The US National Snow and Ice Data Center (NSIDC) has recorded Greenland ice melt surface area data since 1979. It came to my attention in 2019 when there was an alarming spike in the melting surface area. The spike was initially portrayed as extreme event against a central range as in this Economist chart.
A week later, another Economist chart showed the spike in the context of other spikes,, highlighting the challenges of interpreting noisy data.
Since then, I’ve been keeping an eye on this data. Surface melt area is not the only measure to track, but it seems to be the one with the longest available history, presumably because it can be estimated from satellite imagery. It’s concerning that the NSIDC funding has been cut this year; the website says they still expect to collect data, but they won’t be able to analyze it.
2025 update
Checking the NSIDC site today, they still show the same basic daily view with overlaid reference intervals. You can choose which years to show. Here’s 2012 (the year with the most melt area), 2019, and 2025.
While I’m always interested in smoothers and other less-is-more approaches, my first step is usually to try to reproduce the given chart with downloaded data.
This one took some effort but I think I captured all the details and all the data seems to match.
The “effort” was in showing lines for selected years on top of reference regions computed from other years. My approach was to create a “year group” column to categorize each year as base, line, or ignore. Then I added a new column for the reference melt areas with the following formula.
The expression ‘year group == “base”‘ evaluates to 1 for reference years and 0 otherwise, and dividing by 0 will produce a missing value for the non-base years. Then I could compute the needed quantiles based off of the new reference melt area column. A similar formula produced the line melt areas.
It may have been clearer to create separate summary tables and then combine the relevant parts to make a graph, but it’s handy to keep everything in one formula-driven table. For instance, it makes it easy to change the year categories.
My main insight in 2019 was to show cumulative amounts rather than daily amounts to filter out the noise. Here’s the same data as the previous graph (plus two more years) but with cumulative sums within each year.
A cumulative sum is a sneaky way to introduce smoothing into a chart without having any explicit smoothing parameters. After all, a moving average is just a cumulative sum divided by the count, so a cumulative sum is as smooth as a moving average (and it appears more so since the scale is greater). This less crowded chart also allows for direct labeling and showing a few more individual years. The downside is that I had to add interpolated values for the missing days in order for the cumulative sums to be comparable. Fortunately, there weren’t many missing days, so I don’t feel like it’s much of a stretch. And for reasons unrelated to funding cuts, the 2025 data ended earlier than other years, which is easier to see in this view.
Here are the cumulative sum curves for every year, color-coded by decade.
Simplifying even further, we can look at just the total melt area for each year.
While the trend line appears to have leveled out, I doubt it’s any reason to take comfort. It’s still far above the level of the 1980s and melt area is not the same as melt volume. And at the extreme, if the ice sheets are shrinking, there is less surface area available for melting each year (and eventually we’ll have a melt area of zero!). However, we also have replenishing snowfall, and I don’t know that shrinkage is a factor, yet.
Polar Portal
Worried about the NSIDC’s funding cuts, I was happy to find Denmark’s Polar Portal, which has data on melt area and volume. The Polar Portal melt area data only goes back to 2018 (at least what’s available online). They use a different measurement technique, based on meteorological stations throughout Greenland rather than satellite imagery. This provides a rare opportunity to compare two independent measurement approaches.
The Polar Portal data is reported in percent of area rather than km2. They give the total Greenland ice area as 1,700,000 km2, so using that I can convert the NSIDC data to percentages and roughly compare the two data sources for the eight years they both have data.
It’s surprising how different they are but also encouraging that the two methods capture many of the same spikes. However, they differ more in the later years. The Polar Portal notes do caution against year-to-year comparisons due to model updates; nonetheless, here are the yearly totals from the two data sources over time, on a km2 scale.
