Research and Application of Convolution-Based Algorithm for Graded Early Warning of Lightning

Mei Xinjian; Tian Runze

doi:10.16765/j.cnki.1673-7148.2026.03.011

Abstract：Lightning,as a common atmospheric discharge phenomenon in strong convective weather systems,poses significant challenges for accurate early warning due to its transient nature.Within short observation windows,the limited representativeness of lightning location data often fails to accurately re-flect the probability of lightning occurrence in the target area.To address these issues,this study propo-ses a method for constructing a thunderstorm index,based on ADTD lightning location data and lightning stroke current data,which integrates spatial weighting and standardization.This index is then used to achieve graded lightning warnings.A comparative test between this algorithm and the national standard GB/T 38121-2023 algorithm yields the following results:(1)The thunderstorm index effectively over-comes the discreteness limitations of raw lightning data.Areas with high index values show strong spatial consistency with intense radar echo areas,indicating that the index reliably identifies the center of thun-derstorm intensity.Furthermore,the dynamic changes in the thunderstorm index can characterize the spa-tial migration law and intensity development trend of thunderstorm activities.A rapid increase in the in-dex can serve as a key signal for the initiation,development,or movement of thunderstorm cloud clusters toward the center of the target area.(2)Compared to the national standard algorithm,which relies on predefined warning areas and fixed lightning count thresholds,the thunderstorm index algorithm offers greater flexibility.By optimizing spatial weighting parameters or adjusting index thresholds,it can better balance the Probability of Detection(POD)and the Effective Alert Rate(EAR)to meet the needs of different application scenarios.(3)Both algorithms meet the requirement of achieving a 10-minute lead time Probability of Detection(POD10min)exceeding 80%.However,the national standard algorithm ex-hibits relatively low Effective Alert Rates for 10-minute lead time(EAR10min),specifically 50.78%,38.83%,and 19.33%for Level 3,Level 2,and Level 1 warnings,respectively.In contrast,the thun-derstorm index algorithm improves these rates to 60.85%,53.02%,and 45.41%,demonstrating a sig-nificant advantage.(4)Regarding warning timeliness,the thunderstorm index algorithm maintains an Ef-fective Alert Rate for 20-minute lead time(EAR20min)of 53.22%while achieving a Probability of Detec-tion for 20-minute lead time(POD20min)of 80%.This indicates the method's potential for providing lon-ger effective warning times.Future integration of artificial intelligence time-series algorithms is expected to further enhance warning accuracy and effective warning duration.