International conference on innovation advances and implementation of flood forecasting technology THE UNCERTAINTY CASCADE IN FLOOD FORECASTING Keith Beven (1), Renata Romanowicz (1), Florian Pappenberger (1), Peter Young (1) and Micha Werner (2) (1) Environmental Science/Lancaster Environment Centre, Lancaster University, LA1 4YQ, UK (2) WL|Delft Hydraulics, Hydrology and Flood Management, Rotterdamseweg 185, 2629 HD Delft, The Netherlands Abstract A methodology for propagating and constraining the uncertainty inherent in real-time flood forecasting is presented and demonstrated on an application to the River Severn, UK. The flood forecasting system is based on a cascade of rainfall-runoff and flood routing models, developed using stochastic transfer functions with state dependent parameterisations to allow for nonlinearity. The nonlinearities require a Monte Carlo sampling approach to propagation of uncertainty. Model updating and uncertainty constraint as new water level data become available is based on a Kalman filtering approach. The methodology is being implemented into the UK National Flood Forecasting System. Keywords: flood forecasting, transfer functions, National Flood Forecasting System (NFFS), nonlinearity, uncertainty, INTRODUCTION There are many sources of error in making flood forecasts. Such errors mean that all forecasts must be considered uncertain, and that there is a real possibility of getting a forecast wrong, both by not issuing a warning and flood damages being incurred, or by issuing a warning and no flood damages being incurred. It has long been recognised that both types of error will have an effect on the public perception of and reaction to flood warnings (the “crying wolf” problem). Flood forecasting is therefore not just a scientific problem, it is a problem of managing and communicating uncertainty. This involves three critical issues: the representation of different types of uncertainties in the forecasting system; the (preferably optimal) constraint of uncertainty in forecasts by means of real- time data assimilation and updating; and the presentation of forecasts and their associated uncertainties to decision makers and public. This contribution addresses the first two of these problems. The uncertainties in flood forecasting are manifold. There are spatial and temporal uncertainties in the inputs to the system; in the antecedent conditions of the system; in the geometry of the system (including relevant flood defence infrastructure); in the possibility of infrastructure failure; in the characteristics of the system (in the form of model parameters); and in the limitations of the models available to fully represent the surface and subsurface flow processes in flood generation and routing. The importance of the different types of uncertainties will certainly vary with the time (and lead time) of the forecasts, and with the magnitude of the event. There are issues of how to represent these different types of uncertainties. We may have input scenarios based on ensemble predictions from rainfall nowcasting or multiple historical analogues (Buizza et al., 1999; Buizza et al., 2001; Molteni et al., 1996); stochastic uncertainties (based on a variety of assumptions) associated with the calibration of models of runoff generation processes (e.g. Romanowicz and Beven, 2005; Romanowicz et al., 2004); and fuzzy representation of uncertainty in flood inundation predictions (Pappenberger and Beven, 2004; Pappenberger et al., 2005a; Pappenberger et al., 2005b; Pappenberger et al., in press). 17 to 19 October 2005, Tromsø, Norway ACTIF/FloodMan/FloodRelief 1