Hybrid models for hydrological forecasting

High numbers of extreme and unexpected flood events in the last decade lead to an increased interest to build more effective flow forecasting systems. They are necessary to provide flood warnings in order to prevent loss of life and to minimize damage to both properties and livestock.

Traditional conceptual and physically-based hydrological models are widely used but they suffer from certain deficiencies. As a consequence operational forecasting has been exploring a number of novel techniques including the so-called data driven models which in many situations have proven to be accurate. However, they are not based on the physical representation of the ongoing process.

As a step forward in flow simulation and forecasting this dissertation explores the use of integrated solutions with process-based and data-driven models. The models resulting from such integration are referred to as hybrid models. It was proposed to use a hybrid modelling framework, and base it on the principle of “modular modelling”.