Résumé
Growing at the land-sea interface, mangroves are adapted to harsh environmental conditions. This ecotone provides fundamental ecosystem services linked to its double status: forest and coastal ecosystem. However, the extent of mangroves is continuously declining across the globe. Since 2000, there has been increased awareness of the essential role (coastal protection, biodiversity reservoir, provisioning services) played by mangroves, and remote sensing has proven essential for monitoring and mapping these vast, often inaccessible ecosystems. Although various global products on the distribution of mangroves are available at medium resolution, the assessment of their extent (essential for management strategies) is not completely consensual, as recent publications show. Moreover, despite their importance in various topics such as biomass and carbon storage assessment, data, methods and products on mangrove structure at high resolution are scarce. This highlights the importance of accurate, reproducible mapping methods for assessing the area, structure and dynamics of mangroves. In this context, we propose a new methodological framework based on open-source multi-sensor and multi-resolution remote sensing combined to in-situ measurements to map mangroves types at fine scale. On the pilot site of Bombetoka Estuary, Madagascar, this study focuses on building a standardized approach for the processing of Pléiades Very High Spatial Resolution (VHRS) images (50 cm resolution, available through the DINAMIS program) to characterize the structural types of mangroves based on texture information. A map of mangrove extent derived from GMW is used as a mask to apply the unsupervised FOTO (FOurier transform Textural Ordination) method in order to analyze the texture of the mangrove canopy on Pléiades images. Unsupervised Kmeans classification is then applied to extract mangrove classes based on textural information. In-situ measurements of structural variables of mangrove (tree density, tree diameter, tree height, development stage, tree health) from field campaigns carried out in May 2023 and March 2024 were finally used to construct and label the structural typology of the mangroves of Bombetoka Estuary. We were able to extract 5 classes of mangrove structural types based on texture and in-situ variables. Finally, these structural classes were used as training set for the Iota2 method to produce a map of mangrove types at the scale of Madagascar at a resolution of 10 m from Sentinel images.This approach, based on free data and tools currently being made available to the community, is intended to be replicated on other sites in order to assess the reproducibility and genericity of the method.