Satellite Image-Based Multiclass Forest Fire Risk Zone Assessment Using Deep Learning and Geospatial Technology: A Case Study of Similipal Biosphere Reserve, Odisha, India

Abstract

Forest fires are occurring more often and burning up more land as the world gets warmer, which poses a high risk to geological systems, biodiversity, and surrounding ecosystems. Understanding where our forests are at risk is important for being ready and for preventing disasters. Current forest fire monitoring systems can only consider a handful of parameters, and analyzing all the relevant data for a forested area is a manually intensive process that is not practically applicable. The main objective of this research study is to propose a Satellite Image-based Multiclass Forest Fire Risk Zone Assessment Model by integrating Deep Learning and Geospatial Technology. Multispectral satellite images obtained from Sentinel-2 satellites are analyzed in GIS software to derive environmental variables such as vegetation health and surface moisture status. The remote sensing indices namely Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), Moisture Stress Index (MSI) are derived from these images and related with forest fire risk. The features and preprocessed image data were then trained and analyzed using various deep learning algorithms, including Convolutional Neural Networks (CNN) and transformers. The validation results confirmed the effectiveness of forest land cover classification at different fire risk levels and proved that the GIS-based preprocessing and deep learning method can significantly improve the classification accuracy and efficiency. This research framework has the promise of realizing the effective application of forest fire risk assessment, emergency early warning, and disaster reduction and mitigation measures.