Can Sentinel-2 be used to detect invasive alien trees and shrubs in Savanna and Grassland Biomes?

Abstract

The impact of alien plant invasions on ecosystems translates into major economic losses for nations across the globe. Freely available satellite imagery can be used to provide relevant data for effective management of alien plant invasions, especially for developing nations with limited budgets. However, detecting alien herbaceous and shrubby vegetation using freely available imagery has not been widely tested. The aim of this study was to detect invasive alien plants in the Savanna and Grassland Biomes of the uMngeni Catchment. The novelty of this study is the focus on not only alien trees, but also alien shrubs, which tend to be more sparsely spread and are harder to detect. Using the free Google Earth Engine Platform for repeatability, we explore the efficacy of data fusion of Sentinel-2, Sentinel-1 and landform data, various feature selection techniques and two powerful non-parametric classifiers. We found that feature selection techniques result in lower classification accuracies and are therefore not recommended for similar applications. Data fusion, such as combining Sentinel-2, Sentinel-1 and landform data, did not improve classification results. Support Vector Machine shows superiority over the Random Forest classifier. The final classifications indicate, with 80–90 % accuracy, that for 2019 about 17.5 % (736.15–763.75 km2) of the uMngeni Catchment was occupied by invasive alien trees and shrubs (e.g., gums, wattles, pines and Bugweed), either in the form of plantations or invasions. This translates to between 88.3 million and 173 million m3 of water per year in evapotranspiration losses, or 23 % or the current system yield. At a spatial resolution of 20 × 20 m, these outputs are useful for regional or local planning and prioritization. Furthermore, the repeatable approach of this study is a significant advance compared to past efforts that have produced publicly available datasets for the region, and could be employed by other data-scarce developing nations.