Mapping microphytobenthos biomass by non-linear inversion of visible-infrared hyperspectral images

Abstract

This study presents an innovative approach to map microphytobenthos biomass and fractional cover in Bourgneuf Bay (French Atlantic coast) using Digital Airborne Imaging Spectrometer (DAIS) hyperspectral data. Microphytobenthos is a microalgae forming a biofilm on the mudflat. Its spatial distribution is heterogeneous so it varies on a finer scale than that of airborne instrument spatial resolution, leading to a “mixed pixel” problem. Moreover, some microphytobenthic species form, at low tide, a biofilm deposited at the surface of the sediment substrate. The resulting signal is a highly non-linear combination of spectral endmembers due to microscale intimate mixtures. This prevents the use of classical linear unmixing models to retrieve biomass from reflectance spectra. A Modified Gaussian Model (MGM) is therefore used to remove the effects of surface roughness, shadowing and any other unknown processes that contribute to the overall shape (continuum) of the reflectance spectra. Then, relationships between microphytobenthos biomass and spectral shapes are derived from a spectral database compiled from laboratory reflectance spectra of microalgal monospecific cultures with different biomasses. Finally, microphytobenthos biomass and fractional cover are retrieved from the DAIS image by comparing the reflectance spectra of each pixel to a library of synthetic spectra corresponding to combinations of various biomasses and substrate percent cover. This new approach, when compared to more classically used ones such as indices, linear unmixing or spectral distance analysis, is proven to enable a much more reliable determination of biomass despite the large variety of substrates found in Bourgneuf Bay.