Experiments with the Vostok CO2 and T data

My copy of the data is here: t and co2.

The originals are at: ftp://ftp.ngdc.noaa.gov/paleo/icecore/antarctica/vostok/co2nat.txt and ftp://ftp.ngdc.noaa.gov/paleo/icecore/antarctica/vostok/deutnat.txt.

Though I have referred here to "Vostok T", you'll remember of course that it is a proxy T from isotope composition and is subject to correction.

Please note: this is not carefully considered scientific research. This is just me playing with the data!

Also note: Please: if you find any of this interesting, do check my code before believing it: I'm prone to minor errors...

A note from someone who is wise:

See: Cuffey, K.M., and F. Vimeux, Covariation of carbon dioxide and temperature from the Vostok ice core after deuterium-excess correction, Nature, 412 (6846), 523-527, 2001.

In this, they correct the deuterium (temperature) profile to allow for the change in isotopic composition of the ocean, and this gets rid of quite a lot of the annoying disjoint between T and CO2 at coolings.

Note that the data I have here is without this correction.

My IDL pros to work with the data are:

@one (read in and interpolate)
two (calculate running corr)
.run three (plot pic)
@four (plot spectra)
@five (co2 and t ages)
.run six (co2 and t lagged corrs)

Note that since T and CO2 are not on the same age scale they need to be interpolated to a common scale. For ease, I use a scale regular in time.

"Results"

In quotes because this is just playing.

This shows the T (black) and CO2 (blue), interpolated to a common timescale (intervals of 100 years).

Overplotted are n-hundred year running correlations, where n=101 (green), 201 (red), 301 (light blue) , 401 (dark green), 501 (brown). The coloured bar drawn in shows the length of the running correlation.

We see...

Corrs are fairly high most of the time at 50 kyr scale.
On 10 (20) kyr scale correlations are often (sometimes) negative.
For the 140 kyr and 240 kyr terminations, the max corr shifts backwards in time at longer timescales: I would attribute this to the corr responding to the large jump, rather than individual variations. But at the 330 kyr event, the corr stays centered over the event, and is broader at 50 kyr scales.

And lastly, a look at the spectra:

Note logarithmic y-axis. X-axis needs *100 to get to correct scale.

Oddly enough, the 100 kyr peak dominates, as it always does... lines are attempt at red-noise fit, and above them should be sig.

Blue is T, black is CO2. Note that relative strengths reverse at 40/100 kyr peaks.

Notice (below) that T is measured far more finely that CO2. There are 283 CO2 samples in 400 kyr and 3311 for T. See the graph: this shows that the difference in successive CO2 ages is up to 6 kyr and irregular; the diff in T ages rises slowly to about 600 yr.

This has implications for the averaging onto 100 year common timescale.

Finally, lets look at what happens when you allow the T and CO2 to shift in time. We'll pick 20 kyr running correlations and allow shifts of up to 4 kr (40 points).

The top pic shows T (black) and CO2 (blue) again; with the running corr (green) and the "best" running corr achieved, point-by-point, by allowing the CO2 data to slide against the T data thus:

    c1(k+shft)=correlate(t1(j),co21(j+k))

so that -ve k implies T correlated against later (in time, ie closer to now) CO2. The bottom plot shows the value of k that produces the "best" corr. Note that it sometimes "bottoms out" (but never tops out). Fitting a trend line by eye, and ignoring the big jumps, it looks like k goes from -10 now to +10 at 400 kyr; ie a change from T leading CO2 (now) to T lagging CO2 (then) (I do hope I've got that the right way round).

Please, remember again: if you find any of this interesting, do check my code before believing it: I'm prone to minor errors...

wmc@bas.ac.uk / http://www.antarctica.ac.uk/met/wmc/