The last interglacial is the early Holocene on steroids. Changes in the Earth’s orbit caused high latitudes to received increased summer insolation and they were consequently even warmer than the early Holocene. A large fraction of the Greenland Ice Sheet melted and sea levels were several metres higher than today.
These differences between the Holocene and the last interglacial make it an important target for palaeoclimate research that helps us to understand how climate varies under different forcing and testing for how well climate models can estimate past climate. If they can do this well it should enhance our confidence in their predictions for the future.
Unfortunately the last interglacial occurred so long ago that radiocarbon dating, the most widely used dating technique, cannot be used as essentially all the radiocarbon has decayed (the last interglacial was over 20 half-lives of 14C ago, the limit of 14C dating is ~10 half-lives). Without precise chronologies it is not possible to use proxy climate data to reconstruct the spatial and temporal dynamics of the last interglacial and estimate global mean temperature change over time, as can be done for the Holocene. Instead, a snapshot of last interglacial climate is estimated, using the simplifying assumption that the warmest temperature in each record occurred simultaneously. This snapshot estimate is obviously biased, peak warmth is unlikely to be synchronous, and consequentially climate models cannot match this estimate. Cue merriment and misunderstanding from fake climate sceptics.
Bakker & Renssen (2014) have a new paper in Climate of the Past that explores this bias using transient runs of climate models for the last interglacial. For each run, they calculate the maximum global temperature in three ways. First they calculate the global mean temperature and pick the warmest 50 year period. This is the real maximum temperature in the model. Second, the find the warmest temperature for each model grid cell, whenever it occurred in the last interglacial, and take the mean of these. Third, they repeat the second metric with the warmest month temperatures to reflect the seasonal biases of many proxies.
The bias between the proxy data and the climate models’ estimate for last interglacial maximum temperature is 0.67 °C. The overestimation in the climate models of using the mean of the warmest annual temperature at each grid cell is 0.4 ± 0.3 °C, insufficient to explain the proxy-model difference. When the warmest month in each grid cell is used instead of the annual mean, the overestimation is 1.1 ± 0.4 °C. This suggests that the chronological uncertainty and seasonal biases in the proxies are sufficient to explain the 0.67 °C offset between the proxies and the models. The models are performing well at the global scale (at a regional scale there are still problems).
Anthony Watts got confused by the abstract, believing that it shows the opposite of what it really shows, and calls the paper “inconvenient” – code for opposing the consensus.
A new paper published in Climate of the Past compares temperature reconstructions of the last interglacial period [131,000-114,000 years ago] to climate model simulations and finds climate models significantly underestimated global temperatures of the last interglacial by ~0.67C on an annual basis and by ~1.1C during the warmest month.
The ~0.67C is correct – it is the proxy-model difference. The 1.1C is not the underestimation the warmest month, it is the bias caused by chronologically uncertain and seasonally biased proxies. He thinks the paper is showing the models are performing badly whereas it is actually showing them perform fairly well.
This implies that climate models are unable to fully simulate natural global warming, and the error of the underestimation is about the same as the 0.7C global warming since the end of the Little Ice Age in ~1850. Thus, the possibility that present-day temperatures could be entirely the result of natural processes cannot be ruled out in comparison to the last interglacial period.
This is just nonsense, and it doesn’t get any better.
Bakker & Renssen (2014) is not a particularly difficult paper. It’s not very long and there are no equations. Watts read it with his enormous set of biases and completely misinterprets the paper, reporting the converse of what the paper actually finds. I don’t suppose he read past the abstract except to find a juicy figure, if he did, he certainly didn’t understand anything.
There is currently a glitch on the COP website as the PDF for Bakker and Renssen (2014) cannot be downloaded (though Watts obviously has a copy as he shows and mis-describes table 2), so I used the discussion paper to help write this post.