Seabed analysis with Hyperspectral imagery : a sub-pixel approach, taking into consideration precise radiative transfer models of the water column and the bottom. – HypFoM

Hyperspectral remote sensing imagery allows, thanks to the spectral information, to identify on-ground materials. The sea bed analysis involves the additional problem of the water column influence on the photons reaching the bottom, and then observed after reflection and transmission through the water column, with a spectral band limited by the water absorption domain. The water column effect correction involves the physical modelling of the solar radiation flowing into the ocean (i.e. radiative transfer), which depends on the bathymetry and the presence of marine constituents capable to interact with the light rays : suspended matter, organic dissolved colored matter, and detrital matter. Many works have been developed these last years in the oceanic radiative transfer domain in order to propose semi-analytic models taking into account the optical properties of the main constituents. However the joint estimation of the bottom and the water column, despite some works, is still an open problem.
This project owns a main and two secondary objectives.
The main objective is the development of new methods for the sea bed analysis from remote sensing hyperspectral images. We are specifically interested in spectral un-mixing methods, which allow to obtain the proportions of the various materials that fit in each pixel. Here the goal is to jointly determine the pure material spectra, their respective abundances, and the column water composition of an underwater scene. To this end, we must model the water column effects on the mixtures observed in each pixel. We will refine the radiative transfer existing models, in order to include them to the new un-mixing methods. Especially, we will take attention on taking into account the non-linearities inducted by the multiple diffusion generated by the marine particle laying into the water column, onto the spectral mixtures.
A first secondary objective is to obtain maps of water composition, bathymetry and bottom coverage from sites that have been covered (the mouth of the Rhône river, the Côte Bleue, and the Ile of Porquerolles), for the information of the site managers (National Parc of Port Cros).
A second secondary objective is, through simulations, to precise on a experimental base (our airborne data), the right needs in signal to noise ratio, and the impact of the spatial resolution for the coastal zones.
The project resumes as follows :
- From the existing radiative models, implement various inversion methods (semi-supervised, minimum quadratic error, and statistical estimation) in order to have a comparison set. A ground truth campain will be done to form a reference spectral base.
- Take into account the non-linearity effects induced by the multiple diffusion generated by the hydrosols during their interaction with the light into the water column. The radiative transfer modelling will allow to quantify precisely the mechanisms of multiple diffusion into the water column. This part is original in the sense that non-linear effects induced by the hydrosols are not taken into account in the current bathymetry estimation algorithms.
- Incorporate the new radiative transfer model into statistical inversion algorithms that are able to correct the water column effect.
- Develop new methods and new algorithms for spectral un-mixing, consistent with the radiative transfer model which will be incorporated into the un-mixing methods, in order to obtain sub-pixel mapping of the bottom.
- Apply classical methods and the new developed methods onto the available airborne images and compare them. Search for some benthic habitats of interest for marine biology and local collectivities (e.g. Caulerpa Taxifolia).
- Simulate some spatial and signal to noise ratio reductions in order to assess the impact on bottom reflectance and hydrosols optical properties estimation.