Thanks Crandles. My response, quotes edited for legibility.
However, the modeled sea-ice loss in most CMIP5 models is much weaker than observed. Compared to observations, the simulations are weak in terms of their sensitivity to both global mean temperature rise (Rosenblum and Eisenman, 2017) and to anthropogenic CO2 emissions (Notz and Stroeve, 2016).
That makes perfect sense. Arctic sea ice is decreasing faster than the models predict because influence from positive feedbacks like albedo warming, intrusions of hot air and winter cloudiness are overtaking both the global warming signature and the CO2 signature.
This mismatch between the observed and modeled sensitivity of Arctic sea ice implies that the multi-model-mean response of future sea-ice evolution probably underestimates the sea-ice loss for a given amount of global warming.
The first mistake is to associate Arctic Sea Ice with global temperatures. If Antarctica cooled 10C and the Arctic warmed 10C the Arctic would melt with zero degrees of global warming.
To address this issue, studies estimating the future evolution of Arctic Sea Ice tend to bias correct the model simulations based on the observed evolution of Arctic sea ice in response to global warming.
Bias correct... I honestly don't know exactly what that means, but that never stopped me from speculating before, so here we go.
I assume bias correction involves tinkering with parameters and functions until the model produces a better match for the observations. Then the future results are expected to produce better result. If that's the case, I understand the scientific validity and necessity of bias correction. It is a process of perpetual improvement.
Often based on such bias correction, pre-AR5 and post-AR5 studies agree that for 1.5 °C global warming relative to pre-industrial levels, the Arctic Ocean will maintain a sea-ice cover throughout summer for most years
Bias correcting a model that has the wrong shape (strait line vs exponential) will still produce the wrong result. How do we know if the model has the wrong shape? We don't, unless a model with a different shape shows better skill.
But what if a different shape can only be resolved adding so many variable that the model can't be computed? Or what if assumptions taken as invariable because of hundreds of years of climate data are no longer valid when the climate changes? Unknown unknowns.
In particular, the relationship between Arctic sea-ice coverage and GMST is found to be indistinguishable between a warming scenario and a cooling scenario. These results have been confirmed by post-AR5 studies (Li et al., 2013; Jahn, 2018), which implies high confidence that an intermediate temperature overshoot has no long-term consequences for Arctic sea-ice coverage.
So I went looking for the confirmation studies and found what I believe to be Jahn, 2018. This is part of the abstract:
For warming above 2 °C, frequent ice-free conditions can be expected, potentially for several months per year. Although sea-ice loss is generally reversible for decreasing temperatures, sea ice will only recover to current conditions if atmospheric CO2 is reduced below present-day concentrations.
https://www.nature.com/articles/s41558-018-0127-8This paper says that the ice will be gone 2 months during summer, so from mid July to mid September, but there will be no hysteresis. That is quite simply unbelievable and it doesn't even pass a sanity check.
Even more unbelievable is "sea ice will only recover to current conditions if atmospheric CO2 is reduced below present-day concentrations."
The sea ice disappearance might have started because of CO2, but the acceleration of sea ice loss is not because of CO2, as the failure of the models prove. The loss of sea ice is now mostly do to albedo feedback, jetstream destabilization and local Arctic GHG, not CO2. Reducing CO2 back to historic levels will eventually restore the ice, but that will take decades or centuries.