Detection of Cb and TCu clouds using MSG-SEVIRI cloud physical properties and weather radar observations

Deep convective clouds, such as Towering Cumulus (TCu) and Cumulonimbus (Cb) clouds, may pose a serious risk to aviation. For a responsible replacement of human observers at airport weather stations in the Netherlands by automated observation systems of present weather, a Probability Of Detection (POD) of at least 80 % and a False Alarm Rate (FAR) of no more than 20 % is required. Therefore, the POD (58 %) and FAR (70 %) of the present KNMI automated radar-based Cb/TCu cloud detection method are not considered satisfactory. In this study, satellite derived cloud physical properties and High Resolution reflectances in the Visible (HRV) as well as weather radar data are used to develop a new Cb/TCu cloud detection method. The cloud physical properties are derived from the visible and near-infrared channels of the Spinning Enhanced Visible and InfraRed Imager (SEVIRI) onboard a Meteosat Second Generation (MSG) satellite. The detection method is constructed for a MSG box area around Schiphol airport for the daytime summer period. METeorological Aerodrome Reports (METAR) of Cb and TCu clouds are used as ’ground
truth’. The Cb/TCu cloud detection method is performed in two steps. First, a Convective Cloud Mask (CCM) is constructed to produce a hazard map. This map includes pixels which represent potential convective cloud pixels based on a thresholding technique. For the hazard map, the level of risk, which indicates the probability of presence of Cb/TCu clouds at and in the vicinity of the
airport, is determined using a logistic regression model. Predictors for the model have been derived from the cloud physical properties, HRV reflectance and weather radar data. The frequent selection of the HRV derived predictors in the forward stepwise selection method revealed the importance of high resolution satellite data. Therefore, a cloud optical thickness has been derived from the
HRV reflectance. The CCM shows a two-third decrease for the non-events (no Cb/TCu clouds), while over 95 % of the yes-events (Cb/TCu clouds) remain.
The predicted probabilities from the logistic regression model show good Reliability and Resolution and positive skill over sample climatology. Using the Critical Success Index (CSI) and the Bias, a probability threshold is determined
to convert predicted probabilities into predicted group memberships. Combining the results from the CCM with the results from the final logistic regression model, a POD of 65.2 % and a FAR of 35.4 % are obtained for a maximum CSI and a bias of no more than 5 %. These verification scores show a substantial improvement with respect to the scores from the present automated Cb/TCu cloud detection method.