Multitemporal dual-pol Sentinel-1 data to support monitoring of forest post-fire dynamics

Abstract

Approaches based on multitemporal analysis of optical-retrieved vegetation index time series were successfully applied to describe forest disturbances like forest fires; conversely, only few works make use of multitemporal Synthetic Aperture Radar (SAR) data. In this work, a multi-temporal approach based on Sentinel-1 data (S1) is proposed based on CR polarimetric index to monitor forest canopy along the considered period (2016–2019) preceding and following an important fire event occurred in the Piemonte Region (NW Italy) in November 2017. The Pettitt test, applied to the polarimetric index time series, was used for testing fire occurrence date and map burned areas (795 ha) resulting in user’s accuracy of burned area equal to 88%. A trend analysis was also conducted, on ‘burned’ pixels only, to describe tree canopy damage and strength of the consequent recovery process at pixel level using linear trend slope values of cross ratio polarimetric index time series. Finally, a k-means cluster analysis was applied to define classes having the same ecological behaviour with respect to two different criteria: one aimed at mapping type and intensity of damage and a second one aimed at describing the ecological behaviour in terms of resistance and resilience of burned patches. In the study area, the cluster layer called forest damage map classifies about the 22% of burned area as characterized by an early high severity whiled the residual by moderate-low severity levels. The second cluster layer called ecological response map defined the 61% of the burned area as resistant forests, the 20% as resilient forests and the 19% as increasing forest zones. All maps were generated with the aim of supporting post-fire assessment and management with free satellite SAR data.