Abstract:
During extreme rainfall events reliance cannot be placed upon local measurements of rainfall or river flow as hydrometric networks may be destroyed. Therefore, forecasting procedures are increasingly dependent upon remote sensing systems such as radar, satellite-based instrumentation and high resolution Numerical Weather Prediction (NWP) models. In addition, it is necessary to have knowledge of antecedent catchment conditions, conditions at the beginning of an event and a suitable hydrologic modelling structure. However, all these elements introduce uncertainty throughout the forecast chain, and it is necessary to understand the propagation of uncertainty and present it as an integral part of the forecast. In this paper we discuss the preparation of rainfall analyses for use in forecast procedures aimed at warnings an hour or so (nowcasts) to days ahead. The discussion includes procedures for improving the quality of radar and satellite estimates of rainfall, and the development of data assimilation techniques for NWP models. In spite of efforts to specify accurate model initial conditions in a non-linear dynamical system the growth of initial uncertainties in space and time is flow-dependent. To determine the predictability of this flow-dependency, an ensemble of forecasts for small perturbations_in model input conditions may be generated and analysed. Similarly, sources of uncertainty in hydrological forecasts may be addressed using ensemble approaches. The uncertainties in hydrological forecasts need to be incorporated in cost-loss analyses in order to decide what decisions regarding flood mitigation and/or adaptation are best taken to minimize flood losses.
