Duelling models
Aug 28, 2015
Bishop Hill in Climate: WG2

Climate models eh? If nothing else, they are at least the cause of some wonderfully convoluted arguments among climate scientists as they pore over the outputs like psychics gazing at the tea leaves in a saucer.

Recall, if you will, the story over the Sahel. GWPF put out a briefing pointing out that earlier claims that global warming was going to lead to perennial drought there were nullified by a long period of greening. This led to a cod debunking from Desmog, but later to a paper by Dong and Sutton, which claimed, on the basis of a GCM study, that the greening was in fact caused by greenhouse gases (background here).

The latest development is a new paper (£) by Alessandra Giannini of the Earth Institute at Columbia, which critiques Dong and Sutton. Her case is that there is a well-established link between sea-surface temperatures and Sahelian drought:

The generalized warming of the oceans attributed to GHGs, that emerged around 1970, is understood to have exacerbated drought persistence in the 1970s and 1980s3,7,8.

and because Dong and Sutton's computer model doesn't reproduce this link in its hindcast it is not to be trusted:

[O]ne reason why the model used by Dong and Sutton emphasizes the direct effect of GHGs may be its underperformance in reproducing the effect of historical SST on Sahel rainfall. Testing the version of this model made available in CMIP5 finds it does not reproduce the conclusion of SST influence reported in the literature (see refs 3,13, for example) that provides the scientific basis
for seasonal prediction.

Amusingly, the alleged underperformance is highlighted by means of comparing its hindcast, and that of a multimodel mean, to observations.

The multi-model ensemble mean correlation with observations is 0.46, significant at the 5% level. The correlation of the single simulation with HadGEM2 (solid orange line) is 0.30, not significant at the 5% level.

Perhaps a statistician might like to comment on how much weight we should put on a correlation of 0.46 as compared to one of 0.3, particularly in an autocorrelated time series.

So when we learn from reference 3 - also by Gianniani - and reference 13 that the "understood" link between sea-surface temperatures and the Sahelian drought is based on a computer modelling studies in which ocean temperatures were tweaked to see the effect on Sahelian rainfall, all we can do is stand in open-mouthed disbelief.

As everyone knows, climate models are barely capable of doing rainfall, and even less so at a regional level. Yet, if I understand correctly, Giannini is arguing that Sutton's model is wrong (a) because it doesn't reproduce things that other models do and (b) because its hindcast is marginally further away from the observations than the multimodel mean.

Is anyone actually convinced by this?


Article originally appeared on (http://www.bishop-hill.net/).
See website for complete article licensing information.