What Hiatus?

An Eccentric Anomaly: Ed Davies's Blog

It's widely supposed that temperatures over the last 16 years or so (since about 1998) have not risen, much. For example, a discussion of one such case: Global Warming Stopped in 1997, Yeah Right.

Yesterday Tamino published a blog post showing that actually, if anything, it's been generally warmer over this period than would be expected looking at a simple extrapolation of previous years: Global Temperature: the Post-1998 Surprise.

Of all pseudonymous bloggers, Tamino is one of the ones I'd trust most to be thorough and honest in presentation of a result like this. Still, there's always room for miscommunication and misunderstanding and this is, indeed, enough of a surprise to be worth some examination rather than just blind acceptance and repetition.

The first thing that caught my eye was the y-axis labels on the graphs: “Adjusted Anomaly”. We have to ask in what way are they adjusted. Tamino is known to be very much influenced by Grant Foster of Foster and Rahmstorf fame and it passed through my mind that these temperatures were actually their series obtained after adjustment for the effects of ENSO, volcanic aerosols and solar variability.

If that was the case I would have been very surprised if he hadn't said so in the post but there's always that little doubt - maybe he accidentally deleted the relevant paragraph, or something. Therefore, I decided to reproduce one of the graphs for myself.

I've got various datasets, and code to parse them, to hand including HadCRUT4 which is Tamino's first graph so I used that. Actually, in the process I updated to the 4.2 version to get the most recent years' data.

I cobbled together some code: hadcrut4-annual.py, linreg.py, month.py. The first of these contains a reference to the data source.

This produced, amongst others, the following graph:

Modulo graphing styles and the absence of the error ranges on my version that's a pretty good match to Tamino's graph (the second of his labelled HadCRUT4). If he has made any adjustments to the anomalies they're very small and not material.

So, I'm happy with the results and would have no qualms about quoting them. But, there's always room to look a bit deeper to see how robust they are and see if there are any further insights to be had.

(Note that in both Tamino's graphs and mine the x-axis labels correspond to the start of the year whereas the point for the year is plotted in its middle so the label for, say, 1980 appears halfway between the 1979 and 1980 points.)

The choice of 1997/98 as the split is obviously a cherry pick though not really on Tamino's part - it's the obvious choice given the common habit of claiming no/little warming starting in 1998. Still, what of other years? Here I include 1998 in the years used for the linear regression:

Now there are two “hiatus years”, that is, years at or after the split year which are below the temperature of the linear regression model at the split year. In this and following graphs such years are emphasised by having a larger blob.

We can examine a range of years in this way. For compactness an animated graph for split years 1990 through 2010, inclusive:

Yep, 1997 is a cherry-pick in the very weak sense that many other years have a few following hiatus years. The “worst case” is 2004:

which has a hiatus year 8 years after it (2012 is just below the zero line from the linear interpolation ending with 2004).

So, is 2004 the new 1998? No, of course not, with the level of noise (random variability) in the climate there are always going to be years which drop below the previous trend of a few years previously.

If we go back to 1990 we see an almost-as-long hiatus of six years:

which was followed immediately by the steep ramp up to the 1998 El Niño.

So, yes, 1997 is a cherry-pick but a) not really Tamino's and b) not very interestingly so.

What about the choice of the 1979 start year? Tamino chose this as the two satellite data sets he used for the lower troposphere temperatures start in that year but surface records, obviously, go back a lot further.

Let's go back to the start of the modern global warming era using 1974 instead but keeping the 1997 split year:

Nope, not substantially different, though 1999 and 2000 do actually drop below the hiatus line.

Looking at more split years:

And, for completeness, going back even further to 1940:

Now the trend over the last forty years or so is so steep that it's all but impossible to drop below the hiatus line, even the extrapolated warming line is shallow enough to not be in much danger.