Reliability Measures for Sea Ice Motion Retrieval From Synthetic Aperture Radar Images

Abstract

Abstract-Sea ice motion is triggered by wind and ocean currents. Its magnitude and direction can be automatically retrieved using pairs of satellite images acquired over the same area. However, external reference data for validation of drift retrievals, such as tracks from buoys, are sparse. Information about the reliability of the retrieved ice drift field is crucial for applications such as operational sea ice mapping or validation of computer models for simulations of sea ice dynamics. In this paper we introduce an intrinsic measure based on the properties of radar image pairs to assess the reliability of the retrieved ice drift vectors. The proposed method combines different parameters, e. g. correlation coefficient and two textural quantities, to provide information about the suitability of subimage regions for pattern matching. In this way, we generate a quality parameter (called confidence factor, CFA) for the calculated ice drift velocities. The CFA is compared to results obtained by 'backmatching'. The latter requires that the drift field is computed twice using the image pair, first in sequential and then in reversed order. For stable ice conditions, the results show that areas regarded as unreliable by the CFA compare well with the areas revealing larger differences from backmatching.

Impact of Sea Ice Drift Retrieval Errors, Discretization and Grid Type on Calculations of Ice Deformation

“…A displacement error of 0.1 km, as reported in Lindsay and Stern [16], for example, reflects a comparatively high accuracy and, thus, the given PDRE of 0.005 (0.5%) per day is not generally applicable. As indicated in the introduction, adverse conditions such as low contrasts and lacking structures in the SAR intensity images which are used to retrieve the displacement may lead to unreliable results (e.g., [14]). However, up to now, a robust method to quantify the resulting tracking error has not been published.…”

“…The increase of the time interval only makes sense if the trackable sea ice patterns or features are temporally stable. In addition, Equations (13) and (14) should be interpreted in terms of the ratio between a given drift retrieval error and the cell size: the PDRE is large if the drift retrieval error is large relative to the given cell dimension (see Figure 4).…”

A Method to Improve High-Resolution Sea Ice Drift Retrievals in the Presence of Deformation Zones

“…Hollands et al [8] introduced the concept of the confidence factor (CFA) to assess the reliability of the retrieved drift vectors. The CFA is a combination of different parameters that reflect the structural characteristics of each image, which are determined by intensity texture, speckle, and single extreme high-intensity spots due to specular reflections from the ice surface.…”

“…The total CFA is the sum of both. The texture_CFA consists of four parameters for characterizing the ice structure that were already applied by Hollands et al [8] (Table 1). It is determined in each step of the drift algorithm (i.e., in each cascade and the particular step in the resolution pyramid) from the spatial variations of the backscattering intensities inside predefined areas (correlation windows) in images 1 and 2.…”

“…To this end, we combine, modify, and extend the concepts of Hollands et al [8] and Lavergne et al [9], and add them to the drift detection algorithm implemented by Hollands and Dierking [1]. To avoid the replacement of a drift discontinuity by a smoothed velocity transition, we develop an adapted data regularization scheme that identifies linear deformation features (LDFs) in the drift field.…”