FOREST FIRE ANALYSIS FROM PERSPECTIVE OF SPATIAL-TEMPORAL USING GSTAR (p;λ_1,λ_2,…,λ_p) MODEL

Abstract

West Kalimantan is particularly susceptible to the devastating effects of forest fires, among the natural disasters that have a significant impact. One of the indicators that can be used to identify forest fires is the presence of hotspots. The term "hotspot" refers to data that has both spatial and temporal characteristics. Using the Generalized Space-Time Autoregressive (GSTAR) model combined with the Queen Contiguity weight matrix, this research aims to model and forecast the confidence level of hotspots in Kubu Raya Regency and its surrounding areas. We chose the GSTAR model because of its ability to model spatial interactions between locations and temporal change patterns over time. According to NASA FIRMS, the data used in this study were confidence level hotspot data, covering the period from January 2014 to August 2024. To define locations for modeling, the study area was divided into grids measuring degrees. The maximum confidence level value in each grid was used to represent the highest potential fire risk. The research process consists of the following stages: data preparation, stationarity testing, calculation of the Queen Contiguity spatial weight matrix, identification of model orders based on STACF and STPACF plots, and estimation of model parameters to predict hotspot confidence levels. The GSTAR (3;1) model was selected as the best model because it satisfies the white-noise assumption and has a MAPE value of 14.78%. Based on the MAPE, the GSTAR (3;1) model can provide reasonably accurate predictions for the confidence level of fire points over the following three periods. The prediction results indicate a decline in the fire point confidence level across all locations during the following three periods. The findings of this study can support the optimization of resource allocation in the prevention of forest fires.