Artificial neural network and energy budget method to predict daily evaporation of Boudaroua reservoir (northern Morocco)

Abstract

Evaporation is one of the main essential components of the hydrologic cycle. The study of this parameter has significant consequences for knowing reservoir level forecasts and water resource management. This study aimed to test the three artificial neural networks (feed-forward, Elman and nonlinear autoregressive network with exogenous inputs (NARX) models) and multiple linear regression to predict the rate of evaporation in the Boudaroua reservoir using the calculated values obtained from the energy budget method. The various combinations of meteorological data, including solar radiation, air temperature, relative humidity, and wind speed, are used for the training and testing of the model's studies. The architecture that was finally chosen for three types of neural networks has the 4-10-1 structure, with contents of 4 neurons in the input layer, 10 neurons in the hidden layer and 1 neuron in the output layer. The calculated evaporation rate presents a typical annual cycle, with low values in winter and high values in summer. Moreover, air temperature and solar radiation were identified as meteorological variables that mostly influenced the rate of evaporation in this reservoir, with an annual average equal to 4.67 mm•d -1 . The performance evaluation criteria, including the coefficient of determination (R 2 ), root mean square error (RMSE) and mean absolute error (MAE) approved that all the networks studied were valid for the simulation of evaporation rate and gave better results than the multiple linear regression (MLR) models in the study area.