Unsupervised Phenotyping of Forest Fires Using Clustering of Fire Weather Index and Meteorological Variables

Abstract

Due to climate change and human negligence, forest fire problems have become a hot topic in recent years and it is of great importance to be able to prevent them. The aim of this research is to propose a classification model using unsupervised machine learning on the UCI forest fire dataset to identify different fire phenotypes based on the components of the Fire Weather Index (FWI), meteorological variables, weather coding, and burned area characteristics. The model uses K-Means, Gaussian mixture models, DBSCAN (density-based clustering algorithm), and HDBSCAN (Hierarchical Density-Based Spatial Clustering) to optimize the number of clusters. The results obtained show that clustering provides a powerful framework for characterizing forest fire behavior, thereby improving the possibilities for further development in order to suppress and prevent the outbreak of new fires.