Severe thunderstorms pose significant challenges to communities and infrastructure, with timely and accurate prediction increasingly difficult. This study compares a control run (CNTL) with a 3DVAR run by assimilating upper-air observational data, focusing on severe thunderstorm events in Bihar on 25th June, 2020, and West Bengal on 07th June, 2021. It assesses the impact of data assimilation on severe thunderstorm initiation and intensification, considering factors such as Convective Available Potential Energy (CAPE), Convective Inhibition (CIN), rainfall, wind, and relative humidity. Using 0.25 degrees * 0.25 degrees NCEP GDAS FNL data, the model was integrated for 48 h with eight 6-hourly assimilation cycles for 3DVAR for both cases. Bihar exhibited a CAPE of 1800 J/kg, while West Bengal recorded 5000 J/kg, indicating severe thunderstorm conditions. Model performance metrics, including Accuracy (ACC), Equitable Threat Score (ETS), Heidke Skill Score (HSS), Probability of Detection (POD), True Skill Statistic (TSS), Critical Success Index (CSI), and False Alarm Ratio (FAR) were utilised. Assimilating upper air observational data notably improved model-simulated results, rectifying issues such as overestimated rainfall by 20-40 mm in Bihar, correcting overestimations of 100 mm, and underestimating 80-100 mm in northern and southern parts of West Bengal. Additionally, 3DVAR reduced spatial and temporal lags in both cases. ACC approached 1 for Bihar rainfall, while FAR neared 0 for West Bengal CIN during event time in the 3DVAR run. Each model runs showed similar patterns during peak intervals, but the 3DVAR approach generally exhibited closer alignment with the observation, emphasising its effectiveness in enhancing severe thunderstorm prediction accuracy.
