Agriculture is one of the most important factors contributing to the economy of Songkhla province in Thailand. Since the agriculture is highly dependent on the climate and hence on the temperatures, it was the aim of this study to investigate the trends and model the Land Surface Temperature (LST) from January 2001 to December 2018 of Songkhla province. Firstly, simple linear regression was applied and it was found that LST has increased approximately 0.3312 degrees Celsius during the last 18 years. After that the data were divided by 70:30 split to training and testing sets. Then for the predictive model, multiple linear regression and ARIMA (p,d,q) models were fit. Among the possible choices of (p,d,q) parameters in ARIMA, (3,0,0) performed the best. Further, according to their respective root mean squared errors (RMSE), namely 1.3334 and 1.3248, the ARIMA (3,0,0) performed slightly better than multiple linear regression in the training set. However, multiple linear regression performed slightly better than ARIMA (3,0,0) in the test data, with respective RMSEs 1.3249 and 1.3489. In other words, ARIMA gave a better fit, but linear regression gave better predictions. It is worth noting that the performance of a model type varies depending both on context and on the proportions of training and testing sets, so this case study demonstrates a model comparison approach but the results do not allow a generally applicable conclusion of ranking the model types
Keywords
Land Surface Temperature, temperature prediction, Remote Sensing, Linear Regression, ARIMA
NARESUAN UNIVERSITY JOURNAL OF SCIENCE AND TECHNOLOGY