Watts Up With That (WUWT) has a new post called policy implications of climate models on the verge of failure. It discusses a poster presented by Paul C (Chip) Knappenberger and Patrick J. Michaels (Center for the Study of Science, Cato Institute, Washington DC) at the American Geophysical Union (AGU) Science Policy Conference that took place this week.According to the WUWT post, the poster including the following figure. This figure shows how the trend in the various global surface temperature anomaly datasets varies with time interval considered (they considered time intervals from 5 to 15 years). Now, I was a little surprised as I had thought that for time intervals greater than 10 years, the trend was around 0.1oC per decade, whereas this figure suggests it is much smaller. I then noticed that the y-axis was in units of oC per year. Okay, so they’ve changed the units. It has normally been presented as oC per decade, but there’s nothing wrong with changing units as long as it is clear (and, in this case, it didn’t take me long to work out).
The post then goes on to show the following figure (with some additions by me). This compares the observed trends (using a 15 year time interval) with the range predicted using climate models. So the implication is that the comparison is so poor that the models have failed. Well, what I’ve added are the errors in the 15-years GISSTEMP trends (solid vertical black lines). The errors are large, so in fact there is a very large overlap between the observed trends and the model trends. I’ve also added the trend and errors for a 20 year period (1993 – 2013), shown by the vertical dashed red line and the two vertical dashed black lines. The mean now lies almost exactly on the peak of the distribution of model trends and the range now overlaps almost exactly with the range of model errors. To be fair, I don’t quite now where the models come from and what time interval they consider, so am not sure if the 20-year comparison is strictly speaking correct.
So, anyway it appears that Chip and Paddy have specifically chosen short time intervals (15 years or less) during which the mean of the observed trend is low but the error is large. They’ve ignored the error in their analysis and assumed that because the mean of the observed trend lies near the 2σ point on the distribution of model trends that it means that the models have failed. Given the large errors (uncertainties) in the observed trend, however, there is an extremely large overlap between the possible range of observed trends and the range of model trends and so such a conclusion seems somewhat premature. Also, as I pointed out in yesterday’s post, there is at least one climate model that has done a remarkably good job of predicting the evolution of the global surface temperature anomalies between 2001 and 2010.