SENTINEL-1 DATA TIME SERIES TO SUPPORT FOREST POLICE IN HARVESTINGS DETECTION

Abstract

Satellite remote sensing has long been used to monitor forest harvesting with accuracies appropriate for practical mapping across a wide range of forest types by using different sensors. Unfortunately, in Italy, most of the cuts take place in winter where the cloud cover is very high, making it impossible an early detection by optical data. In this framework, synthetic aperture radar (SAR) data such as Sentinel-1 (S1) allows a better land monitoring by penetrating cloud cover. In this work we tested some methods for time series breakpoint detection with the aim of mapping significant forest cover changes in 2019 over an Italian forested area. These maps can be useful tools to support the focusing of field surveys by forest police with the aim of increasing the monitorable areas and decreasing the related field survey costs. Four methods were proposed and compared based on the analysis of SAR polarimetric index time series (Cross Ratio index). In particular, adopted methods search for a breakpoint in the cross-ratio time series assuming it as moment after that forest canopy temporal behaviour significantly change. In general, high overall accuracy and user’s accuracy were found for all methods while producer’s accuracy and K values are lower denoting an underestimation of harvested areas by single method. Conversely, combining all methods into a final classification shows highest user’s accuracy (> 0.9) in detecting forests harvestings when more than two classification methods were adopted.