Developing reservoir evaporation predictive model for successful dam management

• Main Authors: Allawi, Mohammed Falah, Ahmed, Mohammed Lateef, Aidan, Ibraheem Abdallah, Deo, Ravinesh C., El-Shafie, Ahmed
• Format: Article
• Published: Springer 2021
• Subjects: QA Mathematics; TA Engineering (General). Civil engineering (General); TD Environmental technology. Sanitary engineering
• Online Access: http://eprints.um.edu.my/27860/

Evaporation is a primary component of the hydrological cycle, water resources management and forward planning. The succeed management for the dam system is based on the accurate prediction of the reservoir evaporation magnitude. Physical models applied in the prediction of evaporation can encounter obstacles in respect to accurate estimations of evaporation due to the inherent challenges in respect to the mathematical procedure that could fail to address the natural processes and initial conditions that drive the evaporation patterns. To address these limitations, the present study aims to design a new model using the modified Coactive Neuro-Fuzzy Inference System (CANFIS) algorithm to improve feature extraction process in a purely data-driven model. The new approach comprised of the adjustments made to the back-propagation algorithm, allowing the automatic updating of the membership rules and hence, providing the center-weighted set rather than the global weight sets for input-target feature mapping. The predictive ability of the modified CANIFIS model is benchmarked in respect to the conventional ANFIS, SVR and RBF-NN model by statistical performance metrics. To explore its efficiency, the modified CANFIS method is applied for evaporation prediction in two diverse climatic environments. The results revealed the superiority of the modified CANFIS model for evaporation prediction in both Aswan High Dam (AHD) and Timah Tasoh Dam (TTD). The statistical indicators supported the better performance of the modified CANFIS model, which significantly outperforms other proposed models to attain relative error value less than (23% for AHD, 20% for TTD), MAE (12.72 mm month(-1) for AHD, 7.63 mm month(-1) for TTD), RMSE (15.42 mm month(-1) for AHD, 8.53 mm month(-1) for TTD) and a relative large coefficient of determination (0.96 for AHD, 0.91 for TTD).