<![CDATA[Forest and peatland fires in Riau Province, Indonesia, are a recurrent environmental disaster with severe regional and global consequences. Traditional fire danger rating systems often fail to capture the complex interplay of factors driving these events. The advancement of artificial intelligence (AI) offers an opportunity to develop more accurate and dynamic fire risk prediction models. This study aimed to develop and validate a high-performance, AI-powered model for predicting daily forest fire risk at a high spatial resolution across Riau Province by integrating climate, peatland, and land use data. We integrated historical satellite-detected fire hotspots (2015-2023) as the dependent variable. Predictor variables included daily climate data (e.g., temperature, precipitation, wind speed), static peatland characteristics (e.g., depth, type), and dynamic land use/land cover data. An XGBoost (Extreme Gradient Boosting) machine learning algorithm was trained to learn the complex, non-linear relationships between these drivers and fire occurrence. The model’s predictive performance was rigorously evaluated using the Area Under the Curve (AUC) metric. The XGBoost model demonstrated high predictive accuracy, achieving an AUC of 0.93. The analysis revealed that the number of consecutive dry days, peatland depth, and proximity to oil palm plantations were the most influential variables in predicting fire risk. The model successfully generated daily 1-km resolution fire risk maps, identifying specific areas with elevated danger. The AI-powered model provides a robust and significantly more accurate tool for forest fire forecasting in fire-prone tropical peatland landscapes. This approach offers a critical advancement for developing effective early warning systems, enabling targeted resource allocation for fire prevention and mitigation efforts.]]>
Copyrights © 2025
