Over very long periods of time as ice ages come and go, it has been found that temperature leads atmospheric CO2 content by about 800 years. This seems to contradict the IPCC and other views that CO2 causes change in temperature. But we are looking at very different time scales with present changes, so perhaps things happen differently. I decided to examine this question.
The temperature data used is monthly global land-ocean temperature or GHCN, which is available from NOAA. The atmospheric CO2 data used is from Mauna Loa in Hawaii, the longest continuous record of CO2 also available monthly.
When wanting to find the causation when two series are both increasing over time, it is best to look at the rate of change of the variables as this will show clearly which one precedes the other. This first graph shows the rate of change of these two variables monthly over the period 1958 to 2009.
Both monthly series were processed in the same way. The change over a 12 month was calculated, and a 12 month simple moving average of these values was used to avoid all seasonal effects. That data was plotted at the centre of the 23 months values used in the calculation. Because the treatment was the same for both variables, they are directly comparable.
It can be seen that there is generally a good correlation, with nearly all peaks in one variable having similar peaks in the other. When one has a smaller peak such as around 1975, then so does the other. When one has a larger peak around 1973 or 1998, then so does the other. there are one or two minor variations from this.
It is also evident that the red temperature graph generally precedes the black CO2 graph on turning points. This suggests that temperature drives CO2 and not the other way around. A comparison of the two series at different lags gives this second graph.
When the two series are coincident the correlation is quite small, r=0.13, whereas when temperature change 6 months earlier is compared to to CO2 change there is a maximum correlation of r=0.42 which is a high correlation for short period changes which have a high noise content. There is no high correlation for any lag when CO2 precedes temperature, the best being r=0.15 at 42 months.
It seems that, contrary to popular wisdom, temperature changes are driving atmospheric CO2 content changes, with a lag time of 6 months.