Exploring the Potential of Sentinel-2 Broadband Spectral Features for Discrimination of Invasive Lantana Camara L. Among Co-Occurring Species
Loading...
Date
2025-07
Journal Title
Journal ISSN
Volume Title
Publisher
South African Journal of Geomatics
Abstract
Lantana camara L. (LC), an alien invasive plant species, negatively impacts natural habitats globally. In East Africa, it is one of the habitat’s transforming species that requires urgent mapping to support conservation actions. Its spatial distribution has not been adequately established, especially within forest habitats. In this study, we sought to identify Sentinel-2 broadband spectral features that could discriminate LC from co-occurring vegetation in a forest habitat. In-situ leaf-level hyperspectral measurements of LC and co-occurring species, namely, Neonotonia wightii (NW), Cucumis maderaspatanus (CM) and Ocimum gratissimum (OG), were collected in a one-hectare site in the Muringato forest area during the dry and wet seasons using two handheld spectroradiometers covering wavelength ranges of between 340–820 nm and 635–1100 nm. The leaf-level reflectance measurements were used to simulate Sentinel-2 wavebands, which were subsequently used to compute new band combination indices of type Normalized Difference (ND), Simple Ratio (SR), Difference (D) and Inverse Difference (ID) and published Sentinel-2 multispectral indices. The most spectrally significant features were selected using the Boruta and the Guided Regularized Random Forest (GRRF) methods. Jeffries– Matusita (JM) distance analysis was used to quantify the spectral separability of species class pairs using selected spectral features. The findings of this study showed that the selected Sentinel-2 spectral features that produced perfect separability accuracies of ≥97% for LC class pairs consisted of mostly the newly developed band combination indices and, to a lesser extent, the published Sentinel-2 indices. Notably, the separability analysis produced a unique set of spectral variables that accentuated the spectral properties of LC class pairs in both seasons and further pointed to the influence of the seasonal spectral variabilities of the species. Moreover, the Boruta method resulted in the selection of fewer spectral variables (2–12 variables) than the GRRF method, which resulted in the selection of two to19 variables for species class pair separability. Overall, the separability results demonstrate the potential of separating LC from other vegetation with freely available Sentinel-2 image data. The significant spectral variables identified in this study could be used for the seasonal mapping of LC and for further aid in the early detection and targeted management of the species in affected forest habitats
Description
Article
Keywords
Citation
Waititu, J. M. (2025). Exploring the potential of Sentinel-2 broadband spectral features for discrimination of invasive Lantana camara L. among co-occurring species. South African Journal of Geomatics, 14(2).