INTRODUCTION

Land surface parameterization schemes (LSPs), or land surface models (LSMs), are used to determinate the fluxes of radiation, heat, water vapor, and momentum across the land-atmosphere interface in atmospheric general circulation models. The land surface parameterizations have evolved from simple, unrealistic schemes into credible representations of the soil-vegetation-atmosphere transfer system (Manabe, 1969; Sellers et al., 1986; Sellers et al., 1996). Land surface schemes are important for numerical climate predictions. The dependence of the large scale atmospheric circulation and rainfall on land surface boundary conditions has been the subject of discussion for many years. The availability of soil moisture was found to play an important role in the exchange of moisture and energy between the atmosphere and the land surface. Shukla and Mintz (1982) demonstrated that the global fields of rainfall, temperature, and motion strongly depend on the available soil moisture and land surface evapotranspiration with a numerical model. Fennessy and Shukla (1999) demonstrated that the initial soil wetness was important in seasonal predictions with dynamical models. Kanamitsu et al. (2003) found that the soil moisture initial condition was essential for skillful predictions over arid and semiarid regions. Atlas et al. (1993) and Namias (1991) described the potential for soil moisture initialization to aid in the accurate simulation of extreme climate events, such as droughts. Soil moisture availability could strongly influence the atmosphere by altering the partitioning of the outgoing energy flux at the surface into latent and sensible heat components (Delworth and Manabe, 1989; Delworth and Manabe, 1988). GCMs simulations also indicated soil moisture contributes to precipitation predictability (Dirmeyer, 2000), especially in transition zones between dry and humid climates during Northern Hemisphere summer (Koster et al., 2004; Koster et al., 2000). Soil moisture is one of the most important variables that integrates much of the land surface hydrology and plays important roles in ecosystem dynamics and biogeochemical cycles.

The soil moisture would be extremely useful for improving the performance of GCMs. Dirmeyer (1995) stated that the lack of widespread observations of soil moisture continues to hamper efforts to verify and improve GCM simulations. Estimates of large-scale fields of soil moisture could be produced from remote sensing, land surface model, data assimilation and were used for initialization of climate prediction models (Houser et al., 1998; Dirmeyer, 2000; Paloscia et al., 2001). The soil moisture datasets most often used for initialization of GCMs are monthly climatology of soil moisture derived from the applications of simplified evaporation formulations (Mintz and Serafini, 1992; Mintz and Walker, 1993). More recently, data assimilation methods have been developed to initialize the soil moisture states in global climate model using remote sensing data and ground observations (Walker and Houser, 2001). (Reichle et al., 2004) found absolute soil moisture values from the satellite and the model are very different, and neither agrees better with ground observations comparing among three independent global surface soil moisture datasets.

To address the lack of consistent, high-quality datasets of soil wetness, and to understand how different LSMs determine soil wetness, the Global Soil Wetness Project (GSWP) was created (Dirmeyer and Oki, 2002; Dirmeyer et al., 1999). In GSWP, the soil wetness was produced using a dozen different LSMs and a common set of soil and vegetation parameters and meteorological forcing. The model-generated soil moisture data sets were compared with soil moisture observations from grasslands and agricultural regions in Russia, Illinois (USA), China, and Mongolia for a two-year period (1987-1988). None of the models does a good job of producing the actual soil moisture value for any of the regions (Entin et al., 1999).

The Atmospheric Model Intercomparison Project (AMIP) conducted simulations by 30 different atmospheric GSMs for the 10-year period, 1979-1988. The model-generated soil moisture data and simulations of soil moisture with observations from 150 stations in the former Soviet Union for 1979-1985 and Illinois for 1981¨C1988 were compared (Robock et al., 1998; Gates, 1992). The model-generated soil moisture data sets are quite different from the observations, and from each other in many regions, even though they use the same model calculation method.

The intercomparsion of different LSMs had been done in both GSWP and AMIP simulations. However, there are few studies on how model parameters and model structure will affect the soil moisture in large scale simulation. In most of the LSMs in current GSWP, groundwater was not or simplified considered. And the groundwater storage was usually considered as a groundwater reservoir to connect base flow and river channel for calculation of river discharge (Xue et al., 1991; Milly and Shmakin, 2002). There were few investigations on the groundwater influence to predicated soil moisture in large scale. Nijssen et al. (2001) compared modeled soil moisture with observations in Illinois and in central Eurasia with a VIC model and found modeled soil moisture was lower than observed and showed less inter annual persistence than the observations. They suggested that a mechanism for upward diffusion of soil moisture and coupling with a regional groundwater model was needed to simulate soil moisture. And the simulated soil moisture data sets were compared with observations which was highly limited in time and space. There are few catchment scale validations of the simulated soil moisture.

In this study, we explored the soil moisture prediction using a realistic land surface model SIB2 (Sellers et al., 1996) with revised hydrological sub-models. The land surface model is forced with meteorological observations in a semi-arid river basin for a two decades period (1982-2000). The simulated soil moisture data set was compared with observations and also validated with the water balance in the river basin. In section 2, the girded dataset in the study river basin of surface forcing variables is described. The core of the meteorological data consists of daily precipitation and daily minimum and maximum temperatures for the period 1982-2000 for 10×10 kilometers grid cells over the study area. In addition, the dataset contains downward shortwave radiation, and net longwave radiation, derived from the routine meteorological data using standard algorithms. A briefed review of the land surface model SIB2 and the revised hydrological sub-model were described in section 3. In section 4, the simulated soil moisture is evaluated in terms of the season cycle and interannual variability and validate with water balance results in the river basin. Section 5 presents conclusions and discussion.