Extraction of Bohai Sea ice from MODIS data based on multi-constraint endmembers and linear spectral unmixing

Abstract

ABSTRACT A Linear Spectral Unmixing method based on Multi-Constraint Endmembers (LSU-MCE) is proposed to extract the sea ice in the Bohai Sea using the Moderate Resolution Imaging Spectroradiometer (MODIS) images. The method first extracts the spectral features, texture features, shape features, and temperature features of the sea ice and sea water, and then constructs the rules using the features to purify the endmembers, and finally decomposes the pixels using the linear spectral unmixing method. The method can properly detect sea ice including thin ice, drift ice, and melt ponds. Besides, the turbid sea water is separated from sea water as a separate endmember. So the proposed method can eliminate the disturbance caused by suspended sediment and successfully detect sea ice in high sediment waters. The method has good robustness to MODIS images of different satellites (Terra and Aqua) and the stripe noise. The method has been successfully applied to time series data of MODIS and the results show that it has high accuracy for sea ice detection in the Bohai Sea, with better visual results, and higher overall accuracy and kappa coefficient ().