Estimation of Floating Plastic Debris Surface in Inland Waters using Spectral Unmixing with Multispectral Data

Abstract

Unlike hard classification from medium-resolution sensors, spectral unmixing at sub-pixel level offers improved accuracy in estimating the total surface occupied by a material of interest within a given area. While in ideal cases imaging spectrometer data should be utilized for this purpose, we propose to use a limited number of fixed classes in order to perform spectral unmixing from multispectral data to address the specific challenge of estimating the surface area covered by floating plastic debris in inland waters. In that context, working with multispectral data is motivated by extended opportunities to identify and monitor narrow water channels for variable plastic appearances, in terms of extended coverage, spatial and temporal resolution.