The Operational Eta Model Precipitation and Surface Hydrologic Cycle of the Columbia and Colorado Basins

Yan Luo Department of Meteorology/ESSIC, University of Maryland, College Park, College Park, Maryland

Search for other papers by Yan Luo in
Current site
Google Scholar
PubMed
Close
,
Ernesto H. Berbery Department of Meteorology/ESSIC, University of Maryland, College Park, College Park, Maryland

Search for other papers by Ernesto H. Berbery in
Current site
Google Scholar
PubMed
Close
, and
Kenneth E. Mitchell NOAA/Environmental Modeling Center, National Centers for Environmental Prediction, Camp Springs, Maryland

Search for other papers by Kenneth E. Mitchell in
Current site
Google Scholar
PubMed
Close
Full access

Abstract

The surface hydrology of the United States’ western basins is investigated using the National Centers for Environmental Prediction operational Eta Model forecasts. During recent years the model has been subject to changes and upgrades that positively affected its performance. These effects on the surface hydrologic cycle are discussed by analyzing the period June 1995–May 2003. Prior to the model assessment, three gauge-based precipitation analyses that are potential sources of model validation are appraised. A fairly large disparity between the gridded precipitation analyses is found in the long-term area averages over the Columbia basin (∼23% difference) and over the Colorado basin (∼12% difference). These discrepancies are due to the type of analysis scheme employed and whether an orographic correction was applied.

The basin-averaged Eta Model precipitation forecasts correlate well with the observations at monthly time scales and, after 1999, show a small bias. Over the Columbia basin, the model precipitation bias is typically positive. This bias is significantly smaller with respect to orographically corrected precipitation analyses, indicating that the model’s large-scale precipitation processes respond reasonably well to orographic effects, though manifesting a higher bias during the cool season. Over the Colorado basin, the model precipitation bias is typically negative, and notably more so with respect to 1) the orographically corrected precipitation analyses and 2) the warm season, indicating shortfalls in the convection scheme over arid high mountains.

The mean fields of the hydrological variables in the Eta Model are in qualitative agreement with those from the Variable Infiltration Capacity (VIC) macroscale hydrologic model at regional-to-large scales. As expected, the largest differences are found near mountains and the western coastline. While the mean fields of precipitation, evaporation, runoff, and normalized soil moisture are in general agreement, important differences arise in their mean annual cycle over the two basins: snowmelt in the Eta Model precedes that of VIC by 2 months, and this phase shift is also reflected in the other variables. In the last 3–4 yr of the study period, notable improvements are evident in the quality of the model’s precipitation forecast and in the reduction of the residual term of the surface water balance, suggesting that at least similar (or better) quality will be found in studies based on NCEP’s recently completed Eta Model–based North American regional reanalysis.

Corresponding author address: Ernesto Hugo Berbery, Department of Meteorology/ESSIC, 3427 Computer and Space Sciences Building, University of Maryland, College Park, College Park, MD 20742-2425. Email: berbery@atmos.umd.edu

Abstract

The surface hydrology of the United States’ western basins is investigated using the National Centers for Environmental Prediction operational Eta Model forecasts. During recent years the model has been subject to changes and upgrades that positively affected its performance. These effects on the surface hydrologic cycle are discussed by analyzing the period June 1995–May 2003. Prior to the model assessment, three gauge-based precipitation analyses that are potential sources of model validation are appraised. A fairly large disparity between the gridded precipitation analyses is found in the long-term area averages over the Columbia basin (∼23% difference) and over the Colorado basin (∼12% difference). These discrepancies are due to the type of analysis scheme employed and whether an orographic correction was applied.

The basin-averaged Eta Model precipitation forecasts correlate well with the observations at monthly time scales and, after 1999, show a small bias. Over the Columbia basin, the model precipitation bias is typically positive. This bias is significantly smaller with respect to orographically corrected precipitation analyses, indicating that the model’s large-scale precipitation processes respond reasonably well to orographic effects, though manifesting a higher bias during the cool season. Over the Colorado basin, the model precipitation bias is typically negative, and notably more so with respect to 1) the orographically corrected precipitation analyses and 2) the warm season, indicating shortfalls in the convection scheme over arid high mountains.

The mean fields of the hydrological variables in the Eta Model are in qualitative agreement with those from the Variable Infiltration Capacity (VIC) macroscale hydrologic model at regional-to-large scales. As expected, the largest differences are found near mountains and the western coastline. While the mean fields of precipitation, evaporation, runoff, and normalized soil moisture are in general agreement, important differences arise in their mean annual cycle over the two basins: snowmelt in the Eta Model precedes that of VIC by 2 months, and this phase shift is also reflected in the other variables. In the last 3–4 yr of the study period, notable improvements are evident in the quality of the model’s precipitation forecast and in the reduction of the residual term of the surface water balance, suggesting that at least similar (or better) quality will be found in studies based on NCEP’s recently completed Eta Model–based North American regional reanalysis.

Corresponding author address: Ernesto Hugo Berbery, Department of Meteorology/ESSIC, 3427 Computer and Space Sciences Building, University of Maryland, College Park, College Park, MD 20742-2425. Email: berbery@atmos.umd.edu

1. Introduction

The Global Energy and Water Cycle Experiment (GEWEX) Americas Prediction Project (GAPP) has chosen the Columbia and Colorado basins as study areas because of the role the hydrologic cycle plays in the scarce water resources of the western United States. The Columbia River is the third largest river system in the United States. It flows from the North American continent into the North Pacific Ocean, and its 668 000-km2 basin (Fig. 1) covers portions of seven western states and the Canadian province of British Columbia, draining about 85% of the northwestern part of the country. The basin’s climate, strongly affected by complex orography, is partly continental and partly marine. The hydrology of the Columbia River basin is dominated by winter snow accumulation (Leung and Ghan 1998) as the region receives less than 20% of the precipitation during June–August (Pulwarty and Redmond 1997). Thus, the Columbia River is primarily a snowmelt-driven system; it has relatively high runoff per unit area and low reservoir storage relative to the mean annual inflow (Payne et al. 2004).

The Colorado River originates in the Rocky Mountains and flows generally west and south, discharging into the Gulf of California. The Colorado basin (Fig. 1) covers about 637 000 km2 and spreads over the southwestern United States and a small portion of Mexico. Much of the basin is arid, and runoff derives from the high-elevation snowpack over the Rocky Mountains, which contributes about 70% of the annual runoff (Christensen et al. 2004). Thus, the Colorado basin hydrology is also heavily dominated by winter snow accumulation and spring snowmelt. A secondary source of water for the basin is the summer monsoon precipitation, which although it is largest over northwestern Mexico, extends over the southwestern states. The Colorado River system is also one of the most heavily regulated for providing water supply, irrigation, flood control, and hydropower to a large area of the U.S. Southwest (Christensen et al. 2004).

Many of the relevant hydrologic variables for these basins are either not measurable or poorly measured. In some situations, topography and geography distribution have important impacts on the water cycle. First, precipitation is measured at irregular and widely spaced stations in gauges, and these gauges may notably underestimate the precipitation, owing to the undercatch effect of wind on the precipitation, especially snowfall (Groisman and Legates 1994; Adam and Lettenmaier 2003). Second, in the mountainous western United States, most of the long-term precipitation stations are located in valleys (see, e.g., Daly et al. 1994; Roads et al. 1994). Since snowfall increases rapidly with elevation in most mountain areas of the West (Daly et al. 1994), precipitation over complex terrain tends to have systematic biases and needs orographic adjustment. Further, characterization of the surface hydrologic cycle requires adequate long-term records of not only precipitation but also runoff and evaporation, but such records are unfortunately lacking. Given the number of deficiencies that prevent even a qualitative closure of the water and energy budgets from observations alone, model-based four-dimensional data assimilation procedures and forecasts are required to attain more reliable results. Thus, model-generated data is a useful augmentation to observed data. Regional model simulations over the western United States focusing on hydrologic aspects (e.g., Kim and Lee 2003; Leung et al. 2003) have shown that it is possible to achieve a better depiction of the spatial structure and amplitude of precipitation than with global analyses, although this improvement does not translate necessarily to other variables derived from the model’s land surface representation. Moreover, the choice of convective scheme in models has a large influence on surface water terms like runoff, which depends more on individual storm precipitation than on monthly totals (Gochis et al. 2003). Still, a mesoscale model provides comprehensive hydroclimatic output that is a supplement to (but not replacement for) meager observations.

Conversely, understanding the hydrologic cycle is a needed step to improve modeling of seasonal and interannual variability associated with observed soil moisture anomalies. Hydrologic components can also be used to assess the ability of a forecast model to estimate the energy and water balances on river basin scales. Current efforts to better diagnose the hydrologic cycle remain focused on both observations and modeling. For example, Maurer et al. (2002) inferred evaporation, runoff, and soil moisture from the Variable Infiltration Capacity (VIC) land surface model in uncoupled mode using observed precipitation and temperature as input data. The National Centers for Environmental Prediction (NCEP) coupled, operational, mesoscale Eta Model also has a land surface model, known as “Noah,” which has been similarly applied to hydrologic assessments in both coupled systems (Berbery et al. 2003) and uncoupled systems (Mitchell et al. 2004). A recent development with the coupled Eta/Noah system is NCEP’s completion of the North American regional reanalysis (NARR), which consists of 25 yr of Eta Model–based assimilation and forecast products. The NARR system is based on the frozen version of the operational Eta Model and its companion Eta Model–based Data Assimilation System (EDAS) as of April 2003, although some changes were added to optimize the data assimilation system. Unlike the global model reanalyses, NARR assimilates precipitation, which is expected to give better estimates of other hydrologic variables.

The primary purpose of this study is to assess the performance of NCEP’s operational Eta Model free forecasts for studies of the surface branch of the hydrologic cycle over the western United States, in particular the Columbia and Colorado basins. One objective of this study is to help detect potential inaccuracies in the model physical parameterizations and to provide an estimate of the reliability of the surface water cycle, especially over complex terrain. To assist this objective, we first compare three different gauge-based precipitation analyses, which we then use for model validation. Additionally, our model assessment includes comparison with the hydrological fields estimated by the uncoupled, observation-forced, VIC model, described briefly in section 2. The Eta Model is a fully coupled model that takes into account atmospheric and surface processes (through its land surface model), while VIC is an uncoupled hydrologic model. Because the latter, although not free of errors, is forced by observations corrected for topographic effects, we may expect that it can provide additional insight to understand the Eta Model diagnostics. A similar diagnosis of VIC and Eta Model products for the subbasins of the Mississippi was recently presented in Berbery et al. (2003). Although the hydrometeorologic behavior of the Mississippi River basin differs considerably from that of the western basins, some of the model issues are general in nature as addressed later. Others, in particular the errors in the representation of the solid precipitation processes in the model, are much more important for the Columbia basin. Examination of the Eta Model forecasts can reveal characteristic features of the water cycle and point out some of the serious issues that still affect the ability to develop adequate surface water budgets over large-scale river basins.

The analysis of the Eta Model operational products helps us understand how different changes in the Eta Model suite during the late 1990s and early 2000s improved the Eta Model hydrologic estimates of the western basins, and thus such research illustrates the improved model and assimilation behavior to be expected from the NARR. Having in hand the present study and the aforementioned Eta Model–based Mississippi basin study of Berbery et al. (2003) as benchmarks, we plan to carry out a follow-on NARR-based study of these same basins (Mississippi, Columbia, and Colorado).

The Eta and VIC model datasets and observation-based precipitation analyses used in this study are introduced in section 2. The analysis of observed precipitation and the evaluation of model precipitation for the 8-yr period June 1995–May 2003 are presented in sections 3 and 4. Notably, these sections show that the implementation of continuous soil moisture cycling in the EDAS in early June 1998 led to significant improvements in Eta Model surface hydrologic products. With June 1998 in mind, the analysis in section 5 of the basin-average climatology of Eta Model surface hydrology is based on the 5 yr (June 1998–May 2003) following this milestone. The summary and conclusions follow in section 6.

2. Datasets

a. Eta Model products

The primary dataset for this study consists of NCEP mesoscale Eta Model operational forecasts for the period of 1995–2003 over North and Central America. The model domain covers all North and Central America, and extends well over the oceans [past Hawaii to the west and the middle of the Atlantic to the east (see Fig. 1a of Berbery 2001)]. The Eta Model forecasts were initialized with the Eta Model’s own four-dimensional data assimilation system, known as EDAS. The Eta forecast model has been employed at NCEP operationally for over 10 yr, and the EDAS assimilation suite was implemented operationally in April 1995 (Rogers et al. 1996).

The Eta Model and companion EDAS include a land surface model (Ek et al. 2003), known as “Noah,” which applies the energy and water balance equations at every land grid point and produces surface variables such as evaporation, runoff, soil moisture, and snow water equivalent that are consistent with the surface forcing from the atmospheric component of the Eta Model or EDAS. The Noah land model can be executed in coupled or uncoupled mode. Here, we assess the coupled mode, while Mitchell et al. (2004) assess the uncoupled mode.

In this study the 12–36-h Eta free forecasts are used to produce an 8-yr (June 1995–May 2003) climatology of the model precipitation and evaporation (throughout this article the term evaporation is used in lieu of the more precise evapotranspiration). Additionally, estimates of runoff, soil moisture, and snow for June 1998–May 2003 are employed to develop a 5-yr climatology of the surface water cycle for the western basins. We chose to assess the surface hydrologic budget of the free forecasts instead of the assimilation output fields of the EDAS to avoid spinup in the model precipitation. Such spinup is nonnegligible in the first 6–12 h following assimilation, owing to the model’s adjustment to assimilated observations, especially the adjustment of atmospheric divergence and vertical motion to the underlying orography.

We next summarize the changes to Eta Model spatial resolution during the 8-yr study period here. From 1995 to 1997, the Eta Model was run at 48 km and 38 levels, and for that period our diagnostic computations for this study were performed on the model’s native grid. On February 1998, September 2000, and November 2001, the model resolution increased, respectively, to 32 km/45 levels, 22 km/50 levels, and 12 km/60 levels. Each resolution increase implied large increases in the volume of model output, which prevented NCEP from storing it in a public place for retrieval by the users. Hence for model output from February 1998 and thereafter, our analysis was performed on a 40-km output grid, known at NCEP as Advanced Weather Interactive Processing System (AWIPS 212), and used as a standard grid for the GEWEX Continental-scale International Project (GCIP) archive.

From inception of the EDAS in April 1995 until June 1998, the EDAS was configured as a 12-h preforecast suite consisting of only four successive 3-h assimilation periods that culminated in the land and atmospheric initial conditions for the Eta Model forecast. However, land and atmosphere initial conditions for the start of this 12-h EDAS was the NCEP Global Data Assimilation System (GDAS). Beginning in early June 1998, the EDAS configuration was changed to a continuous chain of successive 3-h assimilation periods, referred to here as “continuous cycling,” independent of the GDAS (except for lateral boundary conditions). The land and atmospheric states in the continuously cycled EDAS evolve in a self-contained, uninterrupted fashion as a product of Eta Model physics and assimilated data. However, a fire in the NCEP main computer in September 1999 interrupted the cycling, which was restarted again in December 1999 and has since run continuously. The model has undergone significant changes along the years, in particular in its atmospheric physical package, land surface model, and data assimilation system. The more important changes can be found in Table 1 and a discussion of those before 2002 is presented in Berbery et al. (2003). Model changes at NCEP are grouped; that is, a “bundled” set of changes is put into operations simultaneously on a given date. Thus it is difficult to ascertain what the impact of an individual change in the bundle is on the forecasts.

Of the more recent period, the comprehensive changes to the land surface model (July 2001) and cloud microphysics (November 2001) can be considered major; changes after 2001 include modifications to the subsurface thermal conductivity over patchy and full snow cover (February 2002), and modifications to the land surface physics to avoid slightly negative soil moisture availability under very dry soil conditions (June 2002). Later changes, although listed in Table 1, do not cover the period employed in this study. The complete log of model changes is available at http://www.emc.ncep.noaa.gov/mmb/research/eta.log.html. The reader is referred to Janjić (1990, 1994), Black (1994), Betts et al. (1997), Chen et al. 1997, Rogers et al. (1999, 2001a, b), Berbery et al. (2003), and Ek et al. (2003) for additional discussions of the model changes and their impacts. The operational EDAS began precipitation assimilation (hourly) in late July 2001, using radar-dominated, nonorographically corrected precipitation estimates (Y. Lin 2004, personal communication; http://wwwt.emc.ncep.noaa.gov/mmb/papers/lin/pcpasm/paper.html). Given that radar-based precipitation estimates cannot reliably infer frozen precipitation, the operational EDAS omits precipitation assimilation at any grid point where the model simulation indicates frozen precipitation. (Aside: The NARR assimilates gauge-based orographically corrected precipitation and does not apply any exclusion for frozen precipitation.)

b. Evaluation products

For model validation and assessment, this study applies three different rain gauge–based, gridded precipitation analyses, as well as hydrologic fields estimated from uncoupled, observation-forced executions of the VIC land surface model.

1) Precipitation analyses

Table 2 summarizes the characteristics of the three precipitation analyses, which are all daily analyses based on gauge observations. The methodological differences described below in the production of these precipitation analyses result in noticeable differences among the corresponding precipitation fields, an important aspect to consider when evaluating the Eta Model products. As noted in Table 2, none of the three precipitation analyses employ corrections for snow undercatch by gauges. The winter underestimation of solid and liquid precipitation is a known problem, and some corrections have been suggested (e.g., Groisman and Legates 1994). More recently, Adam and Lettenmaier (2003) developed a method for adjustment of gridded precipitation datasets to account for gauge catch deficiencies, especially of solid precipitation. Their adjustment method yields increases of about 11.7% in global terrestrial precipitation, and probably higher at regional scales. As will be discussed later, this is a notable caveat for both the Columbia and Colorado River basins, which experience significant snowfall in winter.

The first analysis, designated PCPC, is a conterminous United States (CONUS)-only analysis prepared on a 0.25° × 0.25° grid by NCEP’s Climate Prediction Center (CPC) using a Cressman analysis scheme to interpolate the gauge observations to the grid (Higgins et al. 2000). This analysis scheme utilizes a radius of influence, which is a function of the average separation distance between reporting stations and with an upper bound of 200 km that is sufficiently large to ensure that no landmass grid points are left undefined. The PCPC analysis is the only one of the three that omits orographic adjustments. This is of particular concern for the mountainous Columbia and Colorado River basins.

The second analysis, designated POROCPC, is also a CONUS-only analysis prepared by CPC, but on a higher-resolution 0.125° × 0.125° grid (about 14 km), utilizing a different analysis scheme and including an orographic correction procedure that employs the Parameter-elevation Regressions on Independent Slopes Model (PRISM). PRISM is a climatology of monthly mean precipitation data for 1961–90, and observed precipitation, even outside that period, was scaled with these values. The analysis scheme (J. Schaake 2004, personal communication) is an inverse square-distance interpolation algorithm applied with a short influence radius of 50 km. The methodology of PRISM and a general discussion of precipitation gradients with respect to topography can be found in Daly et al. (1994). The PRISM is important for developing more reliable estimates of precipitation in the mountainous western United States.

Until recently, the PCPC and POROCPC analyses for dates during or before December 1998 employed the observations received by CPC in both real time and retrospectively. The combined set of real-time and retrospective observations is designated by CPC as the “Unified Raingauge Data” (URD). On typical days, the URD station count CONUS-wide ranges between 11 000 and 13 000 and includes the observations of the National Oceanic and Atmospheric Administration (NOAA) Cooperative Observer Program (Co-op). For dates after December 1998, the PCPC and POROCPC analyses employ only observations received by CPC in real time, which on typical days number between 6000 and 7000 CONUS-wide and do not include the Co-op data. As stated earlier, unlike the PCPC analysis, the POROCPC analysis contains some missing values (undefined) at some grid points, owing to the relatively short influence radius (50 km) used in its analysis scheme. The smaller observation count in the CPC analyses after December 1998 increases the number of missing points in the POROCPC analysis, and this increase may impact its basin-average values over the Columbia basin (see discussion in section 3a with the support of Table 3). This version of POROCPC, with the December 1998 discontinuity in station density, was assimilated in the production phase of NCEP’s 1979–2003 regional reanalysis, for which the present study represents a benchmark. At the time of this writing, CPC just released new PCPC and POROCPC analyses for 1999–2002 based on the higher density URD. We chose to use these latter datasets, taking into account that the differences with the original version were minimal.

The third set of precipitation analyses, designated PUW, was developed and produced by the University of Washington (on the same 0.125° × 0.125° grid as POROCPC). Original data were first gridded, and, as described in Maurer et al. (2002), scaled to be consistent with the PRISM 12-month climatology using the analysis scheme of Shepard (1984) as implemented by Widmann and Bretherton (2000). This analysis scheme utilizes an inverse square-distance weighting algorithm with a dynamic radius of influence that increases until four reporting stations are attained. The dynamic treatment of influence radius in PUW is sufficient to ensure that no landmass grid points are left undefined. The rainfall Co-op dataset was employed for the United States, while a sparser distribution of rain gauges was employed for Canada (which could lead to a lower quality analysis over that region). The PUW analyses were used as the precipitation forcing for the offline, uncoupled executions of the VIC model described next.

It should be noted that the adjustment for orographic effects, derived from a 1960–90 climatology, should be taken as a first step toward understanding what may be missed if a precipitation dataset (e.g., the widely used PCPC) is employed over mountainous regions without further examination. In the end it would be desirable to have orographic corrections that adjust to the monthly or daily values, rather than to a climatology.

2) VIC model products

The VIC is a macroscale land surface hydrologic model described in detail by Liang et al. (1994, 1996). It balances both energy and water at every grid point over its chosen execution domain at 0.125° × 0.125° grid spacing. The VIC model thereby provides surface variables such as evaporation, runoff, soil moisture, and snow water equivalent depth that are consistent with the external forcing and model surface energy and water balance equations. Maurer et al. (2002) developed a 50-yr (January 1950–July 2000) dataset of observation-based, 3-hourly land surface forcing fields, including the PUW precipitation analyses of the prior section, and used them to drive an offline, uncoupled VIC execution at 3-hourly time steps over the cited period. It should be stressed that while the VIC products are self-consistent and satisfy water and energy balance, they are not free of uncertainties.

In the Mitchell et al. (2004) North American Land Data Assimilation System (NLDAS) study, the VIC model showed the least bias in simulated snowpack when compared to observations over the mountainous western United States (Pan et al. 2003; Sheffield et al. 2003). For this reason, and the aforementioned observation-based nature of the surface forcing used to drive the uncoupled VIC simulations of Maurer et al., we chose the VIC dataset of Maurer et al. as the benchmark here for comparison with the Eta simulations.

This VIC-derived dataset of land surface states and fluxes serves as a reference for a wide variety of studies, especially where many observations are missing and in particular to assess model-predicted land–atmosphere exchanges of moisture and energy. Applications of the VIC model for water and energy budget studies are described in Maurer et al. (2001, 2002). Here, this dataset will be employed for a comparison with the Eta Model forecasts. The VIC dataset of Maurer et al. ends in July 2000, hence necessitating some difference between its averaging period (about 5 yr) and that of the Eta forecasts here (8 yr). From comparisons of the overlap periods (complemented with preliminary analysis of the NARR; not shown) the lack of complete overlap between the two periods does not appear to affect the chief results presented here.

The VIC executions utilized the same 0.125° × 0.125° grid spacing and nearly identical domain of the 3-yr NLDAS project (Mitchell et al. 2004). The NLDAS initiative performed uncoupled executions and comparison of four land models (including Noah and VIC) under common surface forcing and common fields of vegetation and soil class (for the period October 1996–September 1999) and found wide disparities among the four land models in runoff, evaporation, and snowmelt.

All NLDAS models were forced by near-surface meteorology from EDAS, except for 1) the precipitation forcing, taken from the PCPC analyses disaggregated to hourly intervals using radar-based precipitation analyses, and 2) satellite-based solar insolation (Cosgrove et al. 2003). Since NLDAS precipitation forcing is based on PCPC, NLDAS does not benefit from any PRISM orographic correction. This is a chief reason for a low midwinter snowpack bias in all four NLDAS models over the high elevations of the western United States (Pan et al. 2003). In addition to Mitchell et al. (2004), the following NLDAS articles are relevant for the analysis of the land surface water budget: Schaake et al. (2004) for the discussion of soil moisture; Pan et al. (2003) and Sheffield et al. (2003) for evaluations of snow water equivalent and snow-cover extent; and Lohmann et al. (2004) for analysis of runoff and streamflow.

In summary, the choice of VIC products in our article was, first, because of their smaller bias in snowpack; second, because they cover a longer period (January 1950–July 2000) than the NLDAS models (October 1996–September 1999); and third, because unlike NLDAS, VIC used orographically corrected precipitation, which is very important for the hydrology of the western United States.

3. Precipitation estimates

Evaluation of the Eta Model 12–36-h precipitation forecasts over the western United States was done by means of comparison with the observation-based precipitation analyses of section 2b. We thus began with the assessment and comparison of these precipitation analyses.

a. Time-mean basin-scale precipitation

Table 3 shows the precipitation estimates for the two western basins. Unlike PUW and PETA, the PCPC and POROCPC estimates do not include the Canadian portion of the Columbia basin. Therefore, for comparison purposes, Table 3 presents PUW and PETA with and without the Canadian portion. The area averages without Canada’s precipitation can be directly compared to the CPC estimates, while those that include the whole basin (last two rows of Table 3) are consistent with the other surface water terms to be discussed in section 5c. The Columbia-averaged precipitation PUW is significantly larger than PCPC and POROCPC for the period June 1995–May 2000. The mean difference between the two is larger over the Columbia basin than over the Colorado, reflecting that topography effects are larger in the former than the latter. The June 1995–May 2000 average of POROCPC for the Columbia basin is larger than PCPC, but the relation is inverted in the last three years (June 2000–May 2003). This contradicts what would be expected of the orography correction. Moreover, a decreasing trend was noted in POROCPC. As explained in section 2b, fewer rai gauges were included after 2002 in both CPC precipitation analyses. In turn, due to the short radius of influence in the analysis scheme of POROCPC, its interpolation technique may have resulted in an increased number of missing values; however, this does not seem to be the cause of the trend. A preliminary analysis of the NARR (not included in this study) reveals that the decreasing trend for POROCPC was translated to NARR, which can only be noticed in the Columbia and Missouri basins (both have important topographic effects).

The PETA estimate over the Columbia is significantly larger (+27%) than PCPC, and about the same as PUW for the first 5 yr of the sample (the difference with PCPC is reduced to 18% in the last 3 yr). While the closeness to PUW may be fortuitous, it suggests that the large-scale component (the primary cool-season process) of the model precipitation is responding adequately to the complex orography over the Columbia basin, at least in a time-mean area-average sense. Thus we have presented an example of a high-resolution, dynamic free forecast that can better capture orographic precipitation than a naïve analysis (PCPC) of valley-dominated observations, provided the precipitation is not convection dominated. All these estimates were done for the area average of the U. S. portion of the Columbia basin. However, the water budget requires having an estimate for the whole basin. The last two rows in Table 3 show that significant precipitation occurs over Canada’s portion of the basin, which is reflected in larger averages for both PETA and PUW.

In contrast, over the Colorado basin, during June 1995–May 2000, the mean PETA was smaller than PCPC (−32%) and PUW (−41%). In general, there was a dry bias of the model over the southwest United States, which would suggest that in this case the convective parameterization scheme—or its triggering function—are the ones that do not respond adequately. Note that when considering the last 3 yr, the differences are reduced, and actually PETA becomes larger than PCPC or POROCPC. These results highlight the model improvement in precipitation physics.

b. Observed precipitation

Figure 2 depicts the June 1995–May 2000 5-yr annual mean CPC precipitation analysis (PCPC and POROCPC) and the University of Washington precipitation analysis (PUW), as well as their differences, over the western United States (recall that PUW was not available after mid-2000). The three precipitation analyses (Figs. 2a–c) are characterized by large values over the central Columbia basin, along the coastlines (including the Olympic Mountains), over the western slopes of the Cascades and the Sierra Nevada. The differences between the analyses (Figs. 2d–f) are small over flat areas, but the orography correction in POROCPC and PUW is evident over the slopes of the Cascades, the Rockies, and the Sierra Nevada. In general, if the orography-corrected precipitation analyses are taken to be closer to reality, then PCPC underestimates the real precipitation over most of the western United States because of its prevalent complex terrain. In addition, the differences between POROCPC and PUW (Fig. 2f) show that there are important uncertainties still in gauge-based precipitation analyses, despite the correction for orography effects. These differences are of special significance to illustrate the large sensitivity of the monthly precipitation to the topography and northern cold-weather conditions, and they reflect the difficulties of estimating reliably the precipitation over the Columbia basin.

The extent and annual evolution of the differences in precipitation analyses over the Northwest are illustrated further with a Hovmoeller diagram (Fig. 3) for the regional sector between 125° and 115°W at 48°N, where differences in the Columbia basin are largest. The differences of PCPC minus PUW in Fig. 3a are systematically negative and, according to the topography cross section at the bottom of Fig. 3, achieve their largest magnitudes (up to 6–10 mm day−1) during the cold months over the highest terrain (especially the high coastal terrain of the Olympic Mountains). Replacing PCPC with the orographically adjusted POROCPC in Fig. 3b yields a substantial reduction over high terrain in the difference from the orographically adjusted PUW. Comparison of Figs. 3a and 3b highlight the impact of the orographic correction.

c. Eta Model forecast precipitation

The model forecast precipitation is assessed for two different periods to show its progress in the recent years. Figure 4 presents the June 1995–May 2000 averages (as Fig. 2), while Fig. 5 shows the averages for June 2000–May 2003 (no PUW was available, though). The model forecast precipitation for the first period (Fig. 4) captures all the regional features depicted by the precipitation analyses over the northwestern United States, and the intensity over mountains lies between the gridded analyses. In general, the Eta Model has a wet bias in the northern sector (particularly near valleys) and a dry bias toward the south. The larger differences (Figs. 4b–d) are noted in the mountainous west including most of California and the Columbia River basin, where the snowfall is more prevalent. According to Fig. 4c, the biases increase slightly with the orographic correction to the CPC analyses (POROCPC) and become even larger in the case of using PUW (Fig. 4d). In other words, during June 1995–May 2000 the Eta Model tended to produce excessive precipitation when compared to the two CPC analyses, while the differences with PUW show that the positive bias is only found in the western sector of the basin. On the other hand, the difference with respect to all three precipitation analyses reveals a dry bias over the Colorado basin. Similarly, a dry bias is observed over California, with the exception of the Central Valley.

According to Figs. 4b–d, the Eta Model forecast precipitation tends to differ more from all the precipitation analyses (PCPC, POROCPC, and PUW) over the Columbia basin than over the Colorado basin. As it happened with the precipitation analyses, away from the northern region and coastal areas, the differences are less pronounced bringing closer all estimates.

We recall from section 2a that the horizontal resolution of the Eta Model increased from 48 to 32 km in February 1998, then to 22 km in late September 2000, and finally to its present resolution of 12 km in late November 2001. Also at the latter time, the explicit microphysics in the Eta Model was substantially upgraded. Figure 5 presents the results for the period June 2000–May 2003 as a counterpart of Fig. 4. (As several more years pass, it will be interesting to reproduce these figures at the model’s present 12-km computational grid, rather than the 40–48-km output grid resolutions used in the present study.) At first sight the mean field for June 2000–May 2003 (Fig. 5a) resembles the mean of the previous years (Fig. 4a). However, when computing the biases with respect to PCPC and POROCPC (Figs. 5b,c), the reduction in bias magnitude becomes evident. There is a slight wet bias in general, but in most areas it is small and largest values are about 1 mm day−1 (but recall that these precipitation analyses are lower than PUW).

The latter 2 yr of the time series in Fig. 6 provide a foretaste of the improvements in Eta Model precipitation forecasts obtained over rugged orography since November 2001 (albeit still derived in Fig. 6 from coarse-resolution 40–48-km model output fields). Figure 6 shows the Hovmoeller diagram of the differences between the model forecast and the precipitation analyses at 48°N for the entire 8-yr study period. The model precipitation bias depicts a distinct seasonal and topographic variation, showing a dry bias at the highest elevations that is most pronounced in the cold season, and a generally wet (dry) bias at lower elevations in the cold (warm) season (tending in later years to a somewhat wet bias throughout the year at lower elevations). The difference between PETA and PCPC (Fig. 6a) shows a decreasing trend in either the dry or wet biases along the years, particularly after mid-1999, and again after late 2001. This trend illustrates the progressive reduction in forecast error over the Columbia basin resulting from the model changes. In late July 2001, EDAS began assimilating precipitation, which was not orographically corrected. Nevertheless, the model bias with respect to POROCPC (Fig. 6b), although larger than that with respect to PCPC also shows the decreasing trend.

While NCEP tests have shown that the direct impact of precipitation assimilation on Eta Model precipitation forecasts is negligible in a given forecast after 3–6 h of forecast length, there is significant indirect impact owing to the precipitation assimilation improving the short-term and long-term evolution of the soil moisture in the continuously cycled EDAS, which initializes the forecast. The improved soil moisture evolution improves the forecast of land surface fluxes, which in turn impact all the model physical processes. Unlike the precipitation analysis assimilated in the operational EDAS, the NARR assimilates precipitation that has been orographically corrected (POROCPC). Therefore NARR soil moisture evolution and land surface fluxes over mountainous regions like the northwest United States may be superior to that in the operational EDAS (a subject for future analysis).

The amplitude of the cold-season differences between PETA and PUW (Fig. 6c) is larger than that for the CPC analyses (Figs. 6a,b), but all figures show similar seasonal variations and location. One final aspect worth mentioning is the drought in the western United States during the summer of 1996 (U.S. Department of Commerce/NOAA/USDA 1996, 12–13). While the observed precipitation was low, the Eta Model had practically no precipitation at all, as if the now-documented model dry bias during summer in the West was accentuated by the extreme conditions. This behavior of virtually no precipitation in the western U.S. summer drought was not repeated in later years, suggesting that the changes in the model system addressed this limitation.

d. Basin-scale precipitation variability

The Columbia basin area-averaged time series of precipitation are shown in Fig. 7a (POROCPC is not included because it shows variability similar to the other analyses; also, PUW ended in July 2000). Note that CPC analyses do not cover Canada; therefore a portion of the Columbia basin is not taken into account in the PCPC average. Given that precipitation over Canada is not small, this may be another reason for the lower magnitude of the area- average PCPC and POROCPC when compared to PUW. To some degree, the biases noted in the spatial fields tend to compensate each other in the area averages.

The time series of PETA, PCPC, and PUW have a consistently similar evolution, with all showing a larger amplitude of the annual cycle during the first half of the period. Here PETA, although larger than the precipitation analyses (consistently so in the cold season, and less so in the warm season), seems to reproduce consistently the month-to-month variability. Before mid-1999, according to Figs. 7b and 7c, the model biases have large month-to-month variability and discrepancies in magnitude. The large positive differences and large root-mean-square error (rmse) mainly occur in the wintertime over the Columbia basin. In contrast, from mid-1999 onward, the analysis and forecast precipitation show notably closer agreement. From near March 1999 to early 2000 there is a remarkable reduction in the rmse of the model forecast with respect to the CPC precipitation analysis.

Because of relatively lower summer season PETA biases, it is unclear at what precise moment after mid-1999 the model started performing better. Additionally, the fact that model upgrades are done in “bundles” further complicates determining a unique reason for the improvement. Nevertheless, the changes that may have been relevant appear to be circumscribed by the period spring 1999–autumn 2000. According to Rogers et al. (1999), the implementation of the 3DVAR in November 1998 had led to a degradation of the forecasts, but subsequent changes to the 3DVAR in May 1999 to better balance wind and mass fields resulted in a significant improvement in the 48-h forecasts. Our results are consistent with their assertion and suggest that the NARR products, based on a frozen Eta suite from near April 2003, will provide a rather more realistic surface water cycle over the western basins.

The drier nature of the Colorado basin is evident in Fig. 7d. Compared to the Columbia basin, the Colorado basin has a much weaker seasonal cycle and a smaller magnitude of precipitation. Also, the analyses of precipitation (from CPC and UW) and the forecast precipitation (PETA) tend to be much closer over the Colorado basin, again reflecting the fact that despite complex terrain being a factor in precipitation estimation, it is less relevant than over the Columbia. Although the model has a tendency to overestimate the Columbia basin area-averaged precipitation, it underestimates the Colorado basin area averages. According to Figs. 7e and 7f, there is better agreement since late 1999, but unlike in the Columbia basin, where large biases were found during winter, the major discrepancies over the Colorado basin occur mostly during the summer months.

4. The mean annual cycle of precipitation

From the previous analysis, it is clear that the model precipitation biases have well defined annual cycles, which are different for the two basins and the pre- and post-1999–2000 periods. To further illustrate the model performance in different periods, the 4-yr climatology from June 1995 to May 1999 is compared to the 4-yr climatology from June 1999 to May 2003, for the Columbia basin in Fig. 8 and the Colorado basin in Fig. 9. The shaded bands in Figs. 8 and 9 depict the envelope of the spread of the mean annual cycles computed from the three analyses of precipitation (PCPC, POROCPC, and PUW) averaged separately for 1995–99 and 1999–2003 (except for PUW, which is not available in the second period). This means that the envelope does not distinguish interperiod variability. However, the interperiod variability was small compared to the changes in the model precipitation (not shown).

a. The Columbia basin

The mean annual cycles for the Columbia basin are presented in Fig. 8a. Here PETA shows a maximum in December–January and a minimum in August during the two periods, which are consistent in form, if not in magnitude, with observations. The peak near December–January is associated with the large fraction of winter snowfall. For the first 4-yr average, PETA is larger than any of the analyses during the cold season. It also has a slight deficit during the summer months. Nevertheless, during the second period, there is a remarkable improvement in the quality of the Eta forecasts, as PETA falls within the range of the analyses with the exception of two months in springtime. Although the summer negative bias seems reduced as well, the major improvement occurs mostly in the winter, with PETA being reduced by about 2 mm day−1.

To better understand the change in model performance before and after 1999, Fig. 8b presents the mean annual cycle of the convective and large-scale components of the model precipitation during these two periods. Examination of the model’s partitioning of precipitation into large-scale and convective contributions reveals that on an annual basis, their ratio is about 5, which is typical of cold climate and orographically affected basins. During the cold season, the most relevant precipitation is due to large-scale processes, which account for much of the total precipitation. The convective precipitation, on the other hand, is close to zero during winter and achieves a maximum during spring. Although it is not as large during summer, it surpasses the large-scale component during July–August. Given that the convective precipitation is negligible in the winter, the deficiencies in estimating the Columbia precipitation appear to be associated with the large-scale (explicitly resolved) precipitation component.

b. The Colorado basin

The Colorado basin has a two-peak mean annual cycle of precipitation (Fig. 9a), the first one during late winter and probably due to snowstorms over the mountains, and the second in mid- to late summer associated with the onset of the monsoon season. Regardless of the origin, the June 1995–May 1999 average shows that the model had a significant deficit of precipitation from May to November. As in the case of the Columbia basin, the second period (June 1999–May 2003) presents a notable improvement, especially in the warm season, with values well within the range of the precipitation analyses.

Decomposition of precipitation into large-scale and convective components (Fig. 9b) shows that summer precipitation is strongly influenced by convective processes. Therefore, it is likely that the large dry bias is associated with the convective component rather than the large-scale component, which is almost negligible. The changes in the convective parameterization scheme after 1999 were aimed at enhancing the convective precipitation component in summer and led to model precipitation estimates that were much closer to observations. Given the Eta Model’s well-known dry bias over semiarid regions traditionally apparent before and during 1999 (Berbery and Rasmusson 1999), the results suggest an improvement in the forecasts quality after 1999. They also agree with Gochis et al. (2002) who find the region’s precipitation highly sensitive to the convective scheme employed by mesoscale models in general. It also suggests that appropriate adjustments to the precipitation parameterization scheme applied over different regions will be a critical factor in improving the representation of the precipitation processes and thereby the hydrologic cycle in regional mesoscales.

In summary, the dominance of convective processes over large-scale processes in the Colorado basin is in contrast to the dominance of large-scale processes over convective ones in the Columbia basin, affecting the improvement of Eta Model performance in different ways. Although the model precipitation in both basins has topography-related problems, they can be traced to the different scales of the precipitation processes in the Eta Model (explicitly resolved precipitation microphysics versus convective parameterization) acting under significantly different climate regimes.

The quality of the land surface water budgets depends on the reliable estimate of basin-averaged precipitation. In both basins, but more importantly for the Columbia basin, the real magnitude of the model bias cannot be ascertained because of the disparity between the observational estimates. Which precipitation estimates can provide more realistic precipitation depictions for model validation continues to be a subject of debate.

5. Land surface water budgets

Because of the nonexistence of observations of many surface hydrological variables, the spatial and temporal structures of the surface hydrologic cycle from the Eta Model are assessed using VIC’s products. The rationale for choosing the VIC-based products to assess the Eta Model was presented in section 2b(2).

a. Annual mean fields

Figures 10a and 10b depict the June 1995–May 2000 mean annual fields of the Eta and the VIC model evaporation over the western United States. The VIC model evaporation (Fig. 10b) shows greater detail and sharper gradients due to the smaller 1/8° grid spacing in the VIC executions (and the coarse 40–48-km Eta output fields utilized in this study). The Eta Model 12–36-h evaporation forecast (Fig. 10a) has values ranging between 0.5 and 3 mm day−1 with the largest values toward the southeast, and reveals a clear bias that is more evident in the Southwest. The Eta Model tends to have a slightly larger evaporation toward Oklahoma–Kansas and smaller evaporation near Oregon and the coastal areas of Washington State. Over the Columbia basin, evaporation is about 1.5 mm day−1 in the central part and decreasing toward the higher elevations. However, for the later period of June 2000–May 2003 (Fig. 10c), which follows significant upgrades to the Eta Model, the Eta Model 12–36-h evaporation reveals a much closer resemblance to the VIC estimate. These differences are not due to the different period considered, as an examination of the NARR evaporation shows no noticeable interperiod variability (not shown).

Figure 11 presents the mean annual fields of the other surface hydrologic variables as produced by the Eta Model and VIC model parameterizations. In this case, the Eta Model averages are performed for the 5-yr period of June 1998–May 2003, to avoid the earlier period when EDAS (and the land states therein) did not continuously cycle (see section 2a) and the land surface parameterizations in the Eta Model underwent substantial changes (other important changes to the land surface model were made in late July 2001). VIC data are available only until July 2000; therefore the VIC mean fields in Fig. 11 are based on the earlier 5-yr period of June 1995–May 2000. Comparisons for the overlap period (not shown) verify that the dissimilar periods do not alter the discussion that follows. (We preferred not to present VIC averages based on earlier years in the 50-yr VIC time series, in order to avoid possible further sources of difference from encompassing extreme interannual variability associated with decadal regime changes.) As in the case of the evaporation field, the Eta Model has most regional-to-large-scale aspects of the surface hydrologic components in common with the VIC model, like the location of the maxima and their relation to the mountains. Some particular features of each field are discussed next.

On a daily basis since the mid-1990s, the EDAS system has assimilated (via direct replacement) the daily, 47-km, operational global snow depth analysis of the U.S. Air Force, after quality controlling its areal extent of snow cover with the daily, 23-km, operational Northern Hemisphere snow-cover analysis of the National Environmental Satellite, Data, and Information Service (NESDIS) (Ek et al. 2003). Hence the Eta Model 12–36-h forecasts of snow water equivalent depth presented here are not primarily products of the model’s land surface physics. The Eta Model snow water equivalent (SWE) depth has large values exceeding 50–100 mm over mountainous regions corresponding to the Cascades, the Rockies, the Wasatch, and the Sierra Nevada mountains. Consequently, deep snow accumulation is dominant in most of the Columbia basin and northern part of the Colorado basin (Figs. 11a,d). The corresponding magnitude of the SWE depth in the VIC model is larger over the mountain ranges and is more localized (Fig. 11d).

Past studies that have intercompared soil moisture values across land models have shown that large differences in absolute soil moisture often exist, even when the models are supplied with the same surface forcing and soil and vegetation types (Schlosser et al. 2000; Robock et al. 2003; Schaake et al. 2004). This is the case of the soil moisture derived from the Eta Model’s land surface model (Noah) and the VIC land surface model, as illustrated in Mitchell et al. (2004). Koster and Milly (1997) show that such large differences in absolute soil moisture across land models emerge primarily from differences in three model-assigned characteristics: 1) the maximum capacity of soil moisture, and the upper and lower thresholds of soil moisture that control 2) transpiration and 3) surface infiltration of precipitation. On the other hand, the aforementioned studies show that the temporal changes of soil moisture or soil moisture values normalized by annual means or minimum–maximum range show considerably more agreement. Therefore, the soil moisture fields of the two models were normalized by their respective minimum–maximum ranges following the equation SMNORM = (SM SMMIN)/(SMMAX SMMIN), where SM means soil moisture, and NORM, MIN, and MAX refer to the normalized, minimum, and maximum values either in space (Fig. 12) or time (Fig. 13). A discussion on a similar normalization is given in Maurer et al. (2001), while a comparison of soil wetness indices can be found in Saleem and Salvucci (2002).

The relative content of Eta Model–produced soil moisture is presented in Fig. 11b. The southern region is the driest, affecting the southern portion of the Colorado basin, while the Columbia basin is the one with highest soil moisture. The Eta Model soil moisture field reproduces many of VIC’s large-scale soil moisture maxima, including those within the Columbia basin, despite some discrepancies with respect to extent and magnitude. For example, large values are found in VIC’s results over Texas, Oklahoma, and—more relevant for this study—over Arizona. Also, some other small-scale maxima noticed in the VIC model are not well captured in the Eta Model estimates (Fig. 11e), which here are based on 40–48-km coarse-resolution model output. Recall that while the Eta Model grid spacing was changed several times and currently is 12 km, the output utilized in this study was interpolated to a 40 km × 40 km grid, while VIC has a grid spacing of 0.125° × 0.125° (about 14 km).

The Eta Model forecast runoff (Fig. 11c) is largest near the Sierra Nevada slopes toward the Central Valley in California, the Cascades, the Wasatch Mountains, and the Rockies. It can also be seen that runoff in the Columbia basin originates over the northern Rockies and Cascade Mountains, while that of the Colorado basin originates over the southern Rockies, but also the Wasatch Mountains. As with the other variables, the differences between Eta and VIC are mostly in magnitude and extent of the maxima. The resemblance to the VIC estimates is encouraging: the large runoff over the high mountainous areas is clearly seen in both fields (Figs. 11c,f), although the VIC model tends to produce patchy-like patterns with a smaller extent of maximum values. VIC’s larger (but more localized) values were also noted by Lohmann et al. (2004), who also found large intermodel differences in runoff among the four land surface models in NLDAS.

The fields of precipitation, snow accumulation, and runoff and even soil moisture share common locations of their respective maximum centers, implying their close connections. In other words, where there is strong precipitation (heavier snowfall) over the northern high-mountainous regions, the later melting of the deep snow accumulation results in large runoff and increased soil moisture in these same regions.

b. Basin-scale estimates

The multiyear monthly time series and mean annual cycle of the monthly and basin-averaged surface water cycle terms are presented in Figs. 12 and 13 for the Columbia and Colorado basins, respectively. Given the short period in common between the VIC and Eta estimates, the former are presented beginning in June 1995 to give a better sense of the typical amplitude and changes in the corresponding annual cycles. The time series of Eta Model evaporation was reconstructed from the native grid data from 1995 to give a sense of its changes and variability. As stated earlier, the snowpack cover and depth in the EDAS are updated by direct replacement once daily from an external analysis, as described in section 5a. Hence, the initial snowpack state in the initial conditions of each Eta 12–36-h forecast encompasses much of the content of the once daily external update.

The Eta Model’s water equivalent of accumulated snow (Fig. 12a) has nonzero values starting in October, achieves a maximum of about 85 mm in February, and then depletes rapidly through April. The VIC model estimate achieves a substantially larger and later winter accumulation of snowpack than the Eta Model, reaching a maximum of about 170 mm in March, and then it decays slowly and extends even into early summer. The VIC model’s nonzero values during the warm season indicate the presence of not fully melted snow. VIC has subgrid elevation bands and vegetation tiles that enable it to retain mountain elevation snowpack in summer (see Sheffield et al. 2003; Mitchell et al. 2004, section 3.4). Compared to VIC, the Eta Model snow water equivalent depth has a negative bias that is largest during late winter and spring.

The results are consistent with the uncoupled tests of the Noah land model and its comparison with VIC in the NLDAS project. Pan et al. (2003), as summarized also in Mitchell et al. (2004), found that compared to high-elevation snowpack telemetry (SNOTEL) measurements, Noah had the largest low bias in late-winter and early-spring snow water equivalent among the four NLDAS models. In the uncoupled setting of NLDAS, Noah also manifested an early snow depletion bias due to large snowmelt and snow sublimation in late winter and early spring, while VIC had its largest snowpack depletion later in late spring, as is also evident here in our analysis in Fig. 11. Recently, following the evaluation in this study and in the cited NLDAS studies, the early snowpack depletion bias has been eliminated in the Noah land surface component of the Eta Model by 1) identifying and solving a low bias in the formulation for snowpack albedo and 2) introducing a subgrid treatment for patchy snowpack in the calculation of snow sublimation. Unfortunately, the latter improvements in Noah snowpack physics (Noah versions 2.7.1 and later) were not sufficiently refined and tested in time to be implemented into the production of the NARR.

As the Eta Model snow accumulates (Fig. 12a) in the northwest and western mountains, runoff (Fig. 12b) starts increasing slowly and peaks in spring (April) when snowmelt is largest. The Eta runoff then decays to low values in June, and remains low until the following winter. Therefore, the timing of the Eta runoff annual cycle is closely associated with that in snowmelt. The Eta Model runoff peaks 2 months before VIC’s runoff (June), and its maximum is double that in the VIC maximum.

The annual cycle and amplitude of VIC’s runoff is similar to the observed runoff (as derived from streamflow) presented in Fig. 11a of Leung et al. (2003). Additionally, the early Eta Model peak in runoff seems to be common to other mesoscale models as well. For example, Leung et al. (2003) (also in their Fig. 11a) show that simulations with the Regional Spectral Model (RSM) and the fifth-generation Pennsylvania State University–National Center for Atmospheric Research (PSU–NCAR) Mesoscale Model (MM5) have maximum runoff in March, and in fact they precede the Eta Model by 1 month, therefore having a slightly wider gap with either VIC’s runoff or observations.

We next examine soil moisture of both models (Fig. 12c) normalized here for comparison purposes. The normalization is done by taking, for each model, the range between the minimum and maximum values in the basin-averaged time series. The two models show a well-defined mean annual cycle of soil moisture, also achieving a maximum in spring, about 2 months after the maximum in snow, and simultaneous with the maximum runoff. Then, soil moisture decays monotonically until October, due to the increasing evaporation (see Fig. 12d) and reduced precipitation during summer. The VIC model soil moisture, like its runoff, tends to have a later peak (in June) because its snowmelt processes last longer.

Figure 12d shows that although the order of magnitude of the Eta Model basin-averaged evaporation is similar to VIC’s, discrepancies are found in the phasing of its seasonal cycle. The Eta Model estimate is too high (compared to VIC’s) during spring but too low during autumn (except for winter, when the two are at a minimum). This difference manifests itself in the Eta Model calculations by shifting the peak 2 months earlier with respect to VIC model estimates. Notice, however, that in the comparison of four NLDAS models, VIC tended to have somewhat lower evaporation and peak later than the three other models (Figs. 5 and 9 of Mitchell et al. 2004), suggesting that the correct values will be somewhere in between the two estimates.

Interannual variability includes natural variability, and in the case of the Eta Model, the effect of model changes. The time series of the basin averaged variables exhibit changes along the years that may be the result of changes in the model parameterization, but others occur consistently among all variables and observations, suggesting a manifestation of the natural variability. The Eta Model snow depth (Fig. 12e) increases until 2001, but then a decrease in magnitude is noted for the latter two years. The VIC model snow depth does not have the same increasing trend, and moreover, the time series of precipitation (Fig. 7a) does not have that type of behavior. Therefore, it is likely that the interannual variability of the Eta Model snow was mostly dominated by either changes in the external analysis of snow depth and snow cover assimilated daily into the EDAS or the parameterization changes that took place in July 2001, when substantial upgrades were done to the treatment of snowpack physics, including the addition of frozen soil physics.

The Eta Model runoff (Fig. 12f) is largest during the spring of 1999, and the springtime peaks decrease during the following 2 yr; the later years (2002 and 2003) show a slight increase. Soil moisture (Fig. 12g) shows consistently large peaks in 1999 and 2002. In these two cases, the year-to-year evolution of both runoff and soil moisture is similar to that of the observed precipitation (Fig. 7a), therefore making it more likely that changes were at least in part due to natural variability. Along the years there is a slight increase in the amplitude of the evaporation annual cycle (Fig. 12h) that seems to be caused both by increases of evaporation during summer and reduced values of approaching zero during winter.

In comparison with the Columbia basin, the Colorado basin exhibits a fairly weak mean annual cycle of all variables (Figs. 13a–d). As in the Columbia basin, the area-averaged mean annual cycles of the surface variables have a consistent evolution. There is a very close relationship among precipitation and snow accumulation, as well as runoff. Once again in the Colorado basin, a 1–2-month shift in the phase is evident between the Eta Model and VIC model components of the surface hydrologic cycle. Although the Eta Model captures the basic pattern of variability, it tends to highly overestimate the magnitude in the surface hydrological variables, particularly snow accumulation and runoff. It should be noted that in addition to the snow accumulation (Fig. 13a), this basin has a second peak of precipitation due to monsoonal effects during the warm months (Fig. 9a). Therefore, unlike in the Columbia basin, the annual cycles of runoff and soil moisture (Figs. 13b,c) also reflect this second source of water. This is particularly evident for soil moisture. The soil moisture normalization may give the wrong impression that values are “high” during most of the year, but the actual magnitude of the soil moisture in both models is smaller than in the Columbia basin.

In the case of the Eta Model over the Colorado basin, interannual variability again exposes model changes mixed with natural variability. Given the different climate regime that controls the hydrology of this basin, changes do not necessarily follow those in the Columbia basin. Among the more relevant aspects, the peak in snow accumulation in 2001 (Fig. 13e) does not have equivalently large peaks in the other variables. The Colorado basin runoff (Fig. 13f) presents low values in 2001 and 2002.

The soil moisture evolution in Fig. 13g manifests a number of Eta Model land model changes that have occurred over the years, including the continuous cycling of soil moisture that began in the EDAS in early June 1998. This continuous cycling was broken by the NCEP Cray fire in the fall of 1999. Figure 13g vividly illustrates that it took a full year for the Eta Model/EDAS soil moisture to spin up and resume a quasi-uniform annual cycle behavior. We note that the Columbia basin, which has more moisture availability, does not show as significant a spinup effect as the drier Colorado. Finally, evaporation peaks become smaller around 2002 (Fig. 13h). Many of these changes cannot be traced easily to either natural variability or model changes. Rather, the time series are presented here more as an indicator of the impacts from the accumulation of periodic changes, rather than to link each feature in the plots to specific causes.

c. The water balance terms

Table 4 summarizes the components of the surface hydrology (except precipitation, which was discussed earlier with the support of Table 3). According to Table 4, the Columbia basin mean annual evaporation from the Eta and VIC models agree to within ∼14% for the 5-yr average covering June 1995–May 2000, but the Eta Model value is reduced for the last 3 yr (no VIC data are available to verify it, but this value is closer to the average of the previous period). Over the Colorado basin, the Eta Model evaporation is larger than VIC’s by 50% for the first five years of the analysis, but the EETA values are reduced significantly for the latter 3-yr period.

VIC runoff has been adjusted to be consistent with observed streamflow (e.g., Leung et al. 2003). Eta Model and VIC model runoff and snow water equivalent cannot be easily compared directly because they do not have a large period in common. While in both cases, and for the respective basins, the values are of the same order of magnitude, still the differences are considerable. For runoff, this difference is most noticeable in the Colorado basin, where the Eta peak runoff is nearly double that of VIC, while for snow, the biggest difference is in the Columbia basin. The snow water equivalent depth estimated from the Eta Model in the Columbia basin is 43.3 mm smaller than the VIC model, while for the Colorado basin they differ by 1.9 mm. However, both models produce deep snow accumulation and rather wet conditions over the Columbia basin and thinner snow accumulation and drier conditions over the Colorado basin.

The progress illustrated thus far in estimating the surface hydrologic cycle is substantiated further in Fig. 14, which presents the residual of the water balance equation. First, most flat areas are close to balance (no colors) with a residual that is less than 0.5 mm day−1 in magnitude. Imbalances with a positive residual are found over regions with high orography, while a negative residual is found along the northwest coast. As a percentage with respect to precipitation (Fig. 14b) the residual is less than 20% in magnitude over flat areas, increasing to between 20% and 80% over mountains.

The latter large residual over mountains (Fig. 14a) arises not from an omission in the surface water balance calculations internally within the Noah land model, but rather from the source/sink of water arising from the daily update (replacement) of the snowpack state in the EDAS from the external daily snow depth analysis described in section 5a. Calculation of this update term is difficult using the content in traditional archives of Eta Model output, and it was omitted in the computations here for Fig. 14. The omission of this update term will yield the largest residuals during periods and over regions of substantial snowpack accumulation, namely over high mountain elevations. Moreover, the impact of omitting this term is amplified by the Noah model bias toward premature snowmelt (section 5b). Given the latter bias, the daily snowpack update from the external analysis will tend, more often then not, to add water content back into Eta/Noah model’s snowpack state, resulting in a usually positive residual in the storage change (as seen in Fig. 14) when not offset by subtracting the update term. (For future similar studies of the NARR water balance, the fields for calculating the daily snow update term have been included in the mainline public NARR archives.)

The area-averaged residual for the Columbia basin (Fig. 14c) shows a well-defined annual cycle with mostly positive values during spring and slightly negative values the rest of the year. It can also be noticed that in the early years there was a positive bias, and in the later years, it was removed, although individual months still may have large values. However a balance should not be expected on a monthly basis, but in longer-term averages, for example, 12-month averages. When the effect of the annual cycle is removed (by performing a 12-month running mean; heavy line) it is noticed that the residual term was at times almost as large as 2 mm day−1 before the year 2000, but it has since become smaller along the years. While not zero, since mid-2001 the values have remained of the order of 0.5 mm day−1 or less. Given the slow decreasing trend in the residuals, it is possible that there is no unique reason for these improvements, but it can be speculated that the continuous cycling of land and atmospheric states implemented in 1998 may have had a significant impact.

The residual term is also reduced for the Colorado basin (Fig. 14d) from a maximum of about 1.3 mm day−1 before the year 2000 to slightly positive values in the more recent years. Unlike the in Columbia basin, there is no well-defined annual cycle, although relatively smaller values are found in autumn.

6. Summary and conclusions

This study focuses on the operational Eta Model’s successes, improvements, and problems in producing reliable estimates of the hydrologic cycle of basins over the western United States. Previous studies have shown that this is possible to a large extent over areas not significantly affected by orography; however, characterizing all aspects of the hydrologic cycle accurately from observations and model products over complex terrain involves many challenges. Substantial uncertainties remain in the quality of the gridded precipitation analyses, as well as in the validity of surface parameterizations. The main aspects analyzed here are (a) the uncertainties in estimating precipitation either from gridded analyses of rain gauge data or models; (b) the surface water terms; and (c) an assessment of the evolution, resulting from model upgrades, in the surface hydrology of the Eta Model.

The Eta Model 12–36-h forecast precipitation is compared to three gridded precipitation analyses derived from rain gauge observations: the first one is the gauge-based gridded precipitation developed at CPC; the second one, also developed at CPC, includes orography corrections using the PRISM method; the third one, prepared at the University of Washington for input to the VIC hydrologic model, also has PRISM corrections for orographic effects. This correction increases the values of precipitation with altitude. In the case of the Mississippi basin, the differences between uncorrected and corrected precipitation analyses were negligible (Berbery et al. 2003), but toward the west the overall precipitation differences (area average, time average) are of the order of 23% over the Columbia basin, and about 12% over the Colorado basin. These discrepancies between the gridded precipitation analyses pose large uncertainties on the entire estimation of the hydrologic cycle. The Eta Model forecast precipitation has large biases with respect to either estimate in the first part of the period under study, but around 1999–2000 a noticeable reduction is observed in the magnitude of the bias and root-mean-square error. While regional differences are still large, there are compensations within each basin so that the area averages are relatively closer to orographically corrected precipitation (PUW). The origin of the model biases depends on the basin that is considered. Over the Columbia basin biases depend on the large-scale explicit component of the model precipitation during winter; over the Colorado basin, biases depend on the convective precipitation scheme that is responsible for the precipitation during summer.

In the absence of observations, the Eta Model land surface hydrological variables were compared with the VIC model variables generated from uncoupled VIC simulations driven by observation-based surface forcing. However, in the same way that precipitation estimates differ depending on whether orographic effects are taken into account, surface variables depend not only on differences in the surface forcing such as precipitation, but they also differ on the details of the physical parameterizations of such processes as evaporation, infiltration, and snowmelt. The most notable example is that of soil moisture, which had to be normalized in order to do a useful comparison.

The mean annual hydrologic fields of the Eta and VIC models bear encouraging resemblance in shape, location, and scale at regional-to-large scales, but local discrepancies exist mostly related to topography. In both the Colorado and Columbia basins, the Eta and VIC models agree reasonably during the snow accumulation phase, but in spring the Eta Model melts snow comparatively too fast. As a result the Eta Model spring runoff and soil moisture peaks occur about two months earlier than those of VIC. In addition, the Eta Model runoff, when compared to VIC’s, is too high. The amplitude of the mean annual cycle of evaporation of the two models is similar, but again, associated with the phase shift in soil moisture, the Eta Model also has an early peak of evaporation. In summary, the main differences with the VIC model are a phase shift of about 2 months, which results in large differences particularly during spring.

The difficulties in estimating the hydrologic cycle in regions like the Columbia basin are the result of the complex terrain and sparsely sampled observational data. Model parameterizations, which despite great efforts still cannot handle properly these regions of complex orography and physiography, further add to the uncertainties. However, the results presented here suggest that continued improvements have been achieved along the years, best exemplified in such basic terms as the forecast of precipitation and the reduction of the water balance residual term.

In summary, the four major results of this paper are the following:

  1. Although overall averages in the gridded precipitation analyses differ by 5%–10% over the mountain regions of the western United States, within the basins there are large geographical uncertainties.

  2. The Eta Model bias in the early period over the Columbia basin was due to a poor representation of the large-scale precipitation during winter; the bias over the Colorado basin was due to the poor representation of the convective precipitation during summer. Both biases were largely reduced after 1999–2000.

  3. The surface water balance in the Eta Model and its data assimilation system (EDAS) is not closed if one omits the source/sink of water from the daily update (replacement) of the snowpack state in the EDAS from an external snow depth analysis (section 5a). Calculation of this update term is difficult using the content in traditional archives of Eta Model output. Omitting this update term typically yields large residuals in the Eta Model monthly water balance during periods and in regions of substantial snowpack. Nevertheless, the Eta Model water balance residual decreases in recent years, likely from improvements in Eta/Noah physics yielding smaller increments from the daily snowpack update. (The fields for calculating this update term are included in the mainline NARR archives.)

  4. The evolution of snowpack in the coupled Eta/Noah model manifests a substantial early bias in the timing of late-winter to early-spring snowmelt in the western United States. Compared to similar estimates from the uncoupled VIC model, this bias results in a 1–2-month early phase shift in Eta Model timing of the annual peak of snow depth, runoff, soil moisture, and evaporation. Following this study, upgrades to Noah snowpack physics (implemented in the NCEP Eta Model in early 2005) have eliminated this substantial early snowmelt bias (not shown).

Finally, it is important to understand the limitations of our study. The short period in common between the VIC and Eta Models makes it difficult to completely assess their surface water terms. These conclusions will be further investigated as the 25-yr NARR data become available. It is expected that the NARR, with its assimilation of observed precipitation with the PRISM correction, and frozen model configuration will help identify better the structure and intensity of surface hydrologic cycles over the western basins. Careful use of the NARR products over complex terrain should take into account that the CPC orographically corrected precipitation (POROCPC) had a reduction in magnitude after 2000, probably due to the smaller number of rain gauge data, and this will translate into the NARR analyses. We believe that an effort to produce a longer period of Eta regional reanalysis data should allow a much improved description of the surface water budgets.

Acknowledgments

We thank Drs. Ed Maurer and Dennis Lettenmaier for providing the dataset of land surface fluxes and states based on the VIC model, and to Drs. Wei Shi and Wayne Higgins who provided the CPC precipitation datasets. All of them shared many discussions with us on the characteristics of the gridded precipitation analyses. The careful reading and suggestions of three anonymous reviewers helped improve the article and are greatly appreciated. This research was supported by NOAA GCIP (NA76GP0291) and GAPP (NA04OAR4310164) grants.

REFERENCES

  • Adam, J C., and Lettenmaier D L. , 2003: Bias correction of global gridded precipitation for solid precipitation undercatch. J. Geophys. Res., 108 .4257, doi:10.1029/2002JD002499.

    • Search Google Scholar
    • Export Citation
  • Berbery, E H., 2001: Mesoscale moisture analysis of the North American monsoon. J. Climate, 14 , 121137.

  • Berbery, E H., and Rasmusson E M. , 1999: Mississippi moisture budgets on regional scales. Mon. Wea. Rev., 127 , 26542673.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Berbery, E H., Luo Y. , Mitchell K E. , and Betts A K. , 2003: Eta Model-estimated land surface processes and the hydrologic cycle of the Mississippi basin. J. Geophys. Res., 108 .8852, doi:10.1029/2002JD003192.

    • Search Google Scholar
    • Export Citation
  • Betts, A K., Chen F. , Mitchell K E. , and Janjić Z I. , 1997: Assessment of the land surface and boundary layer models in the two operational versions of the NCEP Eta Model using FIFE data. Mon. Wea. Rev., 125 , 28962916.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Black, T L., 1994: The new NMC Mesoscale Eta Model: Description and forecast examples. Wea. Forecasting, 9 , 265278.

  • Chen, F., Janjić Z. , and Mitchell K. , 1997: Impact of the atmospheric surface-layer parameterizations in the new land-surface scheme of the NCEP mesoscale Eta Model. Bound.-Layer Meteor., 85 , 391421.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Christensen, N S., Wood A W. , Voisin N. , Lettenmaier D P. , and Palmer R N. , 2004: Effects of climate change on the hydrology and water resources of the Colorado River basin. Climatic Change, 62 , 337363.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Cosgrove, B A., and Coauthors, 2003: Real-time and retrospective forcing in the North American Land Data Assimilation System (NLDAS) project. J. Geophys. Res., 108 .8842, doi:10.1029/2002JD003118.

    • Search Google Scholar
    • Export Citation
  • 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 , 140158.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ek, M B., Mitchell K E. , Lin Y. , Grunmann P. , Rogers E. , Gayno G. , Koren V. , and Tarpley J D. , 2003: Implementation of Noah land surface model advances in the National Centers for Environmental Prediction operational mesoscale Eta Model. J. Geophys. Res., 108 .8851, doi:10.1029/2002JD003296.

    • Search Google Scholar
    • Export Citation
  • Gochis, D J., Shuttleworth W J. , and Yang Z-L. , 2002: Sensitivity of the modeled North American Monsoon regional climate to convective parameterization. Mon. Wea. Rev., 130 , 12821298.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Gochis, D J., Shuttleworth W J. , and Yang Z-L. , 2003: Hydrometeorological response of the modeled North American monsoon to convective parameterization. J. Hydrometeor., 4 , 235250.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Groisman, P Y., and Legates D R. , 1994: The accuracy of United States precipitation data. Bull. Amer. Meteor. Soc., 75 , 215227.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Higgins, R W., Shi W. , Yarosh E. , and Joice R. , 2000: Improved United States Precipitation Quality Control System and Analysis. NCEP/Climate Prediction Center Atlas 7, U.S. Department of Commerce, NOAA/NWS. [Available online at http://www.cpc.ncep.noaa.gov/research_papers/ncep_cpc_atlas/7/index.html.].

  • Janjić, Z I., 1990: The step-mountain coordinate: Physical package. Mon. Wea. Rev., 118 , 14291443.

  • Janjić, Z I., 1994: The step-mountain Eta coordinate model: Further developments of the convection, viscous sublayer, and turbulence closure schemes. Mon. Wea. Rev., 122 , 927945.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 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 , 878890.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Koster, R., and Milly P C. , 1997: The interplay between transpiration and runoff formulations in land surface schemes used with atmospheric models. J. Climate, 10 , 15781591.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Leung, L R., and Ghan S J. , 1998: Parameterizing subgrid orographic precipitation and surface cover in climate models. Mon. Wea. Rev., 126 , 32713291.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Leung, L R., Qian Y. , Han J. , and Roads J O. , 2003: Intercomparison of global reanalyses and regional simulations of cold season water budgets in the western United States. J. Hydrometeor., 4 , 10671087.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Liang, X., Lettenmaier D. P. , Wood E. F. , and Burges S J. , 1994: A simple hydrologically based model of land and energy fluxes for general circulation models. J. Geophys. Res., 99 , 1441514428.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Liang, X., Wood E. F. , and Lettenmaier E F. , 1996: Surface soil moisture parameterization of the VIC-2L model: Evaluation and modifications. Global Planet. Change, 13 , 195206.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Lohmann, D., and Coauthors, 2004: Streamflow and water balance intercomparisons of four land surface models in the North American Land Data Assimilation System project. J. Geophys. Res., 109 .D07S91, doi:10.1029/2003JD003517.

    • Search Google Scholar
    • Export Citation
  • Maurer, E P., O’Donnell G M. , Lettenmaier D P. , and Roads J O. , 2001: Evaluation of the land surface water budget in NCEP/NCAR and NCEP/DOE reanalyses using an off-line hydrologic model. J. Geophys. Res., 106 , D16,. 1784117862.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 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 , 32373251.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Mitchell, K E., and Coauthors, 2004: The multi-institution North American Land Data Assimilation System (NLDAS): Utilizing multiple GCIP products and partners in a continental distributed hydrological modeling system. J. Geophys. Res., 109 .D07S90, doi:10.1029/2003JD003823.

    • Search Google Scholar
    • Export Citation
  • Pan, M., and Coauthors, 2003: Snow process modeling in the North American Land Data Assimilation System (NLDAS): 2. Evaluation of model simulated snow water equivalent. J. Geophys. Res., 108 .8850, doi:10.1029/2003JD003994.

    • Search Google Scholar
    • Export Citation
  • Payne, J T., Wood A W. , Hamlet A F. , Palmer R N. , and Lettenmaier D P. , 2004: Mitigating the effects of climate change on the water resources of the Columbia River basin. Climatic Change, 62 , 233256.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Pulwarty, R S., and Redmond K T. , 1997: Climate and salmon restoration in the Columbia River basin: The role and usability of seasonal forecasts. Bull. Amer. Meteor. Soc., 78 , 381397.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Roads, J O., Chen S C. , Guetter A K. , and Georgakakos K P. , 1994: Large-scale aspects of the United States hydrologic cycle. Bull. Amer. Meteor. Soc., 75 , 15891610.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Robock, A., and Coauthors, 2003: Evaluation of the North American Land Data Assimilation System over the southern Great Plains during the warm season. J. Geophys. Res., 108 .8846, doi:10.1029/2002JD003245.

    • Search Google Scholar
    • Export Citation
  • Rogers, E., Black T L. , Deaven D G. , and DiMego G J. , 1996: Changes to the operational early Eta analysis/forecast system at the National Centers for Environmental Prediction. Wea. Forecasting, 11 , 391413.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Rogers, E., Parrish D. , and DiMego G. , 1999: Changes to the NCEP operational Eta analysis. NWS/NOAA Technical Procedures Bulletin 454, 25 pp. [Available online at http://www.nws.noaa.gov/om/tpb/454.html.].

  • Rogers, E., Black T L. , Collins W. , Manikin G. , Mesinger F. , Parrish D. , and DiMego G. , 2001a: Changes to the NCEP Meso Eta analysis and forecast system: Assimilation of satellite radiances and increase in resolution. NWS/NOAA Technical Procedures Bulletin 473. [Available online at http://www.emc.ncep.noaa.gov/mmb/mmbpll/eta22tpb/.].

  • Rogers, E., Ek M. , Lin Y. , Mitchell K. , Parrish D. , and DiMego G. , 2001b: Changes to the NCEP Meso Eta analysis and forecast system: Assimilation of observed precipitation, upgrades to land-surface physics, modified 3DVAR analysis. NWS/NOAA Technical Procedures Bulletin 479. [Available online at http://www.emc.ncep.noaa.gov/mmb/mmbpll/spring2001/tpb/.].

  • Saleem, J A., and Salvucci G D. , 2002: Comparison of soil wetness indices for inducing functional similarity of hydrologic response across sites in Illinois. J. Hydrometeor., 3 , 8091.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Schaake, J C., and Coauthors, 2004: An intercomparison of soil moisture fields in the North American Land Data Assimilation System (NLDAS). J. Geophys. Res., 109 .D01S90, doi:10.1029/2002JD003309.

    • Search Google Scholar
    • Export Citation
  • Schlosser, C A., and Coauthors, 2000: Simulations of a boreal grassland hydrology at Valdai, Russia: PILPS Phase 2(d). Mon. Wea. Rev., 128 , 301321.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sheffield, J., and Coauthors, 2003: Snow process modeling in the North American Land Data Assimilation System (NLDAS): 1. Evaluation of model-simulated snow cover extent. J. Geophys. Res., 108 .8849, doi:10.1029/2002JD003274.

    • Search Google Scholar
    • Export Citation
  • Shepard, D S., 1984: Computer mapping: The SYMAP interpolation algorithm. Spatial Statistics and Models, G. L. Gaile and C. J. Willmott, Eds., Reidel, 133–145.

    • Search Google Scholar
    • Export Citation
  • U.S. Department of Commerce/NOAA/USDA, 1996: Weekley weather and crop bulletin. Vol. 83, No. 36, 29 pp. [Available from Climate Predication Center, W/NP52, Attn: Weekly Weather & Crop Bulletin NOAA/NWS/NCEP, NOAA Science Center, Room 605, 5200 Auth Road, Camp Springs, MD 20746; also available online at http://usda.mannlib.cornell.edu/reports/waobr/weather/1996/wwcb3696.pdf.].

  • Widmann, M., and Bretherton C S. , 2000: Validation of mesoscale precipitation in the NCEP reanalysis using a new grid-cell dataset for the northwestern United States. J. Climate, 13 , 19361950.

    • Crossref
    • Search Google Scholar
    • Export Citation

Fig. 1.
Fig. 1.

The topography of the western United States and the boundaries of the Columbia and Colorado basins.

Citation: Journal of Hydrometeorology 6, 4; 10.1175/JHM435.1

Fig. 2.
Fig. 2.

Jun 1995–May 2000 annual mean fields of observed precipitation gridded analyses: (a) CPC precipitation analysis without orographic correction (PCPC); (b) CPC precipitation analysis with orographic correction (POROCPC); (c) University of Washington precipitation analysis with orographic correction (PUW); (d) difference between the two CPC analyses, POROCPCPCPC; (e) difference PCPC PUW; and (f) difference POROCPCPUW. Units are mm day−1, and the contour intervals are indicated in the bar below each panel.

Citation: Journal of Hydrometeorology 6, 4; 10.1175/JHM435.1

Fig. 3.
Fig. 3.

Hovmoeller diagram at 48°N for Jun 1995–Jul 2000 of the differences between the CPC gridded analyses of precipitation and the University of Washington gridded analysis: (a) PCPCPUW and (b) POROCPCPUW. The gray shading and dashed contours denote negative values. Contour interval is 2 mm day−1. The topographic cross section at 48°N is included for reference.

Citation: Journal of Hydrometeorology 6, 4; 10.1175/JHM435.1

Fig. 4.
Fig. 4.

Jun 1995–May 2000 mean of (a) the Eta Model 12–36-h forecast precipitation (PETA) and its difference with the three gridded analyses: (b) PETAPCPC; (c) PETAPOROCPC; and (d) PETAPUW. Units are mm day−1, and the contour intervals are indicated in the bar below each panel.

Citation: Journal of Hydrometeorology 6, 4; 10.1175/JHM435.1

Fig. 5.
Fig. 5.

Jun 2000–May 2003 mean of (a) PETA and its difference with the CPC analyses: (b) PETAPCPC and (c) PETAPOROCPC. Units are mm day−1, and the contour intervals are indicated in the bar below each panel.

Citation: Journal of Hydrometeorology 6, 4; 10.1175/JHM435.1

Fig. 6.
Fig. 6.

(a) Jun 1995–May 2003 Hovmoeller diagram at 48°N of the differences between the Eta Model forecast precipitation and the CPC analysis of precipitation without orographic correction; (b) same as (a), but with the CPC precipitation analysis corrected for topographic effects; and (c) same as (a) but with the University of Washington gridded analysis, which ends in Jul 2000. Contour interval is 1.5 mm day−1. (The blank band indicates missing data due to the fire destruction of NCEP’s computer in 1999)

Citation: Journal of Hydrometeorology 6, 4; 10.1175/JHM435.1

Fig. 7.
Fig. 7.

Jun 1995–May 2003 Columbia basin area-averaged time series of (a) Eta Model forecast precipitation, University of Washington, and CPC not orographically corrected analysis; (b) their difference; and (c) the model’s precipitation rmse. (d)–(f) Same as (a)–(c) but for the Colorado basin. Units are mm day−1.

Citation: Journal of Hydrometeorology 6, 4; 10.1175/JHM435.1

Fig. 8.
Fig. 8.

(a) Mean annual cycle of the Columbia basin area-averaged Eta forecasts of total precipitation during Jun 1995–May 1999 (solid line) and Jun 1999–May 2003 (dashed line). The gray band represents the envelope of the mean annual cycle of the three gridded analyses computed for the two periods separately (see text). (b) Mean annual cycle of the Columbia basin–averaged Eta Model precipitation components: large-scale (PLS) and convective (PCON) for the same two periods as in (a). Units are mm day−1.

Citation: Journal of Hydrometeorology 6, 4; 10.1175/JHM435.1

Fig. 9.
Fig. 9.

Same as Fig. 8 but for the Colorado basin.

Citation: Journal of Hydrometeorology 6, 4; 10.1175/JHM435.1

Fig. 10.
Fig. 10.

Annual mean fields over 5 yr (Jun 1995–May 2000) of (a) coupled Eta Model evaporation and (b) uncoupled VIC model evaporation. (c) As in (a), but for the 3-yr period Jun 2000–May 2003. Units are mm day−1.

Citation: Journal of Hydrometeorology 6, 4; 10.1175/JHM435.1

Fig. 11.
Fig. 11.

The 5-yr annual mean fields for the period Jun 1998–May 2003 of the Eta Model 12–36-h forecasts of (a) water equivalent of accumulated snow depth, (b) normalized soil moisture for the 0–200-cm layer, and (c) total runoff; (d)–(f) same as (a)–(c), but for the VIC model and for the earlier 5-yr period of Jun 1995–May 2000. Normalization of soil moisture was done by taking into account the respective minimum–maximum ranges of the Eta Model and VIC model soil moisture (see text). Units are mm day−1.

Citation: Journal of Hydrometeorology 6, 4; 10.1175/JHM435.1

Fig. 12.
Fig. 12.

Columbia basin area-averaged mean annual cycle and time series of the surface water budget components of the Eta Model: (a), (e) water-equivalent snow depth; (b), (f) runoff plus baseflow; (c), (g) basin-averaged normalized soil moisture; and (d), (h) evaporation. Units are mm day−1.

Citation: Journal of Hydrometeorology 6, 4; 10.1175/JHM435.1

Fig. 13.
Fig. 13.

Same as Fig. 12 but for the Colorado basin.

Citation: Journal of Hydrometeorology 6, 4; 10.1175/JHM435.1

Fig. 14.
Fig. 14.

The residual term of the water balance equation estimated from the Eta Model: (a) the mean field for Jun 1998–May 2003, (b) the same residual as a percentage of the precipitation, (c) the area average for the Columbia basin, and (d) the area average for the Colorado basin. The heavy line in (c) and (d) represents a running mean that spans 12 months to remove the annual cycle. Here, dW/dt is the local change of surface water (soil moisture and snow water equivalent), P is the precipitation, E is the evaporation, and N is the runoff plus the baseflow. Units are mm day−1.

Citation: Journal of Hydrometeorology 6, 4; 10.1175/JHM435.1

Table 1.

Significant Eta Model changes during May 1995–Mar 2004.

Table 1.
Table 2.

Characteristics of the gridded precipitation analyses.

Table 2.
Table 3.

Annual mean precipitation for western U.S. basins. Units are mm day−1.

Table 3.
Table 4.

Annual mean surface water balance for western U.S. basins. All units in mm day−1 except SWE, which is in mm.

Table 4.
Save
  • Adam, J C., and Lettenmaier D L. , 2003: Bias correction of global gridded precipitation for solid precipitation undercatch. J. Geophys. Res., 108 .4257, doi:10.1029/2002JD002499.

    • Search Google Scholar
    • Export Citation
  • Berbery, E H., 2001: Mesoscale moisture analysis of the North American monsoon. J. Climate, 14 , 121137.

  • Berbery, E H., and Rasmusson E M. , 1999: Mississippi moisture budgets on regional scales. Mon. Wea. Rev., 127 , 26542673.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Berbery, E H., Luo Y. , Mitchell K E. , and Betts A K. , 2003: Eta Model-estimated land surface processes and the hydrologic cycle of the Mississippi basin. J. Geophys. Res., 108 .8852, doi:10.1029/2002JD003192.

    • Search Google Scholar
    • Export Citation
  • Betts, A K., Chen F. , Mitchell K E. , and Janjić Z I. , 1997: Assessment of the land surface and boundary layer models in the two operational versions of the NCEP Eta Model using FIFE data. Mon. Wea. Rev., 125 , 28962916.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Black, T L., 1994: The new NMC Mesoscale Eta Model: Description and forecast examples. Wea. Forecasting, 9 , 265278.

  • Chen, F., Janjić Z. , and Mitchell K. , 1997: Impact of the atmospheric surface-layer parameterizations in the new land-surface scheme of the NCEP mesoscale Eta Model. Bound.-Layer Meteor., 85 , 391421.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Christensen, N S., Wood A W. , Voisin N. , Lettenmaier D P. , and Palmer R N. , 2004: Effects of climate change on the hydrology and water resources of the Colorado River basin. Climatic Change, 62 , 337363.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Cosgrove, B A., and Coauthors, 2003: Real-time and retrospective forcing in the North American Land Data Assimilation System (NLDAS) project. J. Geophys. Res., 108 .8842, doi:10.1029/2002JD003118.

    • Search Google Scholar
    • Export Citation
  • 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 , 140158.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ek, M B., Mitchell K E. , Lin Y. , Grunmann P. , Rogers E. , Gayno G. , Koren V. , and Tarpley J D. , 2003: Implementation of Noah land surface model advances in the National Centers for Environmental Prediction operational mesoscale Eta Model. J. Geophys. Res., 108 .8851, doi:10.1029/2002JD003296.

    • Search Google Scholar
    • Export Citation
  • Gochis, D J., Shuttleworth W J. , and Yang Z-L. , 2002: Sensitivity of the modeled North American Monsoon regional climate to convective parameterization. Mon. Wea. Rev., 130 , 12821298.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Gochis, D J., Shuttleworth W J. , and Yang Z-L. , 2003: Hydrometeorological response of the modeled North American monsoon to convective parameterization. J. Hydrometeor., 4 , 235250.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Groisman, P Y., and Legates D R. , 1994: The accuracy of United States precipitation data. Bull. Amer. Meteor. Soc., 75 , 215227.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Higgins, R W., Shi W. , Yarosh E. , and Joice R. , 2000: Improved United States Precipitation Quality Control System and Analysis. NCEP/Climate Prediction Center Atlas 7, U.S. Department of Commerce, NOAA/NWS. [Available online at http://www.cpc.ncep.noaa.gov/research_papers/ncep_cpc_atlas/7/index.html.].

  • Janjić, Z I., 1990: The step-mountain coordinate: Physical package. Mon. Wea. Rev., 118 , 14291443.

  • Janjić, Z I., 1994: The step-mountain Eta coordinate model: Further developments of the convection, viscous sublayer, and turbulence closure schemes. Mon. Wea. Rev., 122 , 927945.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 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 , 878890.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Koster, R., and Milly P C. , 1997: The interplay between transpiration and runoff formulations in land surface schemes used with atmospheric models. J. Climate, 10 , 15781591.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Leung, L R., and Ghan S J. , 1998: Parameterizing subgrid orographic precipitation and surface cover in climate models. Mon. Wea. Rev., 126 , 32713291.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Leung, L R., Qian Y. , Han J. , and Roads J O. , 2003: Intercomparison of global reanalyses and regional simulations of cold season water budgets in the western United States. J. Hydrometeor., 4 , 10671087.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Liang, X., Lettenmaier D. P. , Wood E. F. , and Burges S J. , 1994: A simple hydrologically based model of land and energy fluxes for general circulation models. J. Geophys. Res., 99 , 1441514428.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Liang, X., Wood E. F. , and Lettenmaier E F. , 1996: Surface soil moisture parameterization of the VIC-2L model: Evaluation and modifications. Global Planet. Change, 13 , 195206.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Lohmann, D., and Coauthors, 2004: Streamflow and water balance intercomparisons of four land surface models in the North American Land Data Assimilation System project. J. Geophys. Res., 109 .D07S91, doi:10.1029/2003JD003517.

    • Search Google Scholar
    • Export Citation
  • Maurer, E P., O’Donnell G M. , Lettenmaier D P. , and Roads J O. , 2001: Evaluation of the land surface water budget in NCEP/NCAR and NCEP/DOE reanalyses using an off-line hydrologic model. J. Geophys. Res., 106 , D16,. 1784117862.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 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 , 32373251.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Mitchell, K E., and Coauthors, 2004: The multi-institution North American Land Data Assimilation System (NLDAS): Utilizing multiple GCIP products and partners in a continental distributed hydrological modeling system. J. Geophys. Res., 109 .D07S90, doi:10.1029/2003JD003823.

    • Search Google Scholar
    • Export Citation
  • Pan, M., and Coauthors, 2003: Snow process modeling in the North American Land Data Assimilation System (NLDAS): 2. Evaluation of model simulated snow water equivalent. J. Geophys. Res., 108 .8850, doi:10.1029/2003JD003994.

    • Search Google Scholar
    • Export Citation
  • Payne, J T., Wood A W. , Hamlet A F. , Palmer R N. , and Lettenmaier D P. , 2004: Mitigating the effects of climate change on the water resources of the Columbia River basin. Climatic Change, 62 , 233256.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Pulwarty, R S., and Redmond K T. , 1997: Climate and salmon restoration in the Columbia River basin: The role and usability of seasonal forecasts. Bull. Amer. Meteor. Soc., 78 , 381397.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Roads, J O., Chen S C. , Guetter A K. , and Georgakakos K P. , 1994: Large-scale aspects of the United States hydrologic cycle. Bull. Amer. Meteor. Soc., 75 , 15891610.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Robock, A., and Coauthors, 2003: Evaluation of the North American Land Data Assimilation System over the southern Great Plains during the warm season. J. Geophys. Res., 108 .8846, doi:10.1029/2002JD003245.

    • Search Google Scholar
    • Export Citation
  • Rogers, E., Black T L. , Deaven D G. , and DiMego G J. , 1996: Changes to the operational early Eta analysis/forecast system at the National Centers for Environmental Prediction. Wea. Forecasting, 11 , 391413.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Rogers, E., Parrish D. , and DiMego G. , 1999: Changes to the NCEP operational Eta analysis. NWS/NOAA Technical Procedures Bulletin 454, 25 pp. [Available online at http://www.nws.noaa.gov/om/tpb/454.html.].

  • Rogers, E., Black T L. , Collins W. , Manikin G. , Mesinger F. , Parrish D. , and DiMego G. , 2001a: Changes to the NCEP Meso Eta analysis and forecast system: Assimilation of satellite radiances and increase in resolution. NWS/NOAA Technical Procedures Bulletin 473. [Available online at http://www.emc.ncep.noaa.gov/mmb/mmbpll/eta22tpb/.].

  • Rogers, E., Ek M. , Lin Y. , Mitchell K. , Parrish D. , and DiMego G. , 2001b: Changes to the NCEP Meso Eta analysis and forecast system: Assimilation of observed precipitation, upgrades to land-surface physics, modified 3DVAR analysis. NWS/NOAA Technical Procedures Bulletin 479. [Available online at http://www.emc.ncep.noaa.gov/mmb/mmbpll/spring2001/tpb/.].

  • Saleem, J A., and Salvucci G D. , 2002: Comparison of soil wetness indices for inducing functional similarity of hydrologic response across sites in Illinois. J. Hydrometeor., 3 , 8091.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Schaake, J C., and Coauthors, 2004: An intercomparison of soil moisture fields in the North American Land Data Assimilation System (NLDAS). J. Geophys. Res., 109 .D01S90, doi:10.1029/2002JD003309.

    • Search Google Scholar
    • Export Citation
  • Schlosser, C A., and Coauthors, 2000: Simulations of a boreal grassland hydrology at Valdai, Russia: PILPS Phase 2(d). Mon. Wea. Rev., 128 , 301321.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sheffield, J., and Coauthors, 2003: Snow process modeling in the North American Land Data Assimilation System (NLDAS): 1. Evaluation of model-simulated snow cover extent. J. Geophys. Res., 108 .8849, doi:10.1029/2002JD003274.

    • Search Google Scholar
    • Export Citation
  • Shepard, D S., 1984: Computer mapping: The SYMAP interpolation algorithm. Spatial Statistics and Models, G. L. Gaile and C. J. Willmott, Eds., Reidel, 133–145.

    • Search Google Scholar
    • Export Citation
  • U.S. Department of Commerce/NOAA/USDA, 1996: Weekley weather and crop bulletin. Vol. 83, No. 36, 29 pp. [Available from Climate Predication Center, W/NP52, Attn: Weekly Weather & Crop Bulletin NOAA/NWS/NCEP, NOAA Science Center, Room 605, 5200 Auth Road, Camp Springs, MD 20746; also available online at http://usda.mannlib.cornell.edu/reports/waobr/weather/1996/wwcb3696.pdf.].

  • Widmann, M., and Bretherton C S. , 2000: Validation of mesoscale precipitation in the NCEP reanalysis using a new grid-cell dataset for the northwestern United States. J. Climate, 13 , 19361950.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Fig. 1.

    The topography of the western United States and the boundaries of the Columbia and Colorado basins.

  • Fig. 2.

    Jun 1995–May 2000 annual mean fields of observed precipitation gridded analyses: (a) CPC precipitation analysis without orographic correction (PCPC); (b) CPC precipitation analysis with orographic correction (POROCPC); (c) University of Washington precipitation analysis with orographic correction (PUW); (d) difference between the two CPC analyses, POROCPCPCPC; (e) difference PCPC PUW; and (f) difference POROCPCPUW. Units are mm day−1, and the contour intervals are indicated in the bar below each panel.

  • Fig. 3.

    Hovmoeller diagram at 48°N for Jun 1995–Jul 2000 of the differences between the CPC gridded analyses of precipitation and the University of Washington gridded analysis: (a) PCPCPUW and (b) POROCPCPUW. The gray shading and dashed contours denote negative values. Contour interval is 2 mm day−1. The topographic cross section at 48°N is included for reference.

  • Fig. 4.

    Jun 1995–May 2000 mean of (a) the Eta Model 12–36-h forecast precipitation (PETA) and its difference with the three gridded analyses: (b) PETAPCPC; (c) PETAPOROCPC; and (d) PETAPUW. Units are mm day−1, and the contour intervals are indicated in the bar below each panel.

  • Fig. 5.

    Jun 2000–May 2003 mean of (a) PETA and its difference with the CPC analyses: (b) PETAPCPC and (c) PETAPOROCPC. Units are mm day−1, and the contour intervals are indicated in the bar below each panel.

  • Fig. 6.

    (a) Jun 1995–May 2003 Hovmoeller diagram at 48°N of the differences between the Eta Model forecast precipitation and the CPC analysis of precipitation without orographic correction; (b) same as (a), but with the CPC precipitation analysis corrected for topographic effects; and (c) same as (a) but with the University of Washington gridded analysis, which ends in Jul 2000. Contour interval is 1.5 mm day−1. (The blank band indicates missing data due to the fire destruction of NCEP’s computer in 1999)

  • Fig. 7.

    Jun 1995–May 2003 Columbia basin area-averaged time series of (a) Eta Model forecast precipitation, University of Washington, and CPC not orographically corrected analysis; (b) their difference; and (c) the model’s precipitation rmse. (d)–(f) Same as (a)–(c) but for the Colorado basin. Units are mm day−1.

  • Fig. 8.

    (a) Mean annual cycle of the Columbia basin area-averaged Eta forecasts of total precipitation during Jun 1995–May 1999 (solid line) and Jun 1999–May 2003 (dashed line). The gray band represents the envelope of the mean annual cycle of the three gridded analyses computed for the two periods separately (see text). (b) Mean annual cycle of the Columbia basin–averaged Eta Model precipitation components: large-scale (PLS) and convective (PCON) for the same two periods as in (a). Units are mm day−1.

  • Fig. 9.

    Same as Fig. 8 but for the Colorado basin.

  • Fig. 10.

    Annual mean fields over 5 yr (Jun 1995–May 2000) of (a) coupled Eta Model evaporation and (b) uncoupled VIC model evaporation. (c) As in (a), but for the 3-yr period Jun 2000–May 2003. Units are mm day−1.

  • Fig. 11.

    The 5-yr annual mean fields for the period Jun 1998–May 2003 of the Eta Model 12–36-h forecasts of (a) water equivalent of accumulated snow depth, (b) normalized soil moisture for the 0–200-cm layer, and (c) total runoff; (d)–(f) same as (a)–(c), but for the VIC model and for the earlier 5-yr period of Jun 1995–May 2000. Normalization of soil moisture was done by taking into account the respective minimum–maximum ranges of the Eta Model and VIC model soil moisture (see text). Units are mm day−1.

  • Fig. 12.

    Columbia basin area-averaged mean annual cycle and time series of the surface water budget components of the Eta Model: (a), (e) water-equivalent snow depth; (b), (f) runoff plus baseflow; (c), (g) basin-averaged normalized soil moisture; and (d), (h) evaporation. Units are mm day−1.

  • Fig. 13.

    Same as Fig. 12 but for the Colorado basin.

  • Fig. 14.

    The residual term of the water balance equation estimated from the Eta Model: (a) the mean field for Jun 1998–May 2003, (b) the same residual as a percentage of the precipitation, (c) the area average for the Columbia basin, and (d) the area average for the Colorado basin. The heavy line in (c) and (d) represents a running mean that spans 12 months to remove the annual cycle. Here, dW/dt is the local change of surface water (soil moisture and snow water equivalent), P is the precipitation, E is the evaporation, and N is the runoff plus the baseflow. Units are mm day−1.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 220 67 3
PDF Downloads 95 22 1