Mapping Invasive Plant Species Using Very High-Resolution Imagery and Machine Learning in Lesotho Wetlands

Invasive plant species (IPS) are considered the second largest threat to biodiversity, globally (Dueñas et al. in Biodivers Conserv 27:3171–3183, 2018). Wetlands in Lesotho are such a precious ecosystem under IPS threat. IPS populations exhibit unique characteristics that allow them to outcompete the native species for nutrients, light, water, and space to nourish. Information about the distribution and abundance of the invasive populations is thus very important in assessing the impact of the spread in wetlands. In 2021, the Food and Agriculture Organization of the UN (FAO) used SkySat imagery together with in-situ data to map of presence of IPS in the wetlands of six priority sub-catchments in Lesotho. The SkySat images were initially preprocessed, then vegetation indices were derived and used as input features into the Random Forest (RF) model. The workbench was implemented using Google Earth Engine. Results of the analysis showed an abundance of invasive/native species in wetlands of 37% and 64% respectively. The overall accuracy of the results scored 0.73%. In this work we demonstrate the feasibility for invasive species monitoring using EO data and cloud computing.