Detecting Italian wildfires: a Google Earth Engine workflow for spatial and temporal detection, burned severity and vegetation recovery assessment

Wildfire mapping based on spectral behaviour analysis commonly requires prior knowledge of the fire date and manual selection of suitable pre- and post-fire satellite images. This condition limits reproducibility when multiple events, uncertain fire dates, delayed acquisitions, or strong seasonal vegetation dynamics are involved. This study presents a phenology-aware, rule-based workflow implemented in Google Earth Engine to identify candidate wildfire dates and delineate burned perimeters from Sentinel-2 time series. The method compares the Normalized Burn Ratio (NBR) behaviour for the year under analysis of a target area with a three-year control baseline derived from nearby unburned vegetation with comparable land-cover characteristics. Candidate fire dates are selected where the target area shows an abrupt NBR decline exceeding the expected seasonal decrease observed in the control domain. Pre- and post-fire NBR composites are then generated around the detected date, dNBR is computed, and burned areas are delineated through adaptive Rosin thresholding. The workflow was tested on 25 fire events distributed across Mediterranean and Apennine landscapes in Italy and validated against European Forest Fire Information System reference perimeters. The detected dates were within 5 days of the EFFIS final date for 64% of the events and within 8 days for 80% of the events. Burned-area validation showed generally good spatial agreement, with a median F1-score of 0.82 and median overall accuracy of 0.90. The proposed workflow provides an interpretable and reproducible framework for multi-event post-fire assessment without supervised training data or prior firedate information. This analysis tool could be particularly useful for stakeholders involved in operational forest and land planning, especially in evaluating post-disturbance vegetation recovery.