Classification and Deforestation Monitoring Using Sentinel-1 C-SAR Images in a Temperate Exploited Pine Forest
Abstract
Earth Observation data is often used for land cover classification or change monitoring. It is rarely used for both goals in a single algorithm. The multi-change Cumulative Sum (CuSum) algorithm proposed in this study allows both classification and change monitoring in a single algorithm using Sentinel-1 C-SAR time series. The multi-change CuSum approach allowed to classify pixels belonging to the fused non-forest vegetation and bare soil classes apart from the pixels belonging to new cuts. The distinction of each class is better made using the two polarizations: VV is more accurate for detecting non-forest vegetation (Kappa coefficient of 0.62) and VH for detecting new cuts (Kappa coefficient of 0.65). The algorithm showed an accuracy up to 0.82.