Friday, January 4, 2013

What Statistical Methods to Use?

Two new papers on statistical methods to be used in climate science were brought up for discussion this week.  C. Franzke 2012 in What is Signal and What is Noise in a post on realclimate and Beenstock et al. "AGU Bombshell" over at Wattsupwiththat.  The two papers use "novel" methods to determine if statistically significant trends exist in the temperature records.

I am old school.  If it takes "novel" statistical method to tell if something of "significance" happened, I have either screwed up or need to hire statisticians, plural, to prove it.  More than one statistician because statistics is nearly a black art more than a science.  You can find anything you look for with statistics.

Looking at CO2 and Climate, I was able to find a "signature" of CO2, but only over land and mainly over the higher altitude land areas, that I would consider statistically significant.  I was not able to find a "signature" over the oceans.  Using satellite data, I could find a "significant correlation" between CO2 and temperature, Solar and sea level and using "radiant layer estimates" with satellite data, a fair estimate of the ocean energy imbalance.  All of those are just basic gut checks required since you cannot trust anything, data, models and your eyeballs in complex systems.  That lead me to "What is the Average Global Temperature?"

Since there is a "potential" error in the "average global surface temperature" of around 1.5 to 3.5 C degrees, there has been "land" warming of around 1C degree over the past century and the "noise" in the data is close to the "land" warming over the past century, I would think it is time to phone a statistical friend.  So I blew off the surface as a reliable "metric" and shifted to deep ocean temperatures that while not all that accurate are more stable and make more sense compared with the "more" reliable, IMO, satellite data.   Because of that shift in my frame of reference, I am fairly confident that "sensitivity" to CO2 only using a satellite era baseline is roughly 0.8 +/- 0.2 C degrees and that "sensitivity" to the solar equivalent of a doubling of CO2 is roughly 1.6 +/- 0.4 C degrees.  The solar "sensitivity" would be twice the CO2 only "sensitivity" because the atmosphere would radiantly amplify the ocean absorbed energy.

Using the deep ocean as a reference, the entire game changes.  Internal variability in the deep ocean heat capacity and the location of the bulk of that capacity that changes with deep ocean current and persistent weather patterns, is more than equal to the impact of CO2 only forcing.  This just happens to agree with Toggweilder et al. and a few others that have noticed rather odd methods being required to "defend" the IPCC consensus predictions.

So I believe there are two legitimate "theories", one based on faith more than fact and one based on observation, that have totally different end results.  The question is which is which?

What statistical methods to use will be an important decision in that determination.