Verifying warnings for point precipitation

Abstract Short-term risk forecasts of point precipitation are obtained with COTREC/RainCast, a technique for extrapolation of radar images. The risk forecasts are updated every 5 min for the next 0–2 h. Risk levels are defined for moderate, heavy and extreme precipitation. Warning messages are generated if, at the locations of 23 rain gauges, these risk levels are reached or exceeded. The time-resolved gauge data are used to judge if the warning messages are in time, early or late. Data over a period of 4 months (summer 2002) are used for verification. The largest number of warnings (1790) was obtained for moderate precipitation. About 55% of these warnings were in time, 23% were early and 22% were late. This finding is in a good agreement with the defined risk level for warnings (50%), indicating that the model for calculating the risk factors is reliable. Less warnings in time, and more late warnings were found for heavy and extreme precipitation. Hence, the risk levels need to be lowered for heavy and extreme precipitation, in order to reduce the number of late warnings.