Evaluation of the Efficiency of Artificial Intelligence Models for Simulating Evaporation in Selected Stations in Dry, Semi-Dry and Very-Wet Climates in Iran

In this study, daily simulation models of pan evaporation were developed using a multilayer perceptron neural network, a radial basis function, and an optimized experimental relationship based on the genetic algorithm with 13 input combinations of climate variables, i.e., average daily temperature, maximum and minimum temperature, relative humidity, sunshine hours and wind speed. These models were applied to simulate evaporation in Semnan, Shahrud, and Rasht stations with dry, semi-dry, and very wet climatic conditions, respectively. The best input combination for simulation models was selected based on the quantitative evaluation results; i.e temperature, relative humidity, sunshine, and wind speed which had a correlation coefficient between 0.56 to 0.85, root mean square error between 1.1 to 2.6 mm, and the PBIS coefficient between 1 to 29.5%. The results of the developed models' sensitivity analysis showed that temperature and then relative humidity are sensitive parameters for evaporation simulation. The optimized experimental relationship with the best input combination among the evaporation simulation models had a greater ability to simulate the mean ​​and standard deviation of evaporation, and the model was selected as the best model. Finally, the ability of the best model with the best input combination was confirmed by its application to simulate the evaporation pan in three stations Garmsar (dry), Damghan (semi-dry), and Ramsar (very wet), resulted in correlation coefficients of 0.6 to 0.84, root mean square errors of 1.29 to 3.16 mm and PBIS coefficients of 2.1 to 9.2%.