Optimal time series model for forecasting monthly temperature in the southwestern region of Thailand

Forecasting and describing the dynamic changes of climatic variables are essential in determining the occurrences of extreme climate events. The understanding of these events will help in taking actions to lessen their related effects. This study compares auto-regressive integrated moving-average (ARIMA) and the auto-regressive integrated moving average with exogenous variables (ARIMAX) models in forecasting temperature in Ranong and Phuket, Thailand. The average monthly temperature observations between 2006 and 2016 were collected from the Thai Meteorological Department for the study. The ARIMA and ARIMAX models were then applied to the average monthly temperature using the relative humidity and rainfall as the explanatory variables. Analysis of the root mean square error and the relative root mean square error (RRMSE) values from the models revealed that the methods fitted the data quite well. However, the optimal ARIMAX model obtained in Ranong was the ARIMAX (1,0,0)(0,1,0)12 with RRMSE value of 0.874 did better than the optimal ARIMA model. The relative humidity and rainfall factors were very influential in the models. Also, the ARIMA (3,0,2)(0,1,0)12 model with RRMSE value of 1.113 was observed to be the optimal model in the temperature modelling in Phuket. This model fitted better than the optimal ARIMAX model in temperature modelling at Phuket. The fitted models will be beneficial in numerous applications where the observed temperature records are quite short, incomplete, or lack spatial coverage