COMPLEXITY ANALYSIS OF WILDFIRE FORECASTING TASK WITH MACHINE LEARNING

This article examines the complexity of probabilistic wildfire forecasting using machine learning. A training set including meteorological, satellite, and historical data is analyzed, distributions are visualized, and dimensionality reduction is performed using nonlinear t-SNE and UMAP methods. It is demonstrated that the "fire" and "no fire" classes significantly overlap in the feature space, suggesting the need for complex, nonlinear machine learning methods.