Integration of Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2) data and Lightning Detection Sensor (LDS) data for lightning event prediction

Abstract

Lightning, a complex and potentially destructive atmospheric phenomenon, poses significant risks to public safety and infrastructure resilience globally. In India, the frequency of lightning strikes, particularly during the monsoon season, underscores the importance of early warning systems. The development of accurate detection and timely warning infrastructures is essential to mitigate the impact of lightning events and enhance disaster preparedness. At present 46 lightning detection sensors (LDS) are installed across India, by the Indian Space Research Organisation (ISRO). Each LDS is having 300 Km range detection. The network is designed with a 50% overlap to ensure high geolocation accuracy and maintain redundancy in regions with a strong LDS presence. Though this study is focused on a region with a strong existing LDS network, we recognized that there are regions throughout the nation where the LDS network is weak or nonexistent moreover LDS focuses on detection rather than prediction. To address this disparity, the study has two primary objectives: first, to advance beyond detection by developing models that can predict lightning events before they occur; and second, to enhance lightning detection capabilities in areas where the LDS network is sparse or nonexistent. This research relies on data from the Modern-Era Retrospective Analysis for Research and Applications, Version 2 (MERRA-2) dataset and the LDS dataset, focusing on the region between 86.68°E to 88.52°E longitude and 24.70°N to 26.42°N latitude during June 2022. To improve grid and temporal resolution, data refinement techniques were applied. Following this, statistical analyses, including the Chi-squared test, ANOVA, and Spearman-Rho correlation, were conducted to identify the parameters most strongly correlated with lightning occurrences. Humidity, pressure, and precipitation emerged as the most predictive factors.

Using these parameters, the Extra Trees model with bagging was employed to predict lightning occurrences, and a Random Forest classifier was used to predict lightning intensity based on the number of strikes. These models were validated using additional datasets. The findings from this study have the potential to significantly advance early warning systems, particularly in regions with limited LDS coverage, thereby enhancing resilience to natural hazards such as lightning across a wider area.