Here is an example of a tight-fitting training interval for ENSO, with a gap interval for validation.
Even though we may be over-fitting to the noise within the dual training intervals (with a correlation coefficient nearing 0.95), the validation interval is still at 0.75 and obviously has the same general time-series profile. However the fit to the earlier data prior to 1920 is lower at 0.48.
The timing is perfect for putting the finishing touches on the ENSO and QBO models as I will be presenting at the AGU meeting next month.
[OS11B-04: Analytical Formulation of Equatorial Standing Wave Phenomena: Application to QBO and ENSO ](https://agu.confex.com/agu/fm16/meetingapp.cgi/Paper/120573)
Everything in this abstract still holds and so I continue to be confident in presenting the results. I will either sink or swim on relying on such simple models but I think the findings are rock solid.