Verification of intense precipitation forecasts from single models and ensemble prediction systems

The performance of single models and ensemble prediction systems has been investigated with respect to quantitative precipitation forecasts. Evaluation is based on the potential economic value of +72 h/+96 h forecasts. The verification procedure consists of taking into account all precipitation amounts that are predicted in the vicinity of an observation in order to compute spatial, multi-event contingency tables. A probabilistic forecast from an ensemble can thus be compared to a probabilistic forecast from a single model run. The main results are the following: (1) The performance of the forecasts increases with the precipitation threshold. High levels of potential value reflect high hit rates that are obtained at the expense of a high frequency of false alarms. (2) The ECMWF ensemble performs better than a single forecast based on the same model, even when the resolution of the ensemble is lower. This is true for the NCEP ensemble as well, but only for morning precipitations. (3) The ECMWF ensemble performs better than the 5-member NCEP ensemble running at 12:00 UTC, even when the population of the former is reduced to 5 members. (4) The impact of reducing the population of the ECMWF ensemble is rather small. Differences between 51 members and 21 members are hardly significant. (5) A 2-member poorman ensemble consisting of the control forecasts of the ECMWF and the NCEP ensembles performs as well as the ECMWF ensemble for afternoon precipitations.