Proceedings Article | 19 October 2023
KEYWORDS: Ice, Satellites, Cooccurrence matrices, Random forests, Climate change, Satellite imaging, Remote sensing
Svalbard (Norway) is a “hotspot” of climate change. There, various research activities are taking place, most of them located at the Ny-Ålesund Research Station, on the shore of the Kingsfiord, mainly focused on understanding climate change dynamics and its effects on the Arctic marine and terrestrial ecosystems, including their interconnections. In fact, the fiord is a unique environment where the atmosphere, sea, land and their ecosystems are strictly connected. Among the other ones, Ny Ålesund hosts the Italian Arctic Station “Dirigibile Italia”, a multidisciplinary research facility owned by the National Research Council of Italy (CNR), since 1997. Currently, several scientific projects are developed at the Italian Station, dealing with physics and chemistry of the atmosphere and snow, microbial ecology and evolution, nutrients and ecosystems, biogeochemistry and energy fluxes, clouds, aerosols, gases and remote sensing. One of the important parameters that influences the fiord dynamic is the sea ice occurrence. Sea ice in the fiord can form overnight due to strong wind and disappear as fast. Among other, it is expected that sea ice cover influences the air-sea gas exchange. Various sea ice remote sensing products are available from satellite data. However, they are generalized and do not provide high resolution daily maps. This means that fast ice formation and disappearance is not properly detected, and available products are not detailed enough for local scale analysis. In the literature, the analysis of available sea ice products has been performed to evaluate its usefulness for the local scale applications. Various limitations have been identified, including generalization of sea ice extent, gaps in temporal coverage (e.g., some products are not generated on weekends), mismatches on sea ice extent between different products. To overcome these limitations, usability of Sentinel-1 based products were considered within this study. Ice cover maps from Sentinel-1 data were generated using machine learning algorithms and various feature sets (e.g., GLCM, Hölder exponents). The goal of this algorithm is to develop detailed (with high spatial and temporal resolution) sea ice maps of the Kingsfiord.
