A Transferability Study of b-parameter for VIC Model

One of the important model parameters in the Variable-Infiltration-Capacity land surface model (VIC) is b-parameter that controls the shape of soil moisture capacity curve for a studied area. This b-parameter is generally estimated through model calibration. However, for regional climate model and GCM applications, estimation of the b-parameter through calibration is usually not feasible due to limited observations. In this paper, the STATSGO and CONUS data are used to estimate the b-parameter. Arkansas, Oklahoma and California are selected for this study, where 67, 217 and 307 mapunits are available respectively. The results suggest that the underlying assumption of the shape of soil moisture capacity curve used in VIC model is appropriate for most cases (70%). To apply VIC model to areas where neither observed streamflow data nor STATSGO and CONUS data are available, a framework of using neural network approach is developed to establish a relationship between the b-parameter and statistics of soil properties. In this framework a self-organizing map (SOM) is used to classify the data based on soil characteristics. Then a back propagation neural network is obtained for each cluster to relate the soil characteristics with the b-parameter. The inputs of each network use the same soil statistics as for SOM. The outputs are the estimated b-parameters. Validation results show that the trained neural networks are capable of providing reasonable estimations of b-parameters. This approach provides a promising way to transfer the knowledge of b-parameter from data-rich areas to data-sparse areas.