It’s weird to blog an analysis; it’s sort of like “stream of consciousness curve fitting.” Still, it’s sort of fun. I’m posting results relatively rapidly, and there is no doubt some of these will have errors. I’ll be embarrassed for about 5 minutes after I find them, and then I’ll get over it.
Still, the curious can watch to see the steps I take, and possibly ask questions.
Today, I’m continuuing to try to figure out the best estimate of the climate sensitivity to doubling of CO2 by fitting Mean Surface Temperature Data to a simple physical model. I’ve decided to call my lumped parameter climate model “Planet Lucia”, as it’s such a simple model, and I need a name while fiddling and discussing it here at my blog.
I fiddled a bit with my Excel spread sheet to make it easier to proof-read for errors. Then, I slapped in the monthly GISS Land/Ocean temperatures and monthly forcings I came up with in my last post.
Here’s the graph.
I’d list sensitivity, but as I get closer to having done all the steps to get the “best estimate”, I get paranoid and start to suspect I have a units conversion error, somewhere. I probably shouldn’t even list the time constant. Heck, I could have all sorts of units errors that I need to check using some hand calculations. (That’s the only way I ever find these things. Or who knows? I may have shifted rows by a month in a comparison. Obviously, I need to check. )
Still: The upside is the model describing “climate lucia”, gives a good fit to data. That fit would be unaffected by any units conversion errors.
The correlation for a straight line is R2= 0.40. The correlation for the lumped parameter model is R2=0.70. Of course neither seem as great as the 0.90 I got when processing the annual average data, but the fact is the monthly data are noisy. (Also, likely the simple model misses some physics, and can’t capture some very rapid fluctuations. I still like the model. 🙂 )
After checking for obvious blunders, I’m going to repeat this with HadCrut data. After that, I need to do some sensitivity tests to see how adjusting forcings within uncertainty ranges Gavin suggested changes my results for sensitivity. Then, I’ll try to figure out the uncertainties in the parameters.
Update: I edited the graphic, removing the time constant. As I haven’t fully proof checked for calculation blunders, I thought that wiser. Presumably, any blunders I find and fix would make the model fit better, not worse. 🙂