Using semi-automated classification algorithms in the context of an ecosystem service assessment applied to a temperate atlantic estuary

• Published in: Remote sensing applications Vol. 36; p. 101306
• Main Authors: Afonso, F., Ponte Lira, C., Austen, M.C., Broszeit, S., Melo, R., Nogueira Mendes, R., Salgado, R., Brito, A.C.
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.11.2024
• Subjects: anthropogenic activities; class; coasts; decision making; ecosystem services; ecosystems; environment; environmental management; estuaries; Habitat mapping; habitats; image analysis; Planet scope imagery; Remote sensing; salt marshes; sand; satellites; social welfare; Support vector machines; vegetation; Wetlands; Support vector machines; Habitat mapping; Planet scope imagery; Remote sensing; Wetlands
• ISSN: 2352-9385; 2352-9385
• DOI: 10.1016/j.rsase.2024.101306

The growing anthropogenic pressure near estuarine areas is evidence of the relevance of these systems to human well-being, especially because of their delivery of essential ecosystem services and benefits. Estuaries are composed of a rich large selection of habitats frequently organised in complex patterns. Mapping and further understanding of these habitats can contribute significantly to environmental management and conservation. The main goal of this study was to integrate different data sources to perform a supervised image classification, using remote-sensing products with different spatial resolutions and features. It was focused on the Sado Estuary, located on the Portuguese Atlantic coast. Considering the limitation of using free satellite images to map estuary habitats (i.e. limited spectral range and spatial resolution), this study uses a semi-automated supervised and pixel-based classification to overcome some of the derived classification problems. Support Vector Machine classifier was used to map the estuary for future evaluation of ecosystem services provided by each habitat. High-resolution remote sensing data (i.e., Planet Scope satellite images, aerial photographs) with different spectral and spatial features (3 m and 20 cm resolution, respectively) were used with ground truthing data to train the classifier and validate the derived maps. The first step of the classification identified broader classes of habitats in the satellite images based on visual interpretation of ground-truth data. From this output, aerial images were classified into detailed classes, the same procedure was hindered on the satellite images due to spatial resolution constraints. The sand class had the best overall accuracy (96%), due to its contrasts with surrounding objects. While the vegetation (i.e., pioneer saltmarshes) and algae classes had lower accuracy values (49.6–89.0%), possibly due to being still damp or covered in fine sediment This is a common challenge in transitional systems across land-water interfaces, such as wetlands, where the abiotic conditions (e.g. solar exposure, tides) fluctuate heterogeneously over time and space. The findings presented herein revealed the considerable success of this approach. For the purpose of local decision-making, these are relevant outputs that can be replicated in other regions worldwide. •Habitat mapping underpins natural capital and ecosystem services assessment.•Two maps were created with distinct spatial resolutions (3 m and 20 cm per pixel).•The high-resolution sand class gave excellent accuracy values (96%).•Algal communities and darker vegetation are harder to map accurately (49.6 – 89.0%).