Among severe meteorological hazards, lightning is considered one of the most dangerous; however, its forecast is difficult because the formation of lightning is the result of processes in the clouds that are difficult to model accurately. In this study, we predict lightning activity using the Lightning Potential Index (LPI), which is quite often used to determine areas with expected lightning activity, and analyse its spatial and temporal characteristics. Specifically, we performed simulations of LPI (15 min values) using the COSMO NWP model for 10 selected thunderstorm events of 2018 in Central Europe. We used the model runs with 1- and 2-moment cloud microphysics and with a lead time of 12 h presented in our previous study, while in this study we performed deeper analyses and verified the 15 min LPI values in space and time using ground-based observations of lightning. Our results showed that 2-moment cloud microphysics provide better LPI forecasts which confirms the suggestion of our previous study. The distribution of predicted lightning activity related to the model orography was examined and found consistent with the occurrence of recorded lightning discharges. The Fraction Skill Score analysis revealed that for 2-moment cloud microphysics a skilful forecast was reached at smaller scales than for 1-moment microphysics, namely at scales around 90 km for LPI thresholds 30, 40 and 50 Jkg−1. We also evaluated the forecasts using a performance diagram, which in contrast to other results did not confirm that forecasts using 2-moment cloud microphysical scheme were more accurate than forecasts using 1-moment cloud microphysical scheme. Spatial verification of LPI showed that depending on the distance limit (15–90 km) and the LPI threshold (from LPI > 0 Jkg−1 to LPI > 50 Jkg−1), the probability of lightning discharge occurrence was ca 30–90% and the proportion of successfully predicted lightning discharges reached up to 77%. We consider this result satisfying, though the spatial verification remains challenging. Contrary to spatial verification of LPI, the temporal verification of LPI turned out to be even more efficient (in 70% of cases the time difference between the defined beginnings of forecasted and detected lightning activity was maximum 45 min). In future, we plan to perform lightning prediction in another NWP model, namely the ICON NWP model. We also plan to analyse more thunderstorm events.
Read full abstract