DATABASE-RELATED ACCURACY AND UNCERTAINTY OF PEDOTRANSFER FUNCTIONS

Pedotransfer functions (PTFs) are becoming a more common way to predic t soil hydraulic properties from soil texture, bulk density, and organ ic matter content. Thus far, the calibration and validation of PTFs ha s been hampered by a lack of suitable databases. In this paper we empl oyed three databases (RAWLS, AHUJA, and UNSODA) to evaluate the accura cy and uncertainty of neural network-based PTFs. Sand, silt, and clay percentages and bulk density were used as input for the PTFs, which su bsequently provided retention parameters acid saturated hydraulic cond uctivity, K-s as output. Calibration and validation of PTFs were carri ed out on independent samples from the same database through combinati on with the bootstrap method. This method also yielded the possibility of calculating uncertainty estimates of predicted hydraulic parameter s. Calibration and validation results showed that water retention coul d be predicted with a root mean square residual (RMSR) between 0.06 an d 0.10 cm(3) cm(-3); the RMSR of log (K-s) was between 0.4 and 0.7 log (cm day(-1)). Cross-validation was used to test how well PTFs that we re calibrated for one database could predict the hydraulic properties of the other two databases. The results showed that systematically dif ferent predictions were made when the RMSR values increased to between 0.08 and 0.13 cm(3) cm(-3) for water retention and to between 0.6 and 0.9 log(cm day(-1)) for log(K-s). The uncertainty in predicted K-s wa s one-half to one order of magnitude, whereas predicted water retentio n points had an uncertainty of about 0.04 to 0.10 cm(3) cm(-3). Uncert ainties became somewhat smaller if the PTFs were calibrated on all ava ilable data. We conclude that the performance of PTFs may depend stron gly on the data that were used for calibration and evaluation.