Investigating the effectiveness of transfer functions based on machine learning methods for predicting reference evaporation and transpiration (Case study: Bushehr)

Accurate calculation of reference evapotranspiration is one of the basic tasks to achieve optimal management of water resources. In this research, the evapotranspiration of Bushehr reference was calculated by the Fau Penman Mantis method. Then, temperature methods (Hargreaves Samani and Blaney Cridle) and radiation methods (modified Mc King, Tork and Prestley Taylor) were also used to calculate evaporation transpiration. The results obtained from these methods were compared with the combined Fau Penman Mantis method. The results showed that among the two temperature methods, the Hargreaves Samani method and among the radiation methods, the Prestley Taylor method had closer results to the combined Fau Penman Mantis method. Artificial intelligence, support vector machine, random forest and cubist models were also used to estimate reference evaporation transpiration. The data used included minimum, maximum and average temperature, relative humidity, sunshine hour and wind speed during a thirty year statistical period from 1370 to 1400. In order to check the results of the mentioned models, the standard evaluation criteria of error mean square, absolute mean error and R2 explanation coefficient were used. The results showed that all three models were highly accurate in simulating evaporation transpiration. Cubist model with higher R2 (0.95), the lowest mean squared error (0.87) and the lowest absolute mean error (0.38) was chosen as the best method for evaporation transpiration.