What does it mean when a climate model gets it right?

Roz Pidcock

Do climate models do a good job of forecasting global temperature? Recent coverage in the more climate skeptic press has suggested reasons to be doubtful. But a new paper shows a forecast the Met Office made more than 15 years ago has been “remarkably accurate” in predicting global temperature rise. So what does the fact that some models do better than others tell us about the way climate scientists use them?

A couple of weeks ago, we covered an article in the Mail on Sunday which poured heavy criticism on climate models’ ability to forecast global temperature rise.

The piece argued that because projections of temperature rise made by an important set of climate models are somewhat higher than than we’ve seen in the past decade, climate models are “catastrophically flawed”.

Global temperature is currently tracking the lower end of the temperature range predicted by the latest generation of climate models used by the Intergovernmental Panel on Climate Change (IPCC) – and scientists have speculated about the reasons why.

But as a counter-example, research just published in the journal Nature Geoscience provides an example of one model getting it spot-on. The paper compares a Met Office climate forecast made in 1999 against actual temperature data up to 2012 – and shows they match to within a few hundredths of a degree.

So what does it mean if some models are better than others?

Making a forecast

In 1999, scientists from the Met Office and the University of Oxford used the HadCM2 climate model to forecast how global atmospheric temperature was expected to change up to 2012. To ensure the forecast was as realistic as possible, the team first fitted the model-simulated temperature between 1946 and 1996 to actual measurements.

Sixteen years have passed since 1996. This is about the minimum amount of data needed to test how well the model forecast has performed against observations, the researchers say.

Close match

The Met Office forecast predicted the decade December 2002 to December 2012 would be a quarter of a degree warmer than the decade August 1986 to August 1996. According to a briefing from the University of Oxford, “the latest measurements of global temperature confirm their prediction to within a few hundredths of a degree”.

Image - Good Met Office Forecast _Allen Et Al (2013) (note)

The mean HadCM2 forecast adjusted to fit 1946 to 1996 temperature data (black dashed line) compares well with the 2002 to 2012 decadal mean from observations (red diamond on right). The yellow diamonds are yearly temperatures for the forecast period. The shaded area is the range of uncertainty. Source: Allen et al., (2013)

The actual temperature for each individual year also falls within the uncertainty range the model forecasted. This means the model is able to capture natural factors that affect global temperature from year to year on top of the long term warming trend due to human activity – another way scientists assess model performance.

As Dr Peter Stott, co-author of the new research, tells Carbon Brief, this suggests the forecast is pretty accurate. He says:

“The results show that previous projections showing substantial warming over the course of the 21st century have been borne out by observations so far this century.”

Change of pace

Although the model predicted the right amount of temperature rise by 2012, the speed of warming hasn’t stayed the same over the 16-year forecast period. As lead author Professor Myles Allen explains, a slower rise in atmospheric temperature seen recently is balancing out a period of faster-than-expected rise for a few years around 2000. He says:

“Many people think that recent years have been unexpectedly cool. In fact, what we found was that a few years around the turn of the millennium were slightly warmer than forecast, and that temperatures have now reverted to what we were predicting back in the 1990s.”

In other words, this pattern of faster-than-predicted followed by slower-than-predicted temperature rise means the overall forecast has turned out pretty much spot on. Stott explains:

“Whereas there has been much discussion of the so-called hiatus in global average temperatures since the massive El Niño of 1998 in fact the decade 2001-2010 lies at the centre of the [range] of temperatures forecast in the original paper.”

The fact that such natural fluctuations in the climate can persist for several years – causing temporary deviations from the long term trend – is the reason scientists use longer time periods to assess forecast accuracy.

The study also suggests the recent period of slower warming isn’t inconsistent with future warming. With global temperatures where they are now, the model forecasts a temperature rise of about 0.8 degrees Celsius above 1996 levels by 2040 – that’s nearly 1.5 degrees above pre-industrial levels in total.

Just one test

Of course, while this new paper demonstrates the ability of some climate models to get recent temperature rise right, it’s only one test. As Dr Gavin Schmidt from NASA tells us:

“This is all quite interesting and adds to the evidence supporting the credibility of [global climate models], but of course, in itself, it is not going to change anyone’s mind.”

And as Dr Gareth Jones from the Met Office explains:

“Ideally lots of different projections over different timescales would be tested to be able to validate or reject the [model], something which would obviously take very long periods of time in reality.”

Timely reminder

Nevertheless, these results are useful in illustrating the different aspects of the climate system models need to capture; the long term trend due to global warming, factors that affect the global temperature from year to year and those that can cause deviations from the trend lasting several years.

The reasons why some models do a better job than others is the stuff that climate scientists have been discussing since the first climate models, and probably always will be. Building an accurate climate model is an iterative process. But in the same vein, one good model match isn’t evidence climate scientists have cracked it just yet. 

🗂️ back to the index