it_team4_FOREst 5 January 2024It4forest enthusiasts contributed to a spatial resource assessent including innovative machine learning remote sensing classification, field mapping and ground truthing in Namibia. The large scale regional resource assessment of Manketti trees in the Miombo Forest of Namibia was conducted in the frame of the BioInnovation Africa project in Namibia, in collaboration with the University of Applied Sciences for Sustainable Development Eberswalde and local Namibian partners such as the Namibian Biological Resource Institute (NBRI). The main objective of the study was to identify and map the abundance of Manketti trees, assess their viability and economic potential, and provide spatial data on their distribution and density.
The results of this study will be used to evaluate the accuracy of remote sensing techniques and the predictive quality of the Manketti model, and to compare the efficiency and effectiveness of remote sensing with field-based methods. Therefore, a remote sensing image classification study was conducted in northern Namibia in 2022, using a U-Net CNN classifier trained on high-resolution satellite imagery. The results showed that the approach performed well in open savannah-like forest areas, but had lower accuracy in areas with overlapping trees. A ground verification campaign was
conducted to support the prediction of Manketti tree locations and the mapping of detected trees.
The study found that the CNN algorithm was an effective method for automatically classifying and detecting trees, and was fast enough to be applied to entire countries. The results of the study suggested areas for improvement, such as investigating image normalisation, adding more spectral bands to the training dataset, and continuing with further surveys and resource assessments.