A comparison of regionalisation methods for catchment model parameters

In this study we examine the relative performance of a range of methods for transposing catchinent model parameters to ungauged catchments. We calibrate 11 parameters of a semi-distributed conceptual rainfall-runoff model to daily runoff and snow cover data of 320 Austrian catchments in the period 1987-1997 and verify the model for the period 1976-1986. We evaluate the predictive accuracy of the regionalisation methods by Jack-knife cross-validation against daily runoff and snow cover data. The results indicate that two methods perform best. The first is a kriging approach where the model parameters are regionalised independently from each other based on their spatial correlation. The second is a similarity approach where the complete set of model parameters is transposed from a donor catchment that is most similar in terms of its physiographic attributes (mean catchment elevation, stream network density, lake index, a real proportion of porous aquifers, land use, soils and geology). For the calibration period, the median Nash-Sutcliffe model efficiency ME of daily runoff is 0.67 for both methods as compared to ME=0.72 for the at-site simulations. For the verification period, the corresponding efficiencies are 0.62 and 0.66. All regionalisation methods perform similar in terms of simulating snow cover.