International Journal of Computing and Digital Systems | 2026
Authors: Musleh F.A.; Taha R.G.
DOI: 10.12785/ijcds/1571183932
Journal: International Journal of Computing and Digital Systems
Year: 2026
Publisher: University of Bahrain
Document Type: Article
Open Access: All Open Access; Gold Open Access
Cited by: 0
Accurate prediction of streamflow plays a vital role in sustainable water management, especially in arid and transboundary basins that experience substantial variability and limited hydrological records. The present work focuses on the Flint River at Riverview Plantation (USGS Station 02355662), Georgia, and performs an extensive hydroclimatic assessment of streamflow behavior based on the 2002–2017 dataset. Statistical hydrology tools were employed, including the Mann–Kendall trend test and Sen’s slope estimator for temporal analysis, Pettitt’s test for regime-shift detection, and the Lyne–Hollick digital filter for baseflow extraction. Rainfall–runoff interactions were examined through lag correlation and Pearson coefficients, while Seasonal-Trend decomposition (STL) and Peak-Over-Threshold (POT) analyses were conducted to isolate and characterize hydrologic extremes. To improve forecast accuracy, four supervised learning techniques—Random Forest (RF), eXtreme Gradient Boosting (XGBoost), Feedforward Neural Network (FNN), and Linear Regression (LR)—were designed and compared. Model skill was evaluated through multiple metrics including the coefficient of determination (R2 ), prediction accuracy, and Mean Absolute Error (MAE). Among the tested models, RF exhibited superior performance (R2 ≈ 0.994, accuracy ≈ 99.1%), followed closely by XGBoost (R2 ≈ 0.993) and FNN (R2 ≈ 0.991). LR underperformed (R2 ≈ 0.982) and produced the largest MAE (≈14.15%), demonstrating limited capacity to capture nonlinear streamflow dynamics. The study introduces an integrated workflow that merges comprehensive hydrological diagnostics with advanced machine learning architectures for enhanced streamflow modeling. In contrast to earlier research emphasizing accuracy alone, this framework prioritizes model interpretability using diagnostic analyses such as trend testing, lag-effect quantification, and baseflow decomposition. The proposed methodology is particularly suited to regions with scarce data and high climatic sensitivity, providing a robust and scalable foundation for streamflow prediction and hydrological risk management. © 2026, University of Bahrain. All rights reserved.
Flint River; Georgia; hydrology; rainfall–runoff analysis; river discharge modeling; streamflow forecasting; supervised learning; trend detection; water resource management