Mapping Post-Fire Canopy Condition as Discrete States Using Spectrally Interpretable Remote Sensing

Remote sensing is widely used to assess wildfire effects, with established approaches providing continuous estimates of fire severity from spectral information. This study developed a canopy-condition classification framework that maps post-fire forest canopies into three ecologically interpretable classes: green, scorched, and burned, using Landsat imagery, the Tasseled Cap Transformation, and a Random Forest classifier. Training sample points were obtained through visual interpretation of post-fire Google Earth imagery from eleven wildfire events in Portugal. Model performance was evaluated using stratified group K-fold cross-validation and an independent validation dataset comprising eight additional wildfire events. The classifier achieved an overall cross-validation accuracy of 0.906 and an independent validation accuracy of 0.93, demonstrating good generalization across independent wildfire events. Among the Tasseled Cap components, Greenness was the most important predictor for discriminating canopy condition. Rather than introducing another fire severity metric, the proposed framework provides a straightforward description of the dominant post-fire canopy condition that complements existing fire severity products. The resulting maps offer an additional perspective on canopy fire effects and can support post-fire assessment, monitoring, and management in Mediterranean forest ecosystems.