Unpacking fjord ice in Hornsund, Svalbard

Fjord ice, that includes both sea and glacier ice, is an important part of the fjord microclimate that impacts e.g. water-atmosphere energy transfer, habitat conditions, ocean wave transformation and coastal processes. It also plays a role in ship and snowmobile operations. Understanding the trends in fjord ice extent, duration and timing aids understanding the impact of changing climate on the magnitude of natural hazards (such as coastal flooding and erosion) and improving future predictions.

Satellite images provide high-frequency large-area information on the state of the fjord ice, with Synthetic Aperture Radar (SAR) imagery being unaffected by polar night and weather conditions. Few studies have attempted automating fjord ice detection from satellite imagery, likely due to problems related to the topography influence on the sea state, mixed land/water pixels, presence of rocks and islands and wave breaking in the nearshore.

This study builds on the recent progress of Johansson et al. (2020) who adapted the semi-automated binary ice/open water classification method of Cristea et al. (2020) to Svalbard fjord environment, and Swirad et al. (2024a) who created a near-daily dataset of binary ice/open water maps at 50 m resolution for Hornsund fjord from the entire Sentinel-1 A/B dataset spanning Oct 2014 – Jun 2023. Swirad et al. (2024a) did not find direct relationships between fjord-scale ice coverage and air and water temperatures. Nonetheless, temporal peaks in ice coverage existed in March for the main basin, April for the inner bays and locally in October. The authors associated these with the arrival of pack ice from the Greenland Sea, formation of in situ fast ice and intensification of tidewater glacier calving, respectively.

Speculating that stronger relationships can be found between climate and ice coverage if fjord ice is unpacked into ‘drift ice’, ‘fast ice’ and ‘glacier ice’ we developed an algorithm that splits the ‘ice’ from the binary classification into the three classes using pixel and polygon properties such as continuity in time, location, size, shape and timing. We then explored relationships between ice, meteorological and hydrographic conditions. The dataset was also extended back to Jan 2012 using RADARSAT-2 imagery (Swirad et al., 2024b).

References:

Cristea, A., van Houtte, J., and Doulgeris, A. P.: Integrating Incidence Angle Dependencies Into the Clustering-Based Segmentation of SAR Images, IEEE J. Sel. Top. Appl., 13, 2925–2939, https://doi.org/10.1109/JSTARS.2020.2993067, 2020.

Johansson, A. M., Malnes, E., Gerland, S., Cristea, A., Doulgeris, A. P., Divine, D. V., Pavlova, O., and Lauknes, T. R.: Consistent ice and open water classification combining historical synthetic aperture radar satellite images from ERS-1/2, Envisat ASAR, RADARSAT-2 and Sentinel-1A/B, Ann. Glaciol., 61, 40–50, https://doi.org/10.1017/aog.2019.52, 2020.

Swirad, Z. M., Johansson, A. M., and Malnes, E.: Extent, duration and timing of the sea ice cover in Hornsund, Svalbard, from 2014–2023, The Cryosphere, 18, 895–910, https://doi.org/10.5194/tc-18-895-2024, 2024a.

Swirad, Z. M., Johansson, A. M., and Malnes, E.: Ice distribution in Hornsund fjord, Svalbard from RADARSAT-2 (2012-2016) [dataset], PANGAEA, https://doi.org/10.1594/PANGAEA.969031, 2024b.