Machine learning-based modeling for precise runoff forecasts in hydrological systems

Accurate daily runoff forecasting can provide detailed, data-driven decisions on normal rainfall, flooding, and drought. Therefore, skillful prediction of daily runoff in a region is a challenging task in hydrology. Most studies rely on standalone models, integrated models, or manually tuned hyperparameters, which hardly handle seasonal data. Therefore, this study attempted to integrate standalone models (Extreme Gradient Boosting, Adaptive Boosting, and Gene Expression Programming) with an advanced technique (Particle Swarm Optimization) for daily runoff prediction. Optimized models; PSO-XGB, PSO-Adb, and PSO-GEP were employed using nine years of rainfall data. The Performance of hybrid models was compared to that of classic GEP, XGB & Adb algorithms as the benchmarks. Models performance were evaluated using RMSE, MAE, NSE, R 2 , PBIAS, and the a-20 index. Among all methods, PSO-XGB achieved the highest accuracy (R 2 = 0.96), outperforming hybrid and standalone models. Regarding the novelty approach, a hypothetical user-friendly Software Interface (SI) was developed for hybrid models to deliver its use for real scenario engineering purposes. Additionally, the study also derived a GEP-based equation for practical runoff magnitude.