B. Scheuchl

Potential of RADARSAT-2 data for operational sea ice monitoring

Synthetic aperture radar (SAR) data from RADARSAT-1 are an important operational data source for several ice centres around the world. Whereas RADARSAT-1 is only capable of acquiring data at a single polarization, RADARSAT-2 will be capable of acquiring dual-polarization data in many wide-swath modes (most importantly ScanSAR) and fully polarimetric SAR data in a narrow 25-km swath mode. In this paper, we consider the ice information requirements for operational sea ice monitoring at the Canadian Ice Service and the potential for RADARSAT-2 to meet those requirements. Primary parameters are ice-edge location, ice concentration, and stage of development; secondary parameters include leads, ice thickness, ice topography and roughness, ice decay, and snow properties. Iceberg detection is included as an additional ice information requirement. The dual and fully polarimetric modes of RADARSAT-2 are expected to enhance our ability to measure these parameters. For ice operations, the dual-polarization data are expected to be most useful, as they will provide the required wide coverage in a number of modes. Although ScanSAR is the recommended mode of operation, the properties of other modes are also discussed. To illustrate the expected improvements from polarimetry, we review the conclusions of past work and add some new results using ENVISAT ASAR and simulated RADARSAT-2 data.

Automated sea ice classification using spaceborne polarimetric SAR data

This paper discusses the capability of spaceborne polarimetric C-band SAR data for sea ice detection and classification. Unsupervised classification using polarimetric decomposition and the complex Wishart classifier was performed on SIR-C data acquired off the coast of Newfoundland in April 1994. The algorithm is used for sea ice applications for the first time, and appears promising. In addition to polarimetric classification, three of the measured features were found to have ice edge detection capability: HV-intensity, HH/VV-ratio and anisotropy. These features show a clear separation between sea ice and open water and simple thresholds can be applied.

Sea ice classification using multi-frequency polarimetric SAR data

This paper discusses the capability of the complex Wishart classifier for sea ice and classification using multifrequency, fully polarimetric SAR data. C-, L-, and P-band data acquired by the JPL AIRSAR in the Beaufort Sea was used. Classification using the unsupervised Wishart classifier is a two-stage process. An initial classification is required to seed the algorithm and can be derived using other classification methods. The Wishart classifier then used in iterations where the class means are updated after every step. The convergence of this approach is investigated. The Wishart classifier was found to be extremely dominant so that the classification result after a few iterations depends not necessarily on the initial classification used to derive the first class means. Even an initial classification derived with a random number generator leads to a good result after a few iterations.

Incorporating texture information into polarimetric radar classification using neural networks

Most of the recent research on polarimetric SAR classification focused on pixel-based techniques using the covariance matrix representation. Since multiple channels are inherently provided in polarimetric data, conventional techniques for increasing the dimensionality of the observation, such as texture feature extraction, were ignored. In this paper, we have demonstrated the potential of texture classification through gray level cooccurrence probabilities (GLCP), and proposed an unsupervised scheme using the self-organizing map (SOM) neural network. The increase in separability of the feature space is shown via the Fisher criterion and also verified by increased classification performance. Compared to the Wishart classifier, promising classification results are obtained from the Flevoland data set.

Classification of fully polarimetric single- and dual-frequency SAR data of sea ice using the Wishart statistics

Information on the extent and composition of sea ice is important for shipping and offshore operations. Single-polarization spaceborne synthetic aperture radar (SAR) data are an important information source for ice centres around the world. Next-generation SAR satellites will have the capability to collect fully polarimetric SAR data. In this paper we analyze the differences between polarimetric signatures of sea ice and those of land, test several applications of an unsupervised segmentation based on the Wishart distribution of polarimetric data, and give recommendations for modifications to the segmentation method. Airborne C- and L-band sea ice data are investigated. Surface scattering largely dominates sea ice backscatter. A volume component is present; it only dominates for some regions of selected ice types depending on the radar frequency. Dihedral scattering rarely dominates. The application of speckle filters further reduces the number of pixels with predominantly double bounce. The separation of scattering characteristics using the Freeman–Durden model before image classification provides little additional information. The initialization of the segmentation method with a scattering strength based classifier is very efficient and leads to good results.

Operational sea ice monitoring with RADARSAT-2 - a glimpse into the future

In order to assess the potential of RADARSAT-2, a field-validated airborne and satellite SAR sea ice dataset was collected over young and first year sea ice at multiple polarizations. Some preliminary observations are presented.

Model-based classification of polarimetric SAR sea ice data

This paper discusses the role of scattering decomposition models in the classification of polarimetric SAR sea ice data. The iterative Wishart classifier was applied to 3-frequency airborne SAR data acquired in the Beaufort Sea, and the scattering models were found to be helpful in interpreting the assigned classes. In addition to using the full data set, reduced data sets based on an eigenvector decomposition were investigated for their potential for classification, as the eigenvectors provided a separation of scattering mechanisms. The surface scattering component was found to be the dominant one for this data set, and yielded a classification similar to the full data set.

Potential of RADARSAT-2 for sea ice classification

Polarimetric data acquired by the CCRS CV-580 airborne SAR are used to assess the capability of RADARSAT-2 for operational sea ice classification. The information content of the polarimetric data is illustrated by showing how specific scattering mechanisms are portrayed by the entropy, anisotropy and /spl alpha/-angle features. Ice type classes are derived from the full polarimetric data set using a complex Wishart classifier. The classes are then mapped into 2-D scatterplots to compare the information content between dual and fully polarimetric data. While dual polarimetric data are an improvement over single channel data, it is found that fully polarimetric data are needed to provide accurate ice classification performance.

ENVISAT ASAR AP data for operational sea ice monitoring

The information content of ENVISAT ASAR alternating polarization data is evaluated with respect to operations at the Canadian Ice Service. A data set covering an entire ice season is shown to have a higher information content compared to single polarization data. The potential for automated information extraction is also investigated, in particular problems caused by system noise and its variation over the swath.

Analysis of the influence of NESZ variations on cross-polarized signatures of sea ice

The information content retrievable from ENVISAT ASAR cross-polarization data is evaluated with respect to sea ice monitoring. Low backscatter from sea ice and open water in combination with variations in the NESZ pose challenges for data visualization and classification. A novel classification scheme based on parallel Wishart classifiers for single and dual polariza- tion data is introduced that allows the utilization of cross- polarized information for part of the scene.