Multitemporal river flow discharge prediction: A new framework for integrated environmental management and flood control

Riverine flow estimation is critical for effective water resource management and mitigation planning. Traditional machine learning and deep learning models offer various advantages, but their effectiveness in multitemporal river flow discharge estimation has yet to be fully explored. This study introduces an advanced universal group method of data handling (AUGMDH) model to predict river flow discharge across various temporal scales. The accuracy of the proposed model is compared with that of convolutional neural network (CNN) models in terms of estimating daily, mean monthly, and maximum monthly flow discharge. The AUGMDH model consistently outperforms the CNN models across all major performance metrics, such as the coefficient of determination (R²), Nash-Sutcliffe efficiency (NSE), normalized root mean square error (NRMSE), RMSE-observed standard deviation ratio (RSR), and percent bias (PBIAS), achieving an R² of 0.972 and an NSE of 0.972 for the daily flow, an R² of 0.810 and an NSE of 0.810 for the mean monthly flow, and an R² of 0.819 and an NSE of 0.818 for the maximum monthly flow. Additionally, compared to the CNN approach, the AUGMDH model yields lower AIC values across all the cases (AIC: 37,744 for daily, 2144 for mean monthly, and 2543 for maximum monthly), indicating a better balance between simplicity and accuracy. In terms of uncertainty analysis, the AUGMDH model exhibits lower uncertainty values (i.e., 2.77 for daily flow, 2.31 for mean monthly flow, and 2.46 for maximum monthly flow estimates) than the CNN models do (i.e., 2.78 for daily flow, 2.48 mean monthly flow, and 2.66 for maximum monthly flow estimates). The findings indicate that the AUGMDH model provides a more robust and reliable solution for riverine flood estimation, outperforming the CNN models across all major performance metrics, including accuracy, reliability, and computational efficiency.