Russian Heat Wave - Round 3

Following on from one post that found no evidence that a changed climate was responsible (Dole et al. 2011) for the Russian heat wave of 2010, and another that found exactly the opposite by Rahmostorf and Coumou (2011), I'm going to look at whether it's possible to reconcile the two approaches.

1000 Earths

The approach taken by Otto et al (2012) was to model the Earth's atmosphere as many times as they could, looking for the proverbial needle in the haystack. The needles they were looking for were events that were similar to the 2010 Russian heat wave, and the reason they needed so many models is because they are looking specifically for rare events, so a large number of runs of the climate is necessary to find these statistical outliers. 20 years ago, the only way to do this would have been to convince a large institution to loan you their supercomputer, nowadays an alternative exists: convince a large number of people to download and run a small program using the idle time of their computers. This is a remarkable crowd sourcing achievement, allowing people to run the HadAM3P atmospheric model on their laptops and desktops (which are about as powerful as an early 90s supercomputer by the way). 

Possible clouds over the Earth in 2090
(running on my laptop)

So how does this help work out if the 2010 Russian heat wave is attributable to climate change? The crux of it comes down to two sets of simulations, one for the Earth as it was in 1960s, and one as it was in the 2000s, each containing over 1000 individual simulations (1600 and 1200 to be precise). The occurrence of anomalous events between these two decades can then be compared:

Magnitude and expected return time of
events similar to Russian heat wave.

What the above graph shows is the frequency of events (expressed as a return time, or the average length of time you would expect between events of a given magnitude) plotted against their magnitude, in terms of the temperature of the heat wave that was modelled. It does this for two different decades, with each blue or green circle representing a particular event in the model with corresponding return time and temperature. The Russian heat wave equivalent is shown as the dot that the arrows are pointing towards. There are two important things to take from this graph. First, the return time between an event of the size of the Russian heat wave has decreased from around 100 years to 33 year from the 1960s to the 2000s (as shown by the horizontal red arrow). Second, that the magnitude of a 33 year event has increased by around 1ºC from the 1960s to the 2000s.

Tying things together

How can this help us reconcile the two different findings by Dole et al. and Rahmstorf and Coumou? Otto et al. argue convincingly that what Dole et al. found was that the movement vertically by 1ºC was inside of natural variability, and therefore should not be judged to be caused by climate change. (Note, this is looking at the magnitude of the event.) Alternatively, Rahmstorf and Coumou found that the likelihood of this event had increased (by 5 times), which is equivalent to the horizontal movement of 100 years to 33 years in return time. (looking at expected frequencies this time.) Hence to say they are contradictory is to compare apples with oranges, because they were never analysing the same thing.

All three articles were insightful, the empirical arguments of Dole et al. and the theoretical arguments of Rahmstorf and Coumou both brought a different perspective to the same problem. However, this week's article by Otto et al. ties the both of them together in a most satisfactory manner. I can only hope that the results from the simulation currently running on my computer can help contribute to science like this in the not too distant future!