Lead detection with Sentinel-1 in the Beaufort Gyre using Google Earth
Engine.
Abstract
Sea ice leads are produced from deformational forces, which break apart
the ice surface and expose open water areas or leads. Leads are the
primary regulators of heat in Arctic sea ice during the polar night.
Partially-frozen and re-frozen leads produce smaller heat exchanges than
open leads due to the absence of a warm water surface. Thus detecting
leads is important because they may be used as an indirect way of
estimating air-sea heat fluxes. To quantify winter-time leads, we
utilize Sentinel-1 C-Band Synthetic Aperture Radar (SAR) data to examine
sea ice images through heavy cloud cover. We employ a support vector
machine learning technique in a cloud computation environment (Google
Earth Engine) to detect and quantify lead areas. With the use of
dual-polarization data, we improve the separation of leads from other
elongated features (e.g., ridges) in the Sentinel-1 dataset by adding
altimetry information from ICESat-2. In addition to typical texture
analysis to assess surface roughness, the ICESat-2 ATL-10 data allows us
to train the algorithm by discretizing leads and ridges by their
freeboard values. Performing this method in a cloud environment allows
processing of a large volume of satellite data and converting it into a
time series of leads properties. Overall, our method improves lead
detection with dual-polarization SAR data while simultaneously providing
a big data solution for SAR image processing. The interannual
variability of leads and newly formed ice fractions were found for the
winters of 2017-2020. Finally, we compare the results from previous
studies to validate our cloud-derived sea ice lead detection maps.