September 15, 2012

Made in New Zealand

I don’t know for sure (though I have my suspicions) whether it’s possible to rigorously demonstrate a temperature trend using data just from New Zealand.  I’m sure it’s not possible to demostrate that mammograms save lives using data just from New Zealand.  Fortunately, there’s no real need to do so in either case.

We’ve mentioned the Berkeley Earth Surface Temperature project before.  The project is run by a physicist and former climate-change skeptic, Richard Muller, and has statistical leadership from the eminent David Brillinger.  They have taken a slightly different approach to temperature analyses: they use all available temperature records and weight them for internal consistency rather than selecting a high-quality subset, and their analyses of relationships between temperature and other factors are purely statistical, not based on climate models.  It makes astonishingly little difference (except that they can get better estimates of statistical uncertainty).

Here are two graphs from their results summary page.  The first shows their temperature reconstruction (with gray area for statistical uncertainty) and the three other main reconstructions.  You can see that the statistical uncertainty is quite important back in the early 19th century.

The second shows a simple statistical fit of temperature to CO2 and major volcanic events (which produce tiny particles that cool the atmosphere).   The red line basically has to go up, since temperature and CO2 were low in the past and are high now, but the accurate matching of the trends is a genuine confirmation that you wouldn’t expect if some unrelated factor was causing the temperature increase.    The Berkeley researchers say that they also looked at solar variation (which Muller used to think was a major factor), but it didn’t explain the trends even statistically.

avatar

Thomas Lumley (@tslumley) is Professor of Biostatistics at the University of Auckland. His research interests include semiparametric models, survey sampling, statistical computing, foundations of statistics, and whatever methodological problems his medical collaborators come up with. He also blogs at Biased and Inefficient See all posts by Thomas Lumley »