Oil spill thickness estimation using unmixing methods

The paper presents a new method for estimating oil spill thickness using hyperspectral data. The method relies on the abundance fractions provided by spectral unmixing methods. Given that the materials which compose the spectral signature of pixels presenting oil spills are oil and water, correlation functions between abundance fractions of these endmembers and thickness of oil spills were established using artificial images. The artificial images were created using laboratory measurements of oil spills. Thirteen different types of oil were used for the production of the relevant artificial images, each one presenting oil spread on water surface with six different levels of thickness. Unmixing was performed with the Fully constrained Network Based Method (F-NBM), Fully Constrained Least Square method (FCLS), Orthogonal Subspace Projection (OSP) and Constrained Energy Minimization (CEM). The unmixing results were evaluated using image reconstruction techniques. F-NBM produced the most reliable abundances. Logarithmic equations provided the most reliable oil thickness estimations for the examined oil types. Oil thickness can satisfactorily be estimated using the abundance value of water, without requiring the knowledge of the oil type.