Limited-area model ensemble forecasts for heavy rain events during MAP-IOP 1999 are carried out at ARPA/SMR with the Local Model (LM) of the COSMO Consortium. For future real time applications it is desirable to know already at an early time which forecast out of the ensemble is the best in order to be able to undertake timely prevention measures.

As we are interested in heavy precipitation, satellite and radar remote sensing data exhibit the most relevant weather information. Clouds are detectable in the infrared (IR) and water vapor (WV) channels of METEOSAT and precipitation intensity is represented in the radar imagery. Furthermore, these are independent data sources with high temporal (15 to 30 minutes) and spatial (about 8 km for satellite and 2 km for radar) resolution which allow to trace mesoscale flow details.

For comparing model outputs against observed satellite and radar imagery it is necessary to construct synthetic images. The "Cloudy RTTOV" radiative transfer model (Chevallier, ECMWF), which also takes into account a cloudy atmosphere, is used here to calculate synthetic brightness temperatures in the IR and WV channels of METEOSAT. Synthetic satellite images are therefore constructed using the same color scale used for the observed pictures. Synthetic radar images can be constructed from model output fields by using the formula of Fovell and Ogura, which uses rain water and cloud water to construct radar echoes. However, for a comparison with observations it is necessary to take into account the specific radar characteristics (topography, atmospheric refraction, attenuation by gases and hydrometeors, scanning angles). Complex algorithms have been constructed which calculate a pseudo measurement. In this study however, we only use satellite images and leave synthetic radar images for a further investigation.

For the comparison between observed and synthetic satellite images we use an image matching method developed by Mannstein, which enables to detect and match picture elements at various scales. The results of the matching procedure are displacement vectors representing error measures for all pixels: the longer the displacement, the larger the error. Within complicated systems (cyclones) also the directional variance of the displacement vectors provides an error measure. For real time applications we intend to produce synthetic images already during the model run which provides a warning about severe weather phenomena already at the output time. As time goes by the matching with the observed images allows to trace the errors for every ensemble member and thus to identify the best forecast as the one with the smallest errors as regards to the displacement vectors.

In this work we present synthetic satellite images together with the corresponding displacement vectors from two MAP-IOP episodes.