Using Landsat, Sentinel-1 and -2 and GEE to assess changes in wetland Ecosystem Functional Groups in the Maputaland Coastal Plain of South Africa

Abstract

Consistent monitoring of wetland Ecosystem Functional Groups (EFGs) over large areas and long periods remains challenging owing to uneven availability, varying spatial resolutions, and spectral similarity of some EFGs. A Google Earth Engine (GEE)-based Earth Observation workflow was developed to map and quantify changes in wetland EFGs across the Maputaland Coastal Plain (MCP) in South Africa. The study included Landsat imagery for 1990, 2000, 2006 and 2014; and Sentinel-1 and Sentinel-2 imagery for 2018, 2020 and 2022. The methods involved: • Data collection of historical desktop datasets and in-field validation of six wetland EFG types; • Multiple spectral indices derived from the images were combined with topographical variables and elevation data derived from a Digital Terrain Model, as well as desktop and in-field-collected data points in a classification algorithm, • Recursive feature elimination was used to select the most informative predictors, and • The use of a Random Forest machine learning algorithm and recursive feature elimination for the classifications in GEE. Changes in EFGs were quantified between 1990 and 2022. The classification algorithm can be used recursively, while only the input point datasets should be iteratively validated using fine-scale spatial-resolution data, such as aerial or orthophotos, or in-field validation. The methods improved the delineation of the six wetland EFGs compared to other datasets and contributed to the Red List Assessments of the EFGs.