Mapping estuarine vegetation using satellite imagery: The case of the invasive species Baccharis halimifolia at a Natura 2000 site

Abstract

The invasive shrub Baccharis halimifolia is a threat to the environmental health of many estuarine protected areas throughout Europe. It displaces saltmarsh vegetation and creates monospecific stands that diminish the natural diversity. This work aims to develop a procedure to map this invasive species using satellite imagery. Landsat-8 and Sentinel 2A images are compared, along with three classification approaches (pixel-based, object-based, a mixture of both), to determine which combination yields the best B. halimifolia mapping results. All calculations were made using open-source software, including the ORFEO toolbox for the segmentations in the object-based approach, and the Scikit-learn package for the Support Vector Machines classification algorithm. The pixel-based classifications mapped the invasive species with an accuracy of 70% or higher for both images. The Landsat image had higher accuracy in the overall classification of the vegetation, but the Sentinel image proved better suited for mapping B. halimifolia specifically, due to its higher spatial and spectral resolution. In addition, the procedure was implemented using a Landsat image from 2005, and mapped the invasive species with an ac- curacy of 72% and 88% for producers and users accuracy respectively. The developed procedure represents a valuable tool for restoration projects, allowing for retrospective analyses or relatively low-cost monitoring of B. halimifolia's current distribution.