Geostatistical modeling of spatial variability of water retention curves

Abstract

Abstract. This study compares the performance of two geostatistical approaches, parametric and non-parametric, to evaluate the spatial distribution of water retention curves. Data used in this study were obtained from the Las Cruces trench site database that contains water retention data for 448 soil samples. In a commonly used parametric approach, three standard water retention models, i.e. Brooks and Corey (BC), van Genuchten (VG), and log-normal (LN), were first fitted to each data set. For each model, a cross validation procedure was used to estimate parameters at each sampling location, allowing computation of prediction errors. In a rarely used non-parametric approach, a cross validation procedure was first used to directly estimate water content values for eleven pressure heads at each sampling location and then the three water retention models were fitted using the same automated procedure to compute prediction errors. The results show that the non-parametric approach significantly lowered prediction errors for the VG model, while moderately reducing them also for the LN and BC models.