Ensemble lightning forecasting in the southern foothills of the Alborz mountain range using WRF model

Lightning is a phenomenon of instantaneous electrical discharge over long distances associated with thunderstorms which causes significant human and financial losses worldwide. In this study, the Lightning Potential Index (LPI), which measures the ability to charge within a cloud, was used to predict lightning occurrences on the southern slopes of the Alborz Mountains. This was carried out using a regional WRF model at three nested domains with resolutions of 9, 3, and 1 km. Twelve physics parameterization schemes were utilized in the model with initial and boundary conditions taken from Global Forecast System (GFS), 21 members of Global Ensemble Forecast System (GEFS), and the ECAMWF ERA5 data at a resolution of 0.25 degrees. Additionally, lightning occurrence prediction was enhanced using machine learning methods, including support vector machines (SVM) and random forests (RF). The Earth Networks data was used for the real lightning data. The highest value of the squared correlation R2, or coefficient of determination, was 0.41 with a NRMSE (normalized root mean square error: RMSE divided by the standard deviation) of 0.77 for the GEFS input data set using the Goddard microphysics parameterization, Kain–Fritsch (KF) convection, Dudiha for shortwave radiation, RRTM for longwave radiation, MYJ for boundary layer, and Noah LSM surface layer. The lowest value of R2 is 0.06 with the NRMSE of 0.97 for the GFS input data set and the Morrison–Morr microphysics parameterization, Grell–Devenyi ensemble convection, RRTM for radiation, MYNN boundary layer and NOAH LSM surface layer. In addition to LPI, other quantities related to static instability were also examined for their statistical relation with the number of lightning flashes. The quantities examined included Convective Avaliable Potential Energy (CAPE), Cloud Physics Thunder Parameter (CPTP), K index (KI), Convection Inhibition (CIN) and equivalent reflectivity factor (DBZ). The R2 values for the linear regression between the number of flashes and CAPE, CPTP, KI, CIN and DBZ were 0.14, 0.07, 0.02, 0.03 and 0.07, respectively. Therefore, only CAPE exhibited modest statistical relation with the lightning flashes. After weighted matching and machine learning correction with the support vector machine (SVM) method, the R2 value increased to 0.44 and 0.59, respectively, while the corresponging NRMSE values were 0.75 and 0.65. Given the significant impact of CAPE on the formation of convective clouds, its weight was doubled in applying the SVM. Consequently, R2 for the quadratic regression between the LPI and the actual lightning data has increased to 0.63, while NRMSE decreased slightly to 0.62. Overall, the results suggest that the LPI index is a suitable indicator for predicting lightning occurrence on the southern slopes of the Alborz Mountains, as there is a sufficiently strong statistical relation between the actual lightning data and the LPI index. Moreover, following weighted matching and correction by machine learning, the accuracy of lightning prediction is significantly improved.