On the structure of a semi-distributed adaptive model for flood forecasting

Abstract

Eight different versions of a simple rainfallrunoff adaptive model for on-line flood forecasting are formulated in a framework characterized by an ensemble of spatially uniform units. The main loss from rainfall is considered to be infiltration which is represented by an extended formulation of a point approach utilizing a two-branched scheme for time to ponding and soil infiltration capacity. The differences among the model versions are due to the computation of effective rainfall, the treatment of the transformation process from effective rainfall to direct runoff, and the estimation of surface retention. The various model versions were tested by their application to events observed on a large Italian basin. By a comparison of their forecasts with those of a no-model approach and the estimation of errors in peak runoff, it was found that two of the models were more appropriate and provided flow forecasts sufficiently accurate for forecast lead times up to 6 h.