<![CDATA[Forest and land fires (karhutla) in the tropical peatland ecosystem of South Sumatra pose recurring ecological threats and transboundary haze disasters during every dry season. Existing early warning systems generally rely on satellite hotspot detections without accounting for the direction and rate of fire spread, and remain vulnerable to false alarms caused by persistent industrial heat sources such as refineries, palm oil mill flare stacks, and power plants. This study develops a deterministic, vector-based mathematical model to predict the direction, rate, and hazard-zone geometry of fire spread in near real-time, complemented by a spatial-temporal filtering algorithm that eliminates industrial heat sources. The model derives a propagation bearing from wind direction, a base rate of spread from four environmental factors, and constructs three risk zones as cone-shaped polygons in geospatial coordinates. The model was implemented in the Sumsel Hotspot Monitor system, processing VIIRS and MODIS data from NASA FIRMS. Evaluation using Intersection over Union (IoU) and Dice Similarity Coefficient against real satellite ground truth shows that model performance degrades as the prediction time horizon increases. These results confirm that the model can run at low computational cost and is suitable as an early prediction baseline, although its accuracy still requires further parameter calibration before full adoption by the regional disaster management agency.]]>
Copyrights © 2025
