Guest posting by (or rather, ripped from) Eric Wolff.
It is indeed a very fundamental question about whether the CO2 leads or lags the temperature. If there was somewhere in the ice core record where CO2 increases and temperature does not, then our understanding of the greenhouse effect must be faulty. However, so far we don’t find such a place. [*]
[*] Eric is a scientist, not a lawyer. His words, whilst essentially still valid, were not carefully enough framed. He writes (2012/4): I should have carefully included the words “all othe things being equal” and “significantly” as in: “If there was somewhere in the ice core record where, all other things being equal, CO2 increases significantly and temperature does not, then our understanding of the greenhouse effect must be faulty. However, so far we don’t find such a place”. This is to cover the case of the last 6 kyr-to-preindustrial, where CO2 has risen a bit (though very little, by comparison with iceage-interglacial changes), and T has stayed more-or-less-flat.
Unfortunately in detail the phasing between CO2 and temperature rise is a difficult question to answer. Here is some technical stuff about that which you can ignore if you like, but explains why it is difficult. The reason is that at a given depth in the ice core, the gas has a different age from the ice. the reason for this is that the “temperature” signal is laid down at the surface of the ice sheet. The gas bubbles form only at depth (typically about 100 metres) when the weight of snow on top of a layer has compressed the snow into ice with trapped bubbles. The result is that at a depth (in out idealised ice sheet) of 100 metres: the air is very young (say 30 years, the time for air to diffuse through 100 metres of snowpack) the ice is rather old (if the snow accumulates at 3 cm per year, the ice could be about 3000 years old) and similarly at 200 metres, the air is 3030 years old and the ice 6000 years. This difference between the ice age and the air age can vary, and so it has to be calculated; and there is an uncertainty in that calculation. This all leads to the statement that there is an uncertainty of a few hundred years in the phasing between temperature and CO2.
After this caveat let’s turn to the data, and to the place where we know best this age difference, coming out of the last cold period 20000 years ago. I attach a figure from a paper by Monnin et al that appeared in the journal Science. On the top you see the detail of Antarctic temperature (actually water isotope content representing temperature) and in the middle CO2. You can see that they move absolutely in parallel. A statistical analysis does suggest that temperature slightly leads by a few hundred years and you can see a hint of that, as the curve at the top starts to rise just before the vertical dashed line while the CO2 only starts at the line. However, the fact that they move in parallel for most of the 5000 year increase, even tracking each other in the hiatus period is the characteristic pattern of a chicken-and-egg positive feedback, where the temperature causes the CO2 to rise and the CO2 causes the temperature to rise further. Of course we cannot prove that is what is happening, but it is consistent with that, and the crucial point is that there is no period when CO2 is rising and temperature is not.
Just to be clear: no-one is claiming that temperature cannot change naturally, clearly it does. Thus there is no surprise in the idea that the end of the ice age was kicked off by something other than CO2. however once it started, CO2 appears to be involved in keeping it going. There is still some question whether there is any lag at all (because there is some evidence we may have calculated too large an age difference, think of it as 800 years +/- nearly 800), but if there is a lag what that means is that it takes time for the ocean and land to give up CO2 in response to an initial temperature warming. But that is not an issue for the future since we are actively adding the CO2 to the system ourselves, so that half of the feedback is taken care of.
[Update: its worth pointing out that the 800-y lag stuff is *not* definitively established. Loulergue et al argue otherwise (and since its an open-access journal you can go argue with them if you like!) -W]