Stationary and non-stationary flood frequency analysis of the Surma River Basin, northeastern Bangladesh

The Surma River Basin in northeastern Bangladesh is highly vulnerable to recurrent flash and riverine floods, yet discharge-based flood frequency estimates remain scarce. This study addresses that gap by analyzing 40 years (1985–2024) of annual maximum discharge from two key stations: Kanairghat (SW266) and Sylhet (SW267). Nine probability distributions were evaluated using Product Moments and L-moments, with goodness-of-fit tests (Chi-Square, Kolmogorov-Smirnov, Anderson-Darling) guiding model selection. At Kanairghat, the Generalized Extreme Value distribution with L-moments (GEV-LM) ranked first, capturing the heavy upper tail of flash-driven regimes. At Sylhet, the Log-Pearson Type-III with L-moments (LP3-LM) was optimal, reflecting downstream attenuation by Haor wetlands and channel sedimentation. Trend analysis revealed contrasting stationarity: Kanairghat exhibited a declining signal with a structural break around 1995, while Sylhet remained stationary. Non-stationary GEV modeling projected the 100-year flood at Kanairghat to decrease by roughly 22% by 2030 relative to 1985, underscoring the need to reassess stationary design values. Bootstrap confidence intervals and tail sensitivity analysis showed substantially wider uncertainty bounds at the upstream station. Design water levels consistently exceeded observed rating-curve ranges, indicating that current infrastructure may be under-designed for moderate return periods. The principal methodological contribution is a five-component uncertainty framework that simultaneously quantifies model, estimator, stationarity, sampling, and tail sensitivity uncertainty—a combination not previously applied to discharge-based flood frequency analysis in Bangladesh, offering a fuller characterization of design flood uncertainty than single-component assessments typically provide, within the two-station scope analyzed here. The methodology is directly applicable to other flash-flood-prone regions of Bangladesh, supporting more resilient hydraulic design and flood risk planning.