Is the Arctic September Sea Ice Doomed to Disappear in the 2030’s?

From Climate Etc.

by Frank Bosse

Short answer: NO

A recent paper Kim et al. (2023), hereafter K23, got some media attention, e.g. this article at CNN: “The Arctic may be sea ice-free in summer by the 2030’s, new study warns.”

K23’s key conclusion: “Results indicate that the first sea ice-free September will occur as early as the 2030s–2050s irrespective of emission scenarios.“

How did the authors come to this conclusion? They used the CMIP6 Multi Model Mean (MMM)! In the methods section: “We use multi-model CMIP6 historical and DAMIP simulations performed under different climate forcing combinations…”

DAMIP: “The detection and attribution model intercomparison project (DAMIP v1. 0)”

This is the same approach described in this blogpost  (for the case of the Antarctic overturning oscillation), with the same pitfalls as described by Gavin Schmidt in 2021.

Gavin’s conclusion: “The default behavior in the community has to move away from considering the raw model ensemble mean as meaningful.” Well, this is not simply Gavin-wisdom, but the IPCC AR6 WG1  did not use the MMM, for the same reasons.  This practice has unfortunately not arrived some parts of the community, not to mention the editorial board of “Nature Communications”

To show the impact of the choice of the CMIP6 MMM vs. the not so skewed CMIP5-models, I compared the trend slopes 2020-2050 for both cases for September in the Arctic region:

Fig.1: The September Temperature trend slopes 2000-2050 (K/year) for the CMIP5 MMM (top) and the CMIP6 MMM (bottom). The figure was generated with the KNMI climate explorer.

Note the about 30% steeper trends in CMIP6!

The key figure of K2023:

Fig. 2: Reproduction of parts of Fig.4 of K23.

The authors “scaled” the CMIP6 MMM to the observed Sea Ice Area (SIA) 1979-2019 because the MMM produced too much Arctic Ice in this time. Indeed, the modelled temperatures for 1979-2019 were cooler than the observed data. The reasons are unclear, probably was estimated too much (cooling) aerosol forcing in this time in the forcing data of the CMIP’s.

After the “scaling” (colored lines in Fig2) also the “Sustainable development scenario“ SSP1-2.6 leads to a vanishing SIA (below 1 Mio km²) around 2050, for the SSP2-4.5 it’s almost the same (2045) and  also the August Sea Ice will be doomed before 2060! Astonishing, but highly dubious due to the choice of the MMM.

Here is my method to calculate a possible September SIA below 1 Mio km².  I regressed the NSIDC- Data (September) with brand new ERF (effective radiative forcings) -data from this paper (still Preprint) for 1979 to 2022.

Fig. 3: The regression of the SIA data vs. the ERF data. Note that the variance of the ERF only defines 55% (R²) of the variance of the September SIA.

The resulting Sea Ice Area sensitivity for doubling CO2 (ERF= 3.9 W/m², following IPCC AR6) gives -3.69 Mio km²/2*CO2. The likely 17…83% range: -3,16…-4,13 Mio km²/2*CO2, calculated with the CI of the regression.

About 45% of the SIA is influenced by internal variability, not described in any MMM because all the variability of single models is zero when averaging many model runs as it the MMM does. Therefore the use of any MMM is misleading just like the result of K23.

The calculated trend slope in Fig. 3 gives the result, that the forced part of the September SIA leads to a remaining amount of 1 Mio km² of 5.24 W/m². However, one has to include the internal variability (iv), described in the residuals between the observations and the trend due to the ERF:

Fig. 4: The trend residuals in Fig.3. The highest impact to lower observations was in 2012 with – 1.18 Mio km² below the trend.

I calculated hence the ERF for 2,18 km² remaining SIA to find the year when it’s possible to reach 1 Mio km² also when considering the most negative iv in the last 45 years.  This gives an ERF of 3.92W /m².

Now it’s possible to make estimations in time:

Fig. 5: The ERF data of the SSP over the time. Source

For possible single September ice free years, one can find: never for the SSP1-2.6, about 2060 in the SSP 2-4.5 scenario. A continuous September ice free arctic due to the forcing alone not considering the iv (ERF=5.24 W/m²) that we won’t see, neither in the SSP1-2.6 nor in the SSP2-4.5 W scenario. Compare this result with the result in K23 (Fig.2). My method uses only observed data, and so avoids possible biases in climate models.

A very recent paper (published 12^th of June 2023 in “Nature Climate Change”) also uses a climate model approach like K23, but takes advantage of constrained single selected models and not the MMM. It comes to very similar results like my (only ERF-based) approach: An Arctic Sea Ice Area below 1 Mio km² will be possibly observed not more 10 years earlier (as it finds K23) but 10 years later than 2050, also about 2060.

Conclusion: K23 takes advantage of the rejected CMIP 6 MMM and comes to strongly biased results. It’s overdue for the editorial board of “Nature communication” to check the peer review process to make sure not misleading the audience and the media.