Using satellite images and GEE to monitor post-wildfire forest vegetation response in Monte Serra (Tuscany, Italy)

Wildfires are a cause of forest disturbances, disrupting the structure, composition, and function of forest ecosystems, and changing resource availability or the physical environment at any spatial or temporal scale. Quantifying and characterizing post-disturbance forest dynamics is important to understand how a forest is recuperating over the years in relation both to different levels of undergone damage, the previous health, its resilience, and the forest management interventions carried out after the event. Among the remotely sensed products, time series of satellite images allow obtaining detailed information of vegetation changes and responses after a forest fire by recognizing the spectral signal that forests have in all phenological stages. The study here presented was carried out to characterize and assess the responses of forests, composed of oaks ( ), stone pine ( L.), European chestnut ( Mill.), to a large fire that occurred on 25th September 2018 in 600 ha of Monte Serra (Pisa province, Tuscany region, Italy). We used data from Sentinel-2 Multi-Spectral imager (MSI), characterized by a good trade-off in spatiotemporal resolution (10 m pixel size for Red-Green-Blue and NIR bands, 20 m for SWIR21, and a 5-7 day revisiting). Using a natural colour RGB composition pre-event image, we classified the vegetation into 6 classes and the area affected by the fire into 4 classes of and one after the event. We derived an array of images by a process on the Google Earth Engine (GEE) cloud computing platform for the area of interest with an extension of 4 years, being the period pre-fire January-September 2018 and that post-fire October 2018-December 2021. To obtain a continuous trajectory, starting from the measured NDVI and NBR values, a linear interpolator has been developed and applied to each pixel along the time dimension. Then, from the produced array of images, for every single pixel and for each day of the time series, we calculated the values of the Normalized Difference Vegetation Index (NDVI) and NBR. Applying statistical functions, such as average, median, maximum, minimum, quartile, per each class of vegetation and fire damage, we derived the distribution of the vegetational indices both according to the spatial and temporal dimension (i.e., resampling by month, season, year). Finally, to evaluate the diversity in vegetation responses, we assessed the difference between the distribution curves of the two vegetation indices by measuring their similarity (Bhattacharyya index). The results achieved in this study suggest that the application of the approach here proposed, based on the reconstruction of a continuous trajectory of vegetation indices from Sentinel-2 imagery, represents a valuable way to assess and monitor the ecological responses of forest vegetation to abiotic disturbances.