This article proposes a method for predicting forest fires based on aerial imagery data, which considers frame-by-frame snapshots of the research object along with geospatial coordinates and camera shooting vectors, as well as additional parameters such as soil temperature and humidity. Issues of classifying forests into two main types, coniferous and deciduous, are considered, and the leading causes of fires in these forests are discussed. An architecture for implementing software for collecting and storing a set of parameters of the lower layer of the forest canopy is proposed, allowing for predictive decision-making based on the results of past fire hazard events. Forest condition monitoring is proposed to be organized using unmanned aerial vehicles to obtain images of forest areas for tree type recognition, tree health analysis, and determination of drought index and vapor pressure deficit parameters.