Sea-ice extents recalculated, starting from 2018!

12 September 2020

Every September, the sea-ice extent in the Arctic becomes a major topic of discussion, since it marks the end of the melting period and therefore the summertime minimum extent. Since the beginning of continuous satellite observation in 1979, the daily ice cover in the Arctic and Arctic has been measured extensively, offering a record of the climatic changes in the two polar regions. In this time, the Arctic has proven to be a climate change hotspot where global warming is progressing at nearly twice the speed as elsewhere and the September sea-ice extent has decreased by ca. 14% per decade, i.e., by roughly 50% over the past 42 years. This development has not only produced serious effects on the Arctic’s natural system and ecosystem; it has also affected the climate in the middle latitudes, as can be seen in the increased frequency of extreme weather events in recent years (see the article here).

Accordingly, there is considerable interest in how the Arctic sea-ice extent is presented and analysed, not just in the academic world, but also among the general public – an interest that we have responded to, by creating and operating, since 2013. The portal offers the public and research community scientific data and information prepared in a straightforward format, together with raw data and expert analyses.

What problem did we detect in the data processing?

This year the sea-ice extent, following in the footsteps of 2012 and 2019, has been quite low, and now, at the end of summer, is heading for an extremely low level. According to data recently posted on, this sea-ice retreat seemed likely to match the all-time minimum from 2012, a forecast that sparked major interest among the scientific community and public alike; it also led to intensive discussions among our sea-ice experts. By comparing our results with those of other providers, we determined that our data for 2020 lay below that produced by the other analytical algorithms currently available. This led to a new scientific review of the data analysis process, which currently runs fully automatically, in order to find the root of the discrepancy, and to quantify it.

How is satellite data normally analysed, and which corrections were made?

Those satellites used to investigate sea-ice characteristics measure the microwave radiation of the Earth, employing a variety of frequencies and polarisations – often simply referred to as ‘channels’ – to do so. The microwave radiation primarily comes from the planet’s surface (the land, ocean and sea ice); a small percentage comes from the atmosphere. Since the radiation emitted by seawater and sea ice differs in different channels (frequencies and polarisations), by skilfully combing channels, sea ice can be distinguished from open water. The majority of microwave satellites have between 8 and 12 channels, and there are a variety of channel combinations and methods for calculating the sea-ice cover. The satellites fly over the polar regions ca. 14 times a day, scanning a 1500- to 2000-km-wide strip each time. Accordingly, not only does the satellite data in the different channels have to be converted into sea-ice cover; the 14 partly overlapping strips (referred to as ‘swaths’, ca. 1 GB of data per day) have to be combined to yield a uniform map projection with suitable resolution, landmasses have to be concealed, and coastlines have to be inserted. In addition, atmospheric influences sometimes have to be taken into account: heavy precipitation over the ocean in particular can ‘confuse’ these methods, leading them to detect sea ice where there can’t possibly be any, e.g. on the Baltic in summer, or in the North Atlantic.

Filtering out this ‘false’ sea ice is painstaking work. It took years of development before all of the sea-ice algorithms currently available finally worked automatically and reliably. What’s more, this development work is never truly over, for a number of reasons: satellites are constantly being refined, and the new satellites might well be better, but work differently; there are advances in the land masks and coastline data, not to mention the map projection software; or the software packages being used expire and have to be replaced by newer versions. In May 2018 we had to carry out an extensive update to the sea-ice algorithm that we use, ASI, in order to incorporate updated map projection software (incl. land masks and coastlines), as well as improved methods to compensate for ‘false’ sea ice. Needless to say, after an update like this one, which normally represents an improvement, the algorithm will deliver different, more accurate results. Consequently, following the 2018 update, all sea-ice data since 2002 was recalculated using the new algorithm (ASI version 5.4).

That being said, the overall process is more complex: different sea-ice algorithms, which rely on different satellites and satellite channels, will yield differing outcomes; in many cases, they don’t even have the same resolution. For example, when daily sea-ice charts are used to calculate the sea-ice extent and are presented as curves (time series), the curves differ. The sea-ice extent represents the collective area of all grid cells in an ice chart that are at least 15 % covered by ice. The advantage of using the sea-ice extent is that it’s not prone to errors caused by surface melting in summer; the disadvantage is that, if the grid cell size is changed, it delivers different results (see here). For instance, if the grid cell size is 100 km² , 15 % of which is covered with ice on one side, the full 100-km² cell is counted as part of the sea-ice extent. In contrast, if the grid cell size were 1 km², the amount of sea-ice extent would only be 15 km². As a result, when they are combined, time series produced by different satellites with differing spatial resolutions have to be adjusted to produce a uniform time series.

Nevertheless, all time series of the sea-ice extent show a nearly identical trend, namely a dramatic loss of Arctic sea ice (see also Letter to Nature). But in order to make the various curves mutually compatible (e.g. to combine curves from successive generations of satellite into one long time series), you need to examine the overlap period of two different curves, use one of the curves for reference, and adapt the other to it via offsetting and scaling. Only then is it possible to produce time series longer than the service life of an individual satellite (usually less than 10 years). In our case, the reference time series came from the National Snow and Ice Data Center (NSIDC). After updating our sea ice algorithm in May 2018, the adaptation to the reference time series should have been updated, too. Since that never happened, our time series was effectively skewed too low, a fact that went unnoticed in May, when the sea-ice extent, due to the start of intensive melting, was already rapidly and irregularly declining. Since the sea-ice minima in 2018 and 2019 were essentially ‘average’, the skewing remained undetected. It wasn’t until we got close to this year’s minimum, which is definitely in the low range, that comparisons with other time series and previous years revealed the error.

Once the errors in our processing routine for calculating the sea-ice extent had been identified and corrected, all time series since 1 May 2018 were recalculated, double-checked and then posted on the website again, from 11 September. Older time series are unaffected by the correction. Comparing the results of different algorithms and questioning the accuracy of their results is part of the scientific process, and of quality control. We regret the fact that this error, which stemmed from the scientifically necessary update of the algorithm, didn’t become apparent until the unusually intensive melting this summer. What’s most important, however, is that when errors are identified in scientific outcomes, they are corrected as soon as possible, discussed transparently and reported to the public, regardless of whether or not the timing for doing so is ‘convenient’.

Currently, the Arctic sea-ice extent in September 2020 is ca. 3.8 million km², placing it below the number from 2019, but clearly above the all-time low from 2012. In all likelihood, this September the second-lowest sea-ice extent in the Arctic since 1979 will be reached.

Your team

Dr Gunnar Spreen (IUP Bremen)
Dr Christian Melsheimer (IUP Bremen)
Prof Dr Christian Haas (AWI)
Dr Renate Treffeisen (AWI)
Dr Klaus Grosfeld (AWI)