Innovative Approaches to Flood Forecasting Using Data Driven and Hybrid Modelling

Flood forecasting in rivers and coastal waters demands careful attention both to the reliability of the forecasts and the safety of the decisions made on the basis of the results. Advances in data driven modelling have improved the accuracy of forecasts made using physically based models. The hybrid modelling approach combines the best features of physically based and data driven modelling, either through a combination of their outputs, or using the latter to estimate residual errors and associated confidence bounds of the former. Whereas artificial neural networks are the usual data driven models used for these purposes, increasingly other techniques such as M5 model trees are proving to be as, if not more, powerful because of their focus on localized modelling and higher transparency for practitioners. This paper draws attention to the innovative use of such techniques for flood forecasting in rivers.