Information Theory and Neural Networks for Managing Uncertainty in Flood Routing

Abstract

This paper presents an approach for handling uncertainties arising mainly from ignored or misrepresented processes in physically based models. The approach is based on the application of a parallel artificial neural network (ANN) model that uses state variables, input and output data, and previous model errors at specific time steps to predict the errors of a physically based model. Concepts from information theory are used to discover the relationships between the variables and the model errors, which also serves as a mechanism to detect the predictability of the errors. The resulting information is used to select the best related input data for the error prediction model. The error prediction model is then trained and applied to improve the forecasts made by the physically based model. This approach was applied to a routing model of a 70 km reach of the River Wye, United Kingdom. The results demonstrate that errors from the physically based model show a consistent trend governed by some dynamics of their own, which can be modeled with learning algorithms. Errors were forecasted at different lead times. In all cases the forecasts made by the combined application of both models were more accurate than those made by the physically based model alone. From this it was concluded that, along with proper information analysis techniques, the use of ANN models to predict the forecast errors of physically based models can help to improve significantly the prediction and therefore to reduce the associated uncertainty.