This is a set of links to Context/Earth blog posts and Azimuth Forum posts concerning the Southern Oscillation Index Model (SOIM), used to understand ENSO and El Nino.

## Early Posts setting the stage

- The Southern Oscillation Index Model
- SOIM and the Paul Trap
- The SOIM Differential Equation
- Correlation of Time Series (NINO 3.4)

## Chandler Wobble

- The Chandler Wobble and the SOIM
- The SOIM: substantiating the Chandler Wobble and tidal connection to ENSO

## Quasi-Biennial Oscillation

## Proxy Posts

## Tidal Connection

- Using Tidal Gauges to Estimate ENSO
- An ENSO Predictor Based on a Tide Gauge Data Model
- Two modes to ENSO Variability

## Sloshing

## Comprehensive Posts featuring QBO, TSI, Chandler

- Demodulation and the SOIM
**Final Recap**

White paper "Sloshing Model for ENSO".

And on ARXIV as PDF

## Azimuth Forum threads

- Tidal records and enso
- NINO3 and seasonal alignment
- QBO and ENSO
- Multivariate ENSO Index (MEI)
- ENSO proxy records
- Is there an exact biannual global temperature oscillation?
- Symbolic regression machine learning and ENSO time-series

These forum threads are ones that I either initiated or partially hijacked with SOIM discussions. Look at the entire forum for other interesting ENSO and El Nino discussions (see research focus and recap).

Pingback: Paper on Sloshing Model for ENSO | context/Earth

Pingback: Characterizing Changes in the Angular Momentum of the Earth | context/Earth

Pingback: Raising the Bar on ENSO Model Validation | context/Earth

Hi Paul,

An interesting exercise would be to do the following:

1. SOIM with astronomical forcings only (lunar and planetary)

2.SOIM as in 1, but adding in the Chandler Wobble

3. SOIM as in your white paper (already done)

4. Compare the 3 models above to see if the added complexity really gains very much.

Generally a simpler model is less prone to claims of overfitting, and is likely to give better long term forecasts. In fact the various changes in the frequency of the Chandler wobble over time are a problem because they can only be deduced by using past data. For forward predictions model 1 may be best.

Dennis, Something will eventually click in place.

If a short 18 month fitting interval contains much of what is required to fit the rest of the time series, then the reality has to be close.