Application of textural analysis to map the sea ice concentration with sentinel 1A in the western region of the Antarctic Peninsula

This article proposes the mapping of the concentration of sea ice in the oceanic region of the western Antarctic Peninsula using backscatter coefficients and textural features from synthetic aperture radar (SAR) C-band, horizontal single polarization (HH) images obtained by the Sentinel 1 A satellite sensor. Pure samples of open water and sea ice were obtained from SAR images with the aid of PlanetScope optical images during the austral winter. The statistical technique of logistic regression was applied to identify the best textural features to discriminate the two sampled targets. These selected textural features were used with the backscatter coefficients of SAR images in a maximum likelihood supervised classifier to analyze the images. With sea ice classified in the time series between 2016 and 2019 during winter and austral spring, the concentration of sea ice in the study region was calculated and mapped. The textural features statistically indicated by the logistic regression were gray level co-occurrence matrix (GLCM) Mean and GLCM Variance, resulting in the supervised classification having an accuracy of 86% and precision ranging from 86% to 98% in the validation stage