Using pedotransfer functions in vadose zone models for estimating groundwater recharge in semiarid regions

Process‐based vadose zone models are becoming common tools for evaluating spatial distributions of groundwater recharge (GR), but their applications are restricted by complicated parameterizations, especially because of the need for highly nonlinear and spatially variable soil hydraulic characteristics (SHCs). In an attempt to address the scarcity of field SHC data, pedotransfer functions (PTF) were introduced in earlier attempts to estimate SHCs. However, the accuracy of this method is rarely questioned in spite of significant uncertainties of PTF‐estimated SHCs. In this study, we investigated the applicability of coupling vadose zone models and PTFs for evaluating GR in sand and loamy sand soils in a semiarid region and also their sensitivity to lower boundary conditions. First, a data set containing measured SHCs was used in the simulations. A second data set contained correlated SHCs drawn from the covariance matrix of the first data set. The third SHC data set used was derived from a widely used PTF. Although standard deviations for individual parameters were known for this PTF, no covariance matrix was available. Hence, we assumed that the parameters of this PTF were uncorrelated, thereby potentially overestimating the volume of the parameter space. Results were summarized using histograms of GR for various sets of input parameters. Under the unit gradient flow lower boundary condition, the distributions of GR for sand and loamy sand significantly overlap. Values of GR based on mean SHCs (or GR*) generally lie off the mode of the GR distribution. This indicates that the routinely used method of taking GR* as a regional representation may not be viable. More importantly, the computed GR largely depends in a nonlinear fashion on the shape factor n in the van Genuchten model. Under the same meteorological conditions, a coarser soil with a larger n generally produces a higher GR. Therefore, the uncertainty in computed GR is largely determined by the uncertainty in estimated n by PTFs (e.g., mean and standard deviation). Under the constant head lower boundary condition, upward soil moisture flux may exist from the lower boundary. Especially for regions with shallow water tables where upward flux exists, choosing an appropriate lower boundary condition is more important than selecting SHC values for calculating GR. The results show that the distribution of GR is less scattered and GR is more intense if the constant head lower boundary is located at deeper depths.