Mapping Dominant Forest Tree Species in a highly heterogenous Ghanaian Tropical Forest using Multispectral Remote Sensing Imagery.
Information about tree species distribution in the forest ecosystem is valuable for implementing sustainable forest management. Forest ecology and management practices as well as forest companies require the use of tree species information for operational and tactical resource planning. For instance, higher achievement may be obtained by having tree species information before silvicultural treatments. Furthermore, tree species composition is an indicator of biodiversity, which is vital in forest ecosystem management (Persson & Lindberg, 2018).Tropical forests serve as a host to the largest biodiversity of terrestrial ecosystems and play a fundamental role as far as the carbon cycle is concerned (Vaglio Laurin et al., 2016). Improving the monitoring of tropical forests is an important research issue for implementing climate change-related agreements toward biodiversity conservation.Tree species identification and mapping can be achieved through fieldwork campaigns. However, generally, this practice has some limitations since it is expensive and laborious because of the density of the forest and forest access. Remote sensing, together with automated analysis techniques, has therefore become a prominent approach for species mapping.
This project therefore aims to develop a remote sensing approach suitable for mapping the most dominant tree species in Ghana’s highly heterogenous Tropical Forest. The following specific objectives will be addressed. (1) Evaluate the patterns of forest species distribution and population abundance in the Pra AnumForest Reserve, Ghana. (2) Evaluate the spectral difference among the tree species and (3) Assess the performance of different classification approaches (random forest and support vector machine in a highly heterogeneous tropical forest (Pra Anum Forest Reserve, Ghana).
This study will be carried out in the Pra-Anum Forest Reserve in Ghana which fassls within he moist semideciduous vegetation zone with about 230 tree species, of which 67 are timber species. The forest is dominated by tree species such as Monodora myristica, Hymenostigia afzelli, Nesogodonia papaverifera, Triplochiton scleroxylon, Celtis zenkeri, Funtumia elastica, Ceiba pentandra, among others.
This study will use sentinel-2 and planet imagery to map forest species distribution and population abundance. Spectral features will be extracted from the satellite imagery, both sentinel-2 and planet datasets. The random forest approach will be used for tree species identification using spectral, textural, and vegetation index information. The resulting classification result will go through an accuracy assessment procedure in which overall accuracy, kappa coefficient, and produced and user accuracies will be calculated for the classes (Raczko & Zagajewski, 2017).
To model the patterns of forest species distribution and population abundance in the Pra-Anum forest ecosystem, the response variables will be the mapped tree presence data. This will be regressed with the bioclimatic variables from WorldClim, the topographic feature of land surface derived from STRM digital elevation data, and the vegetation structure derived from TanDem-X as predictor variables.
