Mapping of Monodominant Gilbertiodendron dewevrei Forest Across the Western Congo Basin Using Sentinel-2 Imagery

Tropical rainforests are complex mosaics of different forests types, each with its own biodiversity and structure. Efforts to characterize and map diversity and composition of tropical forests are vital at both local and larger scales in order to improve conservation strategies and accurately monitor anthropogenic threats. However, despite advances in remote sensing, classifying and mapping forest types remains a significant challenge and remotely sensed classifications in the tropics often treat forests as a single category. Here, we used Sentinel-2 data, and a high-quality ground reference dataset, to map monodominant Gilbertiodendron dewevrei forest, a unique forest type in central Africa. We used a random forest classifier, and spectral, vegetation, and textural indices, to map G. dewevrei forest across the Sangha Trinational, a network of national parks in central Africa. The overall accuracy of our classification was 83% when evaluated against an independently sampled reference test dataset, successfully distinguishing this monodominant forest from the spectrally similar terre firme mixed forest present throughout much of the study area. The gray level co-occurrence matrix (GLCM) textural metrics proved the most important factors for distinguishing G. dewevrei forest, due to the homogenous canopy texture created by this monodominant species. In conclusion, our study illustrates that freely available Sentinel-2 data hold promise for mapping distinct forest types in tropical forests, particularly when they exhibit structural and textural differences, as seen in monodominant and mixed forests, and provided that high-quality ground reference data are available.