Update from the Remote Sensing Project

It has been one and half years since joining the ConFoBi project. I appreciated the opportunity to participate in this big project as an associated PhD student. Although there were many obstacles and challenges at the beginning, it is gratifying that everything is going well so far. The spring of 2021 has arrived, this is a perfect time for a little reflection on what happened in the ConFoBi-remote sensing project. The aim of my project is to map forest type, tree species, and deadwood with machine learning and deep learning algorithms using multi-source remote sensing data. The research will allow the monitoring of forest structures and dynamics at the landscape scale (e.g., forest spatial location and landscape traits, tree species distribution, tree mortality).

It is happy to announce that the first part of the project for land cover/use mapping and forest type extraction with Sentinel-2 data has been completed. Some preliminary conclusions, including the choice of the data, performance of the machine learning algorithms, and the classification accuracy, as well as the final land cover classification map (Figure 1), have been achieved. Meanwhile, the second part of the project on the detection of dead trees is already in progress. By using stereo WV-3 images, we have successfully extracted the dead trees in a specific area (Figure 2). The next step is to apply the proposed method to the entire area to obtain an accurate distribution map of dead trees. In addition, we will establish models to further analyze the impact of environmental and climatic variables (such as temperature, precipitation, soil types, forest management methods, etc.) on the distribution of dead trees, so as to provide reliable advice for successfully implementing forest management measures. As an important component of forest structure, tree species classification will also be carried out in the near future. Two accurate tree species distribution maps at landscape and plot-level are expected to be created using sentinel-2 and UAV images. As such, multiple forest landscape-scale indices and parameters will be structured and stored. I expect possible future further collaborations with other projects, and the derived forest landscape-scale indices could be also beneficial to other projects for forest research at multiple levels.

by Xiang Liu (A1)