Probabilistic precipitation forecasts from a deterministic model: a pragmatic approach

Abstract

Precipitation forecasts from mesoscale numerical weather prediction (NWP) models often contain features that are not deterministically predictable and require a probabilistic forecast approach. However, some forecast providers still refrain from a probabilistic approach in operational forecasting because existing methods are associated with substantial costs. Therefore, a pragmatic, low-budget postprocessing procedure is presented that derives probabilistic precipitation forecasts from deterministic NWP model output. The methodology looks in the spatio-temporal neighbourhood of a point to get a set of forecasts and uses this set to derive a probabilistic forecast at the central point of the neighbourhood. For the sake of low implementation costs and low running costs, the procedure does without ensemble simulations, historical error statistics on the operational interaction of a forecaster. The procedure is applied to the output of the mesoscale model LM, the regional part of the operational modelling system of the German Weather Service (DWD). The probabilistic postprocessed forecast (PPPF) outperforms the deterministic direct model output in terms of forecast consistency, forecast quality and forecast value.