Evaluating the potential of Sentinel-2 and landsat image time series for detecting selective logging in the Amazon

Logging in the Amazon is primarily selective, with individual commercial trees cut down, leaving the others untouched. The direct effects of selective logging are difficult to quantify using remote sensing techniques due to the small extent of treefall gaps left after logging, and the quick canopy regrowth that follows. In this thesis, the time series of five vegetation indices (NDVI, NDMI, NBR, EVI, MSAVI) derived from Landsat 7 ETM+, Landsat 8 OLI and Sentinel-2 MSI imagery were evaluated for detecting selective logging features in three study sites in the Amazon. Logging roads and log decks could be identified both manually and automatically using imagery from any of the sensors, given enough history for building a stable seasonal model. Skid trails could not be identified using any sensors. Treefall gaps could not be reliably identified using Landsat imagery, but 43% of treefall gaps in known treefall locations could be manually identified in Sentinel-2 MSI imagery. From the vegetation indices, NDMI was the most sensitive to reflectance changes in treefall gaps after logging, followed by NBR, due to their sensitivity to changes in forest internal shadowing. NDVI was sensitive only to soil and non-photosynthetic vegetation uncovered after logging. Changes in EVI and MSAVI reflected changes in both internal shadows and uncovered soil and non-photosynthetic vegetation, but the magnitude of change was the lowest of all indices tested. This study shows that it is possible to detect treefall gaps in Sentinel-2 imagery, and potentially automate it in the future, so that estimates of selective logging volumes could be based on direct observations of treefall gaps, rather than assuming a correlation between roads or log decks and the volume of timber harvested.

---

Download:

• 
  English-Full (4 MB)