Application of Parzen Densities to Probabilistic Rainfall-runoff Modelling

In this paper a new black box approach for rainfall-runoff modelling at a daily scale is presented. The considered black box model is non-linear regression based on Parzen probability density function. When using only measured rainfall as an input to any black box model there is a serious problem with building in the necessary memory. A standard approach to tackle this issue is to force a black box with a large number of rainfall and runoff variables of the past. In practice however, any regression technique, will have difficulties handling so large (possibly dependent) input set. For that reason, a more hydrological approach is proposed. Two linear reservoirs are used to model the memory. The reservoir constants are found by simple piecewise linear regression. An application to the Beerze catchment in the Netherlands is shown. A good correspondence between measured and estimated runoff is achieved.