Identifying the best season for mapping evergreen swamp and mangrove species using leaf-level spectra in an estuarine system in KwaZulu-Natal, South Africa

Abstract

Swamp and mangrove forests are some of the most threatened forest types in the world. In Africa, these forests are essential in providing food, construction material and medicine to people. These forest types have not sufficiently been mapped and changes in the extent or quality of these habitats can therefore not be effectively monitored. Compared to traditional surveying methods, remote sensing can be used to map these inaccessible areas over regional extents. This study investigated which season would provide the best discrimination of six evergreen tree species, associated with swamp (Ficus Trichopoda), mangrove (Avicennia marina, Bruguiera gymnorrhiza, Hibiscus tiliaceus), wetlands in adjacent woodlands (Syzygium cordatum) and coastal floodplain systems (Ficus sycomorus), using leaf-level hyperspectral data. Leaf spectra were collected from 113 trees for the winter, spring, summer and autumn months between the years of 2011-2012 in the subtropical estuarine system of the uMfolozi, uMsunduzi and St Lucia Rivers, on the east coast of KwaZulu-Natal, South Africa. The classification accuracy for each season was evaluated in the WEKA software using the Random Forest classification algorithm. When the data was upscaled to canopy-level, the results showed that all four seasons produced overall accuracies of > 90%. Spring, summer and autumn produced the highest overall accuracy of 94.7%, whereas the overall accuracy for winter was 89.5%. The results of the leaf-level analysis showed a decrease in accuracy of between 4 – 11% for the four seasons. Similar to other studies, our results showed that the simulated object-oriented approach showed a higher level in accuracy compared to the pixel-level approach. The results of this study showed that evergreen tree species around the uMfolozi, uMsunduzi and St Lucia Rivers in KwaZulu-Natal, South Africa, is highly separable over all four seasons. Further analysis will be done to assess whether the accuracies can be improved for certain species, for example Ficus trichopoda. Similar tests should be done on other tropical and subtropical regions of Africa, to assess whether these trends prevail for other species and regions.