Spatio-Temporal Forecasting of Air Quality in Surabaya, Sidoarjo, and Gresik Using Generalized Space Time Autoregressive Integrated Moving Average (GSTARIMA)

This research employs Spatio-Temporal forecasting using the GSTARIMA) model. The GSTARIMA model is an extension of the STARIMA model, which assumes that parameters are homogeneous across locations. GSTARIMA generalizes the STARIMA model by allowing for heterogeneous parameters between locations at the same spatial lag. This study aims to establish an early warning system for air pollution in Surabaya, Sidoarjo, and Gresik by forecasting pollutant levels, particularly focusing on Particulate Matter 2.5 (PM2.5) concentrations. The data in this study were obtained from the Environmental Agency of East Java Province, covering the period from June 24^th to August 31^st, 2025. Three weights-normalized cross-correlation, uniform, and inverse distance weights-as well as two parameter estimates-OLS and SUR-were used in this investigation. The results demonstrate that lower RMSE values were produced by the GSTAR(11) model with an inverse distance weight and SUR parameter estimation for Surabaya, the GSTAR(11) model with an inverse distance weight and SUR parameter estimation for Sidoarjo, and the GSTAR(11) model with a uniform weight and OLS parameter estimation for Gresik. When using the GSTAR(11) model with an inverse distance weight and SUR parameter estimation, the average RMSE is often smaller than those obtained using other weighting schemes or parameter estimation methods.