Recent Climate Observations Compared to Predictions by Rahmstorf et.al. - a review

Rahmstorf et.al. claims that previous projections of the IPCC have underestimated climate change, particularly temperatures and sea level rise. The conclusion of the brief paper has been used to justify more vigorous action on global warming by the Garnaut review, on the grounds that global warming is ‘worst that we previously thought’. But does Rahmstorf et.al. hold water?

The method used is non-linear fitting of a trend to temperatures. The conclusion is from a marked increase in the trend in recent years beyond IPCC decadal projections. However, there appear to be many statistical black holes.

Lucia has been trying to verify the provenance of the IPCC projections as they are not described explicitly in paper, with no success.

The method is described only as

a nonlinear trend line computed with an embedding period of 11 years and a minimum roughness condition at the end

with a reference to Moore et.al. “New Tools for Analyzing Time Series Relationships and Trends“. The non-linear method appears to be singular spectrum analysis (SSA). This method decomposes a time series into orthogonal components, such a a periodic, linear trend and random noise parts. Rahmstorf et.al. also mentions a non-SSA ‘minimum roughness condition applied at the end’ without further elaboration. Any ad-hoc techniques appled to the end of the series should raise a red flag, as the part of the series of interest is the recent temperatures at the end.

Digging into Moore et.al. we find the ‘minimum roughness condition’ consists of ‘padding the series so the local trend is preserved’. The reference to the padding the data is none other than Mann 2004, “On smoothing potentially non-stationary climate time series“!. Figures. Mann has been outed for various dubious extrapolation activities at ClimateAudit here).

Second, the unique adjustment was not disclosed in the MBH98 footnotes. Worse, the start date of this series was actually misrepresented in the original supplementary information, which listed the series as starting at the “adjusted” start date rather than the true start date. We only noticed the extrapolation when we compared the Mann version to original data. We noted this in MM 2003, but were not then fully aware of the impact.

Amazing what you find with a bit of digging. It seems like the ‘minimum roughness condition’ is simply a fancy name for the extrapolation of the trend of the last few data points in the series. In that case, the conclusions of the paper — that observed trends surpass the IPCC projections — are based on only a few end points, raising the specter of ‘cherry-picking’.

Another opportunity for cherry picking is in the selection of the 11 year embedding interval in the SSA. No justification is provided in Rahmstorf for the choice. Although Moore et.al. does provide the helpful suggestion:

A wise choice of embedding dimension can be made with a priori insight or perhaps more commonly found by simply playing with the data.

Wise choice or parameter tuning?

Not having assess to SSA, as I can’t find it among the R packages at CRAN, I examined the degree that conclusions of Rahmsdorf et.al. might be cherry-picked by looking at how trends vary with length of series. Below is the plot the HadCRU temperature trend for increasing intervals starting from the present, starting with three years (2004, 2005, 2006), then four years (2003, 2004, 2005, 2006) and so on. Below is the code (which I am not proud of but shows you can program R in proceedural style when you want to).

Figure: Above is the plot the HadCRU temperature trend for increasing intervals starting from the present, starting with three years (2004, 2005, 2006), then four years (2003, 2004, 2005, 2006) and so on.

The plot shows clear peaks with a trend of +0.37C per decade at 16 years trend length (1990-2006) and declines to the long term trend of around +0.1C for longer intervals. The elevated trends reported in the paper and shown on the figure are clearly a reflection of local trends in this period. However, there are no significance tests quoted for the trend!. Unlike other examinations of IPCC projections here and here, no attempt has been made to determine if the trends are due to climate variability. As reported, trends are declining, as expected if the increase was simply a random fluctuation.

There are multiple potential problems with Rahmsdorf et.al.:

• Uncertain data provenance
• Series padding
• Parameter tuning
• Lack of significance

These are additional reasons why Garnaut’s Climate Review Interim Report is not convincing. The urgency of the message in the review appears to be influenced by Rahmsdorf’s speculative work, but admits:

The Review team’s report says that they have neither the skills nor resources to re-visit the climate science.

Even a quick review of Rahmsdorf et.al. shows statistical deficiencies and speculation that would probably have blocked publication if it didn’t prop up the IPCC and claims of global warming catastrophe. Does Garnaut mean there are no qualified statisticians willing to work in Canberra?

Code

readCRU< -function(source="http://hadobs.metoffice.com/hadcrut3/diagnostics/global/nh+sh/monthly",temps=2:13,plot=T) {
f<-read.table(source,fill=TRUE)
d<-as.vector(t(as.matrix(f[seq(1,length(f[,1]),by=2),temps])))
#d<-c(d,-0,-0)
ts(d,frequency=length(temps),start=1850)
}

rahmstorf<-function() {
t<-readCRU(temps=14)
trendCy<-rep(NA,100)
for(i in 4:100) {
	x<-1:i
	y<-t[(length(t)-i+1):length(t)]
	l<-lm(y~x)
	trendCy[i]<-l$coefficients[2]
	}
	plot(trendCy,type="l",lab=c(10,5,7))
	trendCy
	}