Automated Event-Based Burned-Area Delineation and dNBR-Based Spectral Severity Mapping for Re-Burn-Aware Recovery Prioritization: A Sentinel-2/GEE Case Study in Zakynthos, Greece
Georgios Stavridis, Konstantinos Poirazidis, Pantelis Apostolopoulos, Panteleimon XofisForest managers require consistent information on wildfire extent, satellite-derived burn effects, vegetation recovery, and sites requiring field inspection. We examined how a consistent Sentinel-2 workflow in Google Earth Engine could link event delineation and re-burn-aware spectral recovery monitoring to post-fire field-verification priorities in Zakynthos, Greece, over the 2016–2025 study period. Here, recovery means change in satellite-derived vegetation indices, not independently measured ecological restoration. A screened daily minimum Normalized Burn Ratio (NBR) workflow supplied the main analyses, while a seasonal-composite workflow was retained as a conservative baseline. Cloud-screened mosaics, persistent burn evidence, and confidence screening reduced 255 candidates to 117 high- or medium-confidence events. Timing diagnostics showed that fixed July–September composites are repeatable but not event-specific post-fire observations for every fire. Differenced NBR classes were treated as indicators of spectral change, not field-validated ecological burn severity. External footprint comparisons with selected Copernicus Emergency Management Service (CEMS) references yielded screened daily workflow intersection over union of 0.867, 0.799 and 0.890 for 2017, 2019 and 2025, respectively; the comparisons were not strictly same-date. Random Forest classification showed 85.86% agreement with NBR-derived recovery classes, with similar results under event-grouped and spatial-block validation. Agreement decreased when the Normalized Difference Vegetation Index (NDVI) and Normalized Difference Moisture Index (NDMI) recovery ratios were excluded, indicating dependence on recovery-related spectral information. High and very high priority zones covered 558.47 ha, but their extent was sensitive to weighting. The outputs support forest-service field checks for regeneration, tree and shrub condition, exposed soil, erosion indicators, and fuel continuity. They do not provide independent ecological validation and are not automatic restoration prescriptions.