kanamitsu M. , 1994: The NMC nested regional spectral model. Mon. Wea. Rev., 122 , 3–26.
Kain, J. S., and Fritsch J. M. , 1993: Convective parameterization in mesoscale models: The Kain–Fritsch scheme. The Representation of Cumulus Convection in Numerical Models, Meteor. Monogr., No. 46, Amer. Meteor. Soc., 165–170.
Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77 , 437–471.
Kanamitsu, M., 2000: Latest GSM/RSM system. Preprints, Second Int. RSM Workshop, Maui, HI, MHPCC.
Kanamitsu, M., Ebisuzaki W. , Woollen J. , Yang S-K. , Hnilo J. , Fiorino M. , and Potter J. , 2002: NCEP–DOE AMIP-II reanalysis (R-2). Bull. Amer. Meteor. Soc., 83 , 1631–1643.
Kim, J., and Lee J-E. , 2003: A multiyear regional climate hindcast for the western United States using the Mesoscale Atmospheric Simulation model. J. Hydrometeor., 4 , 878–890.
Leung, L. R., and Ghan S. J. , 1998: Parameterizing subgrid orographic precipitation and surface cover in climate models. Mon. Wea. Rev., 126 , 3271–3291.
Leung, L. R., and Ghan S. J. , 1999: Pacific Northwest climate sensitivity simulated by a regional climate model driven by a GCM. Part II: 2 × CO2 simulations. J. Climate, 12 , 2031–2053.
Leung, L. R., and Wigmosta M. S. , 1999: Potential climate change impacts on mountain watersheds in the Pacific Northwest. J. Amer. Water Resour. Assoc., 35 , 1463–1471.
Leung, L. R., and Qian Y. , 2003: The sensitivity of precipitation and snowpack simulations to model resolution via nesting in regions of complex terrain. J. Hydrometeor., 4 , 1025–1043.
Leung, L. R., Wigmosta M. S. , Ghan S. J. , Epstein D. J. , and Vail L. W. , 1996: Application of a subgrid orographic precipitation/surface hydrology scheme to a mountain watershed. J. Geophys. Res., 101 (D8) 12803–12818.
Leung, L. R., Mearns L. O. , Giorgi F. , and Wilby R. , 2003a: Workshop on regional climate research: Needs and opportunities. Bull. Amer. Meteor. Soc., 84 , 89–95.
Leung, L. R., Qian Y. , and Bian X. , 2003b: Hydroclimate of the western United States based on observations and regional climate simulation of 1981–2000. Part I: Seasonal statistics. J. Climate, 16 , 1892–1911.
Leung, L. R., Qian Y. , Bian X. , and Hunt A. , 2003c: Hydroclimate of the western United States based on observations and regional climate simulation of 1981–2000. Part II: Mesoscale ENSO anomalies. J. Climate, 16 , 1912–1928.
Leung, L. R., Qian Y. , Bian X. , Washington W. M. , Han J. , and Roads J. O. , 2003d: Mid-century ensemble regional climate change scenarios for the western United States. Climatic Change, in press.
Mahrt, L., and Pan H-L. , 1984: A two layer model of soil hydrology. Bound.-Layer Meteor., 29 , 1–20.
Mass, C. F., Ovens D. , Westrick K. , and Colle B. A. , 2002: Does increasing horizontal resolution produce more skillful forecasts? Bull. Amer. Meteor. Soc., 83 , 407–430.
Maurer, E. P., Wood A. W. , Adam J. C. , Lettenmaier D. P. , and Nijssen B. , 2002: A long-term hydrologically based dataset of land surface fluxes and states for the conterminous United States. J. Climate, 15 , 3237–3251.
Mile, E. L., Snover A. K. , Hamlet A. F. , Callahan B. , and Fluharty D. , 2000: Pacific Northwest regional assessment: The impacts of climate variability and climate change on the water resources of the Columbia River basin. J. Amer. Water Resour. Assoc., 36 , 399–420.
Mlawer, E. J., Taubman S. J. , Brown P. D. , Iacono M. J. , and Clough S. A. , 1997: Radiative transfer for inhomogeneous atmospheres: RRTM, a validated correlated-k model for the longwave. J. Geophys. Res., 102 (D14) 16663–16682.
Pan, H-L., and Wu W-S. , 1995: Implementing a mass flux convective parameterization package for the NMC Medium-Range Forecast Model. NMC Office Note 409, 40 pp. [Available from NCEP/EMC, 5200 Auth Road, Camp Springs, MD 20746.].
Pan, Z., Christensen J. H. , Arritt R. W. , Gutowski W. J. Jr., Takle E. S. , and Otieno F. , 2001: Evaluation of uncertainties in regional climate change simulation. J. Geophys. Res., 106 , 17735–17751.
Reisner, J., Rasmussen R. J. , and Bruintjes R. T. , 1998: Explicit forecasting of supercooled liquid water in winter storms using the MM5 mesoscale model. Quart. J. Roy. Meteor. Soc., 124B , 1071–1107.
Roads, J. O., 2002: Closing the water budget. GEWEX Newsletter, Vol. 12, No. 1, World Climate Research Programme, 12 pp.
Roads, J. O., and Betts A. , 2000: NCEP–NCAR and ECMWF Reanalysis surface water and energy budgets for the Mississippi River basin. J. Hydrometeor., 1 , 88–94.
Stephens, G. L., Ackerman S. , and Smith E. A. , 1984: A shortwave parameterization revised to improve cloud absorption. J. Atmos. Sci., 41 , 687–690.
Tiedtke, M., 1983: The sensitivity of the time-mean large-scale flow to cumulus convection in the ECMWF model. Proc. ECMWF Workshop on Convection in Large-Scale Models, Reading, United Kingdom, ECMWF, 297–316.
Westrick, K., Mass C. , and Colle B. , 1999: The limitations of the WSR-88D radar network for quantitative precipitation measurement over the coastal western United States. Bull. Amer. Meteor. Soc., 80 , 2289–2298.
Surface elevation (m) used in the (a) MM5 and (b) RSM simulations at 40- and 50-km resolution, respectively, for the western United States. Also shown in heavier black lines in (a) are outlines of the CRB and SSJ and a rectangular box corresponding to an area for the moisture flux analysis shown in Fig. 6
Citation: Journal of Hydrometeorology 4, 6; 10.1175/1525-7541(2003)004<1067:IOGRAR>2.0.CO;2
Spatial distributions of atmospheric moisture flux (kg kg−1 m s−1) and flux divergence (10−6 kg kg−1 s−1) at 850 mb in (a) NRA1, (b) NRA2, and (c) ERA during winter (DJF). Moisture flux is indicated by arrows and moisture flux divergence is shown by the contours with intervals of 0.03. The zero contour is highlighted
Citation: Journal of Hydrometeorology 4, 6; 10.1175/1525-7541(2003)004<1067:IOGRAR>2.0.CO;2
Surface elevation (m) of the three global reanalyses; (a) NRA1, (b) NRA2, and (c) ERA
Citation: Journal of Hydrometeorology 4, 6; 10.1175/1525-7541(2003)004<1067:IOGRAR>2.0.CO;2
Height–latitude plots of horizontal wind (UV in m s−1) and relative humidity (RH in %) in the global reanalyses (a) NRA1, (b) NRA2, and (c) ERA along 125°W. Winds are shown in vectors (e.g., westerly pointing eastward, and northerly pointing upward) and relative humidity in contours with intervals of 10%
Citation: Journal of Hydrometeorology 4, 6; 10.1175/1525-7541(2003)004<1067:IOGRAR>2.0.CO;2
Spatial distribution of mean cold season (DJF) precipitation (mm day−1) based on global reanalyses (a) NRA1, (b) ERA, (c) NRA2, and (d) observations (GPCP)
Citation: Journal of Hydrometeorology 4, 6; 10.1175/1525-7541(2003)004<1067:IOGRAR>2.0.CO;2
(upper) Mean DJF moisture flux from each side of the rectangular box (see Fig. 1) within the MM5 domain for each year between 1981–93 based on the global reanalyses, and (lower) DJF moisture convergence and regional mean precipitation within the rectangular box
Citation: Journal of Hydrometeorology 4, 6; 10.1175/1525-7541(2003)004<1067:IOGRAR>2.0.CO;2
Spatial distribution of cold season mean precipitation (mm day−1) based on the three regional climate simulations (a) MM5/NRA1, (b) MM5/ERA, (c) RSM/NRA2, and (d) ⅛° observations
Citation: Journal of Hydrometeorology 4, 6; 10.1175/1525-7541(2003)004<1067:IOGRAR>2.0.CO;2
Mean monthly precipitation (mm day−1) based on observations, regional climate simulations, and global reanalyses in the (upper) CRB and (lower) SSJ basins
Citation: Journal of Hydrometeorology 4, 6; 10.1175/1525-7541(2003)004<1067:IOGRAR>2.0.CO;2
Cold season mean precipitation (mm day−1) and surface elevation (100 m) across two east–west transects at (left column) 47.8° and (right column) 42.6°N based on observations and three regional climate simulations. Also shown are the ratios of precipitation to elevation.
Citation: Journal of Hydrometeorology 4, 6; 10.1175/1525-7541(2003)004<1067:IOGRAR>2.0.CO;2
Surface water budgets in the (left) CRB and (right) SSJ based on the regional climate simulations (top) MM5/NRA1, (middle) RSM/NRA2, (bottom) MM5/ERA. Each plot shows mean monthly precipitation (P), snow reduction rate (SP), soil moisture reduction rate (SM), evaporation (ET), and runoff (R) in mm day−1 averaged over the basins
Citation: Journal of Hydrometeorology 4, 6; 10.1175/1525-7541(2003)004<1067:IOGRAR>2.0.CO;2
Basin mean monthly runoff (mm day−1) based on observations and regional simulations in (upper) CRB and (lower) SSJ
Citation: Journal of Hydrometeorology 4, 6; 10.1175/1525-7541(2003)004<1067:IOGRAR>2.0.CO;2
Difference between the mean 1987 and 1988 JFM precipitation (mm day−1) based on (d) observations and (a), (b), and (c) regional climate simulations
Citation: Journal of Hydrometeorology 4, 6; 10.1175/1525-7541(2003)004<1067:IOGRAR>2.0.CO;2
Spatial correlation coefficients and rms differences comparing pairs of reanalyzed and/or downscaled mean DJF (a) precipitation and (b) 850 mb U-component wind over the western United States within the rectangular box shown in Fig. 1. The rms differences were normalized by the region mean precipitation averaged over all reanalyses and regional simulations. Each small box in the figure corresponds to a comparison of a reanalysis or regional simulation, which are indicated by the two side-by-side boxes. Two datasets that are similar (high spatial correlation and low rms difference) are plotted in locations closer to the origin
Citation: Journal of Hydrometeorology 4, 6; 10.1175/1525-7541(2003)004<1067:IOGRAR>2.0.CO;2
Model configurations used in the MM5 and RSM simulations
Mean cold season (DJF) precipitation and temperature based on the various global reanalyses and regional simulations averaged over CRB and SSJ. The numbers shown inside the parenthesis are percentage bias for precipitation and bias for temperature calculated based on the difference between each dataset and the ⅛° observation