Friday, January 24, 2014

Met Office Predicted 2013's Temperature

A year ago the UK Met Office predicted 2013's average global temperature would be
between between 0.43 °C and 0.71 °C warmer than the long-term (1961-1990) global average of 14.0 °C, with a best estimate of around 0.57 °C.
So how'd they do? Just a little high. The actual 2013 global average was 0.49 °C.

Like all measurements that number has an uncertainty, which comes from the uncertainties of each monthly number. The Met Office conveniently gives those every month, in the last two columns here. They're actually not that small -- typically about 0.15 °C for each monthly anomaly (2-sigma) -- and when I include those the way you're taught in undergraduate lab[*] I get an uncertaintly for the yearly average of ± 0.05 °C.

So the actual temperature[**] was 0.49 ± 0.05 °C compared to a prediction of 0.57 ± 0.14 °C. That's pretty good.

[*] The 95% upper and lower bounds of the uncertainties aren't quite symmetric about the monthly value, which is a pain, so for simplicity I averaged the two bounds to get a monthly uncertainty. The error in doing so, which should be very small, is left (as they say) as an exercise for the reader.

[**] Or rather, this is the actual temperature of the Met Office's model of the world, which is not the temperature of the real world. The model is not reality but an an approximation of reality, and it matters more how it compares to itself (to past temperatures, etc.) than how exactly it simulates reality (as long as the simulation is "good"). Often journalists add the yearly anomaly to 14 °C and write "last year's temperature was 14.49 °C -- and even the Met Office did that in their press release -- but that's not really true and I wish they'd stop it (but understand why it's done). It's the anomaly that matters, and that can be measured, not the "real temperature of the world."

The definition of "good" in the above paragraph is left as another exercise for the reader.


Unknown said...

As a paleogeneticist who needs to use climate models, I know all of the "skeletons" in the closet of climate modelers. As well as more than enough of this science to be annoyed by this kind of misdirection, David.
Anyone can predict NEXT year's average climate temperature THIS year within the error that you find dismissable. We simply predict that it will be the same. Which is what the Met did, adding to it is model. The model component made more wrong than it would have been had we simply left the model component out.
The problem with climate models is not that (if one is willing to cherry pick) they get right some prediction here or there. The problem is that the predictions 10 (or 20) years ago based on what the "political class" has decided is "settled science" have not come close to predicting the 2013 temperature. Some negative feedbacks are clearly at work, feedbacks that the models clearly do not capture, feedbacks that we clearly do not understand.
Which is why any this post is objectionably misleading to non-scientists, or even to scientists who do not work in this field.

Danley Wolfe said...

It's good if you consider that the observed is above the lower error limit of the prediction. Relative to the spread of the error bar, the actual observed is around 20% off the bottom, meaning it's 30% below the midpoint of the error bar. Not great.

David Appell said...

Some negative feedbacks are clearly at work, feedbacks that the models clearly do not capture, feedbacks that we clearly do not understand.

That's not clear at all. It's looking more like natural variability from ocean cycles (especially the PDO) isn't being modeled correctly. For example, the NASA GISS Model E2-R, mentioned here

does a fairly good job up to about the last 90s, when the divergence starts. The graph below that shows it does an even better job of modeling the anthropogenic component, and the one below that that the difference is in the deep ocean, not the upper ocean.

These lectures by Kevin Trenberth give more on this:

Lazarus said...

A copied comment #23 from here;

Just to clarify something, as he states himself, Paul is using data from HadCRUT4 to assess the accuracy of the UK Met Office 2013 global prediction. It should be pointed out that, as the The UKMO made clear at the time, their 2013 forecast was based on the WMO data:

From that link, the WMO data is based on three data sets:

"The three international global temperature data sets are from the Met Office and University of East Anglia (HadCRUT4), NOAA National Climatic Data Center (NOAA NCDC) and NASA Goddard Institute of Space Studies (NASA GISS)."

The WMO figure is the average of these three data sets (based lined to the 1961-1990 period). In 2013 HadCRUT4 (0.486) was the coldest of these. The WMO average for 2013 was 0.502. This is the figure the UKMO forecast should be assessed against.

While this is still below the 'best estimate' figure of 0.57, it's certainly a step closer to it than the HadCRUT4 figure alone suggests. Also, whereas 2013 was the 9th warmest year in HadCRUT4, it was the 4th warmest year using the WMO data set.

The UK Met Office 2013 forecast also states:

".... it is very likely that 2013 will be one of the warmest ten years in the record which goes back to 1850, and it is likely to be warmer than 2012."

Both of those forecasts were correct.

If this is correct then the actual figure would be 0.502 not the 0.49 you quote and even closer to the MO prediction of 0.57.