Agree on the need for better measures of error.

Take correlation coefficient for instance. One would think that the (anti) correlation between Tahiti and Darwin, which comprises SOI, would probably be close to -1, considering how they were picked to best illustrate the standing wave pattern. When one place is up in pressure, the other place is down is perfect anti-correlation.

In fact, the anti-correlation is "only" -0.55, according to the time-series comparison below. One can sense that these two time-series are very likely anti-correlated, but the fact that there are a few regions that don't align drags the coefficient quickly away from -1. There must be other metrics that do a better job of identifying how well two curves match.