A Comparison of Low Flow Estimates in Ungauged Catchments Using Regional Regression and the HBV-Model

Estimates of a low flow index in ungauged catchments calculated by a regional regression model and a regional hydrological model were compared for a study region southwestern Norway. The regression method was based on a relationship between the low flow index and an optimal set of catchment descriptors, established using stepwise linear regression for homogeneous subregions. Subregions were distinguished according to the season in which the lowest flow occurs, summer (May to October) or winter (November to April), and the average July temperature was found to be the best index for determining the low flow season for ungauged catchments. Catchment descriptors characterising the presence of lakes and bogs, in addition to catchment length and indicators of climatic conditions, were found to be important in the regression models. A cross-validation procedure was used to evaluate the predictive performance of the model in ungauged catchments. A gridded version of HBV, a daily rainfall-runoff model was also applied as a regional hydrological model and was calibrated using the average Nash-Sutcliffe coefficient for log-transformed streamflow as the calibration criterion. A comparison of the two methods in 21 independent catchments indicates that the regression method generally gives better estimates of Q (c) in ungauged catchments than does the HBV model, particularly in those catchments with the lowest Q (c) values.