Adaptive Flood Warning and River Management

Recent research carried out in CRES has produced a data-based modelling approach to the adaptive forecasting of river flow or levels using rainfall or upstream flow inputs as leading indicators. This approach, which is based on the estimation of a general transfer function model from historical data, has significant advantages over existing flood forecasting methods: for example, on-line adjustments to the model are made automatically using a sophisticated but fairly simple adaptive parameter estimation technique which updates the model in real time, so producing improved forecasts for systems that exhibit time varying characteristics. Also, estimation of this adaptive parameter also allows probabilistic forecasts to be produced, in contrast to the deterministic forecasts of more conventional systems. In the case of rainfall-runoff forecasting, the transfer function model subsumes other less general models such as the commonly used Input-Storage-Output (ISO) model.