Problems of multivariate inversion for complex cas-2 waters on the example of MERIS-data over the Baltic Sea

Abstract

The use of remote sensing over water regions has the advantage getting frequent information about large areas, and therefore reducing the need of cost intensive ship campaigns. Nowadays it is an important contribution to the regular assessment of the ecological state of our environment. Remote sensing of water areas in the optical range of the electromagnetic spectrum is possible due to the absorption window of the atmospheric and water molecules and the specific scattering and absorption properties of the water constituents. In the open ocean one finds one varying component mainly defining the state of the water, the chlorophyll concentration. Due to the one-component-system the interpretation of the measured spectrum is comparatively simple. After the removal of atmospheric effects color ratios or similar indices can be applied to estimate the chlorophyll concentration with good accuracy. In optically more complex waters, e.g. coastal zones, characterized by more than one component, therefore more complex interpretation algorithms must be applied. The problem leads to a multivariate estimation of the concentration vector from the multispectral measurement. Thus case-2 waters are a challenge for ocean color remote sensing for higher spectral resolution measurements and multivariate inversion techniques. The presentation will focus on model based inversion algorithms and discuss problems related to them, the choice of the specific optical model, radiative transfer calculation and inversion technique. Neural networks are a widespread method for the multispectral interpretation. The presentation will concentrate on an alternate but closely related method, using principal component transformation as a main tool for information extraction. It will be shown the relation to neural networks with its advantages, disadvantages and potential. Results will be demonstrated together with the specific optical model on the example of MERIS-data over the Baltic Sea.