Comparative assessment of metaheuristic-optimized SVR models for river discharge prediction in the Dez watershed

The accurate and reliable prediction of river discharge is fundamentally crucial for effective water resource management, flood control, and sustainable development, necessitating the continuous advancement of high-fidelity, non-linear modeling techniques. Determining river flow remains a core challenge, typically addressed using a variety of methodologies, including numerical, analytical, empirical, and those leveraging artificial intelligence. This research introduces and comparatively assesses a novel hybrid intelligent model, centered on the Support Vector Regression (SVR) framework, specifically designed to enhance river discharge prediction. To achieve this, three distinct optimization and decomposition techniques—Wavelet analysis (WSVR), the Innovative Gunner Algorithm (AIG-SVR), and the Black Widow Spider Algorithm (BWO-SVR)—were integrated into the SVR model. The models were trained and tested using hydrometric data collected between 2012 and 2022 from four stations within the Dez basin, Iran (Tireh Marvakk, Cham Chit, Sezar, and TangPang). The dataset was divided into training (80%) and testing (20%) subsets, and four scenarios of combined input parameters were utilized. Model efficacy was judged using standard statistical metrics: the correlation coefficient R2, Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and Kling–Gupta Efficiency (KGE). The findings consistently indicated that implementing combined input scenarios substantially enhanced the predictive capacity of the models. Crucially, the SVR-Wavelet (WSVR) hybrid model exhibited definitive superiority across all stations. During the validation phase, the WSVR model achieved exceptional performance, with R2 ranging from 0.898 to 0.985 and KGE values between 0.951 and 0.995, clearly outperforming the metaheuristic-optimized variants. In conclusion, these outcomes strongly suggest that strategic input feature engineering via Wavelet decomposition is a more potent and effective methodology for capturing the complex non-stationary nature of river flow than merely optimizing SVR parameters, thus promoting more sustainable river engineering practices.