Ice core CO2 and temperature, lead or lag, ongoing saga
Argument over whether the two long ice cores, EPICA and Vostok are evidence of CO2 leading or lagging temperature has been going on ever since the core data was published. I assume the reader knows this.
A snippet has turned up as a result of an article at Bishop Hill.
Republished images as an aide memoir for the subject
Plots, own work. Pertinence, read on.
The Bishop Hill article is from an anonymous teacher of statistics where a cross-disipline dispute arose.
He set an exercise for a student on ice core data where together they found what was news to them. When this was reported to the climatic people, who are not statisticians, a further revelation appeared… what happens when dogma is challenged.
Changing the topic slightly…
A comment on the BH article mentioned similarity and dropped a link to a discussion paper by statisticians at Exeter University.
Figure 1 above is for resampled data to a common timebase, there are serious quantisation problems.
The discussion paper, please do not take as formal definite, mentions the sample problem at length.
On temperature proxy data
While the mean interval between observations is 138 years, the older observations are considerably sparser than the recent ones, as can be seen in the time plot of these intervals in Figure 3(a). However, the intervals exceed 1000 years on only about 25 occasions, at the earliest dates.
Whereas on CO2
The CO 2 measurements, on the other hand, number only 1098 to cover the same 800,000 year period. This is the concentration of the gas in trapped air bubbles, representing a much smaller proportion of each sample than water. Also, this is a composite series combining data from the EPICA and Vostok sites. The intervals between the observations are plotted in Figure 3(b). Don’t overlook the large di¤erence in both the vertical and horizontal scales in these plots. The mean interval between CO 2 measurements is 729 years, but over 40 of the intervals exceed
2000 years. These gaps presumably correspond to periods of low precipitation, so that a shorter time span yields too small a sample for analysis.
I signal processed to 1ky if I recall correctly. The effect will be similar to the method they use.
May as well mention part of the concluding remarks
(discussion paper, is not to be taken as assertion is so)
However, we should conclude with a cautionary remark regarding interpretation of the findings. For the reasons discussed in Section 4, the rejection of Granger non-causality means that without additional information about the climate system, we are not able to conclude very much about the direct interactions of our observable variables. Consider, for example, using the equations to simulate the effects of a shock to CO 2 levels. This would certainly show a response by temperature, just as a shock to temperature would show a response by CO 2 . However, our inability to impose ceteris paribus conditions means that such an exercise could tell us little
about the outcome of a controlled experiment, such as the injection of anthopogenic CO 2 into the atmosphere.
In other words not a lot can be drawn from the data. I agree although detail work not shown here on the two datasets shows CO2 leading in both cases even though opinion says it doesn’t for EPICA, I found it does, snag is the vague nature much of the time. Seems to way of much science, data is never quite good enough.
Now for external links…
h/t to Bishop Hill and mikep
James Davidson gave a paper on this issue at Ross Mckitrick’s conference in Guelph. It can be found at his website here http://people.exeter.ac.uk/jehd201/research.html. It seems to be hard to say anything definitive.
Jun 16, 2014 mikep
Discussion paper head
Modelling the Interactions in Paleoclimate Data from Ice Cores
University of Exeter
CPTEC/INPE, São Paulo
August 21st 2013
Oh well, conclusions on ice again.