Detailed land cover mapping in a seasonally dry tropical forest landscape using multiple sensor types

Adrian Dwiputra
Detailed mapping of land cover is essential for supporting science-based sustainable landscape management. Despite the importance of land cover mapping in monitoring landscapes dynamics, land cover data are not always available. Even when the land cover data are available, they often lack detailed discrimination between forest types and plantations. This issue was found in a seasonally dry tropical forest landscape in Siem Reap and Preah Vihear, Cambodia. In this thesis, I explored the
more » ... of (1) the fusion of optical and radar data in developing detailed land cover maps and revealing the driver of landscape change, and (2) vertical vegetation structure acquired by the Global Ecosystem Dynamics Investigation (GEDI) mission—a new mission that harnesses a space-borne waveform lidar sensor installed on the International Space Station—to improve the vegetation mapping in the studied landscape. The fusion between radar (Sentinel-1) and optical (Sentinel-2) satellite data slightly improved the land cover classification accuracy (1.6% overall accuracy increase) relative to Sentinel-2-only classification. Between 2015 and 2019, I detected a 247,781.04 Ha dry deciduous forest loss; most were due to logging (147,314 Ha). Land designations, such as the protected areas and the economic land concessions (ELCs), significantly determine land cover change. The classification of vegetation types using GEDI data had 81.9% overall accuracy despite the limited spatial coverage of GEDI data. The GEDI-only classification results could identify the seasonally inundated forests with better accuracy than the land cover map derived from the fusion of optical-radar data. These results demonstrate the potential of structural information acquired by Sentinel-1 and GEDI to improve our ability to identify vegetation types in complex, heterogeneous landscapes.
doi:10.14288/1.0402360 fatcat:cyu5dfigcnefzoj3ft6s6jjgzy