Assessing Peatland Fire Susceptibility Using GIS and Machine Learning in Riau Province, Indonesia

Abstract

Peatland fires represent a recurring environmental disaster in Riau Province, Indonesia, particularly during the dry season. This research aims to analyze the spatial distribution of peatland fire hotspots during the dry months from 2019 to 2023, and to predict peatland fire susceptibility in the dry season using a machine learning approach, while also identifying the most influential environmental and anthropogenic factors. This study employs the Global Moran’s I to measure the spatial autocorrelation of peatland fire hotspot distribution and Forest-Based Classification and Regression (FBCR) to develop a machine learning-based prediction model, integrating spatial, meteorological, hydrological, and anthropogenic data. The results indicate that groundwater level (GWL) is a major contributing factor to peat fires, with a strong correlation between declining groundwater levels and increased fire risk. The FBCR model achieved an accuracy of 70–85%, successfully mapping the spatiotemporal distribution of peat fires, particularly during the dry season, when more than 60% of annual fire events occur. However, variations in the F1-score across different months suggest that the model can be further improved by addressing overfitting and incorporating real-time climate data. Future research should focus on developing ensemble learning or deep learning-based models to enhance prediction accuracy. Integration with high-resolution remote sensing technology and drone-based monitoring systems could further improve the model’s ability to detect fire-prone areas.