Pan Evaporation Modelling With Artificial Neural Network And Multivariate Non-Linear Regression Compared To Empirical Methods Under An Arid Environment

Evaporation, as a key process in the earth’s ecosystems, is a key factor in water resources and hydrometeorological research. Avoiding the expensive evaporation prediction methods, the pan evaporation technique is one of the most widely used and accepted methods. In this study, a multilayer neural network tool (ANN) and a multivariate non-linear regression technique (MNLR) are utilized to estimate the daily and monthly pan evaporation in the arid environment of Kuwait. Estimation results are compared to two empirical methods, the Cuenca and Christiansen models, employing the so-called combination method FAO-56 PM for the same raw data and input variables, daily average air temperature, wind speed, relative humidity and solar radiation. Cross-validation, 10-folds for the daily time scale and 5-folds for the monthly time scale, procedures are implemented to examine the model’s reliability and consistency. Evaluation metrics, like RMSE, R, MAE and NSE are applied to assess model performance. In general, it is demonstrated that the ANN model outperforms the MNLR model, however, the standard deviation among the cross-validated k-folds of the MNLR is lower than that of the ANN model. The monthly performance outperformed the daily output by achieving better statistical indicators with R=0.9084, 0.9092, 0.8961 and 0.9002 than the daily ANN, monthly ANN, daily MNLR and monthly MNLR models, respectively.