Temperature Data Prediction in South Sulawesi Province Using Seasonal-Generalized Space Time Autoregressive (S-GSTAR) Model

Abstract

Indonesia's distinct tropical climate is influenced by its geographic location near the equator and its complex topography, resulting in pronounced seasonal temperature patterns. This study examines the application of the Seasonal Generalized Space-Time Autoregressive (SGSTAR) model to forecast the average air temperature in four regions of South Sulawesi Province: North Luwu, Tana Toraja, Maros, and Makassar. The dataset comprises monthly average temperatures from January 2019 to October 2024, sourced from BMKG's online database. The analysis includes stationarity testing using the Augmented Dickey-Fuller (ADF) test, seasonal pattern identification with autocorrelation function (ACF), and formal seasonal tests such as QS, QS-R, and KW-R. Spatial weight matrices were constructed based on Euclidean distances between regions. The best model was selected based on Mean Square Error (MSE), Root Mean Square Error (RMSE), Akaike Information Criterion (AIC), and adjusted R² criteria. The findings reveal that the seasonal GSTAR model with AR orders (p=3), (ps=4), and (s=12) is the optimal model. Evaluation indicates that the model achieves high accuracy, with forecast errors (MSE and RMSE) below 1°C. This model effectively captures seasonal and spatio-temporal patterns in climate data. The study is expected to serve as a foundation for further development of seasonal GSTAR models for other climate datasets, supporting improved environmental planning and resource management.