I was aware of this story yesterday, but I didn’t like the original plot, (see at the end of this post) since use of straight line linear trends doesn’t accurately reflect the reality of the observation data. While it is often hard to find any reality in climate models, linear trend lines mask the underlying variance. Today, Dr. Spencer has produced a graph that I feel is representative and very well worth sharing, because it does in fact convey an EPIC FAIL speaking directly to the accuracy of an ensemble of climate models. – Anthony
Dr. Roy Spencer writes:
In response to those who complained in my recent post that linear trends are not a good way to compare the models to observations (even though the modelers have claimed that it’s the long-term behavior of the models we should focus on, not individual years), here are running 5-year averages for the tropical tropospheric temperature, models versus observations (click for full size):
In this case, the models and observations have been plotted so that their respective 1979-2012 trend lines all intersect in 1979, which we believe is the most meaningful way to simultaneously plot the models’ results for comparison to the observations.
In my opinion, the day of reckoning has arrived. The modellers and the IPCC have willingly ignored the evidence for low climate sensitivity for many years, despite the fact that some of us have shown that simply confusing cause and effect when examining cloud and temperature variations can totally mislead you on cloud feedbacks (e.g. Spencer & Braswell, 2010). The discrepancy between models and observations is not a new issue…just one that is becoming more glaring over time.
Here is the linear plot from Dr. Spencer’s post yesterday. He writes:
Courtesy of John Christy, a comparison between 73 CMIP5 models (archived at the KNMI Climate Explorer website) and observations for the tropical bulk tropospheric temperature (aka “MT”) since 1979 (click for large version):
Rather than a spaghetti plot of the models’ individual years, we just plotted the linear temperature trend from each model and the observations for the period 1979-2012.
Note that the observations (which coincidentally give virtually identical trends) come from two very different observational systems: 4 radiosonde datasets, and 2 satellite datasets (UAH and RSS).
If we restrict the comparison to the 19 models produced by only U.S. research centers, the models are more tightly clustered:
Now, in what universe do the above results not represent an epic failure for the models?