Evaluation of gene expression programming and Bayesian networks methods in predicting daily air temperature

Air temperature is one of the most important variables in estimating crop water requirement and climatic studies. In recent years, several intelligent models such as Gene Expression Programming and Bayesian Networks have been used to estimate air temperature. The purpose of the present research is to evaluate the accuracy of these two approaches in prediction of air temperature in a specific day (t) using data of one to seven days before, i.e. t-1 to t-7. For this purpose, a 25-years dataset of daily temperature of two stations in northwest of Iran, namely Urmia and Tabriz were collected and used for models performance comparison. The results showed that Gen Expression Programming and Bayesian Networks methods were capable to predict the minimum, mean and maximum air temperature with acceptable accuracy. However, the Bayesian networks method showed relatively better performance comparing to the Gene Expression Programming. The findings revealed that in testing stage of Bayesian networks method for Urmia station, the values of determination coefficient (R2) and root mean square error (RMSE) in the best scenario are 0.92 and 2.5 ◦C for minimum temperature, 0.96 and 1.83 ◦C for mean temperature, 0.96 ◦C and 2.3 ◦C for maximum temperature respectively. The corresponding values of statistical indices for Tabriz station in Bayesian networks method were found to be 0.93 and 2.42 ◦C for minimum temperature, 0.97 and 1.90 ◦C for mean temperature and 0.95 and 2.42 ◦C for maximum temperature. In general the mean temperature was predicted more accurately by both approaches in study stations.