Evaluation the performance of genetic programming in modeling mean monthly temperature in different climates of Iran

Mean monthly temperature is one of the most important parameters in agroclimatic studies and hence several approaches have been proposed for its precise estimation. In this study, the genetic programming approach is used to model monthly mean temperature in selected synoptic stations namely; Mashhad, Sanandaj, Tabriz, Ghazvin and Kermanshah with cold-arid climate and Yazd, Kerman, Zahedan, Bam and Zabol with warm-arid climate. Genetic programming approach was performed in two steps.1.Training and 2.Validation. In first step, the time series with six different patterns were prepared and trained. Then, in the second step, the obtained models were validated using coefficient of determination (R2) and root mean square error (RMSE) indices. Finally, based on these statistics, selected models were proposed for selected stations. The results showed that, genetic programming is an appropriate method for modeling mean monthly temperature. The result also indicated that, model performs better in warm-arid climates. The best results were obtained in 4th pattern of cold-arid and 5th pattern of warm-arid climates. Among the studied stations, Zabol showed the most acceptable results with R2 and RMSE of 0.96 and 1.91ºC, respectively.