If Per Strandberg has the solution to ENSO he really ought to try it on the 1880-1980 interval.

I bring that up again, because I decided to test our model against the ENSO proxy record that goes back to the year 1651. The complete write-up is here:


The significance of the fit over the earlier time spans is not as good as for the recent instrumental record, yet it also appears extremely unlikely to be the result of chance. What I did was to take about 500 sets of synthetic red noise time-series, and then statistically compare the agreement of the model to the real data and to the synthetic data. Since red noise is often thought to characterize ENSO, this is a good way to compensate for over-fitting -- since over-fitting will improve a fit to red noise just as does to the real data, this technique has discriminatory power in separating significance from chance.


Note that the four spans making up the proxy record are all in the high correlation coefficient region. Pulling four straight samples from a set of Monte Carlo red noise runs having each a correlation coefficient above 0.7 is unlikely. This indicates that the model to the current instrumental record likely applies to the past (and seeing one of the fits above 0.8 is a 1 in 500 likelihood just by itself).

This is a confidence booster in substantiating that we are on the right track with the ENSO model.