Detecting and mapping invasive Populus alba species in mountainous ecosystems using Sentinel-2 imagery and random forest classification

Abstract

Mapping invasive alien plants (IAPs) has become essential for land and biodiversity conservation authorities, as these species can transform the areas they invade. Fortunately, advances in remote sensing using publicly available products such as Sentinel-2 have improved this process, especially in hard-to-access mountainous regions. In South Africa, poplar (Populus alba) is among the IAPs of concern and is found in the eastern Free State and elsewhere in the country, but remote sensing has not yet been used to map this species. Using Sentinel-2 imagery and the random forest (RF) algorithm, this study allowed us to: (a) map and distinguish poplar trees from other land covers throughout the year in the eastern Free State’s mountainous region, (b) evaluate influential bands and their combinations in classification, and (c) assess the accuracy of the classification for the first and second halves of the year. The results showed that images from the first half of the year (January–June) had higher classification accuracy (overall accuracy [OA] = 91% and kappa = 0.89) than those from the second half (Jul–Dec) (OA = 87% and kappa = 0.84). Poplar and other classes were separable, with poplar mostly found in riparian areas. The study identified variables such as short-wave infrared-1 (SWIR-1), normalized difference vegetation index (NDVI), blue, poplar detection index-1 (PI-1), modified normalized difference water index (MNDWI), near-infrared (NIR), and PI-3 as key parameters for classifying poplar trees in mountainous regions. Overall, our findings demonstrate that Sentinel-2 bands and indices combined with an RF classifier provide an effective method for mapping poplar invasive trees in mountainous ecosystems.