Mapping the Mangrove Forest Canopy Using Spectral Unmixing of Very High Spatial Resolution Satellite Images
Résumé
Despite the lowtree diversity and scarcity of the understory vegetation, the highmorphological
plasticity of mangrove trees induces, at the stand level, a very large variability of forest structures
that need to be mapped for assessing the functioning of such complex ecosystems. Fully constrained
linear spectral unmixing (FCLSU) of very high spatial resolution (VHSR) multispectral images was
tested to fine-scale map mangrove zonations in terms of horizontal variation of forest structure.
The study was carried out on three Pleiades-1A satellite images covering French island territories
located in the Atlantic, Indian, and Pacific Oceans, namely Guadeloupe, Mayotte, and New Caledonia
archipelagos. In each image, FCLSU was trained from the delineation of areas exclusively related to
four components including either pure vegetation, soil (ferns included), water, or shadows. It was then
applied to the whole mangrove cover imaged for each island and yielded the respective contributions
of those four components for each image pixel. On the forest stand scale, the results interestingly
indicated a close correlation between FCLSU-derived vegetation fractions and canopy closure
estimated from hemispherical photographs (R2 = 0.95) and a weak relation with the Normalized
Difference Vegetation Index (R2 = 0.29). Classification of these fractions also offered the opportunity
to detect and map horizontal patterns of mangrove structure in a given site. K-means classifications
of fraction indeed showed a global view of mangrove structure organization in the three sites,
complementary to the outputs obtained from spectral data analysis. Our findings suggest that the
pixel intensity decomposition applied to VHSR multispectral satellite images can be a simple but
valuable approach for (i) mangrove canopy monitoring and (ii) mangrove forest structure analysis
in the perspective of assessing mangrove dynamics and productivity. As with Lidar-based surveys,
these potential new mapping capabilities deserve further physically based interpretation of sunlight
scattering mechanisms within forest canopy.
Origine : Fichiers éditeurs autorisés sur une archive ouverte