During multiple generations of CMIP models, arctic sea ice trends predictions have gone from being too stable to the right. Because?
The outstanding diagnoses on our page comparison page are currently based on temperature, and in general they show that climatic models have been working well in these trends for decades. But there has been increasing attention to trends in unbeated variables, and there, model performance is more mixed (Simpson et al., 2025). As we have discussed before, the model’s observation discrepancies can arise from three causes: the observations could be wrong (unrealized biases, etc.), the models are incorrect (which can cover errors in forced and physical), or the comparison could be inappropriate.
One of the highest profile ‘failures’ in the modeling coupled in recent decades was the failure of the model projections in CMIP3 (around 2003/4 vintage) to match (Stroeve et al., 2007)And they worsened for the record losses of summer ice in 2007 and then in 2012. With an additional decade, how do you see now?
In a word, the CMIP3 Arctic Marine ice projections were and remained terrible. The prostage exchange rate of the September average of the extension of the arctic sea ice is lower than half of the observation (-4.5 %/decade compared to -11 %/decade for 1979-2024), and there are only five simulations of individual models (of 46) that have a loss rate greater than 10 %/decade (95 %diffusion is is is is is is is is is is is there [-12,-0.7] %/decade). March trends are also in prediction, but to a lesser extent. There is no real ambiguity in the trends observed, nor in the comparison (although the extension is a bit more complicated than the area to be compared), so these discrepancies were very likely due to model failures, an insufficient resolution to capture the polar dynamics of sea ice, too simple, the bies in the simulations of the Arctic Ocean Marino.
The development of CMIP5 models was ongoing since these discrepancies manifest, and there were improvements in the physics and dynamics of sea ice, the increase in resolution and a reduction in general climatic biases. The simulations in CMIP5 were carried out around 2011-2013, and used historical forcing from until 2005, and the scenarios later. Did that make any difference?
Closer, but there is no cigar. The propagation in the CMIP5 models is greater (a function of greater variability), and the observations are now more within the propagation, but the average trend of the September set (-8%/decade) is still too low. But almost 40% of the 107 individual simulations (95% of CI is [-20,-1.4]%/decade) now have losses greater than 10%/decade. March trends are mostly well represented, but there are still great variations in the absolute extension.
There was a longer gap before CMIP6, but those models developed until 2017/8 or so, so the developers were very aware of continuous discrepancies. (Stroeve et al., 2012). Again, there were improvements in sea ice physics, dynamic schemes, forcing (the addition of black carbon impacts on snow and ice albedo, and again, improvements in resolution and base weather.
As a minor side, from 2007 to 2014 there were a series of reviewed statements not peer of a few scientists (Peter Wadhams and Wiesław Masłowski remarkably) that used non -linear statistical adjustments to the marine ice indices observed to predict the essentially free of ice conditions for 2013, or 2016 more or less. These predictions were not based on any physical understanding or model, they were highly criticized by other scientists at that time (I remember a particularly spicy The meeting at the Royal Society in 2014, for example!), And (as expected) were not validated. But this kind of thing is perhaps to be expected when conventional models do not provide credible projections?
Anyway, back to CMIP6. Is the third time a charm?
Actually, this is not bad. The average of the CMIP6 set for the trends of the September area now is -11 %/decade (observed 13 %/decade) and the March trends are perfect. Note that the loss observed in the ‘area’ is slightly greater than the trend in ‘extension’ (13 %/decade compared to 11 %/decade) and I am using the area here because that is what is what it is available. The propagation of September trends is [21,3] %/decade that is slightly stricter than in CMIP5, and 40 %(again) have losses greater than 10 %/decade.
What lessons can be drawn here?
As we have often said, the models are always incorrect, but the degree to which they can be useful must be addressed: by variable or by generation of models or by integrity of the model, etc. The usefulness of the CMIP6 set (and presumably the next CMIP7 models) so that the Arctic Marine ice is clearly higher than the CMIP3 set, but there does not seem to be one thing that should be fixed for it to happen. Rather, an accumulation of improvements, in physics, resolution, integrity, forcing, has led to a gradual improvement in skill (not only in marine ice trends!).
As Simpson et al (2025) pointed out, there is a growing number of climate quality diagnoses that have long enough and emerging change signs, so there is a growing number of tests for model trends. The history of Arctic sea ice comparisons shows that it could be premature to conclude that any specific discrepancy implies that something is fundamentally incorrect, or that climate modeling is in a ‘crisis’ (Shaw and Stevens, 2025)Well, it may be that these discrepancies are resolved in the course of the development of the ‘normal’ model (already measure that the observed signals become clearer). Or no ;-).
Note on sources: CMIP3 (sea, SEP) and CMIP5 (historical, RCP45) The processed data are from Jacob Dörr (Notz et al, 2020) and Alex Jahn (through Julienne Stroeve and Patricia Derepentigny), and the CMIP6 area data are from U. of the Hamburg Data Portal (Courtesy of Dirk Notz). The whole media are in the whole set with a simulation = a vote. In addition, I have not examined CMIP6 models by climatic sensitivity (as I have done for temperatures). These options can make small differences, but not affect the main conclusions.
References
-
IR Simpson, TA Shaw, P. Ceppi, Ac Clement, E. Fischer, Km Greece, Ag Pendergrass, Ja Screen, RCJ Wills, T. Wooolings, R. Blackport, JM Kang and S. Po-Chedley, “confront the trends of the earth systems model with observations”, “,, Scientific advancesvol. 11, 2025. http://dx.doi.org/10.1126/sciadv.adt8035
-
J. Stroeve, Mm Holland, W. Meier, T. Scambos and M. Serreze, “Decline of the Arctic Marine Ice: faster than the prognosis”, Geophysical Research Lettersvol. 34, 2007. http://dx.doi.org/10.1029/2007gl029703
-
JC Stroeve, V. Kattsov, A. Barrett, M. Serreze, T. Pavlova, M. Holland and Wn Meier, “Trends in the extension of the sea ice of the Arctic of CMIP5, CMIP3 and observations”, Geophysical Research Lettersvol. 39, 2012. http://dx.doi.org/10.1029/2012gl052676
-
Ta Shaw and B. Stevens, “the other climatic crisis”, Naturevol. 639, p. 877-887, 2025. http://dx.doi.org/10.1038/s41586-025-08680-1
-
D. Notz and S. Community, “Arctic Sea Ice in CMIP6”, Geophysical Research Lettersvol. 47, 2020. http://dx.doi.org/10.1029/2019gl086749
#Arctic #seafood #planned #trends #time
