One of the significant benefits of using satellite imagery for forest resource monitoring is the consistent availability of recently captured data. This is especially true of the Sentinel-2 platform, which has a revisit interval of five days, allowing multiple observations per month.
Indufor’s harvest tracking process uses such satellite imagery to identify harvest activities. The detection process is iterative; first harvest extents are mapped by separating forest and non-forest using image segmentation and classification techniques. Subsequent classifications are then compared to this base classification, with those areas that have changed to non-forest tagged as harvested.
As new imagery becomes available, harvested areas are confirmed against successive observations to reduce the likelihood of misclassification occurring. At the same time, any new harvest areas are also detected and mapped.
The approach developed allows progressive tracking of harvesting activities and produces outputs that can be displayed via a project dashboard or provided as shapefile boundaries. For this example, we use the refreshed NPI dataset (read about that here) and track forest harvesting over 2019. Over this period 11 of 12 months had a cloud-free Sentinel-2 scene, (where multiple scenes were available for a month the later capture was used). In total, 680 ha of harvesting was detected; of this total, 70% occurred within three months.
The following interactive map provides an overview of the harvest tracker’s output, which includes the harvest extent, and harvest area by month. By clicking on a month within the bar chart zooms to the area detected as harvest and loads the Sentinel-2 scene used to generate that result.