Wednesday, January 09, 2008

Would you bet on the satellite record?

There's been a few amusing goings-on in the blogosphere recently.

First, Chip Knappenberger commented on an old post, to point out that the satellite temperature record had actually shown a negative trend over the last decade (albeit not a "significant" one, and only if he switched his allegiance to the RSS data in contrast to UAH). So although he would have lost his bet, it was quite close. As I commented in reply, it was a bit of a surprise to me that it ended up so close - but a loss is still a loss. WCR has also blogged it here.

Then Roger Pielke Jnr got in on the act, wondering whether the RSS observations posed a problem to climate science. First he thought that the lack of warming was a problem for model predictions, and then that the apparent discrepancy between surface obs (which are in line with the models) and satellite analysis was a problem for those who thought that this had all been sorted out.

Finally, an update from Chip (again in the comments down here) - these satellite data were wrong and are due a correction to warm them up (which will presumably eliminate much if not all of the discrepancy). [Update: RP has emailed to point out that his post predated WCR's.]

It's deja vu all over again! (Check "Lyman" in that post if you don't know what I'm on about.)

Interesting to see again, another example where another apparent anomaly between models and hot-off-the-press data is being resolved in favour of the models. There is a lesson there for those who care to learn it...and those who do not, will no doubt repeat history in the near future.

Of course I'm setting myself up here for a fall - maybe the "cooling" data will turn out to be right after all. Wanna bet?


Chip Knappenberger said...


When I first pointed out that now that the data were all in, we would have lost our bet, I used the RSS data as proof, not out of an “allegiance” switch, but out of my (mis)understanding that in recent months the UAH data had a warm bias in it. In my comment, I noted that I also believed that our bet would have been lost even when the UAH data problem gets worked out. It turns out, that the notion that the UAH data was warm biased recently was due to comparisons with the RSS (which showed more cooling in the past several months than the UAH did). However, after additional work by the UAH team (including incorporating data from a new, orbit-stabilized satellite), it appears that the UAH data has been pretty close all along, and that it now appears (or at least is the opinion of the UAH team) that it is the RSS data which has a recent cool bias (note: the RSS folks have yet to comment on this as far as I can tell). Upon finding this out, I immediately posted a new comment to you updating the situation. How should I have acted differently?

Are you proposing a bet as to whether or not there is a recent cooling bias in the RSS lower troposphere temperature data? As opposed to how the data behave in the current UAH dataset?

It seems, you are taking your betting proposals beyond actual climate trends and are now entering the realm of taking wagers on the accuracy of particular datasets (and thus the ability of particular scientists). This seems to be an odd twist. Perhaps you’ll propose a ranking of researchers based upon their proneness to making errors, and then take wagers (maybe based upon some type of point spread?) as to how many errors will be contained in their next publication? Or what their magnitude may be?

The actual values contained in all sorts of datasets are constantly being tweaked as new and presumably better methodologies/technologies/etc. are being developed. Recently the CRU updated its global surface data from version 2 to version 3 and the warming trend in recent decades increased by a couple hundredths of a degree C. May not seem like much, but it does effect verification stats. A computation error in the GISS data for the U.S. was recently detected and once fixed, the data changed. The same is true for the MSU datasets.

We all tend to fall victim from time to time to trying to get the very latest data values in order to make a particular proclamation about something. Consider that the UK Met Office makes pronouncements about where each year’s temperatures will rank (first, second, third, top-10) in late November or early December, rather than waiting until the data are all in.

In my commenting on the 10-year trend, in evaluating our proposed bet, I waited until the data were in, and then reported on my calculations using the dataset that seemed at the time to be most reliable. After all there seemed to be a problem with the UAH data (that I thought would eventually be fixed) and, further, the RSS data has typically been the warmer of the two MSU TLT data products. So, truth be told, I thought I was taking a conservative route. Again, I ask you, how should I have acted differently?

In retrospect, you may have a response to this particular instance, but how does it apply universally?

How do you ever determine the final answer? A wager that the 20-yr temperature trend in global temperatures would be below 0.20C/dec for the period 1986-2005 would have been a winner using the extant CRU dataset at the end of the wager period (HadCRUT2v) but would have proven to be a loser when the new and improved HadCRUT3v came out a year later. Maybe a newer version in the future will again turn the outcome.

Clearly, arguing over a couple hundredths of degrees here or there is largely irrelevant in the overall picture of climate change, but in the world of wagering, it perhaps has some import.

Note that in reporting the outcome of our WCR proposed wager, I didn’t make any claims as to what this meant to the greater picture of global climate change. Although, indirectly, I suppose, it related to my general belief (which I have tried to propose in a wager to you) that the future course of global temperatures will be closer to lower end of the projected rate of rise than to the higher end. A point to which, I guess, we both agree.


Hank Roberts said...

> a ranking of researchers
> based upon their proneness
> to making errors

Not a new idea, and quite often done in other areas. Turns out the best predictor has been industry sponsorship. Who knew?

Just a couple of examples

Chip Knappenberger said...


In order to help James know where to put his money, which of the two satellite temperature groups has the most industry sponsorship?


James Annan said...


Relax, I wasn't really proposing a bet on the eventual outcome of the satellite record (although I suppose if you are eager, we might be able to work things out). Rather, I was poking a little fun at my own eagerness to assert that the data were wrong.

As for your "allegiance switch", well I found it ironic that on pretty much the same day that Roger posted:

"I am sure that in conducting such a verification some will indeed favor the dataset that best confirms their desired conclusions. But, it would be ironic indeed to see scientists now abandon RSS after championing it in the CCSP and IPCC reports."

you had explicitly chosen to switch from UAH to RSS!

Of course, as I'm sure Roger would agree, there is always an "excess of objectivity" to support such decisions...

EliRabett said...

If memory serves, Drew Shindell had a talk about which to trust, models or data, which in the case of pre-1900 data was pretty clearly not the data.

James Annan said...


Well I have my suspicions even about details such as the 1940s "bump" which is very visible in the AR4 "global ocean" plot of Fig SPM-4 and elsewhere. That was during the war, there were lots of potential changes in ship coverage...and a startling short-lived spike in temperatures that has not been seen since, and which (according to the figure) the models don't come close to reproducing. I saw an interesting presentation last year about revisiting the data coverage (by Philip Brohan at UKMO/HC).

I see now from Deltoid (linkback below) that the UAH data have been fixed...

James Annan said...

Um...that should be RSS data.