Making Low Probability forecasts of High Impact Hydrological Events more useful for Society

Flash floods are a significant societal problem, that rank as the World Meteorological Organisation’s top priority hazard. Pinpointing where and when they will hit is however extremely challenging beyond lead times of an hour or two, even when using state of the art convection-resolving ensembles, due mainly to significant ensemble size limitations. There has been more success in highlighting areas at risk from flash floods by post-processing numerical model output, either from these limited area ensembles, or from global ensembles with parametrised convection, or by blending the two.

A benefit of using global ensembles is that they are much less constrained spatially and in terms of lead times. One successful post-processing approach applied here has been the ECMWF “ecPoint” system. This can deliver finite probabilities for very large, localised totals that ordinarily the raw ensemble system cannot, and should not, predict itself. These have verified very well but could be considered less actionable by users because the probabilities delivered, for a point in a given gridbox, in advance of extreme events, are often very small (e.g. 1-5%). This presentation will outline three developments related to the ecPoint approach that make it more amenable to users by 1) providing an estimate of likely maxima within a gridbox, that 2) tailor better to flash flood risk than purely to rainfall totals by cross referencing a new global point-rainfall climatology, and that 3) demonstrate clear ‘financial’ utility even if probabilities are small, via computations of potential economic value. Case studies will be used for illustration.