Evaluating Hydrological Process Information for Testing Models as Hypotheses Using a Virtual Reality Model

Abstract

Process-behavioural modelling aims to achieve accurate representation of hydrological processes in hydrological models. Evaluation of model performance on discharge alone cannot achieve this; more informative observables are needed. This poses the question of which types of observables might be most informative in discriminating between different potential models. As a way of testing the value of different observables to this end, a method was developed that uses a virtual reality model of a catchment. Based on this virtual reality, ‘observations’ were then simulated by perturbing the ‘real’ values. These observations were used to evaluate the outputs of 1800 simulations with alternative parameter sets, but with the same model structure. The simulated results for the hydrological variables discharge, discharge peaks, discharge components, incremental discharge, cosmic ray neutron sensing soil moisture (CRNS), and total water storage were then normalised with the uncertainties derived for each variable from the virtual reality model. This allowed the individual variables to be compared in terms of their ability to identify non-process-behavioural simulations. For the assumed uncertainties, it was shown that the discharge component separation by the tracer model discriminates between models particularly strongly, while CRNS serves as a complementary variable for evaluation. For all the models tried, however, significant relaxations of the limits of acceptability were required to accept any models as process-behavioural. This appeared to be the result primarily of the perturbations applied to the inputs which resulted in significant model residuals during certain periods. Since in this case there is no significant model structural error, this suggests that hydrological modellers should, as far as possible, assess the uncertainty in the model forcing data as part of the model evaluation process.