Performance of machine learning methods for modeling reservoir management based on irregular daily data sets: a case study of Zit Emba dam

Forecasting the volume of water allocated to drinking water supply (DWS) and irrigation is strategic for efficient and effective planning and management of water mobilized by reservoir dams. The objective of this study is to simulate the total volume allocated (TVA) of water maintaining the minimum level of the reservoir and preventing spillovers. However, accurate and reliable simulation of TVA for optimum use of water resources cannot be achieved without precise and highly performing models. Therefore, this research has examined and compared three machine learning (ML) algorithms namely; random forest regression (RFR), support vector regression (SVR) and multi-layer perceptron neural network (MLPNN), using a database, of eight operating variables at the daily time step, collected over eight years (2009- 2017) at the Zit Emba dam (ZED) reservoir, in Algeria. Seven input combinations were considered and analyzed to find the best input variables for simulating TVA. The results indicate that although all the models, with five inputs, are adequate for modeling TVA, the performance of the RFR is better than the other models giving the correlation coefficient of 0.920, the root mean square error 0.006 hm³, the mean absolute error 0.003 hm³, and the Nash–Sutcliffe efficiency of 0.847. The findings demonstrate that the three ML algorithms are all promising tools for simulating TVA from the reservoir. Accordingly, the accuracy and efficiency of these models emphasize on their importance to be considered in reservoir planning and management.