Data error discovery slows trend of growing Antarctic sea ice

Sea ice, coming to an Antarctic sea near you. Brocken Inaglory, CC BY-SA

The amount of the earth’s ocean surface covered by sea ice has been continually observed by satellites and its extent estimated since 1978. The trend has been for shrinking sea ice in the Arctic and, more recently, expanding sea ice in the Antarctic.

This somewhat counter-intuitive finding has been explained by reference to the ozone hole, currents, and winds. But it seems in part it has a more straightforward origin, stemming from an error in how the data is recorded and processed.

Several mathematical algorithms have been developed to convert raw satellite data into estimates of the area covered by at least some sea ice. While for the most part the results of different algorithms are in reasonable agreement, there are differences. A widely used method is the Bootstrap algorithm, developed at NASA Goddard, which performs particularly well at estimating Antarctic sea ice compared to others.

As satellite sensors have only a limited lifespan, to calculate trends over longer time periods requires stitching together the records from several sensors. However, each sensor is a little bit different, and these differences need to be corrected to create a consistent record over several decades. Slight adjustments in the algorithms are made to ensure the estimates from the new sensor match as closely as possible to the estimates from the old sensor during the period when their records overlap. This can never be done perfectly, but with care disagreements can be minimised. All algorithms, including Bootstrap, do some form of inter-calibration to try to ensure the data is consistent.

The difference is the data

Bootstrap was used to report sea ice extent for the last two IPCC reports, AR4 in 2007 and AR5 in 2013. Between AR4 and AR5, Bootstrap changed from Version 1 to the improved Version 2. The reprocessed data set, V2, shows a magnified sea ice increase over V1. This wasn’t noticed at the time because Antarctic trends, being relatively small in magnitude and with large year-to-year variations, could change fairly substantially from year to year in any case.

However, Ian Eisenman who led our investigation, published in the Cryosphere journal, found that the growth of Antarctic sea ice between the last set compiled with V1 and the first compiled with V2 was greater than could be explained just from adding more data. I helped find and provide the earlier V1 data and he and fellow co-author Joel Norris compared the two versions. We found the discrepancy can be pinned to a sensor calibration change in 1991.

Plot showing significant and until now unnoticed jump from V1 (blue) to V2 (red). Eisenman/Meier/Norris/Cryosphere, Author provided

Our analysis doesn’t discern in which version the inter-calibration error occurred – we simply compared the two versions and noticed the jump. The data set producer, Joey Comiso, has now looked at the issue and is confident that the error lies in V1, meaning the current version in use and used in AR5 is correct. Ian, Joel, and I are interested in looking further into this to confirm the V1 error.

But the end result is the same

It’s important to emphasise that this sort of reprocessing of climate data – any scientific data, in fact – is not just common, but part of the scientific process; better data becomes available, methods are improved, errors are found and corrected. Data sets are never perfect, but as scientists we continually work to improve our data and understanding of it in order to get closer and closer to the truth. I think this paper and the reprocessing of Bootstrap are good examples of this process.

Fundamentally the paper doesn’t change our understanding of Antarctic sea ice. Today, regardless of which version of the data or sensor you use, Antarctic sea ice extent is increasing at a statistically significant rate. The paper documents that the jump in trend numbers reported between V1 and V2 was due to processing and not due to variation in sea ice extent. Our results simply correct the published literature and show that, contrary to what was previously thought, the increase in Antarctic sea ice hasn’t accelerated in the past 15 years, but has been remained consistently positive at moderate levels. Unlike Arctic sea ice, and the Antarctic continental ice sheets, which have been consistently, and substantially, falling.


Next, read this: The Arctic melts, but oceans and ozone hole may cool Antarctica