A hybrid machine learning approach to unravel monsoon variability and meteorological dynamics of Pakistan's 2010 and 2022 historic floods

Pakistan. While large-scale climatic conditions have been studied, localized meteorological dynamics and their role in flooding are not fully understood . This study examined the factors associated with the flood events of 2010 and 2022. The research employed Random Forest Regression (RFR) to evaluate how key climate variables influence different pressure levels, achieving notable predictive accuracy for both years (2010: R² = 0.72, MAE = 0.29, RMSE = 0.41; 2022: R² = 0.68, MAE = 0.24, RMSE = 0.36). Principal Component Analysis (PCA) was employed to simplify the data and highlight important components to enhance the understanding of atmospheric variability. Subsequently, K-means clustering classified monsoon regimes into Active and Break phases, assisting in recognizing time-based patterns in the monsoon cycle. The study presented new perspectives on the localized atmospheric dynamics affecting flood events, even under comparable large-scale climatic conditions. It highlighted the influence of specific meteorological variables such as moisture flux and jet stream patterns. These insights are crucial for refining flood forecasting models, improving regional flood management practices, and providing actionable information to policymakers regarding climate change. • RFR identified critical climate variables affecting extreme rainfall and floods. • PCA and K-means clustering classified monsoon phases and rainfall. • Wind patterns, moisture, and thermodynamics influenced rainfall. • Increased moisture and upper-level conditions stress flood prediction.