Verification of an MSG Image Forecast Model: METCAST

Abstract

Abstract A validation of a very short-range forecast model is presented: the Meteosat Cloud Advection System (METCAST). The model forecasts IR (10.8 μm) images based on Meteosat Second Generation (MSG) data and uses ouput from the Royal Netherlands Meteorological Institute’s [Koninklijk Nederlands Meteorologisch Instituut (KNMI)] NWP model, the High Resolution Limited Area Model (HIRLAM). METCAST advects clouds and takes into account the evaporation–condensation processes in the atmosphere. To assimilate the satellite images into METCAST, an MSG image is converted to a modified image with coarser resolution. The relative performance of METCAST is evaluated, comparing the model results with persistence and a second nowcasting model called CineSat. Two statistical techniques are used to evaluate the forecasts: (a) the computation of the BIAS, RMSE, and Hanssen and Kuiper (HK) discriminant for a cloud mask selected in the modified and forecast images and (b) the contiguous rain areas (CRAs) technique, which permits a decomposition of the mean-squared error (MSE) of cloud clusters in three components: displacement, intensity, and shape. Five months of data, from June to November 2006 (August was not available), are used for this study. METCAST BIAS shows poor skill in comparison to CineSat and persistence. METCAST performs better in terms of the RMSE and HK discriminant. The CRA application reveals that although METCAST has a greater MSE volume component than CineSat, its displacement error component is smaller. Two interesting conclusions can be drawn: METCAST performs well when advecting cloudy pixels, but improvement in the atmospheric physics of the nowcast model may be required.