Regionalization of soil water retention curves in a highly variable soilscape, II. Comparison of regionalization procedures using a pedotransfer function

As measuring soil water retention curves (WRCs) is time-consuming and costly, pedotransfer functions (PTFs) which predict WRCs from the fundamental soil properties hulk density (D(b)), texture, and organic C (C(org)) are in common use. The regionalization of WRCs with a PTF can be performed in two different ways. (1) Interpolate first the fundamental properties, and apply then the PTF to the interpolated data to predict the WRCs. (2) Predict first the WRCs by applying the PTF onto the point-wise measurements of the fundamental data, and interpolate then the WRCs. Both procedures have been tested in a 1.5 km² soilscape with a high variability in parent material and land use. The fundamental properties were measured at the 450 nodes of a rectangular 50 x 50 m grid. The WRCs were measured at seventeen irregularly distributed sites. A new PTF which had been adapted to the soilscape was used to predict the WRCs. Using procedure (1), the spatial variability of each fundamental property could be individually analyzed and accounted for in the regionalization process. Thus, the root of the mean squared differences (RMSD) between the predicted and the observed water contents was 16% lower for procedure (1) than for procedure (2) Considering the effect of land use by a residual variogram method reduced the standard deviation between predicted and observed values of C(org) and D(b) by 11% and 20%, respectively, as determined by cross-validation. The residual method produced more plausible spatial patterns of the soil water content at -300 and -15,000 hPa. As a result of the improved spatial patterns and the decrease in the regionalization error, procedure (1) is clearly superior to procedure (2).