Reservoir Inflow Prediction: A Comparison between Semi Distributed Numerical and Artificial Neural Network Modelling

• Published in: Water resources management 2023-12, Vol.37 (15), p.6127-6143
• Main Authors: Shelke, Mahesh, Londhe, S. N., Dixit, P. R., Kolhe, Pravin
• Format: Article
• Language: English
• Subjects: Artificial neural networks; Atmospheric Sciences; Catchment area; Civil Engineering; Correlation coefficient; Correlation coefficients; Earth and Environmental Science; Earth Sciences; Environment; Geotechnical Engineering & Applied Earth Sciences; Hydrogeology; Hydrologic models; Hydrology/Water Resources; India; Inflow; Land use; management systems; Modelling; Neural networks; Operations management; prediction; Predictions; rain; Rainfall; Reservoir management; Reservoir operation; Root-mean-square errors; Statistical analysis; water; Water inflow; Watersheds
• ISSN: 0920-4741; 1573-1650
• DOI: 10.1007/s11269-023-03646-2

Reservoir inflow is a major component of the reservoir operations management system. It becomes highly essential to predict the accurate reservoir inflow. The lumped models and semi-distributed or fully distributed model implemented to solve a range of specific problems in the prediction of reservoir inflow. The findings in this paper compare a conceptual semi distributed Hydrologic Engineering Centre Hydrologic Modelling System (HEC-HMS) model and an ANN (Artificial Neural Network) based model for the prediction of inflow in the Koyna reservoir catchment, Maharashtra. The performance of the models is assessed using different statistical indicators such as Nash–Sutcliffe Efficiency (NSE), Root Mean Square Error (RMSE), Correlation Coefficient (r) and Mean Absolute Error (MAE). The results confirmed the ability of the semi distributed (r HEC-HMS  = 0.92, RMSE HEC-HMS  = 129.37 m 3 /s, MAE HEC-HMS  = 21.66 m 3 /s, NSE HEC-HMS  = 0.82 and RSR HEC-HMS  = 0.42) and ANN model (r ANN  = 0.85, RMSE ANN  = 176.29 m 3 /s, MAE ANN  = 14.62 m 3 /s, NSE ANN  = 0.69 and RSR ANN  = 0.55) to capture the effect of the complex hydrological phenomenon, variations of land use and soils of watershed. The study illustrates that the semi distributed HEC-HMS model shows moderately better results compared to ANN model. It may be noted that the ANN predicts the reservoir inflow using only one input i.e., rainfall, whereas the HEC-HMS requires exogenous input parameters and plenty of time for model building compared to ANN. This work will have a significant contribution for planning of reservoir operations within the catchment of Koyna reservoir.