Greenland Lake Ice Breakup Detection from Sentinel-1 SAR

The timing of lake ice formation and breakup are relevant climate indicators. In this study, we explore the potential of utilizing Sentinel-1 synthetic aperture radar (SAR) data for identifying the timing of lake ice breakup across Greenland between 2016 and 2022 and assess its latitudinal and vertical gradients. We retrieve average backscatter data of lakes in peripheral Greenland with a surface area > 1km² (n = 1842). Data with a low number of acquisitions for the entire study period (n < 1000) or exhibiting strong uniformal annual characteristics (backscatter difference between 95^th and 5^th quantile < 5dB) are excluded from the analysis. We apply a locally weighted scatterplot smoothing (LOWESS) filter to remove outliers. A dynamic numerical threshold (backscatter decline within 3 consecutive acquisitions > 25% of the annual backscatter range) is applied for each respective year to identify the timing of ice breakup. The study area is divided into 6 main regions of Greenland (N, NE, SE, S, SW, NW) to explore spatio-temporal statistics. The data exhibits a temporal resolution of about 2 days during the relevant period. We validate the breakup detection (n = 10) by utilizing daily time-lapse images of 3 lakes between 2016 and 2020. The detection of the timing from SAR data proves to be conservative (i.e., later) compared to time-lapse camera data and allows characterizing lake ice breakup with a mean error of 7 days. We find that only SAR data in West Greenland (S, SW, NW), i.e., > 43°W and < 70°N, exhibits characteristics for breakup detection (97%, 77% and 57% suitable) while coverage for North and East Greenland (N, NE, SE) lacks necessary radiometric and temporal characteristics (only 2%, 3% and 2% suitable). Our preliminary results indicate that no significant trend (α = 0.05) of breakup timing between 2016 and 2022 can be identified. Annual median DOYs range between June 8 (2019) and July 11 (2022). Ice breakup timing increases with latitude and elevation, however, strong correlations (up to r = 0.81) can only be identified for limited years. Correlations are in the order of 2 to 5 DOY/°lat. and 2 to 7 DOY/100m. Based on these preliminary results, we aim to explore statistical relations in greater detail to assess the role of extreme events and global climate change. Furthermore, we intend to apply this automated algorithm for an analysis of lake ice breakup timing on a global scale.