Nic Lewis has a new, long and rather technical post up at Climate Audit. He's looking at Michael Mann's latest paper in GRL which claims that the standard way of calculating the Atlantic Multidecadal Oscillation is flawed. In short, Nic has shown that Mann is wrong. What follows here is an attempt at a layman's summary for those who fear to negotiate the longer version.
The Atlantic Multidecadal Oscillation (AMO) is an ongoing series fluctuations in the mean sea surface temperature of the North Atlantic ocean. Both northern hemisphere and global temperatures are correlated with the AMO, and around a third of the global warming since the mid 1970s might be due to the strengthening AMO rather than to the increase in greenhouse gases.
Mann's new paper in GRL claims that the normal way of calculating the AMO - by removing the trend from the sea temperature records - is flawed. But his results turn out to depend on a series of clever tricks. He defines the AMO as, in essence, the smoothed difference between model simulations and observations. Of course this isn’t true, because we know that the models aren’t perfect and the difference must therefore also be a function of all the things the models don’t simulate correctly as well. But by ignoring these factors Mann is able to explain away a multitude of sins. So the hiatus in surface temperature rises since the end of the last century is explained by the AMO going negative (in Mann’s estimation at least). And in the final decades of the twentieth century, when surface temperature rises were relatively rapid, in line with the models, the AMO was, by Mann’s definition, broadly neutral. The possibility that the models were wrong but that this was hidden by a positive AMO is brushed under the carpet.
Mann’s main illustrations use a simplified climate model called an EBM, which encapsulates a rather high climate sensitivity of 3°C and a very high transient climate response of 2.8°C. In other words it would be expected to produce very rapid warming. However, Mann still manages to get it to match the observations (he uses the average northern hemisphere temperatures, not global ones) quite closely, by using values for things like the effect of greenhouse gases other than CO2 that are very different to current best estimates and therefore highly implausible. This close fit, particularly after the data has been smoothed, means that Mann's AMO is relatively small from peak to trough. Moreover it is declining in magnitude after 2000, in sharp contrast to the standard AMO graph, which is still rising over this period.
Mann then tries to show that his method is the correct one. He does this by doing a series of simulations, adding a dummy AMO signal (in the shape of red noise) to the EBM-simulated temperature series. He shows that when he processes each of these through his smoothed-difference algorithm he gets out what is in essence a random signal – in other words the dummy AMO but smoothed.
Next he processes the same set of simulated-NH-temperature-plus-red-noise series through his version of the standard AMO methodology (which is based on average NH, not just North Atlantic, temperatures). He shows that after smoothing you then always get the same thing out: something that looks just like the smoothed AMO you get from the observations. Aha! he seems to say, the standard AMO is an artefact – it doesn’t matter what’s in there, you always get the same signal out.
Nic Lewis observes a problem with this, however. There are two keys to understanding what is happening. Firstly, under the standard AMO methodology and with only a couple of cycles of AMO under consideration, the amount of noise that Mann is adding is too small to show up much in the final reckoning. Secondly recall that, after smoothing, there is a very close match between Mann’s simulated temperature series and the observations. So when you process the simulated temperatures plus a bit of noise that is too small have much impact through Mann's version of the standard AMO algorithm you do indeed get something that looks like the AMO calculated on the observations. To all intents and purposes the simulation and observations are the same after smoothing; the noise is a red herring.
As Nic points out, an experiment described in Mann’s own SI makes this point clear. Here Mann shows an alternative set of results in which he simulates the temperatures using a different set of assumptions on volcanic aerosols. This makes the simulated temperatures rather different to the observations. If you then do the experiment adding noise as before, you can see the problem. Using the standard algorithm, the AMO signal extracted for simulated-temperature-plus-dummy-AMO-noise looks very much like that extracted for simulated temperature alone; but the signal extracted from the observations is very different. Clearly, Mann’s argument that the standard AMO signal is an artefact is wrong.