Nonlinear estimation of material abundances in oil-polluted sea-ice images with low background concentrations

The increased oil-spill possibilities in sea-ice areas pose a threat to the marine environment and the ice-area coverage. Contrary to being widely used in monitoring oil spills in open waters, quantitative analyses of oil-spill hyperspectral information in ice-covered sea areas are inadequate. Due to multiple propagating processes, small oil droplets tend to form and mix into ice, resulting in low background oil concentration. For a complex nonlinear mixture, accurately estimating the material abundances remains a challenging yet fundamental task, particularly with extensive detailed and low-energy information. Herein, an abundance estimation method with normalizations in a combined polynomial and sine model is proposed to resolve the over-fitting problems. An energy information–based wavelet package scheme effectively highlights the latent information of the concerned materials. The oil-contaminated sea-ice images were generated with spectral reflectance collected outdoors on sea ice in Bohai Bay, China, in the form of pure as well as mixed spectral signatures. Experimental analyses suggest that the proposed method exhibits superior unmixing performance, particularly in delivering more accurate abundance estimations of different oil concentration levels as low as 10−4 than classic unmixing methods, even in low signal-to-noise data, and can be used to respond to oil spills at high latitudes.