Uncertainty analysis for large-scale prediction of the van Genuchten soil-water retention parameters with pedotransfer functions

Abstract

Hydrological, environmental and ecological modellers require van Genuchten soil-water retention parameters that are difficult to measure. Pedotransfer functions (PTFs) are thus routinely applied to predict hydraulic parameters (θs, ln(α) and n) from basic soil properties (e.g. bulk density, soil texture and organic matter content). This study investigated the spatial variations of van Genuchten parameters via geostatistical methods (e.g. kriging and co-kriging with remote-sensing data) and multiple-stepwise-regression-based PTFs with a limited number of samples (58) collected in Pingdu City, Shandong Province, China. The uncertainties in the spatial estimation of van Genuchten parameters were evaluated using bootstrap and Latin hypercube sampling methods. Results show that PTF-estimated parameters are less varied than observed parameters. The uncertainty in the parameter estimation is mainly due to the limited number of samples used for deriving PTFs (intrinsic uncertainty) and spatial interpolations of basic soil properties by (co)kriging (input uncertainty). When considering the intrinsic uncertainty, 36%, 29% and 47% of measurements are within the corresponding error bars (95% confidence intervals of the predictions) for the θs, ln(α) and n, respectively. When considering both intrinsic and input uncertainties, 86%, 66% and 88% of observations are within the corresponding error bars for the θs, ln(α) and n, respectively. Therefore, the input uncertainty is more important in the spatial estimation of van Genuchten parameters than the intrinsic uncertainty. Measurement of basic soil properties at high resolution and properly use of powerful spatial interpolation approach are both critical in the accurate spatial estimation of van Genuchten parameters.