摘要：Core Ideas Measurements of field‐scale saturated hydraulic conductivity ( K s f >) were performed. Two pedotransfer functions for field‐scale K s f > were developed. A map of field‐scale K s f > at catchment scale was obtained using two approaches. Classical experimental techniques to determine point values of saturated hydraulic conductivity ( K s ) are complex and time consuming; therefore, the development of pedotransfer functions, PTFs, to derive K s from easily available soil properties is of great importance. However, PTFs have been generally developed at the local scale, while hydrological modeling requires K s estimates at larger scales. A small Austrian catchment, where detailed soil characteristics were available, was selected to address this issue. Values of field‐scale saturated hydraulic conductivity ( K s f >), observed in a number of catchment areas by double‐ring infiltrometers, were used to develop two PTFs, one by multiple linear regression (PTF MLR ) and one by ridge regression (PTF R ). Training and validation of the PTFs in the monitored areas indicate that the PTF R provides better outcomes with smaller average errors. This suggests that the ridge regression is a valid alternative to the classical multiple linear regression technique. Predictions of K s f > by the PTFs in the remaining areas, where infiltration measurements were not performed, were also made to obtain a map of K s f > for the whole catchment. Two alternative approaches were used: Method A—soil properties were first interpolated and then the PTFs applied; Method B—the PTFs were first applied to sites with available soil properties and then interpolated. The maps of K s f > obtained by the PTF MLR are not representative of the K s f > spatial variability. On the other hand, the map generated by the PTF R with Method A is consistent with catchment morphology and soil characteristics.
关键词：ANN; artificial neural network; FDF; frequency density function; GCV; generalized crossvalidation; GMER; geometric mean error ratio; MLR; multiple linear regression; PTF; pedotransfer function; VIF; variance inflation factor.
