InsightsHow we workTeamJoin German
Data Readiness Check
back to overview

Data Science Techniques: When will the Arctic be ice free?

Daniel Senftleben, 01. Oktober 2020

“Arctic sea-ice shrinks to near record low extent” (BBC News, 21 Sep 2020), “Another wake-up call: sea ice loss is speeding up” (UN Environment Programme, 28 Sep 2020), “A chunk of ice twice the size of Manhattan has broken off Greenland in the last two years” (CNN, 14 Sep 2020) - Headlines such as these have been piling up in the media in recent weeks. Climate change has manifested in the Arctic like nowhere else on Earth. At the same time, the Artic is also the region with the largest discrepancies in the results of global climate model simulations, making predictions of its future extremely challenging.

In our projects, we also encounter data sources of equally high quality but with analysis results that are in part contradictory. The key here is to not simply average the available data, but to apply weights to the different data sources according to their validity for answering the question at hand. Estimating this validity can be a challenging task by and of itself. A typical approach of back testing is to evaluate the data sources on their ability to predict those parameters that are most relevant to the problem.

So, when will the Arctic be ice free for the first time in millions of years? “Ice free” is commonly defined as a sea ice extent of below one million km², observed in September (yearly minimum). Assuming a so-called “business-as-usual” greenhouse gas emission scenario, projections of 29 of the most renowned global climate models here range from the year 2040 to beyond 2100, with 2076 the arithmetic mean year of an ice-free Arctic (see red lines in Figure 1).

Figure 1. Climate projections of September Arctic sea ice extent from 29 global climate models (gray), with the arithmetic mean (red) and the mean weighted with MDER (blue). For each mean, the vertical lines indicate the first year below 1 million km² of sea ice extent. Gray shading indicates the areas of +/- 1 standard deviation around each mean. Adapted from Senftleben et al. 2020, Figure 7.

The MDER method (Multiple Diagnostic Ensemble Regression, Karpechko et al. 2013) can be used to narrow down this range of uncertainty. First, a complex algorithm iteratively determines the most relevant parameters for the prediction of sea ice extent. From a pool of 15 different diagnostics, MDER has identified two: mean sea ice extent and trend in Arctic air temperatures. The climate models are then weighted according to their observed capability of reproducing these two aspects in the past. With this approach, the span in the projected year of a first-time ice-free Arctic was narrowed down by 15 years. Additionally, the weighted mean is 14 years earlier, in 2062 (see blue lines in Figure 1).

The MDER method is only one of many options to yield deeper insights from the widespread simulation results of the Arctic climate – however, almost all of them point to a more pessimistic outlook on the Arctic’s future than the arithmetic mean suggests. Whereas the observed retreat of Arctic sea ice over the past few decades has been far more pronounced than the one predicted by climate models, this discrepancy has been factored in and corrected by the MDER model.

In the context of our nxt statista projects, these findings demonstrate that it is not enough to simply average available, diverging data sources, even if they seem to be of similar quality. It is crucial to identify the most relevant parameters for answering the question at hand and then assign greater weight to those data sources that more accurately represent these parameters. In an economic context, it is vital to also rely on expert knowledge when selecting possible parameters in order to conduct analyses in the most efficient way.

References

https://www.bbc.com/news/science-environment-54211760

https://www.unenvironment.org/news-and-stories/story/another-wake-call-sea-ice-loss-speeding

https://edition.cnn.com/2020/09/14/europe/greenland-arctic-ice-shelf-intl/index.html

Karpechko, A. Y., D. Maraun, and V. Eyring, 2013: Improving Antarctic Total Ozone Projections by a Process-Oriented Multiple Diagnostic Ensemble Regression. Journal of the Atmospheric Sciences, 70, 3959-3976. https://doi.org/10.1175/jas-d-13-071.1

Senftleben, D., A. Lauer, and A. Karpechko, 2020: Constraining Uncertainties in CMIP5 Projections of September Arctic Sea Ice Extent with Observations. Journal of Climate, 33, 1487–1503, https://doi.org/10.1175/JCLI-D-19-0075.1.

Next article

Fighting COVID-19 with data

The importance of medical data for combating diseases has never been more evident than during the corona pandemic. It is not only about recording the outbreak in as much detail as possible and on a real-time basis; the disease needed to be understood in the first place by exchanging observations and findings, and best-practice benchmarks had to be developed pertaining to specific patient groups and symptoms. This process is far from being completed, especially with regard to long-term health implications.

Thilo Löwe, 2020-10-08
Logo© 2019 - 2020 LSP Digital GmbH & Co. KG
ImprintPrivacy