There are other interesting relationships in the QBO data. [This site]( contains QBO data for various altitudes, 70,50,40,30,20,15,10 hPa (or mbar).

I did a machine learning exercise on the 40 hPa time series in terms of the other series. That involves essentially looking at cuts across the contour map.


What it found was that the 40 hPa was strongly linked as a multiplication of the 30 hPa with the 50 hPa time series -- with a very good correlation coefficient, about 0.83 without filtering.


It also shows positive spikes that split the biennial cycle roughly in half. This is where both the 30 and 50 both have strong positive or negative values. The downward notches are where one or the other crosses zero.

This is what the QBO page says, and it points out a real delay between 30 and 50 hPa

* The time-height section derived from these data in Fig.30 shows the observed structure of the QBO at equatorial latitudes:
* alternating easterly and westerly wind regimes propagate downward with time;
* westerlies move down faster and more regularly than easterlies;
* **the transition to easterlies is often delayed between 30 and 50 hPa**
* easterlies are generally stronger (30-35 m/s) than westerlies (15-20 m/s);
* maximum amplitudes of both phases typically occur near 20-hPa;
* the average period is about 27 months;
* both period and amplitude considerably vary from cycle to cycle.

Now, of course this looks kind of suspicious, so I created my own contour/relief map using the data, see below. What is odd about the view are these strange box-like artifacts that occur along the 40 hPa altitude. I drew one with an enclosing red dashed box below. These appear regularly across the decades.


It is entirely possible that whoever created the data may have tried to average the 40 hPa as (30 hPa + 50 hPa)/2 but inserted a multiplication by mistake?

Otherwise, I can't see this happening, unless the physics says that the 50 hPa is causally related by 40 hPa / 30 hPa, so that the higher altitude can boost the speed as the stratospheric wind direction reverses?

This is either very interesting, or a bust due to data problems. That's what happens with machine learning, as it can find possible artifactual errors -- and these routinely have strong correlations because they are artificially created!