Managing uncertainty in hydrological models using complementary models

Handling uncertainties in a conceptual rainfall runoff model is approached as an error modelling problem. The approach is based on the application of a parallel data-driven model that uses available measured data and previous model errors at specific time steps to forecast the errors of the conceptual model. The average mutual information technique is used to study the relationship between the different variables and the model errors at varying lag times. The resulting information is used to select the most related input data and the lead time at which they can be best applied in the error-forecast model. The method is applied to a conceptual rainfall-runoff model of the Sieve basin in Tuscany, Italy. Artificial neural network models trained to forecast the residuals of the conceptual model at lead times of 1-6 h are applied to forecast the errors and improve the subsequent flow forecasts. The research shows that using a parallel data-driven model to complement the conceptual model produces much better runoff predictions in comparison to using the conceptual model alone.