Investigating MSG-SEVIRI data as an additional predictor source in the KNMI probabilistic (severe) thunderstorm forecasting system

Thunderstorms can be a serious threat to society. In the Netherlands, the Royal Netherlands Meteorological Institute (KNMI) is responsible for issuing an extreme weather warning for severe thunderstorms, based on high total lightning intensity. To help forecasters decide whether they should issue an extreme weather warning, a Model Output Statistics (MOS) system was developed. This system uses logistic regression equations to predict both the probability of thunderstorms and the conditional probability of severe thunderstorms for twelve regions of 90 by 80 km2 over the Netherlands, and makes forecasts for 6-h periods up to 2 days ahead. Predictors are obtained from ECMWF and HIRLAM model output and from ensembles of advected radar and lightning data, the latter only for the 0-6 h forecasts. The system has been operational since 2006 and runs during the warm half year, from mid-April to mid-October.

In this study we investigate an ensemble of Meteosat Second Generation (MSG) data as an additional predictor source for the 0–6 h projections of the MOS thunderstorm forecasting system. Cloud Physical Properties (CPPs), which are derived from MSG data, are advected using vectors derived from previous MSG images. Varying the magnitude and direction of these vectors creates the ensemble. A description is given of the relations between CPP and lightning intensity. Predictors are created based on these relations and investigated as additional potential predictors in the system, besides those used in the current system. CPP predictors are included in 4 of the 8 severe thunderstorm forecast equations. Equations including these predictors generally improve the forecast skill of the system compared to forecast equations without CPP predictors and the (updated) operational system. Another advantage of the new severe thunderstorm forecast equations arises due to their derivation using extended logistic regression. Forecasts can be made using the new system for any arbitrarily chosen lighting intensity threshold. The forecasts prove to be skillful up to very high lightning intensities, much higher than those used in the currently operational forecasting system. The new MOS system currently runs experimentally at KNMI and will become operational if it improves forecast skill over the coming year.