1. Introduction

Soil moisture (SM) is an important mediator in the hydrologic cycle that converts a time series characterized by white noise (precipitation) into a time series characterized by red noise (soil moisture) (Delworth and Manabe 1988; Vinnikov and Yeserkepova 1991). Reliable soil moisture sensors have recently become available but, due to the lack of long-term soil water data, it is difficult to determine how soil moisture varies historically (Georgakakos et al. 1995). If we can estimate soil moisture reliably from long-term climate data, then the role of soil moisture in regional and global water cycles can be better understood. Two decades ago Schmugge et al. (1980) discussed the importance of such SM estimates for hydroclimatological studies in the absence of observed data. Recently, Crawford et al. (2000) suggested that SM parameterizations in numerical weather prediction models are overly simplified. They have also noted that, for example, the Eta Model (Black 1994) of the National Centers for Environmental Prediction (NCEP) and the fifth-generation Pennsylvania State University–National Center for Atmospheric Research (PSU–NCAR) Mesoscale Model (MM5; Dudhia 1993) use a climatological SM value to establish initial conditions.

In this paper we use data from a soil moisture process model to reconstruct and investigate past soil moisture “pseudo” climatologies for selected locations (Fig. 1) and for three land uses of the northern Great Plains (NGP). This study simulates and analyzes SM for multiple land uses for five depths up to 1.2 m at climate scales. This paper also investigates climatic variability of SM at multitemporal scales under contrasting land uses and assesses forcing of land use at these scales. This study is part of our ongoing effort to build and analyze an SM dataset for the NGP. We also raise the possibility that this model applied on a grid-by-grid basis could be used to initialize mesoscale meteorological and climatological models. The SM model was applied and successfully validated by Robinson and Hubbard (1990, RH hereafter), and is currently offered as a tool for irrigation scheduling in the Northern Great Plains. In previous studies, the SM model was applied to investigate the impacts of land use change on SM and evapotranspiration (ET) in the NGP (Mahmood and Hubbard 2002a), the role of irrigation on near-surface temperature records (Mahmood et al. 2002), the partitioning of energy (Mahmood et al. 2001), and the timing of precipitation in drought assessment (Meyer et al. 1993a,b). A sensitivity analysis of the SM and ET for soils heterogeneity is also completed (Mahmood and Hubbard 2003).

The RH model validation was completed for five predominant and representative land uses in the Great Plains at nine locations extending from eastern to western Nebraska (Robinson and Hubbard 1990). The validation was also provided for six soils layers up to the depth of 1.8 m. Land uses included in the validation were grass, corn, wheat, sorghum, and soybeans. The validation sites are representative of the east–west hydroclimatic gradient of the Great Plains and included several climatological zones. The data were collected over two growing seasons (April through September and/or October) in 1986 and 1987. This was already an improvement over Cherkauer and Lettenmaier (1999), for which the variable infiltration capacity (VIC) model validation was completed for one location, and also over Crawford et al. (2000), where validation was conducted for 1 month of data for grass only (July 1997). In addition, validation of the RH model was completed over a larger area than the previous studies (cf. Crawford et al. 2000). Our assessment shows that performance of the RH model is also superior to many land surface schemes (LSSs). Section 3 of this paper includes a detailed discussion on the RH model performance. For additional information on the model and its validation, the authors urge readers to consult Robinson and Hubbard (1990), Camargo (1993), and Camargo et al. (1994).

For this study, as noted above, the model is applied to three predominant land uses, namely, irrigated and rain-fed corn, and rain-fed grass. Irrigated corn has a longer growing season (May through late September and/or early October) compared to rain-fed corn (May through late August and/or early September) and grass (May–August). Water consumption also varies according to land use and differences in plant life cycles. Application of water for irrigated corn notably modifies natural root zone SM and hydrology (Mahmood and Hubbard 2002a). Thus, rain-fed and irrigated land uses with the same plants provide significantly different hydrological conditions. The RH model applications were repeated for three sites in Nebraska, representing a decreasing precipitation gradient from east to west, typical of the Great Plains (Fig. 2). The model estimates of soil water are for the top five layers for a period from 1982 to 1999. The five layers represented in the model are from 0–2.5, 2.5–30.5, 30.5–61, 61–91.5, and 91.5–122 cm, respectively. Vegetation- and crop-specific growing degree days (GDDs) were used for realistic representation of plant growth. GDD is calculated to determine crop phenology, which is associated with the potential for water consumption at different growth stages.

2. Background

A significant number of research projects have been undertaken and completed over the last two decades, focusing on the relationship between SM and climate. Hong and Pan (2000) reported a strong positive feedback between initial SM and simulated seasonal precipitation. The simulations were conducted by using the NCEP Regional Spectral Model (RSM). Model simulations show that soil water storage eventually effects moisture distribution within the boundary layer atmosphere and its structure (Hong and Pan 2000). Wang and Kumar (1998) suggested that interannual variations of SM may play a role in seasonal predictability of surface climate anomalies. They have found a strong correlation between SM and surface temperature anomalies. Dirmeyer (1999, 2000) also noted that incorrect SM inputs reduced the accuracy of simulated precipitation anomalies. Huang et al. (1996) reported from their study that SM is a better predictor of temperature over large areas during the summer season, compared to precipitation. A number of sensitivity studies for several LSSs were conducted under the Global Soil Wetness Project (GSWP) and the Project for Intercomparison of Land-Surface Parameterization Schemes (PILPS) (e.g., Qu et al. 1998; Morrill et al. 1999; Pitman et al. 1999; Dirmeyer 1999). PILPS undertook a phase-by-phase approach to compare and improve estimates of water and energy flux by the LSSs (Chen et al. 1997; Wood et al. 1998; Liang et al. 1998; Lohmann et al. 1998). Lohmann et al. (1998) noted that simulated changes in mean seasonal cycles of soil moisture storage agreed qualitatively with observations. In other words, it will be difficult for researchers and forecasters to confidently quantify soil moisture using these models.

To overcome some of the uncertainties in the design of LSSs, and to increase the accuracy of the estimated flux and storage of energy and water and to produce global SM data, a new project, the Land Data Assimilation System (LDAS) was undertaken (Mitchell et al. 2000). The success of this project remains to be seen. The Atmospheric Model Intercomparison Project (AMIP) reported that none of the AMIP models captured interannual variations in SM (Robock et al. 1998). The SM estimates provided by the revised AMIP models also show no improvement in capturing seasonal variations in SM (Srinivasan et al. 2000). The GSWP used 10 LSSs to produce global SM data. It is noted that none of the models produced satisfactory estimates of SM for any regions (Entin et al. 1999). Robock et al. (2000) made a similar observation. Entin et al. (1999) observed that model biases vary from region to region. Therefore, a simple approach would not be sufficient (Entin et al. 1999). Entin et al. (1999) suggest that future research projects should include experiments with a longer time scale.

Hollinger and Isard (1994) analyzed a measured SM dataset from Illinois. The data were collected from 17 sites in Illinois and the SM was measured biweekly from March through September, and monthly during the winter up to a depth of 2 m. They found latitudinal and longitudinal gradients of SM variations related to season and reported variations of water-holding capacity due to soil texture. Vinnikov and Yeserkepova (1991) analyzed measured SM data from the former Soviet Union for the period of 1972–85. The data used in this study were collected from a depth of 1 m. For warm months the data were measured 3 times a month, and for cold months measurements were made once a month. Vinnikov and Yeserkepova (1991) found a positive linear trend in SM over a number of regions in response to increasing regional precipitation. They also noted that measured SM data did not agree well with GCM-simulated estimates. Entin et al. (2000) estimated temporal and spatial scales of SM variations, using measured data from China, Mongolia, the Soviet Union (Vinnikov and Yeserkepova 1991), Illinois (Hollinger and Isard 1994), and Iowa. Wittrock and Ripley (1999) used a Canadian dataset to determine the regional and spatial pattern of SM during autumn. The SM data used in the above studies were collected at various time scales, including every 10 days, once a month, and once a year (e.g., Wittrock and Ripley 1999). In the recent past, Nijssen et al. (2001) and Maurer et al. (2002) developed an SM dataset for the globe and the conterminous United States, which can be used by various climate models. The Maurer et al. (2002) SM dataset is used by Maurer and Lettenmaier (2003) to forecast runoff in the Mississippi basin and found promising results.

It is apparent from the above that the absence of long-term SM data with high temporal measurement density has precluded meteorological and climatic interpretations. The unavailability of data also has prevented a better grasp of how SM behaves at different depths within the top 2 m of the soil profile. A lack of simultaneous assessment of SM variations under a number of prevalent land use/land cover types has been a hurdle to understanding linkages to climate. In addition, insufficient discussion on SM variability at different temporal scales under multiple land uses has prevented a more realistic representation of landscape. This study provides daily SM estimates for five layers in the soil profile for three land uses and SM variability at different time scales under different land uses. One of the advantages of using the RH model lies in the fact that it is fine-tuned to local (the NGP) land use and hydroclimatic conditions. In addition, due to large-scale modification of natural grasslands to agricultural land use in the Great Plains, it is essential that we investigate SM of croplands along with grasslands. Furthermore, Qiu et al. (2001) emphasized that the relationship between land use and soil moisture needs to be investigated under a variety of conditions. The present study addresses this concern.

GCIP (2000), Leese (2000), and Leese et al. (2001) noted that development of long-term soil moisture datasets for climatological studies and that understanding the role of land memory processes in predictability is a high priority for future research. The United States Global Change Research Program (USGCRP) recognizes that variations in the water cycle at climatological scales are a key research issue for the future (USGCRP 2001). Our long-term goal is to address these objectives of the USGCRP. Note that this study will be conducted within a Global Energy and Water Cycle Experiment (GEWEX) Americas Prediction Project (GAPP) area (Mississippi watershed).

3. RH SM model and its performance evaluation

It is noted above that the RH model was validated, and its performance evaluated, during the two growing seasons of 1986 and 1987 (Table 1) for five land uses at nine sites (representing varied soil conditions), for six layers up to 1.8-m depth in Nebraska (cf. RH; Camargo 1993; Camargo et al. 1994). Tables 1, 2, and 3 and Fig. 3 present detail of the model validation from RH. The sites are located in Nebraska (5), South Dakota (2), and Wyoming (2) (Table 1). Neutron probes were used to measure soil water at six depths: 15, 45, 75, 105, 135, and 165 cm. It was assumed that the data from each depth represent a soil layer of 30 cm. Hence, these points are the midpoints for each of the 30-cm soil layers. In addition, each of the nine locations represents varying root zone soil characteristics, for example, clay soil to sandy soil. If we combine location and land use, SM data were collected from 20 different land surface conditions (Table 1). Table 2 presents model evaluation statistics. Note that the index of agreement, also known as the “d” index (Legates and McCabe 1999; Willmott 1981), can be expressed as follows:

View Expanded

where O and P are observed and predicted values, respectively. The d index takes into account differences in the observed and model-simulated means and variances and, thus, penalizes the model for consistent over- or underestimation, even though the correlations may be high (Legates and McCabe 1999). This index varies from 0.0 to 1.0, with higher values indicating better agreement between modeled and measured values. It is found that for most of the land uses the d index is greater than 0.90 (Table 2). The spread of the d index is between 0.70 to 1.0 (Table 2). The r² estimates also reported greater than 0.90 values for most (13 out of 20 cases) of the land uses (Table 2). The highest and the lowest r² estimates are 0.99 and 0.4 (0.67 is the second lowest), respectively. The observed and predicted SM values range between 23.4–64.5 and 22.6–62.5 cm, respectively (Table 2). Variances (standard deviation squared) for observed and predicted SM values range between 1.7–10.8 and 1.5–10.2 cm, respectively. High d index values, high r² estimates, and a high degree of agreement in distribution of observed and predicted SM values and their variances clearly indicate that the model performance is satisfactory for various land uses with varying soil physical properties. Figure 3 shows the distribution of measured and observed SM and precipitation for five land uses at five locations. Small dots indicate measured SM and the thin line shows modeled SM. Again, it is apparent that the model is estimating SM satisfactorily. Precipitation events and the response of measured and modeled SM are also in close agreement. This suggests that phase and amplitude of modeled and measured SM are also in agreement.

Performance of the RH model was also evaluated for its estimation of SM at various depths in the root zone. Table 3 presents an example of model evaluation results at various depths in the root zone of corn at North Platte, Nebraska. It is found that the index of agreement (d) and r² estimates for observed and modeled SM for six layers ranged between 0.88–0.98 and 0.92–0.98, respectively (Table 3). Observed and predicted SM for these six layers of the root zone ranged between 5.5–6.2 and 5.6–6.2 cm, respectively (Table 3). Variances of the observed and predicted SM for these layers ranged between 1.7–2.0 and 1.1–2.1 cm, respectively. In summary, based on these evaluation results, it can be said that the RH model is satisfactorily estimating SM for various soil layers of the root zone.

A further assessment of the RH model performance was conducted by Camargo (1993) and Camargo et al. (1994). The model was validated for sorghum land use at Mead, Nebraska, in 1990 and 1991 where eight irrigation treatments were applied in a split plot design. The treatments were formulated based on plant phenological development where irrigation was applied and withheld in all possible combinations for three growth stages. Irrigation water was applied whenever plant available SM depleted to 50% of field capacity in the 0–90-cm layer. These experiments were conducted for the 1990 and 1991 growing season, and SM data were collected. Figure 4 shows results from model-estimated and observed SM. It is apparent from Fig. 4 that the RH model is estimating SM satisfactorily. The d index and r² values for model estimates compared to the observed SM range from between 0.78–0.96 and 0.64–0.93, respectively. Note that most of the d index and r² values for model evaluation are above 0.90 and 0.80, respectively. For the total of 16 treatment designs (8 + 8 treatment designs over 2 yr), 9 times the d index exceeded 0.90, and 14 times it exceeded 0.80. On the other hand, for the same number of treatments over 2 yr, r² reached over 0.80 for 12 occasions. The performance of the model, in regard to estimates of SM in the five soil layers (as assigned by Robinson and Hubbard 1990), was positive for 1990 and 1991. The d index and r² estimate, reaching 0.95 and 0.96, indicate satisfactory model performance. Recent applications show that the model is simulating soil water satisfactorily (Mahmood and Hubbard 2002a; Mahmood et al. 2001).

Based on the detailed quantitative comparison of model-estimated and measured SM, we suspect that the RH model performance is superior to Crawford et al.'s (2000) model, and also to many LSSs used in GSWP, AMIP, and PILPS. The other models, including LSSs, have not been evaluated in a comprehensive manner with multiple land uses, locations, and soils under significantly different hydroclimatic forcings. Validation is also completed for six depths from the surface to root zone. Four SM data-collection campaigns were undertaken to evaluate the RH model.

Crawford et al. (2000) conducted their validation by using only 1 month of data and for one land use. They (Crawford et al. 2000) have noted that r² between observed and simulated SM for 25-, 60-, and 75-cm depths ranged between 0.27 and 0.40. For comparable depth ranges, RH recorded r² between 0.94 and 0.98. Camargo et al. (1994) also found similar results and noted that r² values ranged between 0.73 and 0.96. Thus, performance of the RH model is significantly superior. Crawford et al. (2000) noted that their model does not perform well in simulating SM at deeper layers. The evaluation of the RH model shows that its performance is superior and satisfactory for deeper layers (Table 3).

Our survey of literature shows that detailed quantitative assessment of the LSSs is notably deficient. Use of some of the well-established model evaluation methods was not employed and, thus, exact reliability of the LSS-estimated SM is unknown. Despite the availability of observed SM data for validation, most of the studies only show graphics on observed versus modeled SM and provide only qualitative assessment of the model performance. For example, Chen and Mitchell (1999) only plot observed and NCEP land surface model (LSM)-simulated SM and do not provide any quantitative assessment on reliability of the model-estimated values. A discussion on seasonal changes in SM is not adequate. From visual examination we can state that large under- and overestimation of SM is prevalent for Chen and Mitchell's (1999) simulation. Similar qualitative assessment of LSS validation was provided for some of the major SM modeling efforts under the umbrella of GSWP, AMIP, and PILPS (e.g., Matsuyama et al. 1999; Zhang et al. 1999). In the absence of quantitative evaluation, graphic presentation of observed and modeled data indicates a significant amount of disagreement. Matsuyama et al. (1999), in the application of Japan Meteorological Agency's Simple Biosphere Model (SiB) found a consistent underestimation of estimated SM, compared to observed SM. After evaluating GSWP SM simulations, Entin et al. (1999) correctly noted that none of the models do a satisfactory job in estimating SM. Dirmeyer et al. (1999) agreed with this observation.

Schlosser et al. (2000) also qualitatively assessed performance of 21 LSSs for Valdai, Russia. This evaluation was conducted for only one land use, namely, grass, as opposed to our quantitative evaluation for five land uses at nine locations. Chen et al. (1997) “evaluated” performance of the 23 LSS schemes without comparing simulated values with measured values during the Cabauw, Netherlands, experiment. Schlosser et al. (1997) evaluated the performance of the Simple SiB (SSiB) LSS for Valdai, Russia. The variance of model-estimated and observed SM data shows a large disagreement and departure from each other. Our model evaluation shows close agreement between estimated and observed SM (Tables 2 and 3). Previously, Robock et al. (1995) have shown bias associated with SSiB and the bucket model–estimated SM values. The bias was associated with incorrect specification of field capacity (Robock et al. 1997). Another LSS, namely, a two-layer VIC(-2L) (Liang et al. 1996) was evaluated, but the location and model evaluation statistics presented are minimal. This limited evaluation failed to convey the exact reliability of the model.

It can be concluded, for LSSs used in GSWP, PILPS, and AMIP, that 1) the quantitative assessment of the accuracy of modeled SM is absent or severely limited, 2) the qualitative assessment is prevalent, 3) the “validation” is largely conducted for one land use—grass, 4) the validation data used are for only one location, despite the availability of data from additional sites, and 5) the departures of model variance (only one study applied this statistic for evaluation) and estimates from observed data were large. As noted above, for the RH model, we have completed a detailed quantitative assessment (Tables 2 and 3; Figs. 3 and 4). All of the model evaluation statistics applied suggest that the model performance is satisfactory under various land uses, root zone soil conditions, and hydroclimatic regimes (Tables 2 and 3). Based on the model evaluation statistics, we are able to indicate how well the model is performing and how reliable the model estimates are. Visual assessment is not sufficient, as shown by many LSS evaluation exercises. Therefore, based on the results from quantitative assessment and the accuracy of calculated SM values, we conclude that the RH model will provide satisfactory estimates of SM for the study area.

For this study, the RH model is applied for three predominant land uses and associated soils characteristics in Nebraska, namely, irrigated and rain-fed corn, and rain-fed grass. Three sites in Nebraska were selected to simulate the SM climate under varying land use and moisture regimes. The Mead, Clay Center, and McCook sites (Fig. 1) follow an east–west gradient from moist (Mead) to very dry (McCook) conditions. There were three applications per site (three types of land use and one soil type). The setting allows us to estimate and understand potential responses of SM in more than one moisture regime. Daily weather data for the RH model simulations were provided by the automated weather stations at these locations. Site-specific soil input data for the model application are available from a soils database maintained by the High Plains Regional Climate Center (HPRCC). These data include bulk density, soil texture, wilting point, field capacity, and saturation point for each soil layer (Tables 4 and 5).

In this paper total plant available root zone water (SM) is presented as water depth, and individual layer SM is presented as volumetric water content (mm³ mm⁻³). In the following discussions, we chose to emphasize temporal distribution of SM over 1 yr by using daily estimates. An analysis of seasonal variations of SM is not discussed here.

4. Applications of SM model and results

5. Discussion and concluding remarks

The present study investigated the daily SM climatology at three locations under three predominant land uses, namely, irrigated corn, rain-fed corn, and natural grass. Because these land uses of Nebraska are predominant, point-scale application provides insight into SM variations for the three land uses. The vertical scale of the applications is 1.2 m. Large-scale modeling studies have yet to provide a clear understanding of these issues partly due to their spatiotemporal scale and due to the unavailability of input data at this scale. The estimation of SM through the vertical soil profile with input data describing the local soil–plant–hydrologic continuum at three locations offer us an opportunity to grasp the hydrologic process at the point scale. Because these point-scale estimations are representative of larger-scale land use and soil conditions, we speculate that the results could be insightful for parameterization of large-scale SM calculation. Simultaneous estimates of SM for three land uses for each of the three locations and the analyses of simulated data provide a quantitative assessment of the response and impacts of predominant land cover and prevailing long-term hydroclimatic conditions. Due to large-scale modification of natural grasslands to agricultural land use in the Great Plains, it is essential that we investigate SM of crop lands along with grasslands.

As indicated above, this investigation provides a quantitative assessment of daily SM dynamics and a mechanism to quantify what is known at a general level: SM begins depletion with the start of the growing season and reaches its lowest at the height of the crop growth, at the same time ET reaches a maximum, annual total evapotranspiration is higher for irrigated corn, and rain-fed corn also evapotranspires at a higher rate compared to grass. Compared to grasslands, irrigated and rain-fed corn farming elevated and lowered plant available soil moisture through adding and mining water from the soil profile, respectively. Long-term averages suggest that lower soil layers (fourth and fifth) contain less water compared to upper soil layers (first and second) for all three land uses and at all three locations with variable moisture regimes. The type of vegetation affects depletion of SM at various depths via the differences in length of growing season and rooting depth. Unlike grass, SM variability in the top two layers under irrigated and rain-fed corn was lower compared to the third–fifth layers. Interannual modulation of SM resulted in such variability (Figs. 15a,b).

Analysis of SM data at different temporal scales indicates that at the daily time scale, plant phenology plays an important role in modulation of SM variability. It is found that SM variability increases with progression of the growing season and decreases as the plants mature. At the monthly time scale, SM variability also follows the pattern that resembles the variability of the daily time scale. However, as expected, changes in SM variability from one month to another are “smooth” as opposed to “step changes.” It is found that land use and soils influence SM variability at daily and monthly time scales. The forcing of hydroclimatic condition (wet vs dry) is overwhelming. At all time scales, wet conditions reduced SM variability. At the annual time scale, forcing of large-scale climate modulation and associated wet or dry conditions influence SM variability. It is found that excessively wet conditions reduced SM variability and the differences between SM relative to land use, while extremely dry condition increased SM variability and forcing of land use. In other words, the influence of land use is dependent on the “normality” or “degree of anomaly” of climatic conditions.

The general pattern in SM depletion and recharge estimated by this model is comparable to Georgakakos et al. (1995) and Georgakakos and Bae (1994). This study presents a more detailed SM estimate by dividing the soil profile into five layers instead of one and two layers, as presented by Durre et al. (2000) and Georgakakos et al. (1995), respectively. Therefore, the movement of soil water can be followed on scales not resolved by these previous models, giving us the opportunity to understand annual distribution of SM within the soil profile. Scott et al. (2000) reported that the extent of SM recharge in the root zone during winter in the semiarid southwestern United States may vary depending on precipitation amount. Our study also finds similar recharging during wintertime; obviously, the extent of recharge varies from year to year. Scott et al. (2000) identified large interannual variability in deeper root zone recharge. In our study, we have large interannual variability in SM at lower layers of all land uses except for the fifth soil layer under shallow-rooted grass. In other words, roots mine the water from the lower layers and, over a long period of time, SM variability is higher in the deeper layers despite different climate and soil regimes. In addition, larger amplitudes of root zone total SM under irrigated and rain-fed corn indicate their vulnerability to current hydroclimatic conditions.

The results of this study have raised a number of critical questions related to forcings of land use and soils on SM variability as they interact with climate. What is the threshold anomaly that reduces or enhances SM variability? What is the variation in this threshold as land use and soils changes? The authors suggest that additional research needs to be conducted to quantify these threshold conditions for predominant land uses. In our opinion, findings can be important to diagnose modeling results and improve predictive skills.

This study demonstrated the changes in the magnitude of variability of SM under different land uses at multitemporal scales. An important question is how can we transfer this knowledge to mesoscale and climate models. In other words, we need to develop methods to incorporate these near-surface variabilities in the models. In addition, how sensitive is the system to these SM variabilities? How much modulation of weather and climate can be expected at various spatial scales by these variabilities? This brings up the question of feedback between SM and the atmosphere at multiple spatiotemporal scales. Are these issues adequately addressed during model formulation? Are these issues being explicitly addressed during current modeling activities? Wang and Kumar (1998), Dirmeyer (1999, 2000), and Huang et al. (1996) noted that SM can be an important predictor of surface temperature and precipitation. Thus, we suggest that specific modeling studies should be undertaken to determine SM variability at multitemporal scales under heterogeneous land use, and their impacts on weather and climate and their predictability. Such studies would hopefully determine the sensitivity of the atmosphere to SM variability and, also, the behavior of the mesoscale and climate models.

This study has demonstrated the potential future use of the RH soil water balance model to estimate SM of the past at various depths and under a number of land covers. We plan to produce a gridded SM database from point simulations and to use these data to study land memory processes in the NGP, water cycling, and variations of SM at various spatiotemporal scales. It is also suggested that this SM data can be used further for various mesoscale and climate modeling studies. In our case, there are advantages of gridding from point data, as opposed to cell-based modeling. We have already collected these data for 150+ application sites and are prepared to collect data for an additional 50 locations. These sites also represent predominant soil types of the local county. The soils database for cooperative weather-recording locations is also under way. At the same time, detailed cell-by-cell data on soil physical properties, as required by a grid-based model, are not available. Based on the spatial density of the model application sites and the detailed representation of root zone soil characteristics for predominant soils, we have concluded that the land surface condition of the study area has been described in detail (horizontally and vertically), correctly, for the RH model applications. Therefore, gridding from point data will provide sufficiently accurate estimates of SM and ET.

In related studies, methods were developed for estimating the solar radiation and relative humidity of the past for the NGP (Mahmood and Hubbard 2002b; Hubbard et al. 2002). Availability of soil and weather data with reliable solar and relative humidity estimates will permit us in the near future to estimate SM of the past for a large number of locations in the NGP. These locations will include automated and cooperative weather-observing sites and, thus, will provide a higher spatial density for SM model applications. The SM data produced from this work will be invaluable for future investigations in the areas of hydrology and land–atmosphere interactions.

We believe the results presented in this paper indicate the importance of developing reliable parameterization of soil water balance in various climate models. The grasslands of the North American Great Plains and other parts of the world were modified rapidly and vastly over the last century due to human economic activities, including farming. Therefore, it is imperative that we understand the temporal changes in the soil water condition in the soil profile under dominant land uses over large regions, along with natural conditions. This type of study will improve our overall understanding of soil water distribution within the soil profile over a time period.