[MASK]: MEaSUREs ITS_LIVE Greenland Monthly Ice Masks, Version 1

[chadagreene]

Overview

• Product name and version: MEaSUREs ITS_LIVE Greenland Monthly Ice Masks, Version 1
• Purpose of your product: mass balance
• Geographic region: Greenland
• Year(s) covered by product: 1985-2022 (the dataset begins in 1972 but is poorly constrained before Landsat 5, so use caution before 1985)
• Literature Citation: Greene, C.A., Gardner, A.S., Wood, M. et al. Ubiquitous acceleration in Greenland Ice Sheet calving from 1985 to 2022. Nature 625, 523–528 (2024). https://doi.org/10.1038/s41586-023-06863-2
• Data doi: doi.org/10.5067/579TO87M7IZB
• GitHub repo containing code used to generate the data.

Here's an animation of the final product:

Upstream

These data products were used to generate our monthly ice masks:

Here's how many line-kilometers of terminus position data we used for each glacier, in each month:

The description below is copied and pasted from the Methods section of the published paper:

We use 237,556 manually derived and AI-derived glacier terminus picks from 1972 to 2022, obtained from the sources described below. We focus our analysis primarily on the years since 1985, during which time 236,328 terminus picks were acquired. Although data coverage is generally poor before 1985, we include all available observations to help constrain the state of the ice sheet at the beginning of our analysis period. Extended Data Fig. 6 shows the temporal distribution of the acquisition times of all terminus picks, which were selected from the following datasets:

• AutoTerm: we use 153,281 terminus positions from the AutoTerm dataset55,56, including 153,250 positions acquired since 1985. AutoTerm provides data from several optical and radar satellite sensors, spanning nearly four decades and includes winter data in recent years. Through visual inspection, we found that AutoTerm performs particularly well at some of the 295 glaciers it covers; however, data quality clearly suffers at other glaciers. Terminus-position accuracy is often dependent on satellite sensor and corresponds reasonably well with error estimates that are provided with the AutoTerm data. For our purposes, we inspected all AutoTerm picks visually to manually determine separate error thresholds for each of the 295 glaciers, such that we eliminate all data corresponding to error values that are associated with obvious outliers or asynchronous behaviour. For this reason, we use only 153,281 of the 278,239 terminus positions available in the full AutoTerm dataset.

• MEaSUREs weekly to monthly: we use the 21,990 weekly to monthly terminus positions18,57 collected using the Sentinel-1 synthetic aperture radar since January 2015. Although the full dataset contains 23,676 terminus positions, we only use positions whose quality flag is 0.

  • MEaSUREs Annual v2: we use 3,437 terminus positions from the MEaSUREs v2 dataset58, including 2,987 picks acquired since 1985. We only use the highest-confidence data, with quality flags 0 or 2. Quality flags 1 and 3 correspond to uncertain picks or Landsat-7 SLC-off images and are not used in our study. We also eliminate redundant data by discarding any positions obtained from the same images used in MEaSUREs weekly to monthly data.

  • CALFIN: we use 19,835 terminus positions from the CALFIN dataset59,60, including 19,665 picks acquired since 1985. In this subset, we have discarded any CALFIN picks in which MEaSUREs Annual v2 manual picks are available for the same satellite image.

  • TermPicks: we use 39,013 terminus positions from the TermPicks dataset61,62, including 38,436 picks acquired since 1985. We discard any TermPicks data in which MEaSUREs Annual v2 manual picks are available for the same satellite image.

We note that the AutoTerm dataset includes a large amount of Landsat imagery that is also included in the MEaSUREs Annual v2, CALFIN and TermPicks datasets, meaning that there is some redundancy and probably some discrepancies between the various methods of terminus-position picking. We find that AutoTerm provides the most comprehensive record overall, but the width of fjord walls tend to be defined more narrowly in AutoTerm than in other datasets, meaning that the full widths of glaciers are sometimes not captured in AutoTerm. Also, in some cases, the bounding boxes of the AutoTerm picks seem to cut off the full extents of calving-front migration.

Internal

Here are the datasets we used:

• Observations of terminus positions were combined from the five datasets listed in the section above. We find that human and AI-derived terminus positions both contain errors and often disagree with each other or even themselves. To reduce "flicker" (as ice can sometimes magically appear or disappear if a terminus position pick is incorrect) we constrain the advance rate of the ice masks from one month to the next using observed velocities.

• Masks are from Mouginot 2019. The text below is copied and pasted from the methods section of our paper:

We use 260 named Glacier catchments for the GrIS78. To account for terminus activity that may have occurred beyond the extents of the predefined glacier catchments, we extrapolate each catchment downstream following flowlines from our velocity grid. Each catchment is then dilated by 5 km to fill any gaps between extrapolated flowlines and fjord walls or neighbouring catchments. Our extrapolated catchment delineations are shown in Extended Data Fig. 11.

• Geographical region (Greenland, Antarctica, sub-region, peripheral, individual glacier)

• Type: raster

• Resolution if raster: 120 m

• Projection information: EPSG 3413

Downstream

• The dataset is currently in review at NSIDC. The metadata in the data file here may change before publication, but the ice mask data itself will stay the same.