Analysis of polarimetric signatures of Arctic lead ice using data from AIRSAR and RADARSAT

Abstract: In December 2004 the JPL airborne synthetic aperture radar (AIRSAR) flown on the NASA DC-8 acquiredfully polarimetric data in the Beaufort Sea at C, L and P-band. This work presents the backscattercoefficients (σHH, σVV, σHV), copolarized ratios (σHH/σVV) and copolarized phase differences (ϕHH-VV) at thethree frequencies from various sea ice types and in particular different young sea-ice types formed inrecently frozen leads. Two weeks of RADARSAT imagery from the same region as the AIRSAR data wasused in order to identify when and where leads were formed and the age of the newly formed icecontained within the leads. By using a known empirical relationship based on freezing degree days, thethickness of a sea-ice layer could be estimated from its age.Several stages of new and young ice were identified and examined by using RADARSAT imagery to trackback in time to find sea-ice openings: lead ice 1-2 days old, lead ice 2-3 days old, lead ice 9-14 days old,lead ice/first year ice older than 15 days and multi-year ice. Furthermore, narrow cracks with uniquesignatures, hypothesized to be a few hours old, found in several 1-2 days old leads have been included inthe analysis. In addition to report the polarimetric signatures of these ice types, this study seeks to findwhich combination of polarimetric parameters that best differentiates the primary ice types as well asthe stages of new and young ice within the leads.Ice that is thought to be less than one day old has the lowest backscatter coefficients for C- and L-band.This ice type also has significantly larger copolar phases for C- band than the other ice types and is theonly ice type that shows negative phase differences for L-band, in correspondence with previousreported results for very young ice. Furthermore, it has the lowest observed C-band copolar ratios. Ice 1-2 days old is characteristic for the large variability of the significantly negative P-band phases. Ice up tothree days old has the lowest C-band copolar ratios. Using copolar ratios and phase to discriminate iceof age 9-14 days and older than 15 days was difficult, but L-band backscatter appears to give morecontrast. The analysis of the phase information shows large variations between the thin ice types but itis hard to draw unambiguous conclusions. However, generally the phases for thin ice have much largervariability and typically have an offset relative to multi-year ice.The best separations of thin ice are found by using C-band channels and for combinations with differentpolarizations. C-band crosspolarized backscatter coefficient combined with C-band copolarized ratio hasbeen determined as giving the largest separation causing less misclassification, using a discriminantclassifier. Ice less than one day and multi-year ice are the ice types that separate best from the othertypes while the other ice types more or less overlap for most combinations. Severe mixing between icetypes occurs for combination of different L-band channels.In future work the calibration quality of L- and P-band needs to be assessed. More time must be spenton trying to define different ice types and how to pick ice samples minimizing the effect ofinhomogenities in the different polarimetric channels. Methods to determine normalized classificationaccuracies, independent of the number of samples of each class, must be considered.