Automated iceberg detection using Landsat: method and example
application in Disko Bay, west Greenland
Abstract. Over the last two decades, the flux of icebergs into Greenland's fjords and coastal waters has increased, concurrent with changes in mass loss and dynamics of Greenland's marine-terminating outlet glaciers. Icebergs impact fjord circulation and stratification, freshwater flux, and ecosystem structure and pose a hazard to marine navigation and infrastructure, yet they remain a relatively understudied component of the ice-ocean system. Icebergs are easily detected in optical satellite imagery, but manual analysis to derive an iceberg size distribution time series is time prohibitive and partially cloudy scenes pose a challenge to automated analysis. Here we present a novel, computationally simple machine learning-based cloud mask for Landsat 7 and 8. This mask is incorporated into a larger iceberg delineation algorithm that allows us to extract iceberg size distributions, including outlines of individual icebergs, for cloud-free and partially cloud-covered Landsat scenes. We applied the algorithm to the Landsat archive covering Disko Bay, west Greenland, to derive a time series of iceberg size distributions from 2000–2002 and 2013–2015. The time series captures the seasonal signal in ice cover resulting from the annual cycles of sea ice formation and breakup and calving of Jakobshavn Isbrae, the dominant source of icebergs in Disko Bay. We note a change in this annual signal during the latter time period, likely a direct result of changes in the calving regime of Jakobshavn Isbrae. During 2000–2002, Jakobshavn Isbrae's floating ice tongue disintegrated and disappeared, transitioning the glacier from low energy, tabular iceberg calving to high energy, full thickness calving and the production of more small icebergs. This transition is also evident in the change in the number of small (~ 225 m2) icebergs found on the bay, which increased during the latter period. The change in the number of small icebergs also led to increasingly negative power law slopes fit to the iceberg size distribution time series. This work suggests that iceberg size distribution time series may provide useful insights into changes in calving dynamics and the physics of iceberg decay, while aiding marine and navigation safety in iceberg laden waters.