In a recent post, I noted the discrepancy between the UK Met OFfice contribution to IPCC AR5 and observations (as many others have observed), a discrepancy that is also evident in the “initialized” decadal forecast using the most recent model (HadGEM3). I thought that it would be interesting to examine the HadGEM2 hindcast to see if there are other periods in which there might have been similar discrepancies. (Reader Kenneth Fritsch has mentioned that he’s been doing similar exercises.)
In the figure below, I’ve compared HadCRUT4 (anomaly basis 1961-1990) to the Met Office CMIP5 contribution (red), converted to 1961-90 anomaly.
Figure 1. IPCC CMIP5 contribution (HadGEM2 RPC45 average) vs HadCRUT4.
There is a persistent over-estimate over the first half of the 20th century, particularly in the 1920s. Nor does the Met Office model adequately replicate the temperature increase of the early part of the 20th century. In its CMIP contribution, the average temperature in the first decade (1900-1910) was -0.116, almost identical to the average temperature from 1960-70 (-0.111), as compared to an increase of 0.35 deg in HadCRUT4 (from -0.518 to -0.161).
We often hear about the supposed success of current GCMs in hindcasting 20th century from first principles. Nonetheless, quite aside from the developing discrepancy in the recent period, the apparent inability of the Met Office model used for their IPCC submissions to model the early 20th century suggests a certain amount of salesmanship in the success proclamations.
Update: An overplot of the UK Met Office CMIP5 contribution (HadGEM2) onto graphic showing distribution of CMIP5 runs.
Figure 2. Overplot of UK Met Office CMIP5 contribution on CMIP5 distribution (from Ed Hawkins here.) Slight extension of HadCRUT4 to current (thin black).
Here is another overlay of the UK Met Office CMI5 contribution onto graphic from Stott et al 2013 shown at Hawkins’ blog here. I’ve extended the current temperatures from the 2010 shown in the original graphic. One of Hawkins’ readers had asked him to do so, but Hawkins begged off saying
Couple of thoughts – firstly I think the observations would still be inside the dashed lines, just, but I didn’t make the plot so I can’t add them to check.
I didn’t make the original plot either, but it’s not that hard to an overlay (see code below.) With the updated data, observations are outside the dashed lines.
Figure 3. Overplot onto http://www.met.reading.ac.uk/~ed/bloguploads/stott_2013.png. Code is as follows (retrieval of HadCRUT4 is done elsewhere.)
loc="http://www.met.reading.ac.uk/~ed/bloguploads/stott_2013.png"
dest="d:/temp/temp.png"
download.file(loc,dest,mode="wb")
imgs=readPNG(dest)
par(mar=c(1,1,1,1))
plot(0:1,type="n",xlim=c(1850,2060),ylim=c(-1.2,2.4),xlab="",ylab="",axes=FALSE)
rasterImage(imgs,1841.2,-1.38,2064,2.37)
abline(v=seq(1800,2050,50),lty=3)
abline(h=seq(-1,2,1),lty=3)
points(2010.5,.217,pch=19,col=2)