Decoupling of Rainfall and Flood Magnitude in the Niger–Benue Confluence: Evidence from Three Decades of Monthly Hydrometeorological Data

Abstract Flood risk at tropical river confluences is conventionally viewed as primarily rainfall-driven. However, emerging evidence suggests that flood magnitude can decouple from rainfall trends due to anthropogenic modifications within the basin. This study investigates the relationships among rainfall, river stage, and discharge at the Niger–Benue confluence in Lokoja, Nigeria, and evaluates their influence on flood occurrence and artificial neural network (ANN) forecasting performance. Monthly hydrometeorological data (1990–2025, April–October) were analyzed using Theil–Sen trend estimation with Hamed-Rao modified Mann–Kendall tests, lag cross-correlation, and a multilayer perceptron ANN. Variable importance was assessed through permutation importance, SHAP analysis, ablation studies, and partial dependence plots. Results revealed statistically significant positive trends in all variables: rainfall (+13.71 mm/year, p = 0.024), water level (+36.60 units/year, p = 0.023), and discharge (+24.20 m³/s/year, p = 0.010). Cross-correlation analysis showed a very strong contemporaneous relationship between rainfall and discharge (r = 0.954 at lag 0) and strong discharge–stage coupling (r = 0.939 at lag 1). The ANN achieved high predictive performance on the independent test set (accuracy = 0.959, recall = 1.000). Permutation importance identified discharge as the dominant predictor (ΔNSE = 1.316), far outweighing water level and rainfall. These findings demonstrate that flood magnitude at the Niger–Benue confluence is increasingly driven by discharge thresholds rather than concurrent rainfall. The results challenge traditional rainfall-centric flood paradigms and support a shift toward discharge- and stage-based early warning systems in anthropogenically modified tropical basins. Keywords:   rainfall–flood decoupling, Niger–Benue confluence, trend analysis, artificial neural networks, SHAP, permutation importance, tropical hydrology, flood forecasting.