Floe Size Mapping from Satellite SAR Images and Icewatch Observations in the Beaufort Sea during Autumn 2015
Abstract
An approach for automatic detection of the sea ice type in the MIZ from RADARSAT-2 SAR images with HH polarization and resolution of50 m has been developed and tested. The approach is based on texture analysis using the GLCM (Gray-Level Co-occurrence Matrix) method and several additional functions based on the estimates of the averaged gradient tensor. A machine learning technique (Support Vector Machine, or SVM) is applied to imagery of ice taken for the region of the Beaufort Sea in autumn 2015, with observations of ice type from two ship cruises used as ground truth. It is found that this method shows promise, but the training requires more collocations than is practical at present-specifically, the ubiquitous inhomogeneity of ice presents a challenge for colocation, as it limits the training set.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 26, 2019
- Accession Number
- AD1078812
Entities
People
- Erick Erick Rogers
- Gleb G. Panteleev
- Hui Shen
- Julia Crout
- Luc Rainville
- Max I. Yaremchuk
- Tamara Townsend
Organizations
- United States Naval Research Laboratory