Tenuous Correlation Between the Snow Depth or Sea Ice Thickness and C- or X-band Backscattering in Nunavik Fjords of the Hudson Strait

Kurzfassung

Sea ice is an essential part of the Arctic Ocean and is extremely affected by the global warming. Over the past 30 years, the Arctic has warmed at roughly twice the rate as the entire globe. The Kangiqsujuammiut (people of Kangiqsujuaq, Nunavik, in Canada) observed a later sea ice freeze-up in the fall (Nickels et al., 2011), less snow on the ground, earlier sea ice breakup in spring, changes in travel routes and more variable winds (Cuerrier et al., 2015). These changes influence immensely the Inuits live and have consequences on their personal safety and access to travel and marine wildlife due to ice instability and weather unpredictability (Laidler et al., 2006). Recent work by Yackel et al. showed that, prior to melting onset, the daily variance in C- and Ku-band backscattering is correlated with relative snow thickness (Yackel et al., 2019). The majority of these sea ice applications rely on the C-band (Zakhvatkina et al., 2019). Though, the X-band can adequately discriminate between newly formed ice and its surroundings (Johansson et al., 2018) and is more sensitive to melt onset and surface roughness than the C-band (Eriksson et al., 2010), as well as changes in top-layer snow salinity (Nandan et al., 2017). As for the snow cover, Gill et al. noted that a positive correlation between C-band HH backscattering and snow depth is best observed in cold (-7.9°C) than warm conditions (-0.4°C) (Gill et al., 2015). In contrast, Nandan et al. rather observed that the HH backscattering was greater from a thin snow cover of 4 cm than from snow 8 or 14 cm deep, which they attributed to steeper salinity gradients in thin snow causing enhanced surface scattering (Nandan et al., 2017). They also reported that the HH polarization was more sensitive to snow depth variations than its VV counterpart. This study investigates the correlation between snow depth and sea ice thickness based on field measurements and the SAR backscatter from remote sensing data from 2015 to 2018 focusing the three Nunavik fjords of the Hudson Strait, Canada. The objectives are to 1) characterize and explain snow depth and ice thickness distributions in Salluit, Deception Bay, and Kangiqsujuaq over three winters (2015-2018) and 2) investigate the empirical relationship between the C- and X-band backscattering coefficients and snow depth and ice thickness. The in-situ data were collected in January-February and April-May of 2016, 2017, and 2018, for each site, except for April 2016 where bad weather prevailed in Salluit. Snow depth was measured using a meter-rule and ice thickness using a 2.5 cm diameter Kovacs ice auger and measuring tape. The measurements were performed at 20 to 30 target locations arranged in a grid-like pattern over each fjord. The remote sensing data were complemented by a time series of Radarsat-2 Wide-Fine quad-polarization data with a repeat period of 24 days and an incidence angle of 35° - 38° and TerraSAR-X StripMap dual-polarization (HH/VV) data with a repeat period of 11 days with an incidence angle of 38°. The factors driving snow depth and ice thickness distribution in Salluit, Deception Bay, and Kangiqsujuaq include: (i) greater precipitation in the latter which consistently leads to deeper snow and thinner ice than in the other sites, (ii) freeze-up timing which was earlier in 2015 and led to thicker ice, and (iii) the orientation of dominant winds relative to fjord length which occasionally leads to downwind snow depth gradients. The ice in Deception Bay was 50 cm thinner on average than 25 years ago. In cases of landfast and undeformed first-year sea ice, we conclude that the necessary conditions for detecting a correlation between sea ice thicknesses above 30 cm and SAR HH backscattering in the C- or X-band are a snow cover thinner than 10 cm and slightly rough ice formed from dynamic processes. In terms of usability for predictions, the correlations we observed in these conditions were poor at best for the C- or X-band, respectively, with r-squared values of 0.2 and 0.6. In cases with snow thicker than 20 cm on average or with very smooth ice, no correlation could be detected with ice thickness. No correlation above 0.3 was observed between snow depth and backscattering. The Bayesian linear regression analysis proved to be useful in categorizing each case according to their most likely hypothesis out of the ones we tested. We were unable to reproduce results showing a correlation either between ice thickness and the co-polarization ratio in the C-band or between snow depth and the HH backscattering coefficient in either frequency. Backscattering in the C- and X-bands was either different or similar depending on the type of ice. Over smooth ice formed from thermal freeze-up, backscattering from both bands is dominated by volume scattering and their HH backscattering coefficients present a 5 dB difference. Over slightly rougher ice formed from consolidated nilas patches, surface scattering dominates at both frequencies. Their HH backscattering coefficients differ by 1 dB or less.