Tuesday, March 26, 2013

The State of Climate Change - Part 2

Part 1

Synopsis:  We have seen that climate people do not follow the same rigour with error, as the other physical sciences.  In particular, they have plotted different resolutions on one time plot.  This is anathema to earthquake science since it would always show increased earthquake activity, due to the recent installation of numerous seismometers.


We cannot have a time plot without somebody wanting to extrapolate into the future.  It is human nature. Over the years we have seen many methods of doing this.


The simplest thing to do is attach a straight dotted line to the end of the time plot.  This is frequently done, although nobody would ever admit it.  Thus, we had 'Momentum Investors' and other such fools.  No matter what they say, when you see that dotted straight line, it is momentum.

Spurious Cross-Correlation

This has plagued earthquake people for time immemorial.  The basic recipe:

Take one time plot

Search the world for another time plot that matches this.  You have an infinite number of choices.

Associate the two in some manner.

For earthquake prediction, everybody found some correlation with clouds, temperature, small seismic activity, animals, etc.  By the sweat of their brows in the initial choosing, the match is perfect.  We have also seen this with stock investing.  This perfect match, however, always falls apart tomorrow.

I have found, over the years, that you cannot argue against this.  The only thing you can do is run away, like from the mythical killer rabbit.  You might mention, as I have done with climate, that there is no physics behind this correlation, but it will fall on deaf ears.

A variation on this is 'Hind-Cast Modelling', which is computer modelling where the material properties are derived from the cross-correlation.  In other words, you calibrate by making sure the computer results match the past.  I have run into this numerous times, and I am bruised and bloodied by it.

Things properly done

If you want to extrapolate properly, you have to get down and dirty with the physics.  This is done all the time with car-crash modelling, where the material properties are derived from lab testing.  With earthquakes, it involves getting down to the actual grain-to-grain properties of rock, which again can be derived from lab testing.  To extrapolate and scale up, you have to ensure that you have a self-similar system, and you are following the standard scaling laws.  You then come up with a time-slice finite differences computer model.

To do climate science right, you must get down to the physics, which unfortunately is mostly unknown.  For example, what are the energies of the big forcing functions?  How stable are they?  Are there non-linear interactions?  What is the role of a single molecule of CO2 at various elevations?  Although the task may seem daunting, the more physics, the merrier!

No comments: