USING PROCESS-BASED MODELS AND PEDOTRANSFER FUNCTIONS FOR SOIL HYDRAULIC CHARACTERISTICS TO ESTIMATE GROUNDWATER RECHARGE IN SEMI-ARID REGIONS: IS THIS A RIGHT APPROACH?

Process-based models are commonly used for estimating spatial distributions of groundwater recharge (GR), but they require parameters that have heterogeneous and nonlinear nature, especially soil characteristics (SC). As an attempt to address heterogeneity of vadoze zone and lack of field data (e.g., Keese et al., 2005; Small, 2005), pedotransfer functions (PTF) are used (Schaap et al., 1998, 2001), which are based on statistical analyses of existing soil databases. However, accuracy of this method was rarely questioned in spite of significant uncertainties of PTFs and highly nonlinearity of SCs. Schaap et al. (1998)'s PTFs provide a mean and standard deviation for each parameter of the van Genuchten model. To assess if using mean SCs for sand and loamy sand yields plausible results for GR, we started with hydroclimatic and phenological dataset at the GDEX site in the Nebraska Sand Hills. Then, using HYDRUS-1D, we calculated GR for perturbed values of each of five parameters in the van Genuchten model, namely mean ± standard deviation, which yielded 3⁵=243 combinations of SCs and corresponding 243 GR values at the GDEX site. This allowed addressing the question: does using mean SCs yield adequate GR estimates? Considering nonlinearity of SCs, one could hypothesize the opposite. The results yielded counterintuitive, diverse, and important conclusions on GR simulations: (1) On average, GR was higher in sand under free drainage conditions, but loamy sand may also produce comparable high GR; (2) Although GR for sand varied by an order of magnitude comparing to three orders for GR of loamy sand, GR for sand was more scattered and exhibited a bimodal distribution; whereas, GR for loamy sand exhibited a unimodal distribution and was clustered towards the low-GR tail; (3) If groundwater table depths (GTD) were considered, GR distributions became more spread with decreasing GTDs, and the PTF-based method is less applicable for shallow GTD; (4) The simulation results of two-layer soil profiles (i.e., loamy sand over sand) showed significant impacts of vertical soil textural variability on GR distributions. For regions with a complicated sedimentary history, the needs of detailed information on SD distribution cannot be ignored.