Dataset for Machine Learning-based Wildfire Severity Prediction in Castilla-La Mancha, Spain

This repository contains the comprehensive dataset used to develop and validate machine learning models for predicting potential wildfire severity in the Castilla-La Mancha region of Spain. The data encompasses fire events that occurred between 2011 and 2022. This dataset is provided to ensure the full reproducibility of the results presented in our manuscript, "SEVERIA: A CONCEPTUAL ONLINE TOOL TO ASSESS POTENTIAL BURN SEVERITY IN MEDITERRANEAN FORESTS", which is currently under peer review. The full citation for the article will be added here upon its final publication. The source code is available on GitHub: Repository Link. All raster data (.tif files) are provided in the ETRS89 / UTM zone 30N coordinate reference system (EPSG:25830). File and Directory Structure: raster_layers_clm.zip: A compressed file containing eight raster layers (.tif format) representing the static topographic and vegetation variables for the entire Castilla-La Mancha study area. These variables served as the primary inputs for model training. The included layers are: Elevation Orientation (Aspect) Mean_vegetation_height Erodibility Flammability Fuel_type Slope Total_canopy_cover fires.zip: A compressed archive containing the data for individual fire events. The internal directory structure is organized by province and subsequently by individual fire. Each fire-specific subdirectory contains: A raster layer (.tif) of the observed fire severity, calculated as the differenced Normalized Burn Ratio (dNBR). A .csv file containing meteorological variables (Climate Anomaly, Vapor Pressure Deficit, and Wind Speed) recorded on the day of the fire from all available weather stations in the region. Three raster layers (.tif), one for each meteorological variable, spatially interpolated across the fire perimeter using an Inverse Distance Weighting (IDW) algorithm from the station data. pred_clm_15-07-2023.tif: A full-coverage raster prediction (.tif) of potential fire severity for the entire Castilla-La Mancha region for a specific date (July 15, 2023), generated using the final selected model. experiment_1.zip: A compressed archive containing supplementary files from the project's "Experiment 1". This includes the trained machine learning models that exceed GitHub's file size limits.