Improving hydrologic model hypothesis testing by accounting for rainfall and discharge uncertainties

Abstract:

Hydrology has been at the forefront of developments that improve the way we evaluate model simulations and quantify the resultant prediction uncertainties. A wealth of approaches have resulted from this research, varying in their statistical rigour and assumptions, which has sparked controversy between different factions. Recently, however, there appears to be a growing consensus across these approaches that there is a need to characterise, quantify and account for the effect of observational errors in the performance measures we use to evaluate models. This is particularly the case when we use different catchment information to evaluate model behaviour and our conceptualisations of catchment processes. This presentation will explore this issue by building on our recent efforts to quantify rainfall and discharge uncertainties in a generalised approach. We incorporate these uncertainties within a ‘limits of acceptability’ framework for hypothesis testing through model rejection. We apply this framework within a multi-model framework across a number of catchments to evaluate any systematic model failures outside these limits utilising time step based analyses. We show how constraining model structures and behaviour using observational uncertainties provides a coherent platform for hypothesis testing of different process representations.