Limitations in Interpretable Top-Down Effective Rainfall-Runoff Modelling

Abstract

The effective rainfall-flow module of a "top-down" hydrological model has linear, fixed dynamics, giving a two-exponential instantaneous unit hydrograph (IUH) often interpreted as summing slow- and quick-flow components. Such components suffer uncertainty due to ignorance of rainfall variation between samples and ignorance of delay in the IUH. If they are obtained from linear-in-parameters models, ill conditioning in the conversion amplifies the uncertainty. The extent of these uncertainties is analysed, with examples. The paper also considers alternatives to performance assessment of such models by Nash-Sutcliffe efficiency. Rather than employing the sample mean as an output-prediction benchmark, they use almost equally simple predictors taking the flow correlation structure into account. Again numerical examples are given.