Bishop Hill Climate heroism
Aug 1, 2012 Climate: Statistics More mathematically inclined readers will be interested in a discussion paper by Jonty Rougier, a colleague of Tamsin Edwards' at Bristol. I met Jonty while I was at the Met Office earlier in the year and found him very engaging as well as having a very sharp mind.
The paper is rather mathematical for me. It discusses the difficulties of getting a calibration from systems that are sensitive to initial conditions and where there is an attractor, describing the difficulties as almost intractable. It then goes on to list the further complications that are found in environmental systems and in particular in paleoclimate, and concludes:
When these additional complications are added to the intractability of palaeoclimate reconstruction (climate de finitely has sensitive dependence on initial conditions and an attractor), that enterprise must be seen as heroic in the extreme, and we must expect the uncertainties to be very large indeed. But, somewhat surprisingly, they are not; e.g., as shown in the celebrated hockey stick, which was used so much in the Third Assessment Report of the IPCC (Houghton et al., 2001); Montford (2010) provides a readable if slightly hair-raising account.
Reader Comments (52)
This is about paleoclimate reconstructions rather than about GCMs, but it does indicate the extent to which climate scientists may feel free to jiggle the datasets until the get the results looking as desired:
Neukom tells Phil Jones in Climategate email how hard he works to get just the right recipe of SH proxies
Wow! Rampant confirmation bias indeed, at least the potential is clearly there and pervasive.... This is regarding a previous paper (on which P. Jones ended up as a co-author). It is a most interesting exchange as it indicates that both Neukom and Phil Jones are well aware of how to run tests until one gets the desired list of proxies. Searching only for statistical "skill" or also ripe with potential confirmation bias??
Thank you to Paul Matthews for alerting me to a very sloppy piece of writing in this draft. I did not express my own views at all well, and nor did I do justice to the stated views of M&W.
I expressed some reservations about the M&W study in my published comments that followed the paper. My main concern is whether it is possible to derive reliable measures of uncertainty from the 'backwards' approach which reverses the dominant causal direction by regressing temperature on the sensor's response to climate variation. The framework I describe in my draft paper goes in the 'forward' direction. A strong case can be made for constructing one's statistical model in the forward direction; this is a subject that has been carefully explored by Judea Pearl.
But I do not expect the problems that I describe in my draft paper to feature in climate reconstructions for the last millennium, when considered at the hemispheric spatial scale and decadal temporal scale. On this scale I would treat climate as a smooth process, as done by John Haslett and colleagues, in
J. Haslett, M. Whiley, S. Bhattacharya, M. Salter-Townshend, J.R.M. Allen, B. Huntley, F.J.G. Mitchell and S.P. Wilson, Bayesian palaeoclimate reconstruction, Journal of the Royal Statistical Society, Series A, 169, (3), 2006, p395 - 438.
The weakness of this approach is that it does not permit the inclusion of changing forcing, such as orbital variations in solar insolation, and atmospheric CO2 concentrations. So it may break down for longer time-periods where these become an important feature. The approach in my draft paper (and the work of Michel Crucifix and colleagues on glacial cycles) does incorporate these: it is the long time-periods (eg 400kyr) which enable the dynamics to cause problems for the statistics.
I will revise this passage in the draft paper before publication. Thanks again to Paul and other discussants for correcting me.