Towards high resolution estimates of East Antarctic pack ice thickness

The area of the Southern Ocean covered by sea ice is relatively well described by a longterm time series of satellite observations spanning over 35 years. Throughout the seasonal cycle, Antarctic sea-ice cover varies between approximately 3 x 10‚ÄövÖ‚àÇ km¬¨‚⧠at the end of summer to approximately 20 x 10‚ÄövÖ‚àÇ km¬¨‚⧠during winter. This has profound implications for ocean-atmosphere heat and momentum exchange and also for global oceanic circulation induced by sea-ice production and melt. However, the thickness distribution of Antarctic sea-ice is poorly described. Without detailed knowledge of the distribution of sea-ice thickness (and hence volume) it is difficult to assess the magnitude of its influence, and its feedback within the ocean-ice-atmosphere system. Current estimates of the sea-ice thickness distribution derived from satellite-based altimetry rely on empirical models describing the relationship between sea-ice topography, snow depth and sea-ice draft. The input to these empirical models include approximations for the density of snow, sea ice and seawater, and relationships between snow depth and the total freeboard (ice + snow) of sea ice derived from sparse in-situ observations. Further, observations from comparatively lower resolution satellite altimetry remain insensitive to small scale features that often characterise sea ice. This research estimates high-resolution sea-ice thickness from airborne nadir-looking photography and Light Detection and Ranging (LiDAR), deployed over East Antarctic pack ice between 2007 and 2012. It presents a detailed analysis of empirical models for snow depth estimation derived from in-situ observations. For altimetry without coincident methods to remotely sense the snow/ice interface, using an empirical relationship between snow depth and measured elevation is the only available means to estimate snow depth. This represents a novel direct comparison between empirical snow depth models and in situ observations at scales of less than hundreds of metres. Examination of sources of uncertainty in altimetry observations follow, with the development of a method for deriving the uncertainty of each point position in a LiDAR swath. This allows the rigorous propagation of uncertainties from the airborne instruments through to snow depth and ice thickness estimates. A unique comparison of draft estimates from airborne LiDAR with sea-ice draft mapped by upward looking SONAR from an underwater robot forms a key component of this research. SONAR-based ice drafts are used to empirically derive parameters for the ice thickness model applied to airborne LiDAR observations. Using the same small patch of sea ice, relationships between surface and under-ice features are explored, providing a unique insight into how well ice drafts derived from surface topography characterise the under-ice environment. The empirically derived model parameters from this exercise are then applied to a larger-scale dataset, which is compared to near-coincident ship-based visual sea-ice observations (using the ASPeCt protocol). Using in situ and sonar observations collected by an Autonomous Underwater Vehicle (AUV) to inform parameter choices for modelling sea ice thickness from LiDAR altimetry, a close match between sub-floe scale ice draft from AUV and LiDAR-derived estimates was obtained. The ice density parameter used in the model was artificially high, at 915.6kgm¬¨‚â• compared to observed values which ranged from 800 kg m¬¨‚â• to 870 kg m¬¨‚â• on SIPEX-II. The distribution of ice keels associated with surface ridges was broadly similar in both AUV observations and LiDAR-derived estimates, although keels modelled using surface topography are narrower and deeper than keels present in the AUV observations. Smallscale topography was also added at the ice-ocean interface for draft estimates from airborne LiDAR, reflecting the influence of snow dunes at the surface. Importantly, co registration of in situ, AUV and airborne observations shows clearly that in situ observations captured only the thinnest and least deformed region of the surveyed ice floe. At a broader scale, the knowledge gained from validating sea ice thickness estimates over a single floe was applied to a 120 km flight leg with a near-coincident ASPeCt ship-based ice thickness observations. The distribution of ice thickness estimates from airborne LiDAR exhibited a shallow peak in the 3m to 4m range, where ship-based observations were predominantly showing ice thicknesses between 1 m to 2 m. This pattern is also seen for the smaller floe-scale study, suggesting that in situ observations for regions of deformed ice off East Antarctica may only represent one tail of the sea-ice thickness distribution. The implications for the use of in situ observations in tuning algorithms for satellite-based estimates of sea-ice thickness are clear, suggesting a bias toward thin ice especially in regions of heavy deformation. This work advocates for the combination of multiple observation types and techniques given their ability to offer insight into biases that affect a particular method at a particular spatial scale. Technologies spanning in situ, airborne and satellite platforms have a strong and integrated future in sea-ice observation.