Evaluation of Hybrid Model ANN-PSO and Different Data Mining Methods in Estimating Monthly Evapotranspiration in Two Different Climates

Evapotranspiration is one of the main components of any region's water balance. Its accurate estimation is very necessary for hydrological studies, designing irrigation and drainage systems, and planning irrigation systems. In this research, the M5 tree model, M5 Rules, K Star, Rep Tree, artificial neural network model, and ANN-PSO neural network hybrid model in the estimation of reference evapotranspiration in two different climatic regions in Markazi Province based on the FAO Penman-Monteith model were evaluated. The data used included minimum and maximum temperature, average relative humidity, and wind speed at a height of two meters and sunny hours as input to the model from the synoptic stations of Delijan and Tafresh between 2004-2021, and evapotranspiration calculated by FAO Penman Monteith method was used as the output of the models. To evaluate the models, the Root Mean Square of Error (RMSE), Mean Absolute Error (MAE) and Correlation Coefficient (R) indices were used. The analysis of the results showed that for both stations, ANN-PSO neural network hybrid model had the best performance with RMSE=0.3115, MAE=0.2441 and R=0.9989 for the Delijan station and RMSE=0.2915, MAE=0.2355 and R=0.9989 for Tafresh station and tree model M5 with RMSE=0.2793, MAE=0.2398 and R=0.9967 for Delijan station and RMSE=0.2803, MAE=0.2306 and R=0.9969 for Tafresh station. Also, the Rep Tree model had the weakest performance among the models examined in this research. Considering that the tree model provides simple, linear and more comprehensible relationships for estimating reference evapotranspiration in addition to optimal accuracy, this model is recommended for estimating evapotranspiration in this region.