Sea Ice Pressure-Ridge Risk Maps Evaluated with an Advanced Feature Tracking and Strain Calculation Method
Rabus, Bernhard; Ma, Andy

Synthetic aperture radar (SAR) already has been established and is extensively used for deriving information on sea ice properties and conditions. Most information products such as ice concentration or wind and current maps for polar waters can be derived with methods that (after potential multi-image calibration) operate on single SAR images as input. In this paper we present a robust method that operates on temporal high resolution (few hours to 1 day) pairs or short sequences of SAR images to derive sea ice movement and strain map time series. As for all sea ice information products there is a trade-off between image spatial resolution and coverage: We show that even for coarse image resolution - RADARSAT-2 ScanSAR with 50 m ground resolution but over 300 km swath coverage our method can be used reliably to map convergence zones and pressure ridging for a variety of sea ice environments ranging from choked floes in a closed channel/fjord to open free flowing pack ice.
To derive the sea ice strain maps from a ScanSAR image pair our method employs a feature-based tracking algorithm to derive velocity vectors; followed by a sparse strain map calculation, and final product visualization The first step is a precise geo-registration of the image pair, which brings stationary features (on the coast or inland) to align within better than 0.1 resolution cells. Displacements and distortion of the sea ice then manifests as absolute feature shifts in the image pair.
The geo-registration step includes relative image to image registration (in SAR coordinates) using tie points from the master image, rational functions from the slave image, and calculated ground control points for warping the slave image onto the master image. The displacement and velocity information are derived using a variant of the Scale-Invariant Feature Transform (SIFT) algorithm, which detects sets of local features in each of the images separately. The SIFT features of the master image are then matched individually, checking for feature consistency in a corresponding neighborhood of specified size in the slave image feature space. This feature tracking algorithm is superior in deriving reliable ice velocity vectors because (other than more widely used area based matching) it is robust both to changes in brightness, as a result of open water roughness and floe wetness changes between master and slave images, as well as affine transformations of the sea ice (including translation, rotation, and strain). The resulting 2D ice velocity vectors are defined on an irregular grid. Post analysis of the velocity vectors in a circular (2D) spatial neighborhood is used as an efficient method to detect and filter out outlier measurements. Velocity vector components are triangulated and the 2D sea ice strain tensor is derived on a spatial grid (typical grid sizes are 1 to 4 km). Suitable thresholds for compressive isotropic strain (related to ice pressure) and deviatoric compressive strain, respectively, are used to identify areas prone to ice ridging and other hazardous ice conditions that affect the safety of shipping and oil and gas exploration and recovery operations. Suitable color-coding is used for visualization of the identified pressure and drift zones in the final information product (Sea Ice Pressure-Ridge Risk Map)
We demonstrate our new method to derive Sea Ice Pressure-Ridge Risk Maps from RADARSAT-2 SacnSAR Narrow image sequences for a number of test sites in the Canadian Arctic (Lancaster Sound, Victoria Strait, Beaufort Sea, and Labrador Sea). In future work we plan to incorporate additional ocean-meteorological data for a predictive extension of sea ice velocity vectors and pressure-ridge risk maps that can be used for short-term (tactical) forecasting of hazardous ice conditions.