1. Introduction
In the semiarid regions of the southwestern United States, much of the water supply for seven states primarily begins as snowpack in the Upper Colorado River basin (UCRB). With sparse vegetation in the basin, changes in temperature and precipitation lead to direct responses in the water budget (particularly storage in snowpack and runoff), thus greatly affecting the water supply. A complete understanding of the water budget is critical, as changes can have major socioeconomic impacts. Though there is still doubt about how climate change will affect precipitation trends in the basin, temperature increases are likely. Seager et al. (2007) have shown that rising temperatures associated with climate change could increase the frequency of severe droughts in arid and semiarid regions, while other studies have shown shifts in the timing of peak runoff (Miller and Piechota 2008; Hamlet et al. 2007) and reduction in runoff (Christensen and Lettenmaier 2007; Hoerling and Eischeid 2007; McCabe and Wolock 2007) over the region. Several studies have also pointed to positive feedbacks that could have an effect on the basin’s water supply. Groisman et al. (1994) showed a positive feedback between warming temperatures and snow cover extent, and Entekhabi et al. (1992) highlighted several positive feedbacks (as a result of precipitation deficits propagating down through all components of the water budget) that prohibit the generation of precipitation and result in prolonged droughts.
In addition to sensitivity due to feedbacks and climate change, the UCRB is also sensitive to climate variability. Several studies have identified a correlation between hydroclimatic variability in the region and large-scale climate forcings such as the El Niño–Southern Oscillation (ENSO) and Pacific decadal oscillation (PDO; Hurkmans et al. 2009; Kim et al. 2005; McCabe and Dettinger 2002). Other studies have found strong relationships between UCRB climate variability and sea surface temperature (SST) variability in other ocean regions not related to the ENSO and PDO (Aziz et al. 2010; Switanek et al. 2009; Tootle and Piechota 2006). With the sensitivities of the water budget to climate change, climate variability, and land–atmospheric feedbacks, it is essential to have a thorough knowledge of the variability of each water budget component and how they relate to each other in the region’s hydrologic cycle.
Ideally, one would study the different components of the water budget using perfect observations taken at a continuous and high spatial and temporal resolution. This is unfortunately impossible. Though in situ observations are the most direct way to take measurements of the water variables, they are only point measurements that cannot be easily (or accurately) extrapolated to other areas throughout a larger basin (Kustas et al. 1991). In addition, direct measurements may not yield the most accurate and representative datasets. For example, snowpack measurements cannot be made above tree line (because of wind contamination) where the heaviest amounts of snow can accumulate throughout the winter; precipitation observations tend to be more concentrated in populated areas; soil moisture measurements are sparse given the hard soils found throughout much of the basin; and evaporation measurements rely on some set of assumptions that may only apply at a local scale. The weaknesses of in situ observations point to the necessity of utilizing model, reanalysis, and satellite-derived datasets.
Reanalysis and satellite-derived datasets have an advantage over in situ observations in that they provide spatially and temporally consistent global measurements of all atmospheric and surface hydrologic variables. However, each gridded dataset comes with its own set of weaknesses. Reanalysis data, though initially forced by direct or remotely sensed observations, are continuously modified and confined by model physics. Each reanalysis product has its own set of model parameterizations and data assimilation methods to simulate land and atmospheric variables. Reanalysis products are also released at a relatively coarse spatial resolution that is ideal for global climate studies and may not be as ideal for regional analysis. Satellite-derived datasets tend to be available at much higher spatial resolutions, but they each rely on different algorithms to translate raw retrievals into water variables. These generalized (i.e., imperfect) models and algorithms add some degree of uncertainty to each gridded dataset.
In this study, a suite of in situ, reanalysis, and satellite-derived datasets will be compared for each component of the surface and atmospheric water budgets. The objectives of this study are to gain a better understanding of the UCRB’s hydroclimatic variability and to identify the most ideal datasets to use for each water budget component. These objectives will be met through analysis of each water budget component and by attempting water budget closure. This paper is organized as follows. Section 2 introduces details about the datasets. Section 3 analyzes the consistency among the datasets and the variability of each water budget component. In the discussion, section 4, a balance of the surface and atmospheric budgets is performed. Conclusions are presented in section 5.
2. Data and methodology
The UCRB, with an areal size of about 2.6 × 105 km2, is located in the western United States (Fig. 1). Table 1 details the resolutions, time periods, and references for the datasets that are used to analyze the water budget components. Data are monthly totaled and presented in terms of water years (October–September). For the gridded datasets, a bounding box is defined over the UCRB, and only grid points inside the bounding box are used. Each water budget component is averaged over all grid points in the UCRB, resulting in one basin-averaged value over time. This allows for easier comparisons across differing spatial resolutions among the datasets.
UCRB outlined in gray with blue dots representing COOP stations, red triangles representing SNOTEL sites, and the two streamflow sites labeled as green dots. Elevation contours are shown in color at 250-m intervals, triangulated from the station elevations.
Citation: Journal of Hydrometeorology 14, 3; 10.1175/JHM-D-12-0119.1
Summary of the different datasets used, including their temporal and spatial resolutions, the time period used for this study, and the reference for the dataset.
a. In situ measurements
1) Precipitation
Locations for the National Weather Service Cooperative Observer Program (COOP) network stations are shown in Fig. 1. Because the COOP data are not spatially consistent, the question of representativeness is raised. To obtain one basin-averaged precipitation value over time, one could average the 105 stations together for one value. However, this might not be reasonable for a basin where precipitation is heavily dependent on elevation. Analysis of long-term averaged monthly precipitation at all sites shows that precipitation amounts can vary by over an inch between the lower and higher elevation stations. To test the representativeness of the COOP stations, a frequency distribution of the elevations in the basin (found by taking an average elevation for every ¼° by ¼° grid box in the basin) is compared to the frequency distribution of the 105 COOP station elevations (Fig. 2). The spatially continuous gridpoint elevations are normally distributed with a mean elevation of 2149 m. The COOP station distribution is also normal, but shifted to lower elevations, with a mean elevation of 1921 m. Higher elevations are not well represented for COOP station data; therefore, basin-averaged precipitation could be an underestimate. Instead of averaging the stations, an average precipitation is calculated for each 150-m elevation bin. These are then weighted based on the frequency distribution of the gridded elevations and totaled for one basin value.
Frequency distribution of elevations (150-m bins) in the UCRB for ¼° by ¼° grid points (gray), COOP stations (maroon), and SNOTEL sites (blue). Dashed lines represent average elevation for each dataset.
Citation: Journal of Hydrometeorology 14, 3; 10.1175/JHM-D-12-0119.1
Previous studies have shown the usefulness of incorporating elevation influences when spatially gridding station precipitation. External drift kriging and cokriging have been used to statistically interpolate precipitation based on elevation (Haiden and Pistotnik 2009). Daly et al. (1994) found that the Parameter-Elevation Regressions on Independent Slopes Model (PRISM) dataset (released by the PRISM Climate Group at Oregon State University) performed better than any kriging method over the Willamette River basin in Oregon. When spatially interpolating precipitation from point data in mountainous regions, elevation rather than distance is very important, and the use of elevation-dependent data (such as PRISM) is ideal. For this study, it is only necessary to define one basinwide precipitation value (monthly or annually); therefore, spatially gridding the data first may be needlessly and computationally intensive. Additionally, the weighting technique (which captures the elevation dependence) removes the need for linear interpolation and simply uses the actual precipitation values from the COOP stations.
Figure 3 shows the comparison of annual precipitation using a simple basin average compared to using the weighting technique. The weighting technique results in annual totals that are 10%–25% higher than the basin-averaged precipitation. When comparing one year of monthly basin-averaged precipitation, weighted precipitation, and averaged PRISM precipitation (not shown), PRISM values match more closely with the weighted totals than with the basin-averaged totals. However, the PRISM totals are 10%–20% higher than the weighted totals for many of the months. The biggest discrepancies occur during months when the higher-elevation station precipitation totals are anomalously less than at the lower elevations. The weighting technique captures this anomaly, whereas a technique using a linear interpolation (PRISM in this example) did not.
Annual accumulated precipitation using a simple basin average of the 105 COOP stations compared to weighting the COOP stations based on elevation.
Citation: Journal of Hydrometeorology 14, 3; 10.1175/JHM-D-12-0119.1
2) Evapotranspiration
Direct measurements of evapotranspiration (ET) in the UCRB are not common. The Colorado Agricultural Meteorological Network (CoAgMet) provides ET data at 18 sites in the basin (all in western Colorado). These sites are located in lower-elevation locations that are irrigated during the warm season. Reference ET measurements at CoAgMet sites are calculated using the Penman–Monteith or Kimberly–Penman equations (which rely on maximum and minimum temperature, relative humidity, daily wind, and solar radiation measurements) and are based on a standardized reference crop (alfalfa or short grass). Weather data required for the ET measurements are collected in well-irrigated grass areas and therefore are only accurate for croplands. Land cover data over the region (not shown) indicate that irrigated/cultivated croplands only cover a small percentage of the UCRB, so on a regional scale, reference ET measurements will not be representative—and, therefore, are not used in the analysis.
3) Snow water equivalent
Locations for the Natural Resources Conservation Service (NRCS) Snowpack Telemetry (SNOTEL) sites are shown in Fig. 1. Similar to the COOP dataset, SNOTEL data are not spatially continuous and may not be representative of the entire basin. Snow water equivalent (SWE) values are also heavily dependent on elevation. Since the goal of the SNOTEL network is to monitor snowpack in the mountains, most stations are high elevation. The frequency distribution of the SNOTEL elevations, which is higher than that of the gridded elevations, has a mean elevation of over 2923 m (Fig. 2). Calculating a simple average of all SNOTEL sites would result in an overestimate of basin-averaged SWE. A similar methodology as presented in the precipitation section (weighting the values based on elevation bins) was used to weight the SNOTEL SWE values across the basin. Although the methodology generally reduces the SWE peaks by approximately 50%, it also changes the interannual variability in those peaks (possibly as a result of the linear interpolation that is required for the lower-elevation bins that contain no actual SNOTEL sites). Maintaining the actual interannual variability is very important for comparison with the other water variables in the basin. Therefore, the simple basin average of SNOTEL SWE values will be used.
4) Soil moisture
Soil moisture measurements in the UCRB are limited. The NRCS maintains the Soil Climate Analysis Network (SCAN) dataset. There are 12 SCAN sites in the UCRB (all in eastern Utah), and most do not have data for more than five years. The majority of the sites are at elevations below 2100 m; thus, they are not representative of the entire basin.
5) Streamflow and reservoir storage
Raw monthly streamflows (in cubic feet per second) are provided by the U.S. Geological Survey. Monthly data are obtained for two sites along the Colorado River: the gauge at the Colorado–Utah (CO–UT) state line and the gauge at the base of the watershed at Lee’s Ferry, just below Lake Powell (Fig. 1). Because this basin is an entirely closed system, the only runoff out of the basin is from the Lee’s Ferry gauge and the several different transmountain diversion tunnels to the east side of the Continental Divide. The sum of all annual transmountain volume diversions is more than an order of magnitude less than the annual volume runoff at Lee’s Ferry, so this study assumes that volume runoff recorded at Lee’s Ferry represents total runoff for the UCRB.
Data from eight major reservoirs, provided by the U.S. Bureau of Reclamation, are used to calculate the water storage of the UCRB, with a total capacity of 39.5 km3. Storage volumes are sensitive to climate variability, with actual volumes fluctuating between 30% and 60% of capacity over the past decade.
b. Reanalysis and gridded datasets
1) Precipitation
The National Aeronautics and Space Administration Modern-Era Retrospective Analysis for Research Applications (MERRA) utilizes satellite rain retrievals over the ocean as input to the Goddard Earth Observing System Model, version 5 (GEOS-5), data assimilation system; however, Reichle et al. (2011) states that this has little effect on the system over land. The precipitation estimates over the UCRB are not the result of the data assimilation of surface rain gauge data, but rather the assimilation of atmospheric data and model physics. Globally, MERRA reproduces annual precipitation rates and spatial variability fairly well (Bosilovich et al. 2011; Reichle et al. 2011), though it tends to underestimate midlatitude precipitation.
The European Centre for Medium-Range Weather Forecasts Interim Re-Analysis (ERA-Interim, hereafter ERA-I) also shows improved skill from older generation reanalyses at reproducing global precipitation patterns (Bosilovich et al. 2011), partly as a result of the modification of the convection scheme in the model (Dee et al. 2011). Precipitation estimates from ERA-I are also based on model physics forced by temperature and humidity in the assimilation scheme.
The National Centers for Environmental Prediction’s Climate Forecast System Reanalysis (CFSR) combines the Noah land surface model with the Global Forecast System (GFS) atmospheric model and is the only reanalysis product to include a fully coupled ocean model. Over land, daily observations of precipitation are combined with gauge analysis and Climate Prediction Center (CPC) Merged Analysis of Precipitation (CMAP) 5-day mean precipitation to drive the land surface model, making it the only reanalysis product out of the three to rely on precipitation observations in its precipitation dataset.
2) Evapotranspiration
The land surface model in the GEOS-5 data assimilation system (used by MERRA) divides the land surface variables into catchments instead of grid boxes (Decker et al. 2012). MERRA also uses the highest time resolution for inputs into the land surface model, thus making it likely to be the most ideal dataset to use for ET. The ERA-I land surface analysis requires 6-hourly inputs of temperature and relative humidity to model near-surface fluxes at a 3-h interval (Decker et al. 2012). Land surface assimilation for the CFSR only occurs once a day. Reanalysis ET measurements have not been validated over the UCRB, but other regions have been studied. When compared to flux towers, all reanalyses showed good agreement at many locations, though all had a tendency to overestimate latent heat flux (Decker et al. 2012). Mueller et al. (2011) showed that in the Mississippi River basin, reanalysis land surface fluxes tended to be greater than satellite-derived land surface fluxes.
3) Water vapor divergence
Trenberth et al. (2000) were one of the first to examine global divergence using reanalysis data and found that reanalysis could reliably reproduce the global monsoon and the divergent circulation. Since then, several regional studies have specifically examined wind and humidity data from reanalysis (the two variables needed to calculate vapor divergence). Decker et al. (2012) found that ERA-I winds had the lowest error and bias when compared to flux tower observations across North America; they also found that the CFSR performed the worst of the reanalysis products. Over the Tibetan Plateau, Bao and Zhang (2013) found that ERA-I and CFSR had large positive biases and errors in humidity, while Kennedy et al. (2011) found that MERRA had a slight dry bias in humidity compared to Atmospheric Radiation Measurement Program (ARM) sites in the U.S. Midwest. Both studies also pointed to different biases in the reanalyses at lower levels compared to the middle and upper troposphere.
4) Soil moisture
The Climate Prediction Center Soil Moisture V2 dataset (CPC Soil) uses a one-layer leaky bucket model, with temperature and precipitation as inputs, to model soil moisture (Fan and van den Dool 2004). Though difficult to validate over the UCRB, it has done reasonably well at simulating soil moisture when compared to the limited observations available (Huang et al. 1996; Fan and van den Dool 2004). It is likely to compare well with precipitation observations over the UCRB since it is forced by precipitation from the U.S. Historical Climate Network (USHCN).
The MERRA reanalysis includes a root-zone soil wetness variable, which represents the wetness of the top meter of soil (as a fraction). It has been shown to compare well with ERA-I soil moisture (Rienecker et al. 2011). Reichle et al. (2011) have pointed out that MERRA precipitation estimates may sometimes be underestimated due to less intense precipitation rates and that solar radiation could be overestimated, thus leading to an overall underestimate of soil moisture. Additionally, Rienecker et al. have pointed to systematic differences in high-latitude soil moisture because it cannot be adjusted by evaporation and runoff during long periods of the year with frozen conditions.
c. Satellite-derived datasets
1) Precipitation
High-resolution global precipitation datasets are widely used in studies and are also useful for comparison with in situ data. For higher-latitude land regions, satellite-derived precipitation datasets are limited. Microwave precipitation estimates are not as accurate over land (particularly over mountainous regions), and land infrared (IR) precipitation estimates are not provided over regions with snow cover owing to the brightness temperature of snow closely matching the brightness temperature of cloud tops. Over the UCRB, snow cover is an issue for over six months of the year, limiting the amount of satellite precipitation observations. The Tropical Rainfall Measuring Mission (TRMM) Multisatellite Precipitation Analysis (TMPA) dataset is unique among many high-resolution satellite-derived precipitation products as it combines precipitation estimates from multiple satellites with gauge data to produce precipitation data for the lower and middle latitudes (Huffman et al. 2007; Sapiano and Arkin 2009). Because it utilizes gauge analysis, it provides data even during snow cover times and is therefore suitable for this study.
2) Evapotranspiration
Some recent studies have done global comparisons of satellite- and model-derived land heat flux products (e.g., Jimenez et al. 2011; Mueller et al. 2011). This study uses several of the higher-resolution latent heat flux datasets presented in the aforementioned articles. The University of California, Berkeley, dataset (UCB) combines satellite retrievals with Priestley–Taylor estimates, which transform potential ET values to actual ET values. The Max Planck Institute for Biogeochemistry (MPI) utilizes direct eddy-covariance measurements from flux towers (FLUXNET) and globally grids the data using an approach called model tree ensemble. The Paris Observatory (PAO) dataset empirically derives latent heat fluxes using satellite-derived reflectances, emissivities, backscatter, and temperature as inputs.
Jimenez et al. (2011) have shown that, globally, reanalyses produce the highest ET values (with the exception of the UCB dataset). They also found good agreement between MPI and PAO datasets. While they found good spatial consistency among all the products, the magnitudes of ET vary widely from product to product. It was found that the mountainous and desert regions (where ET values are relatively smaller) displayed the greatest amount of variability between products. Therefore, it will be important to see how these products differ over the UCRB.
3) Snow water equivalent
Storage in the form of SWE is measured by the Advanced Microwave Scanning Radiometer for the Earth Observing System (AMSR-E) on the Aqua satellite. The data have a coarse resolution for SWE, which is highly variable in mountainous terrain. De Lannoy et al. (2012) showed that for low elevation flat areas, AMSR-E values can exceed nearby SNOTEL SWE values by an order of magnitude, while the higher elevation peak SNOTELs can exceed the AMSR-E values by an order of magnitude.
4) Total water storage
The Gravity Recovery and Climate Experiment (GRACE) satellites measure changes in the gravity field caused by changes in mass. Over land, these mass changes are primarily the result of changes in terrestrial water storage (ΔTWS). Land data processed by the University of Texas Center for Space Research (CSR, Release 04) were processed from the raw gravity fields using spherical harmonics (Swenson and Wahr 2006). The mass of the atmosphere is removed using ECMWF fields, and then the data are destriped and smoothed using a 300-km Gaussian filter. The data are ideal for basins 200 000 km2 or larger (Zaitchik et al. 2008). Land water content from the Noah land surface model in the Global Land Data Assimilation System (GLDAS) can be compared to the CSR measurements over land (Rodell et al. 2004b). Both datasets are available through the Jet Propulsion Laboratory’s GRACE website (http://grace.jpl.nasa.gov).
Changes in terrestrial water storage (ΔTWS) over land can be the result of changes in soil moisture, deep groundwater, reservoir storage, or storage in SWE. Using in situ, gridded, and satellite data, ΔTWS can be estimated by calculating basin-averaged monthly volume changes in SWE, reservoir storage, and soil moisture and then taking the sum of the three. This can then be compared to ΔTWS from CSR and GLDAS.
3. Water budget components
a. Precipitation
The in situ precipitation time series is overlaid with MERRA, ERA-I, CFSR, and TMPA precipitation time series in Fig. 4. Based on the location and topography of the UCRB, two seasonal peaks would be expected: the first during the cold season, when the higher elevations receive their largest amounts of precipitation, and the second in late summer, coinciding with the onset of the North American monsoon (NAM). The northernmost fringes of the NAM results in increased precipitation over the southern portion of the UCRB (Vera et al. 2006).
(top) Monthly and (bottom) annual precipitation averaged over the UCRB, comparing COOP observations with reanalysis and satellite datasets.
Citation: Journal of Hydrometeorology 14, 3; 10.1175/JHM-D-12-0119.1
When looking at the in situ monthly precipitation time series (Fig. 4, top), no consistent seasonal cycle with two peaks is apparent. In this case, using one value for the basin could be wiping out the signal. A simple k-means cluster analysis of every COOP station’s precipitation time series separated the basin into northern and southern regions, but analysis of the separate cluster precipitation time series yielded minimal differences (not shown). However, when the COOP stations were binned according to elevation and then averaged over each bin, larger seasonal differences became apparent (Fig. 5). The highest-elevation bin’s seasonal cycle shows a precipitation maximum in July–September, another maximum in December, and a third maximum in April. The middle and lower elevation bin seasonal cycles show a precipitation maximum for August–October. It appears as if all elevations are in some way influenced by the NAM (though the lower elevations lag the higher elevations), and the higher elevations are more influenced by larger cold-season snowfall accumulations.
Long-term average monthly precipitation, using COOP observations, for three different elevation bins.
Citation: Journal of Hydrometeorology 14, 3; 10.1175/JHM-D-12-0119.1
There is good agreement in seasonal and interannual variability among the different gridded precipitation products and the in situ precipitation. The in situ and gridded datasets all show a wet–dry–wet–dry pattern. Focusing on water years 2004–08 (Fig. 4, bottom), the even years (WY2004, WY2006, WY2008) show relatively lower precipitation accumulations, and the odd years (WY2005, WY2007) show higher precipitation accumulations.
The reanalysis datasets tend to overestimate the monthly and annual in situ precipitation in the UCRB by around 10%–20%. This is consistent with the Reichle et al. (2011) study that found MERRA overestimated precipitation in the western United States. The higher-resolution satellite precipitation product (TMPA) underestimates in situ and reanalysis precipitation by more than 40%. The IR rain-retrieval method used by TMPA may possibly be missing lighter precipitation events in mountainous areas, thus resulting in this underestimation (Hirpa et al. 2010). Additionally, the included gauge analysis in TMPA, provided by the Global Precipitation Climatology Centre (GPCC), does adjust for elevation, but could still be underestimating higher elevation precipitation in the basin. Based on the precipitation time series, the CFSR is least correlated with in situ precipitation, and MERRA is highest correlated. The CFSR dataset relies on CMAP precipitation observations in its land surface model, and the CMAP precipitation time series is very similar to the TMPA time series over the UCRB (not shown). Therefore, the poorer performance of the CFSR precipitation over the UCRB is not due to the initial observational input, but rather from some other internal physics within the model, and should be considered when using CFSR data in future studies over the region.
b. Evapotranspiration
Reanalysis and satellite-derived evapotranspiration over the UCRB are compared in Fig. 6. All of the gridded products show similar seasonal cycles (with a summer peak coinciding with warmest temperatures). Reanalysis ET are in good agreement with MPI and PAO ET, while UCB ET estimates can be 25%–40% higher. Unlike satellite-derived ET, reanalysis data show a consistently repeating double seasonal peak in ET. Looking at long-term monthly averaged MERRA ET (Fig. 7), the first ET peak generally occurs in the late spring, immediately following the heaviest precipitation accumulations and as temperatures are rapidly warming. Lower ET values follow during the drier summer months, and then a second ET peak occurs in the late summer, coinciding with the arrival of the NAM. All products (with the exception of the CFSR) agree on interannual variability (Fig. 6, bottom), which is strongly coupled to precipitation, with WY2005 and WY2007 showing relatively higher ET values and WY2004, WY2006, and WY2008 showing lower ET values.
(top) Monthly and (bottom) annual evapotranspiration averaged over the UCRB, comparing reanalysis and satellite datasets.
Citation: Journal of Hydrometeorology 14, 3; 10.1175/JHM-D-12-0119.1
Long-term average monthly ET averaged over the UCRB using MERRA data.
Citation: Journal of Hydrometeorology 14, 3; 10.1175/JHM-D-12-0119.1
It is difficult to identify one evapotranspiration product as being more representative than the others for the basin since reference ET is the only basis for comparison in the UCRB. Mueller et al. (2011) pointed out that good agreement among several products may not necessarily mean that they are more accurate but that these products share similar model physics and data forcings. Additionally, one product may stand out as an outlier, but that may not mean it is erroneous (Mueller et al. 2011). The CFSR ET will not be used further in this study as it does not exhibit an accurate pattern in the interannual cycle, and its strong positive bias during the cold season (compared to all the other products) may be inaccurate as it is assumed there is very little evaporation during the cold season. MERRA ET will be used for further analysis as it captures the interannual variability, and MERRA precipitation is most closely correlated with in situ observations. However, since reanalysis ET could possibly be overestimating actual evapotranspiration (Jimenez et al. 2011), the MPI ET product will also be used for further analysis.
c. Storage volume changes
One of the largest contributing components to changes in surface storage, and the largest source of seasonal and interannual variability, is water stored as snowpack. In situ snow water equivalent is shown in the top of Fig. 8. At the beginning of the water year (October) virtually no snow is present in the basin. Progressing through time, SWE builds up to its maximum around April or May, then quickly melts off and typically disappears by June.
Monthly snow water equivalent averaged over the UCRB, using (top) SNOTEL observations and (bottom) AMSR-E retrievals.
Citation: Journal of Hydrometeorology 14, 3; 10.1175/JHM-D-12-0119.1
According to in situ SWE, WY2005 and WY2008 were the highest accumulation years in the analysis, while WY2004 and WY2007 show relatively lower peak SWE values. The variability observed in the SWE peaks does not match the wet–dry–wet–dry pattern observed in the precipitation and ET time series. WY2004 was drier (and WY2005 was wetter) according to precipitation and SWE, but, where precipitation shows a higher accumulation in WY2007, peak SWE is lower. This is the result of an anomalously wet October that led to a large amount of nonfrozen precipitation accumulations that did not translate into storage in snowpack. In WY2008, when peak SWE was high, precipitation was lower as a result of a weak monsoon season at the end of the water year. It is interesting to note that ET is more closely coupled with precipitation and does not respond to higher SWE peaks.
The bottom of Fig. 8 shows AMSR-E satellite-derived snow water equivalent. Similar to previous studies in the region (De Lannoy et al. 2012), in situ SWE peaks are almost an order of magnitude higher than AMSR-E estimates. AMSR-E provides spatial continuity, which is not provided with the in situ measurements that are only available at the high elevations. AMSR-E may provide more realistic magnitudes on a basin-averaged scale and is therefore useful for calculating storage volume changes. However, it fails to capture the observed interannual variability. Henceforth, satellite-derived SWE will be used for surface water budget calculations, but in situ SWE will still be used to analyze the interannual variability in the basin.
When comparing gridded soil moisture products with soil moisture at several of the SCAN sites in eastern Utah, large discrepancies are evident in the overall magnitudes of soil moisture and in the seasonal variability. The gridded products themselves show similar seasonal cycles and interannual variability, but the MERRA magnitudes are 2 to 3 times larger. The introduction of more soil moisture measuring sites in the UCRB (and longer periods of record) is needed in order to properly validate these gridded soil moisture products. Since that is impossible for this study, only the gridded data are analyzed. In Fig. 9, changes in soil moisture are added to changes in SWE and reservoir storage over time. Adding soil moisture can result in a 50%–100% increase in surface storage changes, with MERRA soil moisture resulting in larger seasonal changes than CPC soil moisture. Annually (not shown) total soil moisture storage displays the same interannual variability seen in the precipitation and ET time series. Soil moisture changes also respond quickly to precipitation anomalies, which is to be expected as precipitation observations are used to force the CPC soil moisture model. As it is difficult to ascertain which soil moisture product is more accurate for the basin, both will continue to be evaluated.
Comparison of surface storage changes in the UCRB for Δ(SWE + reservoir) (black), Δ(SWE + reservoir + MERRA soil) (green), and Δ(SWE + reservoir + CPC soil) (purple).
Citation: Journal of Hydrometeorology 14, 3; 10.1175/JHM-D-12-0119.1
In Fig. 10, storage volume changes (combining AMSR-E SWE with reservoir storage and soil moisture) are compared with changes in GLDAS and CSR satellite-derived ΔTWS. The seasonal cycle is similar for all products, with water storage gains (losses) during the cold (warm) season. The magnitude and variability of the GLDAS and AMSR-E time series are well correlated, while the CSR time series displays too much month-to-month variability.
(top) Monthly and (bottom) annual storage volume changes for the UCRB comparing observations (SWE + reservoir + soil moisture) with satellite ΔTWS estimates from GRACE.
Citation: Journal of Hydrometeorology 14, 3; 10.1175/JHM-D-12-0119.1
Annual changes in storage volume are compared in the bottom of Fig. 10. The annual variability of (reservoir + CPC soil moisture) storage changes matches well with the GLDAS annual changes in ΔTWS. Both show very little change in storage for WY2004 and WY2007, large gains in WY2005, and storage losses in WY2006. WY2005 and WY2006 are consistent with in situ SWE peaks and precipitation variability—there was a gain in storage during the wetter year with a higher peak in situ SWE (WY2005) and a loss in storage during the drier year with a lower peak SWE (WY2006). When in situ SWE peak and precipitation are out of phase (WY2007), storage changes are closer to zero. Though the annual magnitudes of (reservoir + MERRA soil moisture) are closer to GLDAS magnitudes, the interannual variability is not as well correlated with GLDAS as the (reservoir + CPC soil moisture). The interannual variability in the CSR time series does not match with any of the other products.
d. Runoff
Variations in streamflow across the UCRB are the result of a combination of manmade regulations and seasonal variability. The Colorado River at the CO–UT state line typically reaches peak streamflows between May and June of each water year (shortly after snowmelt begins). Farther downstream, at Lee’s Ferry, peak streamflows typically occur a little later between June and August. A secondary seasonal peak is observed at the Lee’s Ferry site from January to March.
Though flows on the Colorado River are regulated year-round, natural interannual variability can still be detected when analyzing the water-year accumulated volume runoff (Fig. 11). Upstream, at the CO–UT state line, low accumulations during 2002 can be linked to the major drought in the early twenty-first century that affected the entire southwestern United States. Annual runoff accumulations have been increasing since that period. Downstream at Lee’s Ferry, accumulated runoff stays fairly consistent interannually, showing more of a manmade signal than a signal from natural variability.
Annual volume runoff on the Colorado River at the CO–UT state line and at Lee’s Ferry.
Citation: Journal of Hydrometeorology 14, 3; 10.1175/JHM-D-12-0119.1
Runoff variability at the CO–UT state line site is better correlated with peak SWE values rather than precipitation. Higher runoff years (WY2005 and WY2008) correspond to higher peak SWEs, whereas the lower runoff year of WY2007 corresponds to a lower peak SWE (even though annual precipitation was high that year). This close relationship between SWE and runoff should be expected: the overall purpose of the SNOTEL program (and as a result, the ultimate placement of the sites) was to better predict the upcoming springtime runoff in the western river basins (NRCS NWCC 2012). Because of this deliberate choosing of SNOTEL sites (to capture the deeper snow depths), SWE interannual variability is better correlated with runoff variability than with precipitation variability.
e. The divergence of water vapor
Calculated ∇ · Q over time from the different reanalysis products are shown in Fig. 12. The MERRA and ERA-I show similar seasonal variability. Both show atmospheric divergence over the UCRB in the spring and summer months and atmospheric convergence during the winter (Fig. 12, top). This atmospheric seasonal cycle is consistent with the surface water variables and with the climate of the basin. The CFSR shows more erratic month-to-month variability and rarely exhibits atmospheric convergence.
(top) Monthly and (bottom) annual accumulation atmospheric ∇ · Q averaged over the UCRB using reanalysis data.
Citation: Journal of Hydrometeorology 14, 3; 10.1175/JHM-D-12-0119.1
Annually, the CFSR suggests that water vapor is always diverging out of the UCRB region, while MERRA shows several years of divergence (Fig. 12, bottom). As all of these years showed surface runoff, it does not make sense to have atmospheric divergence for an entire year. ERA-I is the only reanalysis dataset to consistently show atmospheric convergence with variability only in the magnitude of the convergence. ERA-I displays less atmospheric convergence during years with less precipitation (WY2004, WY2006, and WY2008) and greater atmospheric convergence during the wet year of WY2005. Though WY2007 was a high precipitation year, the lower snowpack would mean less convergence during the winter and, thus, would lead to a year with less atmospheric convergence according the ERA-I.
4. Discussion
a. Surface and atmospheric storage estimates
Estimated ΔS is calculated using in situ precipitation, MPI ET, and accumulated runoff at Lee’s Ferry. When MERRA ET was input into the calculation, ΔS was too often negative. As MERRA ET appears to be an overestimate (especially when combined with in situ precipitation) and the use of MPI ET leads to a better balance, the MPI ET will continue to be used. Estimated ΔS is compared with the CSR and GLDAS products in Fig. 13 (top). Again, the CSR shows much more month-to-month variability than what is estimated. CSR magnitudes tend to be 50% greater than estimated ΔS, while GLDAS magnitudes tend to be 50% less. Based on the seasonal cycle and interannual variability, GLDAS appears to better represent surface water storage changes over the UCRB. Rodell et al. (2004a) found that GLDAS data are in better agreement with JPL-processed GRACE data (Chambers 2006). Tang et al. (2010) concluded that the data processed by Chambers (2006) significantly underestimated true values in the seasonal cycle of terrestrial water storage (particularly over the Klamath and Sacramento River basins). This should be considered when using GLDAS data over the UCRB.
Comparison of estimated storage changes (in situ P − MPI ET − R, black line) (top) with GLDAS (blue) and CSR (green) and (bottom) with change in (AMSR-E SWE + reservoir + CPC soil moisture, maroon dashed line).
Citation: Journal of Hydrometeorology 14, 3; 10.1175/JHM-D-12-0119.1
The bottom of Fig. 13 shows the comparison of estimated ΔS with storage changes calculated from changes in (SWE + reservoir + soil moisture). With the use of AMSR-E SWE and CPC soil moisture, calculated changes in storage are well correlated with estimated ΔS in terms of magnitude, seasonal cycle, and interannual variability.
In the atmosphere, estimates of precipitation, ET, and ∇ · Q can be used to estimate changes in atmospheric storage, using Eq. (1). Estimated 
Annual accumulation of 
Citation: Journal of Hydrometeorology 14, 3; 10.1175/JHM-D-12-0119.1
b. Long-term storage
Another way to compare how the different products balance is to total them over long periods. When totaling estimated ΔS (P − ET − R) from WY2003 through WY2008 for the UCRB, there is an overall increase in storage of +176 mm. Yet, calculated storage changes in (reservoir + soil moisture) over the same period only yield an increase of +21 mm. Both the GLDAS and CSR products show a loss in storage for the same time period, of −14 and −11 mm, respectively. When looking at long-term atmospheric changes, estimated 
c. Connecting the surface and atmosphere
Over a long enough period of time, atmospheric convergence/divergence over an area will equal the runoff and change in storage at the surface. This can help connect the atmospheric water budget with the surface water budget and shed more light on the different products, as shown in Table 2. Over the same long-term period (WY2003 through WY2008) total ∇ · Q from ERA-I shows overall atmospheric convergence, while MERRA shows overall atmospheric divergence. Since surface runoff is positive over that long time period, atmospheric convergence would have had to occur, which MERRA does not show.
Long-term water changes in the atmosphere and the surface, WY2003–WY2008. The bold values denote a balance between the atmosphere and the surface. A positive value in the atmosphere denotes convergence or addition of storage, while a negative value denotes divergence or loss of storage. At the surface, a positive value denotes excess water to runoff or a gain in storage.
Table 2 shows that adding estimated ΔS to the runoff results in surface changes that are greater than the atmospheric convergence (by ERA-I). Adding (reservoir + soil moisture) or the CSR and GLDAS to the runoff results in surface changes that are much less than the atmospheric convergence. Estimated ΔS could be too large owing to a possible underestimation of the MPI ET product. A systematic increase of 5% to the MPI ET yields a better balance. In Table 2, adjustment of ΔS (and adding to runoff) yields a much closer value to atmospheric convergence. The difference between the two would leave a surplus of 13 mm in the atmosphere. This can be accounted for in the adjusted 
The issues that remain are with CSR, GLDAS, and MERRA. Both CSR and GLDAS show a loss in surface storage for a time period when a gain in surface storage is expected. Adding the next two years to each (WY2009 and WY2010) results in a switch in sign for CSR. GLDAS numbers stay fairly consistent, with a loss in storage of approximately −10 mm, while CSR switches and shows a gain of approximately +10 mm. There is not enough data (as the GRACE satellites were only launched in 2002) to tell if CSR had issues in early years and the data improved for later years. However, even with it switching to show surface storage gains over a long period, the CSR still has issues with month-to-month and interannual variability that cannot be overlooked. The GLDAS has compared better with other products in terms of variability, but with its long-term storage losses it is either underestimating gains in the winter or overestimating losses in the summer. This disagreement between estimated ΔS, GLDAS, and CSR should be further explored. Discrepancies could possibly be due to the smaller basin size and the coarse resolution of the data. Fersch et al. (2012) found that GRACE estimates also had limitations in basins with smaller water storage changes (less than 25 mm month−1). It is beyond the scope of this study to determine how the internal physics of the GLDAS and CSR algorithms affect the estimates over the UCRB, but it should be considered for further research.
Finally, the main issue with MERRA is with its long-term ∇ · Q. When compared with the surface runoff, its ∇ · Q estimates are far off. When calculating a 5-yr running mean of ∇ · Q from WY1999 through WY2008, MERRA consistently shows long-term divergence, while ERA-I always shows convergence, which is more realistic for the UCRB.
5. Conclusions
Utilizing in situ, reanalysis, and satellite-derived datasets, each component of the surface and atmospheric water budgets is analyzed for the Upper Colorado River basin. In general, all products capture the seasonal cycle of each water budget component. Table 3 provides an inventory of the various datasets for each water budget component. Reanalysis datasets tend to overestimate in situ precipitation while satellite-derived precipitation underestimates. In this region, where there is a fairly dense network of in situ observations that are consistent over time, in situ precipitation data is suggested for future studies. Based on the temporal and spatial resolution and its consistency with other components in the water budget calculations, the MPI is suggested for evapotranspiration measurements in the UCRB. Additionally, CPC soil moisture and AMSR-E snow water equivalent are also consistent among each other in water budget calculations. As ERA-I provides the only ∇ · Q that is consistently convergent on an annual and long-term basis, it is suggested for future studies of atmospheric water budgets over the UCRB.
Summary of the various datasets for each water variable component: precipitation, evapotranspiration, soil moisture, surface storage changes, and water vapor divergence. The suggested dataset for each component is denoted with a ✓*.
In terms of interannual variability, there is a strong relationship between precipitation, ET, and soil moisture, while total runoff is more strongly correlated with SWE peaks. This key finding indicates that water years with greater amounts of rain but lower snowpack will result in increased ET lost to the atmosphere and lower surface runoff.
Using estimates of precipitation, ET, and runoff that are consistent among each other, the estimated change in surface storage is calculated. The magnitude and interannual variability of estimated ΔS is comparable to GLDAS and to calculated changes in (AMSR-E SWE + reservoir + soil moisture). When estimates of precipitation, ET, and ∇ · Q are used to estimate changes in atmospheric storage, there are generally small changes in annual atmospheric storage.
Several key results were found when analyzing long-term storage changes (shown in Table 2).
The estimated long-term ΔS shows a large gain in surface storage, while long-term changes in (reservoir + soil moisture) lead to a small gain in surface storage. Both the CSR and GLDAS actually show surface storage losses for the same time period.
Long-term ∇ · Q is convergent for ERA-I and divergent for MERRA. Adding surface runoff to either estimated ΔS or changes in (reservoir + soil moisture) yields a surplus of surface water. This means atmospheric convergence occurred (i.e., the ERA-I ∇ · Q yields more accurate results).
Increasing MPI ET by 5% leads to a closer balance between ERA-I ∇ · Q and surface runoff + estimated ΔS. The 5% increase also leads to a better balance in
Acknowledgments
This research is funded by a grant from the NOAA National Integrated Drought Information System. The authors wish to thank Nolan Doesken for his input and advice, Janice Bytheway and Paula Brown for their assistance working with MERRA and ERA-I data, and the two anonymous reviewers for their comments and suggestions.
REFERENCES
Aziz, O. A., Tootle G. A. , Gray S. T. , and Piechota T. C. , 2010: Identification of Pacific Ocean sea surface temperature influences of upper Colorado River basin snowpack. Water Resour. Res., 46, W07536, doi:10.1029/2009WR008053.
Bao, X., and Zhang F. , 2013: Evaluation of NCEP–CFSR, NCEP–NCAR, ERA-Interim and ERA-40 reanalysis datasets against independent sounding observations over the Tibetan Plateau. J. Climate, 26, 206–214.
Bosilovich, M. G., Robertson F. R. , and Chen J. , 2011: Global energy and water budgets in MERRA. J. Climate, 24, 5721–5739.
Chambers, D. P., 2006: Evaluation of new GRACE time-variable gravity data over the ocean. Geophys. Res. Lett., 33, L17603, doi:10.1029/2006GL027296.
Christensen, N. S., and Lettenmaier D. P. , 2007: A multimodel ensemble approach to climate change impacts on the hydrology and water resources of the Colorado River basin. Hydrol. Earth Syst. Sci., 11, 1417–1434.
Daly, C., Neilson R. P. , and Phillips D. L. , 1994: A statistical-topographic model for mapping climatological precipitation over mountainous terrain. J. Appl. Meteor., 33, 140–158.
Decker, M., Brunke M. , Wang Z. , Sakaguchi K. , Zeng X. , and Bosilovich M. G. , 2012: Evaluation of the reanalysis products from GSFC, NCEP, and ECMWF using flux tower observations. J. Climate, 25, 1916–1944.
Dee, D. P., and Coauthors, 2011: The ERA-Interim reanalysis: Configuration and performance of the data assimilation system. Quart. J. Roy. Meteor. Soc., 137, 553–597.
De Lannoy, G. J. M., Reichle R. H. , Arsenault K. R. , Houser P. R. , Kumar S. , Verhoest N. E. C. , and Pauwels V. R. N. , 2012: Multiscale assimilation of Advanced Microwave Scanning Radiometer–EOS snow water equivalent and Moderate Resolution Imaging Spectroradiometer snow cover fraction observations in northern Colorado. Water Resour. Res., 48, W01522, doi:10.1029/2011WR010588.
Entekhabi, D., Rodriguez-Iturbe I. , and Bras R. L. , 1992: Variability in large-scale water balance with land surface–atmosphere interaction. J. Climate, 5, 798–813.
Fan, Y., and van den Dool H. , 2004: Climate Prediction Center global monthly soil moisture data set at 0.5° resolution for 1948 to present. J. Geophys. Res., 109, D10102, doi:10.1029/2003JD004345.
Fersch, B., Kunstmann H. , Bardossy A. , Devaraju B. , and Sneeuw N. , 2012: Continental-scale water storage variation from global and dynamically downscaled atmospheric water budgets in comparison with GRACE-derived observations. J. Hydrometeor., 13, 1589–1603.
Fisher, J. B., Tu K. P. , and Baldocchi D. D. , 2008: Global estimates of the land-atmosphere water flux based on monthly AVHRR and ISLSCP-II data, validated at 16 FLUXNET sites. Remote Sens. Environ., 112, 901–919.
Groisman, P. Ya., Karl T. R. , Knight R. W. , and Stenchikov G. L. , 1994: Changes of snow cover, temperature, and radiative heat balance over the Northern Hemisphere. J. Climate, 7, 1633–1656.
Haiden, T., and Pistotnik G. , 2009: Intensity-dependent parameterization of elevation effects in precipitation analysis. Adv. Geosci., 20, 33–38.
Hamlet, A. F., Mote P. W. , Clark M. P. , and Lettenmaier D. P. , 2007: Twentieth-century trends in runoff, evapotranspiration, and soil moisture in the western United States. J. Climate, 20, 1468–1486.
Hirpa, F. A., Gebremichael M. , and Hopson T. , 2010: Evaluation of high-resolution satellite precipitation products over very complex terrain in Ethiopia. J. Appl. Meteor. Climatol., 49, 1044–1051.
Hoerling, M., and Eischeid J. , 2007: Past peak water in the Southwest. Southwest Hydrol., 6, 18–19.
Huang, J., van den Dool H. M. , and Georgakakos K. P. , 1996: Analysis of model-calculated soil moisture over the United States (1931–1993) and applications to long-range temperature forecasts. J. Climate, 9, 1350–1362.
Huffman, G. J., and Coauthors, 2007: The TRMM Multisatellite Precipitation Analysis (TMPA): Quasi-global, multiyear, combined-sensor precipitation estimates at fine scales. J. Hydrometeor., 8, 38–55.
Hurkmans, R., Troch P. A. , Uijlenhoet R. , Torfs P. , and Durcik M. , 2009: Effects of climate variability on water storage in the Colorado River basin. J. Hydrometeor., 10, 1257–1270.
Jimenez, C., and Coauthors, 2011: Global intercomparison of 12 land surface heat flux estimates. J. Geophys. Res., 116, D02102, doi:10.1029/2010JD014545.
Jung, M., Reichstein M. , and Bondeau A. , 2009: Towards global empirical upscaling of FLUXNET eddy covariance observations: Validation of a model tree ensemble approach using a biosphere model. Biogeosciences, 6, 2001–2013.
Kelly, R. E., 2009: The AMSR-E snow depth algorithm: Description and initial results. J. Remote Sens. Soc. Japan, 29, 307–317.
Kennedy, A. D., Dong X. , Xi B. , Xie S. , Zhang Y. , and Chen J. , 2011: A comparison of MERRA and NARR reanalyses with DOE ARM SGP data. J. Climate, 24, 4541–4557.
Kim, T. W., Valdes B. , Nijssen B. , and Roncayolo D. , 2005: Quantification of linkages between large-scale climatic patterns and precipitation in the Colorado River Basin. J. Hydrol., 321, 173–186.
Kustas, W. P., and Coauthors, 1991: An interdisciplinary field study of the energy and water fluxes in the atmosphere–biosphere system over semiarid rangelands: Description and some preliminary results. Bull. Amer. Meteor. Soc., 72, 1683–1705.
McCabe, G. J., and Dettinger M. D. , 2002: Primary modes and predictability of year-to-year snowpack variations in the western United States from teleconnections with Pacific Ocean climate. J. Hydrometeor., 3, 13–25.
McCabe, G. J., and Wolock D. M. , 2007: Warming may create substantial water supply shortages in the Colorado River basin. Geophys. Res. Lett., 34, L22708, doi:10.1029/2007GL031764.
Miller, W. P., and Piechota T. C. , 2008: Regional analysis of trend and step changes observed in hydroclimatic variability around the Colorado River Basin. J. Hydrometeor., 9, 1020–1034.
Mueller, B., and Coauthors, 2011: Evaluation of global observations-based evapotranspiration datasets and IPCC AR4 simulations. Geophys. Res. Lett., 38, L06402, doi:10.1029/2010GL046230.
NRCS NWCC, cited 2012: SNOTEL data network: General information. [Available online at http://www.wcc.nrcs.usda.gov/snow/about.html.]
Peixoto, J. P., and Oort A. H. , 1992: Physics of Climate. Springer, 520 pp.
Reichle, R. H., Koster R. D. , De Lannoy G. J. M. , Forman B. A. , Liu Q. , Mahanama S. P. P. , and Toure A. , 2011: Assessment and enhancement of MERRA land surface hydrology estimates. J. Climate, 24, 6322–6338.
Rienecker, M., and Coauthors, 2011: MERRA: NASA’s Modern-Era Retrospective Analysis for Research and Applications. J. Climate, 24, 3624–3648.
Rodell, M., Famiglietti J. S. , Chen J. , Seneviratne S. I. , Viterbo P. , Holl S. , and Wilson C. R. , 2004a: Basin scale estimates of evapotranspiration using GRACE and other observations. Geophys. Res. Lett., 31, L20504, doi:10.1029/2004GL020873.
Rodell, M., and Coauthors, 2004b: The Global Land Data Assimilation System. Bull. Amer. Meteor. Soc., 85, 381–394.
Saha, S., and Coauthors, 2010: The NCEP Climate Forecast System Reanalysis. Bull. Amer. Meteor. Soc., 91, 1015–1057.
Sapiano, M. R. P., and Arkin P. A. , 2009: An intercomparison and validation of high-resolution satellite precipitation estimates with 3-hourly gauge data. J. Hydrometeor., 10, 149–166.
Seager, R., and Coauthors, 2007: Model projections of an imminent transition to a more arid climate in southwestern North America. Science, 316, 1181–1184.
Swenson, S. C., and Wahr J. , 2006: Post-processing removal of correlated errors in GRACE data. Geophys. Res. Lett., 33, L08402, doi:10.1029/2005GL025285.
Switanek, M. B., Troch P. A. , and Castro C. L. , 2009: Improving seasonal predictions of climate variability and water availability at the catchment scale. J. Hydrometeor., 10, 1521–1533.
Tang, Q., Gao H. , Yeh P. , Oki T. , Su F. , and Lettenmaier D. P. , 2010: Dynamics of terrestrial water storage change from satellite and surface observations and modeling. J. Hydrometeor., 11, 156–170.
Tootle, G. A., and Piechota T. C. , 2006: Relationships between Pacific and Atlantic Ocean sea surface temperatures and U.S. streamflow variability. Water Resour. Res., 42, W07411, doi:10.1029/2005WR004184.
Trenberth, K. E., Stepaniak D. P. , and Caron J. M. , 2000: The global monsoon as seen through the divergent atmospheric circulation. J. Climate, 13, 3969–3993.
Vera, C., and Coauthors, 2006: Toward a unified view of the American monsoon systems. J. Climate, 19, 4977–5000.
Zaitchik, B. F., Rodell M. , and Reichle R. H. , 2008: Assimilation of GRACE terrestrial water storage data into a land surface model: Results for the Mississippi River basin. J. Hydrometeor., 9, 535–548.
