A textural approach to snow depth distribution on Antarctic sea ice

Understanding the distribution of snow depth on Antarctic sea ice is critical to estimating the sea ice thickness distribution from laser altimetry data, such as from Operation IceBridge or ICESat-2.  One major uncertainty in converting laser altimetry data to ice thickness is knowing the proportion of snow within the surface measurement. Snow redistributed by wind collects around areas of deformed ice, but it is not known how different surface morphologies affect this distribution. Here, we apply a textural segmentation algorithm to classify and group similar textures to infer the distribution of snow-ice ratios using snow surface freeboard measurements from Operation IceBridge (OIB) campaigns over the Weddell Sea. We find that texturally-similar regions have similar snow/ice ratios, but not similar snow depth measurements. This allows for the extrapolation of nadir-looking snow radar data using two-dimensional surface altimetry scans, providing a two-dimensional estimate of the snow depth. Using a convolutional neural network on an in-situ dataset, we find that local (~20 m) snow depth and sea ice thickness can be estimated with errors of < 20%, and that the learned convolutional filters imply that different surface morphologies have different proportions of snow/ice within the measured surface elevation. For the OIB data,  we show that at slightly larger scales (~180 m), snow depths can be estimated using the snow surface texture, and that the learned filters are comparable to standard textural segmentation filters. We also examine the statistical variability in the distribution of snow/ice ratios across different years to determine if snow distribution patterns on sea ice exhibit universal behaviour, or have significant interannual variations. These results suggest that surface morphological information can improve remotely-sensed estimates of snow depth, and hence sea ice thickness, compared to current methods. Such methods may be useful for reducing errors in Antarctic sea ice thickness estimates from ICESat-2.