Explainable Unsupervised Models for Forest Fire Detection

Abstract

In 2023, Greece faced its worst wildfire season, with nine major fires causing unprecedented environmental damage of 1470.31 km2.In this research we are aiming to find a low-resource-consuming, unsupervised methods with suitable feature set to automatically detect forest fire. Preliminary research shows usefulness of featureless bands , and fire indices. Existing upsupervised methods based on multispectral satellite images bare based on feature thresholding and may lack generalization. Properties in focus while for unsupervised fire detection: 1. No pretrained model, but a discovery approach 2. Featureless band information in Sentinel-2 , saving the cost and overhead of feature selection, computation 3. Certainty of modeling 4. Experiments with state-of-the-art Fire Indices (FI) and categorization of FI by image acquisition time