A multivariate spatial model for soil water profiles

Pedotransfer functions are classes of models used to estimate soil water holding characteristics based on commonly measured soil composition data as well as other soil characteristics. These models are important on their own but are particularly useful in modeling agricultural crop yields where only soil composition is known. In this article, an additive, multivariate spatial process model is introduced that offers the flexibility to capture the complex structure typical of the relationship between soil composition and water holding characteristics, thus defining a new form of pedotransfer function. Further, the uncertainty in the soil water characteristics is quantified in a manner to simulate ensembles of soil water profiles. Using this capability, a small study is conducted with the CERES maize crop model to examine the sources of variation in the yields of maize. Here it is shown that the interannual variability of weather is a more significant source of variation in crop yield than the uncertainty in the pedotransfer function for two specific soil textures.