资源环境科技发展态势分析平台

Lightning is considered one of the most severe meteorological hazards. Currently, many Numerical Weather Prediction (NWP) models make operational use of the Lightning Potential Index (LPI), which enables one to determine areas prone to lightning. In this study, we investigated the ability of LPI to forecast lightning by comparing forecasts of LPI among four configurations of the COSMO NWP model. The four configurations of the COSMO NWP model differ in horizontal resolution (1.2 and 2.2 km) and cloud microphysical scheme (1- and 2-moment cloud microphysics). We evaluated binary forecasts of LPI for varying lead time (1 -10h) against the observed number and peak current of lightning flashes recorded by EUCLID network. The evaluation was performed for eight area sizes (4.8 km x 4.8 km to 240 km x 240 km) around model grid points and for 8 days in 2018 when thunderstorms occurred in Central Europe. To assess the success of lightning prediction, we evaluated LPI forecasts using the Area under the Receiver Operating Characteristic (AROC). Results show that the forecasts almost always outperformed the random forecast (AROC = 0.5). As expected, the prediction was more successful for models having higher resolution and the models comprising 2-moment cloud microphysics. The results of the evaluation of LPI forecasts compared to the number of observed lightning were similar to those compared to the sum of the observed peak current. However, differences in results among the four configurations of the model were larger in the latter case. Based on the performed evaluation, we confirm that LPI is a suitable tool for implicit forecasting of lightning in the COSMO NWP model.