1. Introduction
Surface soil moisture plays an important role in controlling the exchange of land–atmosphere water and energy fluxes (Walker and Rowntree 1977). It is also an important indicator of near-surface hydrologic conditions and so it is commonly used for drought monitoring (Oglesby and Erickson 1989; Quiring and Papakryiakou 2003) and climate forecasting (Koster and Suarez 2001). Field measurement of soil moisture is time-consuming and expensive, making it difficult to monitor at a regional scale. As a result, there are few observational soil moisture datasets, and most applications (e.g., drought monitoring) that require soil moisture information rely on model simulations.
Accurately simulating soil moisture is difficult because of the complex interactions between atmospheric and land surface conditions and soil characteristics (Robock et al. 1998). The atmospheric conditions that affect surface soil moisture include precipitation, temperature, wind speed, solar radiation, and upper-level atmospheric circulation. Precipitation plays a dominant role in modifying soil moisture conditions, and soil moisture in turn influences precipitation. Model simulations have demonstrated that precipitation patterns are affected by soil moisture anomalies (Koster and Suarez 2001; Koster et al. 2004; Mo and Juang 2003). Temperature and soil moisture are also strongly coupled. Soil moisture controls the partitioning of energy into sensible and latent heat fluxes, and consequently influences near-surface air temperatures (Huang et al. 1996).
Soil and vegetation characteristics affect soil moisture (Noilhan and Planton 1989). Soil characteristics determine the hydraulic conductivity, field capacity, permanent wilting point, and albedo of the soil. In addition, soil properties affect the partitioning of precipitation into infiltration and surface runoff and determine the total amount of water that can be stored in the soil profile. Vegetation type and growth stage affect soil moisture by altering rooting depths and the rate of root water uptake (Zuo et al. 2004). When vegetation grows it needs to absorb the water from surrounding soil, which decreases soil water content. Root density distribution varies with depth; therefore, dividing subsurface into several layers is necessary to accurately model the vertical distribution of soil moisture.
Due to the complexity of the soil–vegetation–atmosphere transfer scheme (SVAT), it is difficult to model all of the relevant climatic, edaphic, and environmental factors. Appropriate simplifications are needed to develop soil moisture models. It is often believed that complex models are preferable because they can more closely approximate reality. However, complex models typically require more input data, which limits their applicability. Simple models are advantageous because they require less input data and are easier to use. Intermodel comparisons are commonly used to validate existing models and provide insights into improving the accuracy of model simulations (Chen et al. 1997; Koster and Milly 1997).
Given the lack of in situ soil moisture observations, a variety of models have been developed to simulate soil moisture. These land surface models can be broadly classified as either coupled or uncoupled models (Dirmeyer et al. 2004). Coupled atmosphere–land surface models include general circulation models (GCMs) such as those evaluated for the Atmospheric Model Intercomparison Project (AMIP; Robock et al. 1998; Srinivasan et al. 2000) and reanalysis models such as the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis (Kalnay et al. 1996), the NCEP–Department of Energy (DOE) reanalysis (Kanamitsu et al. 2002), and the 40-yr European Centre for Medium-Range Weather Forecasts (ECMWF) Re-Analysis (ERA-40; Simmons and Gibson 2000). The accuracy of the soil moisture simulations for these three reanalysis products has been evaluated by Li et al. (2005) and Lu et al. (2005).
Uncoupled land surface models are also commonly used to simulate soil moisture. These models have varying complexities, but all rely on meteorological data to provide the upper boundary conditions (Dirmeyer et al. 2004). Uncoupled land surface models have been evaluated individually (Robock et al. 1995), in model intercomparison studies like the North American Land Data Assimilation System (NLDAS; Robock et al. 2003; Schaake et al. 2004), and against coupled and satellite-derived soil moisture estimates (Dirmeyer et al. 2004). The Project for the Intercomparison of Land-Surface Parameterization Schemes (PILPS) evaluated uncoupled land surface models and found that they can produce markedly different water and energy budgets even when the initial land surface conditions and atmospheric forcings are carefully controlled. For instance, Desborough et al. (1996) compared 13 land surface models and found that the simulations of soil evaporation and total evaporation varied greatly between models because of the different equations used in each model. To investigate the causes of these differences in performance, Koster and Milly (1997) applied a revised monthly water balance to investigate a hierarchy of land surface models and concluded that the interactions between evaporation and runoff processes can explain much of the intermodel variability. In one of the largest and most recent model intercomparison studies, Guo and Dirmeyer (2006) evaluated 11 different land surface models for a 10-yr period as part of the Second Global Soil Wetness Project (GSWP-2). These models were evaluated at sites in the former Soviet Union, the United States, China, and Mongolia using observations from the Global Soil Moisture Data Bank (GSMDB) (Robock et al. 2000). Median correlations ranged from 0.3 in Mongolia to 0.8 in Illinois, and median root-mean-square error (RMSE) ranged from 40 to 110 mm of soil water in the upper 1 m of soil. If the plant-available soil moisture is assumed to be 150 mm in the upper 1 m of soil, the errors in the simulated monthly soil moisture values ranged from ∼25% to 65%. This indicates that estimation of actual soil water content is still a problem for GSWP-2 models. Guo and Dirmeyer (2006) also concluded that the differences in skill between regions are much greater than the differences between models.
Results from all of the soil moisture model evaluation and intercomparison studies referenced above indicate that even if models are driven with the same meteorological data, they often produce different results because each model uses a unique land surface scheme for simulating soil moisture. Generally, most models are able to simulate the annual cycle well, but they tend to have difficulties simulating interannual variability (particularly in regions with strong seasonal soil moisture cycles; Dirmeyer et al. 2004) and actual soil water content (Guo and Dirmeyer 2006). There is also significant spatial variability in model performance; most models tend to simulate soil moisture more accurately in Illinois than in other GSMDB sites around the world (Dirmeyer et al. 2004; Guo and Dirmeyer 2006). This results partly from the quality of the meteorological observations needed to force the model (e.g., simulation accuracy proportional to rain gauge density) and partly from the inability of models to simulate all aspects of the hydrological cycle in each region (e.g., inability to handle frozen soils or accurately simulate other important processes).
All these studies have shown that model performance varies by location and that intermodel comparisons are useful for evaluating and improving model performance. However, to date, there have been no direct comparisons of the Variable Infiltration Capacity (VIC), the Decision Support System for Agrotechnology Transfer (DSSAT; Ritchie and Otter 1985), and the Climatic Water Budget (CWB; Thornthwaite 1948; Thornthwaite and Mather 1955) models. These three models have different complexities and therefore require different input data. Of the three models, VIC is the most sophisticated land surface model and it requires the most input data. CWB is the simplest model and only requires monthly temperature and precipitation data. These three models were evaluated to answer the following questions: Does the most complex model provide the most accurate predictions of soil moisture? Can simple models replace complex models without significantly reducing prediction skill? In this paper, the performance of these three models was evaluated and compared using observed soil moisture data from three Soil Climate Analysis Network (SCAN) sites: Bushland, Texas; Prairie View, Texas; and Powder Mill, Maryland. These three sites have different topographic, climatic, and edaphic conditions. Section 2 describes the structure of the three models; section 3 discusses their input data requirements. Modeling results and discussion are presented in section 4, and conclusions are given in section 5.
2. Model descriptions
a. VIC hydrological model
The Variable Infiltration Capacity hydrological model was first developed as a single-layer land surface model by Wood et al. (1992), and it was later expanded to a two-layer model by Liang et al. (1994). The VIC model is a semidistributed hydrological model that is capable of representing subgrid-scale variations in vegetation, available water holding capacity, and infiltration capacity (Liang et al. 1996a, b, 1994). The influence of variations in soil properties, topography, and vegetation within each grid cell is accounted for statistically by using a spatially varying infiltration capacity. VIC utilizes a soil–vegetation–atmosphere transfer scheme that accounts for the influence of vegetation and soil moisture on land–atmosphere moisture and energy fluxes, and the fluxes are balanced over each grid cell (Andreadis et al. 2005). The model has been utilized in basin-scale hydrological modeling, continental-scale simulations associated with NLDAS, and global-scale applications (Abdulla et al. 1996; Cherkauer and Lettenmaier 1999; Nijssen et al. 1997; Wood et al. 1997). A thorough evaluation of VIC was undertaken as part of NLDAS, and the results indicated that soil moisture is generally well simulated by the VIC model (Robock et al. 2003).
The model divides the subsurface into three soil layers. Each layer is characterized by a variety of parameters, such as bulk density, infiltration capacity, saturated hydraulic conductivity, soil layer depths, and soil moisture diffusion parameters. The land surface is described by approximately 11 land cover types. The vegetation types are characterized by their leaf area index (LAI), canopy resistance, and relative fraction of roots in each of the soil layers. Roots can extend to layer 1 (usually ∼10 cm) or deeper layers depending on vegetation and soil type. Bare soil (no vegetation cover) can also be simulated by the model.
The evapotranspiration from each land cover type is simulated using vegetation-class specific potential evapotranspiration ETp, canopy resistance, aerodynamic resistance to the transfer of water, and architectural resistance coefficients. In this model, the ETp includes evaporation from the canopy layer of each vegetation class, transpiration from each vegetation class, and evaporation from bare soil. Total evapotranspiration ETp over each grid cell is calculated as the area-weighted sum of these three components for each vegetation class. For each land cover type, there is a single canopy layer and three soil layers. The top layer is the most dynamic layer of the soil column, and it rapidly responds to daily weather (e.g., precipitation and temperature). Soil moisture in the lower layers of the soil varies more slowly (i.e., it has a lagged response to weather). A detailed description of the VIC model is provided by Liang et al. (1994, 1996a, b). The water balance mode of the VIC model was used in this study.
b. DSSAT soil moisture model
DSSAT requires information about the soil water content for the lower limit of plant water availability (i.e., the lowest volumetric water content at which plants can extract water, which corresponds closely to the permanent wilting point), the drainage upper limit (i.e., the highest volumetric water content of a soil after thorough wetting and gravity drainage, which is closely related to field capacity), and field saturation (i.e., the volumetric water content of a soil when all pores of the soil are filled with water) to calculate processes such as root uptake, drainage, and soil evaporation. These soil parameters are necessary for all layers of the model because of the heterogeneity in the subsurface. The depth for each layer must be specified. In general, each layer should be approximately 20 cm deep for the top layers and approximately 30 cm for lower layers, with a total number of 7–10 layers (Ritchie 1998). Several of the soil inputs are only required for the soil surface; these include the albedo of the soil, the limit of first stage soil evaporation, the runoff curve number, and the drainage coefficient. These variables are used to calculate the various components of the water balance in Eq. (1) (Ritchie 1998).
Daily runoff is computed in the DSSAT model using a modified United States Department of Agriculture (USDA) Soil Conservation Service curve number method (Williams et al. 1984). Soil water drainage is estimated based on a “tipping bucket” approach. The amount of drainage is calculated using the drainage coefficient, layer depth, volumetric water content, and the drained upper limit of soil water content. Upward unsaturated flow is approximated using a normalized soil water diffusion equation operating on a daily time step (Ritchie 1998).
Evaporation from the soil surface, root water uptake, and plant transpiration are based on methods developed by Ritchie (1972). The calculation of potential evapotranspiration (ETp) can be calculated using one of four options within DSSAT (Sau et al. 2004). The Priestley and Taylor (1972) method was employed in this research. Once ETp has been calculated, it is partitioned into potential soil evaporation and potential plant transpiration based on the fraction of solar energy reaching the soil surface and the LAI (Jones et al. 2003). Calculation of actual soil evaporation is based on a two-stage process, including the free soil evaporation and soil-limiting evaporation stages. The actual soil evaporation is the minimum of the free soil evaporation and soil-limiting evaporation on a daily basis. The actual plant transpiration is considered to be the minimum of the potential plant transpiration and potential root water uptake. The potential root water uptake is estimated by calculating a maximum water flow to roots in each layer and summing these values.
c. Climatic water budget
A modified version of a well-established climatic water budget model (Mather 1978; Thornthwaite 1948; Thornthwaite and Mather 1955) is the third method used to simulate soil moisture in the study area. The CWB is a one-dimensional model that calculates the daily or monthly changes in soil moisture storage due to evaporation, precipitation, infiltration, and runoff. The CWB assumes that the subsurface can be represented as a single soil layer. Soil moisture storage will increase whenever precipitation exceeds climatic demand for water (e.g., ETp). When the climatic demand is greater than precipitation, soil moisture storage will be depleted. Thus, the difference between precipitation and actual evapotranspiration is the estimated soil moisture change for a specific period of time.
The CWB model requires precipitation and temperature data as well as data for describing soil properties. Among the soil properties, the available water capacity data (AWC) of the soil is a key component in calculating soil moisture because it represents the maximum amount of water that plants can extract from the soil. AWC represents the difference between the field capacity (FC) and the permanent wilting point (PWP). PWP is a function of soil porosity and pore size; it represents the lower limit of plant-available soil moisture (Miller and White 1998).
3. Data description and methodology
a. Soil moisture data
Observed soil moisture data were obtained from three SCAN sites: Bushland (35°10′N, 102°6′W) and Prairie View (30°5′N, 95°59′W), Texas, and Powder Mill (39°1′N, 76°51′W), Maryland. For convenience, hereafter we use BL, PV, and PM to represent the Bushland, Prairie View, and Powder Mill sites, respectively.
BL is located in the Texas high plains where the climate can be defined as continental steppe. This climate type is typical of interiors of continents and is associated with relatively dry conditions. The average annual precipitation is approximately 482 mm, and the wettest period occurs between July and October (Larkin and Bomar 1983). BL is characterized by native, undisturbed rangeland that has never been plowed. The dominant vegetation is blue grama and buffalo grass and the dominant soil type is a well-drained dark brown silt clay. This site is relatively flat (1% slope).
PV is located in eastern Texas and it has a subtropical climate with humid, hot summers (Larkin and Bomar 1983). The average annual precipitation is roughly 1062 mm; May is the wettest month. This SCAN site is in an agricultural field that grows watermelons, and the dominant soil type is a moderately well-drained fine sandy loam. PV is a relatively flat (1% slope) site.
PM is located in the coastal plains of Maryland and has a continental climate with average annual precipitation of approximately 1100 mm (Quiring 2004). July and August are the wettest months. This site is covered by mixed grasses and the dominant soil type is a well-drained dark brown sandy loam. PM is a moderately sloped (4%) site.
All of the Natural Resources Conservation Service (NRCS) SCAN sites use Hydra Probe sensors to measure soil water content. The Hydra Probe sends an electromagnetic signal (50 MHz) into the soil. The reflected wave is associated with the electrical properties of the soil and can be used to determine the soil water content, conductivity, and salinity of the soil (www.stevenswater.com). The measured response (or dielectric permittivity) of the soil is related to soil water content using a calibration equation (Bosch 2004; Seyfried and Murdock 2004). A number of studies have evaluated the accuracy of the calibration equations and the Hydra Probe’s intersensor variability (Seyfried and Murdock 2004; Seyfried et al. 2005). For example, Bosch (2004) demonstrated that the Hydro Probe measurements using laboratory-calibrated equations were within 0.04 cm3 cm−3 of the observed water content and Seyfried et al. (2005) found that the average difference in soil moisture content varies from 0.027 (silt) to 0.053 cm3 cm−3 (clay) between soil specific calibrations and a general multisoil calibration. Seyfried et al. (2005) also examined intersensor variability by testing 30 sensors in four different fluids and found that the maximum coefficient of variability (CV) is 1.5% for individual sensor measurements. These studies indicate that the Hydra Probe can provide reliable and accurate measurements of soil water content under a variety of soil types and surface conditions (Seyfried et al. 2005).
The SCAN soil moisture data are measured hourly at depths of 5, 10, 20, 50, and 100 cm (available at http://www.wcc.nrcs.usda.gov/scan/). These hourly measurements were aggregated to a daily mean value. In total, 4 yr of observational data from BL and PM were used for evaluating the VIC and DSSAT models (2004 and 2005 for BL; 2002 and 2004 for PM). For convenience, 2004BL, 2005BL, 2002PM, and 2004PM are used to refer to the four simulations for the VIC and DSSAT models. The monthly averaged data (1997–2004) from PV and BL were only used for evaluating the CWB model because it simulates soil moisture using a monthly time step. For VIC and DSSAT, daily simulated soil moisture was compared to the in situ measurements. Both the model-simulated and observed soil moisture data were aggregated to compare soil moisture in the top 50 cm of the soil. The measured soil moisture in the top 50 cm was calculated by averaging the observations at 5, 10, 20, and 50 cm. For DSSAT and VIC, weighted soil moisture in the top 50 cm was calculated by assuming that soil moisture is vertically homogeneously distributed within each layer. This method has been applied in other similar studies (Robock et al. 2003). A preprocessing procedure was used to remove all possible outliers in the soil moisture measurements. In addition, the first two years (1995 and 1996) of SCAN data from PV and BL were eliminated because of instrumental problems.
b. Climatological data
The VIC model was driven using station-based measurements of daily maximum and minimum temperatures and precipitation. Additional meteorological and radiative forcings, such as vapor pressure, shortwave radiation, and net longwave radiation, were derived using established relationships with maximum and minimum temperatures, daily temperature range, and precipitation (Kimball et al. 1997; Thornton and Running 1999).
The DSSAT model also requires daily precipitation and minimum and maximum temperature data; these data were obtained from the same weather stations as VIC. The modeled solar radiation data used in the VIC model were also used in the DSSAT model. Thus, VIC and DSSAT were driven by the same meteorological and radiative forcing data.
Only monthly precipitation and minimum and maximum temperature data are required to run the CWB model; these data were obtained from the Precipitation-elevation Regressions on Independent Slopes Model (PRISM) dataset (Daly et al. 1994). The dataset contains 110 yr of monthly precipitation and temperature data at ∼4-km resolution (1895–present).
c. Soil and vegetation characteristics
Soil characteristics, including saturated hydraulic conductivity, porosity, soil water content at field capacity and wilting point, soil depths, and available water capacity (AWC), were available from the NRCS. These soil characteristics can also be calculated from soil texture and organic content data using the Rawls and Brakensiek (1985) method. Table 1 shows measured (from NRCS) and calculated (using Rawls and Brakensiek 1985) field capacity, wilting point, and available water holding capacity at BL and PM. Previous studies have demonstrated that proper parameterization of soil properties, such as hydraulic conductivity and wilting point, has a large impact on the accuracy of the model simulations (Xue et al. 1996, 1997). To account for the uncertainties in the soil parameters, four different runs were conducted at BL and PM, namely VIC-1 (using calculated soil parameters), VIC-2 (using measured soil parameters), DSSAT-1 (using calculated soil parameters), and DSSAT-2 (using measured soil parameters). All of the parameters required for VIC and DSSAT were set to be as identical as possible.
The additional soil parameters required by VIC, such as the variable infiltration curve parameter and maximum velocity of flow, were obtained from the calibrated regional simulations performed by Maurer et al. (2001, 2002). They selected a number of basins where observed streamflow data were available and then calibrated the VIC-simulated streamflow by adjusting soil parameters describing soil depth, baseflow drainage, and infiltration capacity of the soil layers (Maurer et al. 2001, 2002). This procedure is described in greater detail by Maurer et al. (2001, 2002).
The initial soil moisture for all of the model runs was set at field capacity. Land cover and vegetation parameters were derived using the global vegetation classification developed by Hansen et al. (2000). Because the vegetation at the BL and PM SCAN sites is mixed grass, the DSSAT grass module (CROPGRO-Bahia) was used.
d. Error analysis
e. Sensitivity analysis
The accuracy of the soil moisture simulations are influenced by the soil parameter estimates (Xue et al. 1996, 1997). Undoubtedly these parameters are associated with a variety of errors and uncertainties. A sensitivity analysis was employed to quantify the impact of parameter uncertainty on model performance. A sensitivity analysis is useful for determining how model parameters influence model results and for identifying parameters that have the greatest influence on model performance (Gebremichael and Barros 2006; Yildiz and Barros 2007). The first stage of the sensitivity analysis involved varying each parameter independently of the others and measuring how it influenced model performance. A total of 16 parameters were selected from each model for the sensitivity analysis and each parameter was varied by an arbitrary constant ±20%. Based on simulations at BL and PM using data from 2004, the five most significant parameters were further examined using a factorial analysis approach. Rather than employ the traditional “change one parameter at a time” approach, the factorial design method was employed (Box et al. 1978). This method of sensitivity analysis accounts for the interacting effects of model parameters (Gebremichael and Barros 2006; Yildiz and Barros 2007). Compared with Monte Carlo simulations, this method is simpler and less computationally intensive. It has been applied to many different environmental models (Barros 1996; Liong et al. 1995). The half-fraction factorial design of 5 parameters (25) was employed, resulting in 16 simulations for each model.
4. Results and discussion
a. VIC and DSSAT soil moisture simulations
Because both VIC and DSSAT models were run on a daily time step, the performance of these two models can be directly compared at BL and PM (Figs. 1 and 2). DSSAT and VIC both did well in simulating the annual cycle of soil moisture and daily patterns of the wetting and drying in response to weather conditions, as evidenced by the relatively strong correlations. The correlation between the model-simulated and measured soil moisture ranged from 0.51 to 0.95, with an average of 0.76 (Table 2).
Figures 1 and 2 show that all simulations produce similar daily variations in soil moisture, but the absolute magnitude of simulated soil moisture is quite different. DSSAT was slightly more accurate than VIC in simulating the actual soil water content in the top 50 cm of soil because it had a lower MAE and a higher E for 3 of 4 simulations (2002PM, 2004PM, and 2004BL). The coefficient of efficiency E was negative for 5 of 8 VIC simulations, which indicates that the observed mean soil moisture value was a better predictor of soil moisture conditions than VIC. VIC overestimated the actual soil water content in 7 of 8 simulations. This systematic bias has been found in other studies (Robock et al. 2003; Sheffield et al. 2003). However, VIC was more strongly correlated with observed soil moisture than DSSAT in 6 of the 8 simulations and it had a higher average correlation (0.80 versus 0.72).
At both BL and PM, the two VIC simulations (VIC-1 and VIC-2) are nearly identical, but the two DSSAT simulations (DSSAT-1 and DSSAT-2) differ from each other. In fact, changing the soil parameters did not significantly change VIC simulated soil moisture in the top 50 cm of the soil; only simulated soil moisture below 50 cm was slightly different (not shown). This suggests that DSSAT is more sensitive to the specified soil parameters than VIC.
The results also demonstrate that there is significant interannual and spatial variability in model performance. At PM, the 2002 and 2004 mean measured (mean VIC) soil moisture at PM was 0.122 (0.294) and 0.189 (0.320), respectively. VIC overestimated the magnitude of soil moisture by more than 100% and this is reflected in the low E values: −5.94 (2004 VIC-1) and −5.54 (2002 VIC-1) (Table 2). However, VIC was much more accurate in simulating the soil water content at BL (particularly in 2005). This spatial variability in VIC model performance has also been observed in other studies (Guo and Dirmeyer 2006). It can partially be attributed to differences in soil texture at the BL and PM sites. At the PM site the soil is sandy loam, which has higher hydraulic conductivity and lower field capacity and wilting points (Table 1) than the clay loam soil at the BL site. As a result, the observed soil moisture content at PM is significantly lower than at BL. This difference in soil water content between the two sites is not well captured by VIC.
The DSSAT soil moisture simulations using the calculated soil parameters (DSSAT-1) generally simulated the observed soil moisture more accurately than those using measured soil parameters. Table 2 shows that DSSAT-1 simulated soil moisture most accurately at PM in 2002. This simulation had the highest E (0.73) and the lowest MAE (0.02). A more detailed examination of the results revealed seasonal differences in the accuracy of the model simulations. In particular, DSSAT simulated soil moisture more accurately during the growth season than during other seasons (not shown). This is likely because DSSAT was primarily designed for simulating soil moisture for agricultural applications, and therefore it has been extensively tested and evaluated using growing season data.
Like VIC, DSSAT model performance also varied significantly from year to year. Model performance during the 2005BL simulation was significantly better than the 2004BL simulation, especially in terms of the strength of the correlations. At the PM site, the differences in model performance between 2002 and 2004 may result from the antecedent moisture conditions because 2003 was one of the wettest years on record in Maryland (Quiring 2004).
b. CWB soil moisture simulations
As can be seen from Fig. 3, the soil moisture for both BL and PV were poorly simulated by the CWB model. CWB simulated monthly soil moisture and treated the whole soil profile as a homogeneous unit, which is not practical given the heterogeneity of soil characteristics (Table 1). The annual cycle of soil moisture can be predicted in most cases by the CWB model, as evidenced by the correlations (>0.5). The scatter of points shows that CWB tends to underestimate the soil moisture content in the wettest months (Fig. 4). This might be caused by the constraint of the upper limit of soil moisture (field capacity). Results show that the highest observed soil moisture content exceeds the field capacity specified in the model; thus, changing the field capacity might allow the model to more closely replicate actual soil moisture conditions. Another cause of the poor model performance is that the model assumed that half of the surplus water (the difference between FC and soil water content) was converted to streamflow and half was held over to the next time step (i.e., the next month). This simple method of handling runoff and storage is not physically realistic and has a negative impact on the accuracy of the model simulations. The CWB model also does not account for net groundwater fluxes, which might be an important factor for upward or download flow recharge. Therefore, the poor simulation of soil moisture is not surprising. Our results suggest that CWB is not an appropriate model for simulating soil water content.
c. Comparison of VIC, DSSAT, and CWB model performance
Soil moisture simulations from three different models were evaluated and compared to determine which model simulates soil water content most accurately. All of these models have been used for simulating soil moisture, but they vary in their data requirements and level of complexity. The VIC model is the most complex and data intensive of the land surface models that were evaluated in this study because it accounts for subgrid-scale variability in vegetation and infiltration. VIC is commonly used for simulating hydrology and land surface processes at basin to global scales. DSSAT is a model of moderate complexity that is commonly used for simulating crop growth and evaluating the impact of various agricultural management decisions at field to regional scales. CWB is a simple model that is used to simulate the water balance at basin, regional, and global scales.
Of the three models, DSSAT and VIC simulated soil moisture in the top 50 cm at BL and PM with similar levels of accuracy (Table 2). Due to the small sample size, a paired t test was applied and demonstrated that the differences in the model performance statistics (E, MAE, r) between VIC and DSSAT are not statistically significant. Therefore, although there are some differences in model performance between VIC and DSSAT, both models demonstrated similar skill in simulating soil moisture in the upper 50 cm of the soil and the performance of both models varied significantly in time and space.
CWB was able to simulate the annual cycle and interannual variability of soil moisture. However, it could not accurately simulate the actual soil water content, as demonstrated by the large MAE and negative E at both the PV and BL sites. These performance issues, coupled with the coarse temporal resolution (monthly) and vertical resolution (a single layer) make CWB, despite its simplicity, the least desirable of the three models. Therefore, CWB was excluded from further consideration, and the sensitivity analysis and detailed examination of model differences will focus on VIC and DSSAT.
d. VIC and DSSAT evapotranspiration simulations
Insights into the differences in model performance between VIC and DSSAT can be gained by examining the other components of the water balance. Although both models were driven using the same meteorological and radiative forcing data, the evapotranspiration rates for the two models are quite different. Figure 5 shows the monthly actual evapotranspiration rate for BL and PM. It is evident that the DSSAT-simulated evapotranspiration is in good agreement with VIC-simulated evapotranspiration at PM. However, there are significant differences between DSSAT- and VIC-estimated evapotranspiration at BL. For the BL site in 2005, the DSSAT-simulated evapotranspiration rate peaked in June, but the VIC evapotranspiration rate was highest in April. VIC-simulated evapotranspiration was much less than DSSAT evapotranspiration for 2004BL. The estimated annual evapotranspiration also shows remarkable intermodel differences (Table 3). These differences can be attributed to the different schemes used by the two models to calculate evapotranspiration. DSSAT uses the Priestly–Taylor equation; VIC uses the Penman–Monteith equation. Previous studies have suggested that the performance of evapotranspiration equations varies in space (Sau et al. 2004). This is in good agreement with Desborough et al. (1996), who found that the difference between model-estimated soil evaporation rates results from the different methodologies used to calculate it. Because VIC and DSSAT also utilize different methodologies to partition potential evapotranspiration into soil evaporation and plant transpiration, this influences the vertical distribution of the simulated soil moisture. Ideally, the DSSAT and VIC evapotranspiration rates should be compared with observed data to evaluate model performance; however, the lack of observed data makes it impossible in this study.
e. VIC and DSSAT drainage and runoff simulations
Table 3 shows the estimated annual drainage (subsurface flow) and runoff (overland flow). It is clear that there are large differences in modeled drainage and runoff at the two SCAN sites (Figs. 6 and 7; Table 3). At BL, drainage and runoff combined account for less than 10% of annual precipitation. This indicates that most precipitation infiltrates the soil and very little water drains out the bottom of the soil profile. Drainage and runoff do not have a major impact on soil water content at this location. However, drainage and runoff are more important at PM; together they account for 18%–40% of annual precipitation. Both models generally simulated more runoff at PM than at BL (Table 3) because of a combination of factors including differences in the amount of precipitation, precipitation intensity, soil type (hydraulic conductivity), and slope.
Generally, both DSSAT and VIC produced a similar pattern of monthly runoff, although the amount of runoff differed between the two models (Fig. 6). Although DSSAT- and VIC-estimated annual runoff varied by more than a factor of 2 in some years, the absolute difference in mean annual runoff between the two models was less than 35 mm for all simulations.
Both models simulated more drainage at PM than at BL (Table 3). The differences in drainage can be attributed to differences in the depth of the soil profile, soil texture, and annual precipitation. The soil is much shallower at PM (129 cm) than at BL (229 cm) and it has a coarser texture (sandy loam) than BL (silty clay). PM also received more precipitation than BL. These factors help explain why annual drainage at PM was approximately 10 times greater than at BL. Generally, both DSSAT and VIC produced a similar pattern of monthly drainage pattern, except for 2004PM (Fig. 7). The absolute difference between DSSAT and VIC-estimated annual drainage was relatively small at BL (<18 mm) and it was relatively large at PM (∼100 mm).
Although there are significant differences in the amount of drainage and runoff simulated by the models (especially at PM), no consistent pattern was evident. During some simulations, DSSAT predicted higher amounts of runoff (drainage) than VIC, and during others VIC simulated more runoff (drainage) than DSSAT. Accurate simulation of drainage and runoff processes is necessary to get the soil water balance right, but the lack of observational data makes it impossible to evaluate how well the models are doing.
f. VIC and DSSAT sensitivity analysis
1) VIC
Uncertainty in model parameters can have a significant impact on the response of the model. A total of 16 VIC soil parameters were selected for the sensitivity analysis: binfilt (variable infiltration curve parameter); Dsmax (maximum velocity of baseflow); Ds (fraction of Dsmax at which nonlinear baseflow begins); Ws (fraction of maximum soil moisture at which nonlinear baseflow occurs); Ksat (saturated hydraulic conductivity); expt1, expt2, and expt3 (parameters describing the variation of Ksat with soil moisture); Wcr_Fract 1, Wcr_Fract 2, and Wcr_Fract 3 (parameters describing the fractional soil moisture content at the critical point); and Wpmp_ Fract 1, Wpmp_Fract 2, and Wpmp_Fract 3 (parameters describing the fractional soil moisture content at the wilting point). Each parameter was varied by an arbitrary ±20%. The model was run once using the upper value of the parameter (+20%) and once using the lower value of the parameter (−20%). Table 4 shows the difference in the mean annual soil water content between the simulations using the upper value and lower value of each parameter. In the majority of simulations, varying a parameter from 120% to 80% of its prescribed value produced a change in mean soil moisture of less than 1%. Note that even though the differences in mean soil moisture are small, they are statistically significant because of the even smaller standard deviation in the differences. Generally the sign and magnitude of the changes in mean soil moisture were relatively consistent across both sites, although PM tended to be more sensitive to changes in the parameters than BL. The five most significant parameters for both sites were Ds, Dsmax, Ws, expt1 and expt2. These parameters were selected for further sensitivity analysis using the factorial analysis approach (Box et al. 1978).
Table 5 shows the design of the half-fraction factorial analysis and the resulting mean soil moisture. The plus sign denotes the prescribed value of a parameter plus 20% and the minus sign denotes the prescribed value of a parameter minus 20%. The largest decrease in mean soil moisture was obtained from simulation 4 for each site, with a concurrent increase of Ds and Dsmax and decrease of Ws, expt1, and expt2. Simulation 4 decreased mean soil moisture by 6% at BL and 13% at PM. The largest increase in mean soil moisture was obtained from simulation 13, with a concurrent decrease of Ds and Dsmax, and increase of Ws, expt1, and expt2 (the opposite of simulation 4). Simulation 13 increased mean soil moisture by 4% and 10% at BL and PM sites, respectively. Three of these parameters are related to how the model simulates baseflow (Ds, Dsmax, and Ws) and the other two parameters (expt1 and expt2) control how the unsaturated hydraulic conductivity in the top two layers of the soil changes as a function of soil water content. In general, the values of Ds, Dsmax, and Ws control the threshold below (above) which linear (nonlinear) baseflow occurs according to the Arno model conceptualization (Franchini and Pacciani 1991). Accordingly, the decrease of Ds and Dsmax and the increase of Ws will lower the threshold value, which will eventually decrease the rate of baseflow and increase the soil moisture content. The variables expt1 and expt2 are the Brooks–Corey exponents for layer 1 and layer 2 and control the soil water retention curve. Higher values of expt1 and expt2 will make the soil water retention curve closer to the typical retention curve of clay soil and thereby decrease the rate at which water infiltrates and moves through the soil.
The sensitivity analysis shows that VIC is quite stable because relatively large changes in individual parameters and groups of parameters resulted in relatively small changes in mean soil moisture (<10%). The results also suggest that the model response is more sensitive at PM than at BL, which might be caused by the different climatology. This agrees with Demaria et al. (2007), who found that parameter sensitivity was more strongly controlled by climatic gradients than by changes in soil properties.
2) DSSAT
The same methodology was used to evaluate the sensitivity of DSSAT. Up to 16 soil parameters were selected for the sensitivity analysis: the SCS runoff curve number, the drainage coefficient, FC 1 to FC 7 (the soil moisture content at field capacity in layers 1 to 7), and WP 1 to WP 7 (the soil moisture content at the wilting point in layers 1 to 7). Each parameter was varied by an arbitrary ±20%, although because there are only five soil layers at PM site, no sensitivity analysis was undertaken for WP 6, WP 7, FC 6, and FC 7 (Table 6). The model was run once using the upper value of the parameter (+20%) and once using the lower value of the parameter (−20%). Table 6 shows the difference in the mean annual soil water content between the simulations using the upper value and lower value of each parameter. In the majority of simulations, varying a parameter from 120%–80% of its prescribed value produced a change in mean soil moisture of less than 5%. Generally the sign and magnitude of the changes in mean soil moisture were relatively consistent across both sites, although the changes tended to be larger at PM than at BL (Table 6). This is similar to VIC, which also showed greater sensitivity at PM. The five parameters that had the largest influence on mean soil moisture were the runoff curve and the field capacity of soil layers 2, 3, and 4 and the deepest soil layer (i.e., layer 5 at PM and layer 7 at BL). These parameters were further evaluated using the factorial analysis approach.
Table 7 shows the design of the half-fraction factorial analysis and the resulting mean soil moisture for BL and PM. Results indicate that the largest decrease in mean annual soil moisture at both sites was obtained from simulation 2, which corresponds to a higher runoff curve number and a decrease in the field capacity in four of the soil layers. Simulation 2 decreased soil moisture by 11% and 23% at BL and PM, respectively. The largest increase in mean soil moisture was obtained from simulations 7, 11, and 13. The average soil moisture content of all three simulations was 7% and 11% higher than the reference values at BL and PM sites, respectively. These simulations were associated with a lower runoff curve number and an increase in the field capacity of three of the soil layers. Not surprisingly, increasing the runoff curve number led to decreases in soil moisture because it reduced the amount of infiltration and increased the amount of overland flow. Decreasing the field capacity of the soil reduced soil moisture because the field capacity directly controls the amount of water that can be held against the pull of gravity in each soil layer.
The sensitivity analysis shows that DSSAT is more sensitive to changes in the model parameters (at least for the subset of parameters that were tested) than VIC because both the sensitivity analysis of the individual parameters and the groups of parameters produced much larger changes (up to 22%) in DSSAT-simulated soil moisture. This finding is also supported by the soil moisture simulations that were previously reported, which showed much larger differences between the DSSAT-1 and DSSAT-2 simulations than between VIC-1 and VIC-2 (Figs. 1 and 2). The results also suggest that DSSAT is more sensitive to changes in model parameters at PM than at BL.
5. Conclusions
The results of this study indicate that DSSAT and VIC more accurately simulate soil moisture than CWB. The overall accuracy of VIC and DSSAT soil moisture simulations, as measured by the correlation coefficient and RMSE, compare favorably with the values reported in other model intercomparison studies (cf. Guo and Dirmeyer 2006). In this study, model complexity was not a perfect predictor of model performance. Although the limitations of the least complex model (CWB) were readily apparent, the performance of the model of moderate complexity (DSSAT) was statistically indistinguishable from the most complex model (VIC).
The analysis revealed significant spatial variations in model performance. For example, VIC simulated soil moisture more accurately at BL than at PM. These spatial variations in model performance have been identified in other studies (Dirmeyer et al. 2004; Guo and Dirmeyer 2006). Model performance also varied significantly from year to year. Both VIC and DSSAT simulations at BL were significantly more accurate in 2005 than in 2004. In 2005, the correlation between DSSAT soil moisture and the observations increased approximately 80%, and E increased from 0.23 to 0.41. In addition, DSSAT also exhibited intra-annual variations in model performance because it tended to simulate soil moisture more accurately during the growing season. These variations in model performance demonstrate that it is difficult to develop a model that can accurately simulate soil moisture under a variety of edaphic and climatic conditions.
The primary cause of differences in model performance is the different land surface scheme used by each model. For example, although VIC and DSSAT were run using the same meteorological and radiative forcing data, there was significant intermodel variability in the simulated evapotranspiration, drainage, and runoff. These differences are important because the models need to be able to simulate all aspects of the soil water balance to accurately predict soil water content. However, given the lack of observation data, it was only possible to verify the accuracy of the DSSAT and VIC soil moisture simulations. Previous studies have also demonstrated that differences in model formulation and land surface schemes have a significant impact on simulated soil moisture (Dirmeyer et al. 2004; Robock et al. 2003).
Some of the variation in model performance is also likely due to model differences in the number of soil layers and the layer depths. Soil properties can change dramatically over short distances. The one-layer CWB model does not account for any of the vertical changes in soil properties, which greatly reduces model performance. VIC and DSSAT used the same soil properties obtained directly from soil surveys conducted at the SCAN sites. DSSAT can divide the subsurface into up to 10 layers, but VIC only uses three layers. This allows DSSAT to more accurately account for the vertical heterogeneity in soil properties. In addition, DSSAT was designed for agricultural applications and therefore has been extensively tested and evaluated at sites similar to those used in this study. VIC was primarily designed for hydrological applications; accordingly, previous evaluations have emphasized the accurate simulation of streamflow.
The sensitivity analysis focused on examining the sensitivity of VIC and DSSAT to a selection of soil parameters required by each model both because soil parameters have a large and direct influence on the soil moisture simulations and because soil properties are extremely spatially heterogeneous. In addition, these models are often applied in areas where detailed soil surveys are not available and therefore the soil parameters must be estimated from relatively coarse datasets [e.g., NRCS State Soil Geographic (STATSGO) data]. The sensitivity analysis demonstrated that both models are sensitive to changes in the model parameters (VIC was most sensitive to changes in Ds, Dsmax, Ws, expt1, and expt2; DSSAT was most sensitive to changes in the runoff curve number and the field capacity in four of the soil layers), although it appears that overall DSSAT is more sensitive than VIC. The factorial analysis revealed that changing a number of parameters simultaneously can have a greater influence on the model than varying an individual parameter. Our results also showed that model sensitivity varied by location; the sensitivity analysis produced larger changes in soil moisture at PM than at BL. Therefore, model sensitivity is a function of changes not only in soil parameters but also in the climate. Generally, the results of the sensitivity analysis demonstrated that a portion of the systematic error in the soil moisture simulations may be attributable to uncertainties in the model parameters. However, the sensitivity analysis did not attempt to quantify the uncertainty in each of the parameter estimates (an arbitrary change of ±20% was applied to all parameters).
Our results suggest that model complexity is not always a good indicator of model accuracy. The accuracy of soil moisture simulations varies not only as a function of the land surface scheme utilized by the model but also as a function of location and time. Therefore, it is important to undertake a careful evaluation of model performance, using data from the location and time period of interest, to determine which model is most suitable for simulating soil moisture conditions at a given location.
Acknowledgments
This research was partially supported by a contract from the Texas Water Development Board. The authors thank Dr. Timothy Hawkins for providing the code for calculating the CWB. The authors would also like to thank Dr. Ana Barros and other anonymous reviewers for their comments and suggestions, which improved this paper.
REFERENCES
Abdulla, F. A., Lettenmaier D. P. , Wood E. F. , and Smith J. A. , 1996: Application of a macroscale hydrologic model to estimate the water balance of the Arkansas–Red River basin. J. Geophys. Res., 101 , D3. 7449–7459.
Andreadis, K. M., Clark E. A. , Wood A. W. , Hamlet A. F. , and Lettenmaier D. P. , 2005: Twentieth-century drought in the conterminous United States. J. Hydrometeor., 6 , 985–1001.
Barros, A. P., 1996: An evaluation of model parameterizations of sediment pathways: A case study for the Tejo estuary. Cont. Shelf Res., 16 , 1725–1749.
Bosch, D. D., 2004: Comparison of capacitance-based soil water probes in coastal plain soils. Vadose Zone J., 3 , 1380–1389.
Box, G. E. P., Hunter W. G. , and Hunter J. S. , 1978: Statistics for Experimenters: An Introduction to Design, Data Analysis, and Model Building. Wiley, 653 pp.
Chen, T. H., and Coauthors, 1997: Cabauw experimental results from the project for intercomparison of land-surface parameterization schemes. J. Climate, 10 , 1194–1215.
Cherkauer, K. A., and Lettenmaier D. P. , 1999: Hydrologic effects of frozen soils in the upper Mississippi River basin. J. Geophys. Res., 104 , D16. 19599–19610.
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.
Demaria, E. M., Nijssen B. , and Wagener T. , 2007: Monte Carlo sensitivity analysis of land surface parameters using the Variable Infiltration Capacity model. J. Geophys. Res., 112 .D11113, doi:10.1029/2006JD007534.
Desborough, C. E., Pitman A. J. , and Irannejad P. , 1996: Analysis of the relationship between bare soil evaporation and soil moisture simulated by 13 land surface schemes for a simple non-vegetated site. Global Planet. Change, 13 , 47–56.
Dirmeyer, P. A., Guo Z. C. , and Gao X. , 2004: Comparison, validation, and transferability of eight multiyear global soil wetness products. J. Hydrometeor., 5 , 1011–1033.
Franchini, M., and Pacciani M. , 1991: Comparative analysis of several conceptual rainfall runoff models. J. Hydrol., 122 , 161–219.
Gebremichael, M., and Barros A. P. , 2006: Evaluation of MODIS gross primary productivity (GPP) in tropical monsoon regions. Remote Sens. Environ., 100 , 150–166.
Guo, Z., and Dirmeyer P. A. , 2006: Evaluation of the Second Global Soil Wetness Project soil moisture simulations: 1. Intermodel comparison. J. Geophys. Res., 111 .D22S02, doi:10.1029/2006JD007233.
Hansen, M. C., Defries R. S. , Townshend J. R. G. , and Sohlberg R. , 2000: Global land cover classification at 1-km spatial resolution using a classification tree approach. Int. J. Remote Sens., 21 , 1331–1364.
Huang, J., van den Dool H. M. , and Georgakakos K. P. , 1996: Analysis of model-calculated soil moisture over the United States (1931–93) and applications to long-range temperature forecasts. J. Climate, 9 , 1350–1362.
Jones, J. W., and Coauthors, 2003: The DSSAT cropping system model. Eur. J. Agron., 18 , 235–265.
Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77 , 437–471.
Kanamitsu, M., Ebisuzaki W. , Woollen J. , Yang S-K. , Hnilo J. J. , Fiorino M. , and Potter G. L. , 2002: NCEP-DOE AMIP-II Reanalysis (R-2). Bull. Amer. Meteor. Soc., 83 , 1631–1643.
Kimball, J. S., Running S. W. , and Nemani R. , 1997: An improved method for estimating surface humidity from daily minimum temperature. Agric. For. Meteor., 85 , 87–98.
Koster, R. D., and Milly P. C. D. , 1997: The interplay between transpiration and runoff formulations in land surface schemes used with atmospheric models. J. Climate, 10 , 1578–1591.
Koster, R. D., and Suarez M. J. , 2001: Soil moisture memory in climate models. J. Hydrometeor., 2 , 558–570.
Koster, R. D., and Coauthors, 2004: Regions of strong coupling between soil moisture and precipitation. Science, 305 , 1138–1140.
Larkin, T. J., and Bomar G. W. , 1983: Climatic Atlas of Texas. Texas Department of Water Resources, 151 pp.
Legates, D. R., and McCabe G. J. , 1999: Evaluating the use of “goodness-of-fit” measures in hydrologic and hydroclimatic model validation. Water Resour. Res., 35 , 233–241.
Li, H. B., Robock A. , Liu S. , Mo X. , and Viterbo P. , 2005: Evaluation of reanalysis soil moisture simulations using updated Chinese soil moisture observations. J. Hydrometeor., 6 , 180–193.
Liang, X., Lettenmaier D. P. , Wood E. F. , and Burges S. J. , 1994: A simple hydrologically based model of land surface water and energy fluxes for general circulation models. J. Geophys. Res., 99 , D7. 14415–14428.
Liang, X., Lettenmaier D. P. , and Wood E. F. , 1996a: One-dimensional statistical dynamic representation of subgrid spatial variability of precipitation in the two-layer variable infiltration capacity model. J. Geophys. Res., 101 , D16. 21403–21422.
Liang, X., Wood E. F. , and Lettenmaier D. P. , 1996b: Surface soil moisture parameterization of the VIC-2L model: Evaluation and modification. Global Planet. Change, 13 , 195–206.
Liong, S. Y., Shreeram J. , and Ibrahim Y. , 1995: Catchment calibration using fractional-factorial and central-composite-designs-based response surface. J. Hydraul. Eng., 121 , 507–510.
Lu, C. H., Kanamitsu M. , Roads J. O. , Ebisuzaki W. , Mitchell K. E. , and Lohmann D. , 2005: Evaluation of soil moisture in the NCEP–NCAR and NCEP–DOE global reanalyses. J. Hydrometeor., 6 , 391–408.
Mather, J. R., 1978: The Climatic Water Budget in Environmental Analysis. Lexington Books, 239 pp.
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. 17841–17862.
Maurer, E. P., Wood A. W. , Adam J. C. , Lettenmaier D. P. , and Nijssen B. , 2002: A long-term hydrologically based dataset of land surface fluxes and states for the conterminous United States. J. Climate, 15 , 3237–3251.
Miller, D. A., and White R. A. , 1998: A conterminous United States multi-layer soil characteristics data set for regional climate and hydrology modeling. Earth Interactions, 2 .[Available online at http://EarthInteractions.org.].
Mo, K. C., and Juang H-M. H. , 2003: Relationships between soil moisture and summer precipitation over the Great Plains and the Southwest. J. Geophys. Res., 108 .8610, doi:10.1029/2002JD002952.
Nijssen, B., Lettenmaier D. P. , Liang X. , Wetzel S. W. , and Wood E. F. , 1997: Streamflow simulation for continental-scale river basins. Water Resour. Res., 33 , 711–724.
Noilhan, J., and Planton S. , 1989: A simple parameterization of land surface processes for meteorological models. Mon. Wea. Rev., 117 , 536–549.
Oglesby, R. J., and Erickson D. J. , 1989: Soil moisture and the persistence of North American drought. J. Climate, 2 , 1362–1380.
Priestley, C. H. B., and Taylor R. J. , 1972: On the assessment of surface heat flux and evaporation using large-scale parameters. Mon. Wea. Rev., 100 , 81–92.
Quiring, S. M., 2004: Developing a real-time agricultural drought monitoring system for Delaware. Publications in Climatology, Vol. 57, University of Delaware, 104 pp.
Quiring, S. M., and Papakryiakou T. N. , 2003: An evaluation of agricultural drought indices for the Canadian prairies. Agric. For. Meteor., 118 , 49–62.
Rawls, W. J., and Brakensiek D. L. , 1985: Prediction of soil water properties for hydrologic modeling. Proc. Symp. on Watershed Management in the Eighties, Denver, CO, American Society of Civil Engineering, 293–299.
Ritchie, J. T., 1972: Model for predicting evaporation from a row crop with incomplete cover. Water Resour. Res., 8 , 1204–1213.
Ritchie, J. T., 1998: Soil water balance and plant stress. Understanding Options for Agricultural Production, G. Y. Tsuji, G. Hoogenboom and P. K. Thornton, Eds., Kluwer Academic, 45–58.
Ritchie, J. T., and Otter S. , 1985: Description and performance of CERES-Wheat: A user-oriented wheat yield model. ARS Wheat Yield Project, USDA Rep. ARS-38, 159–175.
Robock, A., Vinnikov K. Ya , Schlosser C. A. , Speranskaya N. A. , and Xue Y. , 1995: Use of midlatitude soil moisture and meteorological observations to validate soil moisture simulations with biosphere and bucket models. J. Climate, 8 , 15–35.
Robock, A., Schlosser C. A. , Vinnikov K. Ya , Speranskaya N. A. , Entin J. K. , and Qiu S. , 1998: Evaluation of the AMIP soil moisture simulations. Global Planet. Change, 19 , 181–208.
Robock, A., Vinnikov K. Ya , Srinivasan G. , Entin J. K. , Hollinger S. E. , Speranskaya N. A. , Liu S. , and Namkhai A. , 2000: The Global Soil Moisture Data Bank. Bull. Amer. Meteor. Soc., 81 , 1281–1299.
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.
Sau, F., Boote K. J. , Bostick W. M. , Jones J. W. , and Minguez M. I. , 2004: Testing and improving evapotranspiration and soil water balance of the DSSAT crop models. Agron. J., 96 , 1243–1257.
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.
Seyfried, M. S., and Murdock M. D. , 2004: Measurement of soil water content with a 50-MHz soil dielectric sensor. Soil Sci. Soc. Amer. J., 68 , 394–403.
Seyfried, M. S., Grant L. E. , Du E. , and Humes K. , 2005: Dielectric loss and calibration of the Hydra Probe soil water sensor. Vadose Zone J., 4 , 1070–1079.
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.
Simmons, A. J., and Gibson J. K. , 2000: The ERA-40 Project Plan. ERA-40 Project Report Series 1, 63 pp.
Srinivasan, G., Robock A. , Entin J. K. , Luo L. , Vinnikov K. Ya , and Viterbo P. , 2000: Soil moisture simulations in revised AMIP models. J. Geophys. Res., 105 , D21. 26635–26644.
Thornthwaite, C. W., 1948: An approach toward a rational classification of climate. Geogr. Rev., 38 , 55–94.
Thornthwaite, C. W., and Mather J. R. , 1955: The water balance. Publications in Climatology, Vol. 8, University of Delaware, 104 pp.
Thornton, P. E., and Running S. W. , 1999: An improved algorithm for estimating incident daily solar radiation from measurements of temperature, humidity, and precipitation. Agric. For. Meteor., 93 , 211–228.
Walker, J., and Rowntree P. R. , 1977: Effect of soil moisture on circulation and rainfall in a tropical model. Quart. J. Roy. Meteor. Soc., 103 , 29–46.
Williams, J. R., Jones C. A. , and Dyke P. T. , 1984: A modeling approach to determining the relationship between erosion and soil productivity. Trans. ASAE, 27 , 129–144.
Willmott, C. J., and Matsuura K. , 2005: Advantages of the mean absolute error (MAE) over the root mean square error (RMSE) in assessing average model performance. Climate Res., 30 , 79–82.
Willmott, C. J., Rowe C. M. , and Mintz Y. , 1985: Climatology of the terrestrial seasonal water cycle. J. Climatol., 5 , 589–606.
Wood, E. F., Lettenmaier D. P. , and Zartarian V. G. , 1992: A land-surface hydrology parameterization with subgrid variability for general circulation models. J. Geophys. Res., 97 , D3. 2717–2728.
Wood, E. F., Lettenmaier D. , Liang X. , Nijssen B. , and Wetzel S. W. , 1997: Hydrological modeling of continental-scale basins. Annu. Rev. Earth Planet. Sci., 25 , 279–300.
Xue, Y. K., Zeng F. J. , and Schlosser C. A. , 1996: SSiB and its sensitivity to soil properties: A case study using HAPEX-Mobilhy data. Global Planet. Change, 13 , 183–194.
Xue, Y. K., Sellers P. J. , Zeng F. J. , and Schlosser C. A. , 1997: Comments on “Use of midlatitude soil moisture and meteorological observations to validate soil moisture simulations with biosphere and bucket models”. J. Climate, 10 , 374–376.
Yildiz, O., and Barros A. P. , 2007: Elucidating vegetation controls on the hydroclimatology of a mid-latitude basin. J. Hydrol., 333 , 431–448.
Zuo, Q., Meng L. , and Zhang R. , 2004: Simulating soil water flow with root-water-uptake applying an inverse method. Soil Sci., 169 , 13–24.
Measured vs calculated field capacity, wilting point, and AWC at BL and PM.
Model performance statistics for the VIC, DSSAT, and CWB soil moisture models.
Annual drainage, runoff, and evapotranspiration (all in mm) simulated by DSSAT and VIC at BL and PM.
VIC Factorial analysis for the BL and PM sites. Here, Δ is the change in simulated mean soil moisture (cm cm−1 × 100) caused by increasing the parameter from 80% of the estimate to 120% of the estimate. Nonzero denotes significant difference in the mean at 5% level; binfilt is the variable infiltration curve parameter; Ds is fraction of Dsmax at which nonlinear baseflow begins; Dsmax is maximum velocity of baseflow; Ws is fraction of maximum soil moisture at which nonlinear baseflow occurs; Ksat is saturated hydraulic conductivity; expt is a parameter describing the variation of Ksat with soil moisture; Wcr_Fract is fractional soil moisture content at the critical point; and Wpmp_Fract is fractional soil moisture content at the wilting point. The number in parameter names indicates the number of layers. Asterisks indicate the five most significant parameters that are common for both sites; these parameters are used for further factorial analysis.
Design of the 25 half-fraction factorial sensitivity analysis for the BL and PM sites for the VIC-3L model. Columns 2–6 show parameter names for the five parameters in the analysis. A plus sign indicates that the parameter was set at 120% of the estimate; a minus sign indicates 80% of the estimate. The last row as reference shows the resulting mean soil moisture based on the chosen model parameter estimates. The percent values in parentheses indicate the change relative to the reference values in the last row.
DSSAT factorial analysis for the BL and PM sites. Here, Δ is the change in simulated mean soil moisture (cm cm−1 × 100) in the total zone by changing the parameter from 120%–80% of the estimate. Runoff curve is the SCS runoff curve number, Drainage coeff. is the drainage coefficient, FC 1 to FC 7 indicates the soil moisture content at field capacity in layers 1 to 7, and WP 1 to WP 7 indicates the soil moisture content at the wilting point in layers 1 to 7. Asterisks indicate the five most significant parameters that are common for both sites; these parameters are used in the factorial analysis (note that FC 5 and FC 7 are for PM and BL, respectively).
Design of the 25 half-fraction factorial sensitivity analysis for the BL and PM sites for DSSAT model.