Abstract The Tropical Rainfall Measuring Mission (TRMM) satellite provides simultaneous radar and lightning observations, offering critical insights into severe convection across regional and global scales; however, its snapshot nature constrains research on thunderstorm evolution. To address this, establishing a clustering framework is essential for understanding the stage‐dependent relationship between lightning and radar echo structural features and for improving lightning parameterization schemes. Using 16 years of TRMM observations from tropical Africa, the global hotspot for lightning and thunderstorms, we developed a machine learning‐based K‐means clustering method that classifies snapshot thunderstorms by their radar echo structure. Based on this classification, compact storms were further grouped into three clusters, which effectively capture the physical processes of thunderstorm evolution and are consistent with the known life cycle characteristics; therefore, they are identified as Pre‐Mature, Mature, and Post‐Mature stages. Mature‐stage storms are, unsurprisingly, the strongest. In comparison, the other two stages, reflecting the trade‐off between convective intensity and horizontal scale, Pre‐Mature storms show smaller horizontal scales yet higher lightning density (per unit area) despite comparable lightning flash rates, while Post‐Mature storms exhibit the reverse (larger scales accompanied by lower lightning density). Accordingly, the stage‐dependent relationships between lightning activity and convective intensity are examined. Similar correlation coefficients, but distinct linear fitting parameters, are observed across stages, revealing stage‐specific electrification mechanisms and emphasizing the necessity of stage‐dependent analysis. By emphasizing stage‐dependent analysis, this work advances the understanding of severe storm dynamics over tropical Africa and extends the utility of snapshot observations from precipitation radar onboard low‐Earth‐orbit satellites.
Wu et al. (2026) studied this question.