Fig. 1: Overall framework of the study.
a Wildfire data describing individual events in terms of fire-related characteristics such as size, perimeter, duration, and average expansion are collected from products derived from Moderate Resolution Imaging Spectroradiometer (MODIS) satellite observations. b The data are processed and consolidated into a raster dividing the world into a grid with a resolution of 1∘ × 1∘. Annual statistics and features are calculated for each cell, generating numerical (e.g., average fire frequency per time period) and spatial datasets. c Statistical methods to analyze multidimensional data are combined with unsupervised learning in order to discover similar groups of cells sharing fire-related characteristics. No explicit spatial components are included in this step. d Climatic and socio-economic layers are introduced for each cell in the grid. e Spatial density plots are generated for each pyrome, detecting the regions of the world with more observations, assumed to spatially frame a specific regime. The detected fire pyromes and regimes are characterized using climatic and demographic data. An evaluation of the influencing factors is performed for the most relevant areas. A temporal analysis to determine the trends and seasonality patterns of fire activity is also conducted. f All results and generated datasets are deployed on cloud services and a public-access repository, along with the scripts to reproduce all steps of the study.
