The growth of frazil ice marks the initial stage of sea ice formation. It often develops at the ocean surface when katabatic winds drive ice away from the coast and leave the ocean exposed to the freezing atmosphere. Brine is rejected as the sea water freezes, leading to the production of high salinity shelf water. This dense water sinks beneath the surrounding seawater and can flow under ice shelves. This triggers ice shelf basal melting near the grounding line deep within the ice shelf cavity. Freshwater is released, which mixes with the high salinity shelf water and rises up along the ice shelf base, leading to the formation of supercooled ice shelf water. As this water emerges out of the ice shelf cavity, pressure is relieved, leading to a positive feedback loop where the freezing point is raised even further. This results in the ice shelf water becoming in situ supercooled, which is alleviated by the growth of frazil ice crystals. The outflow of ice shelf water and subsequent production of frazil ice crystals is a significant contributor to Antarctica's coastal sea ice mass balance. In combination with latent heat polynyas, ice shelf water-induced sea ice production contributes to around 10% of the total Antarctic sea ice budget from an area that makes up only 1% of Antarctica’s maximum sea ice extent. Antarctic sea ice extent has experienced an unprecedented decline in recent years, following a positive trend that was at odds with climate models. Changes in both the input of ice shelf water and offshore wind patterns may contribute to these sea ice trends. Despite their significance, ice shelf/ocean/sea ice interactions are poorly understood, owing to the logistical difficulties of taking in situ measurements.

This thesis used satellite remote sensing of the Victoria Land coast in the Western Ross Sea to detect summer frazil ice abundance, which is a potential indicator for the emergence of ice shelf water. This involved classifying sea ice types from Landsat 8, MODIS, and Sentinel-1 data, with a random forest machine learning algorithm in Google Earth Engine. These classified images were then used to generate heat maps of sea ice abundance, alongside sea surface temperature maps that were created from thermal imagery. Sentinel-1 radar image data was unable to detect frazil ice sufficiently, as the HH backscattering of frazil ice matched that of wind-roughened sea surfaces, while its HV backscattering was below the Noise Equivalent Sigma Zero. Landfast sea ice could not be adequately distinguished from pack ice, as their backscattering values largely overlapped. Sea ice types were more successfully classified with the optical sensors, although the lower spatial resolution of MODIS led to significant spectral mixing. Frazil ice appeared most abundant in the Terra Nova Bay polynya area, next to the Nansen Ice shelf and Drygalski Ice Tongue, where consolidated sea ice was scarce due to strong offshore katabatic winds, while the North Victoria Land coast had little frazil ice but abundant consolidated sea ice. This seems to indicate that frazil ice forms more readily when other sea ice types are absent and is also more easily detected by satellites under these conditions. Frazil ice was also detected in McMurdo Sound, but in lower abundance than in Terra Nova Bay. There is a significant amount of ice shelf water here, which may have contributed to frazil ice formation. Spectral mixing could explain why frazil ice appears less abundant in McMurdo Sound where there is a large amount of landfast sea ice and pack ice, even in summer.