Unmixing water and mud: Characterizing diffuse boundaries of subtidal mud banks from individual satellite observations

Mapping of subtidal banks in mud-dominated coastal systems is crucial as they influence not only shoreline and ecosystem dynamics but also economic activities livelihoods local communities. Due to associated spatiotemporal variations suspended particulate matter concentrations, mudbanks are often confined by diffuse rapidly changing boundaries. To avoid inaccurate representations these remote sensing images, it necessary unmix distinctive reflectance signals into representative landcover fractions. Yet, extracting mud fractions, order characterize such boundaries, challenging because the spectral similarity between subtidal- intertidal features. Here we show that an unsupervised decision tree, used derive spatially explicit spectrally coherent image endmembers, facilitates robust linear unmixing on image-to-image basis, enabling separation We found resulting abundance maps represent cross-shore gradients vegetation, water fractions present at coast Suriname. Furthermore, confirmed possible separate land, initial estimate zones individual images. Thus, signatures end-member candidates, determined from relevant index histograms within estimates, consistent. These results demonstrate information well-defined spatial neighbourhoods detection boundaries with a approach.