Spatio-temporal Bayes model for estimating the number of hotspots as an indicator of forest and land fires in Kalimantan Island, Indonesia

Abstract

Forest and land fires often occur on the island of Kalimantan and have a widespread impact on neighboring countries. One indicator of forest and land fires is hotspot. Climate factors play an important role in determining hotspot patterns and trends in a location, which often fluctuate and are difficult to predict. This research aims to predict the number of hotspot spatially and temporally in the next month on Kalimantan Island and analyze the influence of local climate on hotspot events. The Bayesian Conditional Autoregressive method with Integrated Nested Laplace Approximation and optimal weight selection using Getis-Ord G are used to increase prediction accuracy. The distribution of hotspot is assumed to follow the Negative Binomial distribution. The research results show that the best model uses an additive approach and interaction with explanatory variables with a Deviance Information Criterion value of 97,799.8. Predictions from this model have a Root Mean Square Prediction Error of 7.08 and an Average Absolute Prediction Error of 0.63. However, the model still has limitations in predicting extreme events. Climatic factors such as low rainfall, long days without rain, high air temperatures, and low humidity contribute significantly to the increase in the number of hotspot in Kalimantan.