Model prediction of Arctic sea ice loss - 2000 (L) vs. 2040 ®. From BBC news. Click to Enlarge.Now that the IPCC has released the summary of its upcoming study report Climate Change 2007: The Physical Science Basis , strongly stating that global warming is man-made, it is still important for scientists to clearly enumerate any issues surrounding the accuracy and reliability of their models.
Because if they don’t, global warming deniers, or those who believe that we need still more time for convincing proof will focus on the slightest inaccuracy in a model’s prediction in classic red herring fashion. (In a way similar to anti-evolutionists, they will neglect the hundreds of accurate predictions and claim that the one that doesn’t quite fit calls for total abandonment of a theory.)
Climate models are by their very nature prone to chaotic behavior. This behavior must be accounted for when using climate models for any type of prediction. An excellent article on how chaotic models are handled has been provided recently by Cecilia Bitz of the Univ. of Washington. In her Real Climate article Arctic Sea Ice decline in the 21st Century, Bitz describes work she did with colleagues Marika Holland and Bruno Tremblay, which culminated in a paper for the Dec. 2006 Geophysical Research Letters. I want to focus on two aspects of Bitz’s commentary: the presence of chaos in the climate models, and the overall accuracy of the modeling process. Bitz recounts:
It is common practice to run climate models multiple times with slight variations to the initial conditions. Because the system is chaotic, the natural variability in each run is random and uncorrelated from one run to the next. When an ensemble of runs is averaged, the natural variability is reduced in the ensemble mean, and it is easier to detect a significant trend.
With a chaotic system there is sensitive dependence on initial conditions (SDIC). This is the cause of the "random and uncorrelated" variability in each model run. Note that this variability is known to be present in the models, and therefore a methodology of handling this variability must be adopted, which has typically been accomplished by the ensemble averaging. Any trend that survives this averaging is then by definition "significant." Such a trend in the output of a climate model could be a rapid temperature change, or, in the case described by Bitz and her colleagues, the rapid disappearance of arctic sea ice.
The natural question then is how accurate a prediction is the noted "significant trend"? The only way to establish accuracy for a prediction is to wait and see if it comes true. So the operative way to describe a prediction is more on the lines of "reliability" or "confidence". Perhaps the best way to develop confidence in a modeling technique is to compare model results with previous history. Bitz’s climate model does a good job of matching past observations:
The trends in the seven ensemble members for 1979-2006 span the trend in the observations: Some members retreat a little faster and some a little slower, as expected from the random natural variability in the runs. The model also reproduces the mean and variance of the observations with good fidelity.
Given the care in which the modeling has been handled, what about future predictions of the model? Here is Bitz’ response. Note how circumspect she is about exactly what will happen when, as opposed to the confidence in something that probably will happen during a given time frame:
There is considerable uncertainty in future model projections, and Figs 2 and 4 illustrate why it would be better not to focus too much on the year 2040, which to our dismay was highly publicized. The more important message from models is that all but a few outliers predict enormous sea ice retreat this century. At least a few respectable models predict a nearly ice-free Arctic by midcentury, with a retreat that may be punctuated by rapid events.
Read Bitz’s full article for all of the details, including whether anything can be done to reverse the trend in loss of sea ice. I recommend doing this before tackling the full IPCC report on the physical basis for Climate Change, where there should be ample references to model ensemble averages. (Note: The full report of the 2001 IPCC study on the Physical Basis of Climate change is available online. Find out more about climate model methodology by following the Model Evaluation link.)
And definitely keep Bitz’ explanations in mind when you read the inevitable complaints against the IPCC report as being purely political as opposed to scientific.