Decoupling rainfall occurrence and intensity: a two-stage GLM analysis of global climate drivers in Northeastern Bangladesh

The Haor Basin in northeast Bangladesh is a sensitive wetland ecosystem vital for national food security, yet highly vulnerable to pre-monsoon flash floods. To analyze 44 years of daily precipitation dynamics (1981–2024), this study applies a two-stage generalized linear model using high-resolution CHIRPS v2.0 data. This method separates precipitation into wet-day occurrence and wet-day intensity to evaluate the distinct impacts of global teleconnections and local atmospheric persistence. Results reveal a statistically significant long-term wetting trend in wet-day occurrence (Odds Ratio = 1.006 per year; p = 0.003). Furthermore, the El Niño-Southern Oscillation strongly suppresses rainfall occurrence during the monsoon peak (Odds Ratio = 0.865), whereas the Madden-Julian Oscillation increases precipitation probabilities (Odds Ratio = 1.070). Rainfall intensity, however, remains decoupled from these large-scale modes and is instead controlled by strong local autoregressive memory (Lag-1 Multiplier = 1.8). Overall, these findings support a gatekeeper hypothesis: large-scale climate states modulate convection initiation, while local thermodynamics govern storm intensity.