Characterization of Arctic Sea Ice Melt Pond Dynamics with Remote Sensing

Studies suggest that the Arctic is warming twice as fast as the mid-latitude regions, losing more than 80,000 km2 of its ice cover annually. One unique feature which demarcates rapidly melting regions of sea ice is melt ponds. These shallow pond structures accelerate melting by creating positive feedback through changes to the surface energy balance and are reported to cover up to 50% of the sea ice surface during the peak melt season. Yet, these features are not adequately represented in Earth System Models (ESMs). We use Sentinel-2 and WorldView-3 imageries coupled with Machine Learning classification algorithms to estimate the surface area fraction of melt ponds through multiple melt seasons. First, we employ a Random Forest classification algorithm to isolate the melt pond regions and estimate their total fraction with respect to the entire study region. Next, we develop a time-series of melt pond surface area fraction using images during the thaw-freeze season (May to August) from 2018 through 2021. Preliminary results from analysis of Sentinel-2 images indicate melt pond fractions range from 0% to 19%, exhibiting considerable variability between both inter and intra-melt seasons. Results from the lower 10 m resolution Sentinel-2 image are within 2% of the higher 2 m resolution WorldView-3 counterpart, indicating that despite the difference in spatial resolution, the two are comparable. The results from this research will be useful in understanding the evolution of melt ponds through melt seasons and validating models of melt ponds that can be built into ESMs.