I Found a Solar Wind Signal in Polar Motion Stability
At the beginning of June, I started testing something that is very easy to eyeball and very hard to establish statistically… do changes in the solar wind line up with changes in Earth’s polar motion?
I wanted to know whether unusual polar motion states really cluster around heliospheric boundary changes, or whether the timing could just be coincidence.
For the Earth side part of this test, I used Craig Stone’s (@nobulart on Twitter) DRIFT dashboard and API.
Craig’s DRIFT system takes IERS polar motion data and pulls out different ways of describing how that motion is behaving like its direction, speed, curvature, phase, and how stable or unstable the motion is. The variable that ended up getting interesting here was the Coupling Stability Index. For this first test, I was looking for times when that stability measure became unusually low.
Building the dataset
I lined everything up on the same hourly timeline from: 01 Jan 1996 to 09 June 2026. That gave me:
On the DRIFT side, I pulled several of Craig’s diagnostics, including things like drift direction, phase, angular velocity, curvature, manifold departure, (R(t)), and Coupling Stability.
On the solar side, I used NASA OMNI data.
That includes measurements of the interplanetary magnetic field and solar wind near Earth, including:
IMF (B_x), (B_y), and (B_z)
magnetic-field strength
IMF phi angle
solar-wind speed
proton density
dynamic pressure
Kp
ap
Dst
SYM/H
I also added ICME intervals and some simple indicators for high speed streams and geomagnetic activity.
The point was to avoid asking one vague question like, “Does solar activity correlate with polar motion?”
I wanted to test specific solar wind conditions against specific DRIFT states.
I originally wanted to test actual HCS crossings
My first idea was pretty straightforward. I wanted a list of known heliospheric current sheet or sector boundary crossings, then I wanted to see whether unusual DRIFT states clustered around them.
Of course, this resulted in a problem (as research goes….)
I tried to download the WSO/Svalgaard sector boundary catalog, but the connection failed with an SSL handshake error. :(
I had a backup method in the code using changes in the sign of OMNI (B_x), but that proxy ended up being way too broad.
The resulting boundary window covered:
At that point, almost everything is “near a boundary,” which makes it pretty useless, and so, not surprisingly, that version didn’t show anything interesting.
Empirical p ≈ 0.42
Basically null.
So, I don’t consider that a successful independent HCS test. That still needs to be done correctly with a real boundary catalog.
The more interesting result came from the IMF direction itself
Instead of relying on that failed HCS download, I built another proxy directly from the observed IMF phi angle in the OMNI data.
The interplanetary magnetic field around Earth falls into two broad sector orientations. I divided the phi angle measurements into those two broad states and marked the times when the field changed from one sector to the other.
I also required at least 24 hours between accepted flips so I wasn’t counting every little back and forth change as a new transition.
That left:
One correction from my original analysis notes : I originally defined these as “±6 hour” and “±24 hour” windows.
Looking back at the actual code, that isn’t quite right. The main test was looking at the hours after a detected sector transition.
So the actual question is does unusually low DRIFT coupling stability show up more often shortly after an IMF sector transition?
What counts as “low stability”?
I defined low Coupling Stability as the lowest 5% of the full record. The cutoff was:
That gave:
Those are hours, not 13,340 completely independent physical events. Polar motion behavior tends to persist for stretches of time.
Then I compared the timing
First, I looked at the short post transition window. Among the low stability DRIFT hours, about:
occurred shortly after an IMF sector transition
For the rest of the DRIFT record, the same sector transition state occurred about:
Now, I know that doesn’t look huge at first glance, but the ratio works out to:
So, when DRIFT coupling stability was unusually low, we were about 10% more likely to also be in this IMF sector transition window.
The broader post transition window gave almost the same answer:
≈ 9.6% enrichment
It’s interesting… but time series data can be sneaky. The important part was testing whether this could happen by accident.
Solar wind data and polar motion data both contain long term patterns, cycles, and periods where similar states cluster together. So, I didn’t want to randomly shuffle individual hours and pretend every measurement was independent. Instead, I used a circular-shift test.
I kept the solar wind series intact, with all of its normal timing and clustering, but shifted the whole thing to a different place in time relative to DRIFT. Then, I recalculated the relationship.
I did that 1000 times.
That allows me to test: If these two time series had the same internal structure but the timing between them were wrong, how often would I get an enrichment this large by chance?
The average result from those shifted datasets was basically no relationship:
97.4th percentile of the null distribution
Empirical p = 0.029
So, only about 3% of those shifted timelines produced a relationship at least this strong. That’s the part that caught my attention.
The broader post transition window gave almost exactly the same result:
Empirical p = 0.030
So, this wasn’t just one oddly specific time window. Then, I tried to make the boundary definition stricter, and this is where the result gets more complicated.
I rebuilt the sector transition proxy but required 48 hours between accepted transitions instead of 24. That should remove more of the noisy back and forth sector flipping. When I did that, the signal got weaker.
For the short window:
p = 0.133
And for the longer window:
p = 0.054
So, the relationship is clearly sensitive to exactly how I define a sector transition. This is why I want to follow this a little closer.
Solar Cycle Effect
I also broke the result apart by solar cycle. And, the relationship was definitely not equally strong across all three.
For the short post-transition window:
Solar Cycle 23
p = 0.049
Solar Cycle 24
p = 0.015
Solar Cycle 25 (partial record)
p = 0.631
Cycle 24 was clearly the strongest. And, the broader transition window showed the same basic pattern.
Cycle 24: RR = 1.138
Cycle 25: RR = 1.091
And Cycle 24 again stood out against its own shifted null tests:
Even with the stricter 48 hour boundary definition, the broader Cycle 24 result stayed interesting:
p = 0.033
I’m being careful with Cycle 25 because we’re still in the middle of it. So, we can’t yet say what the relationship means in Cycle 25.
For now, I’d just say the result appears to be sensitive to solar cycle conditions, with the strongest signal showing up in Cycle 24.
Did the solar wind transition actually come first?
This was another thing I wanted to check. If the heliospheric boundary was directly triggering the DRIFT change, I’d expect the sector transition to happen more often before the low-stability state than after it.
That wasn’t what I found.… The timing was pretty symmetrical.
About:
had a sector flip in the earlier window
compared with:
in the corresponding later window
So, I don’t think this result supports a simple: solar boundary crossing → polar motion changes story. At least, not yet.
What it does show is that low DRIFT stability and this IMF sector-transition environment appear together more often than expected.
Why? I don’t know yet.
They could be physically connected. They could both be responding to some larger state. Or, there could still be some structure in the data that I haven’t controlled for.
Another signal popped up
While I was doing this, another DRIFT variable caught my attention. Extreme values of Craig’s (R(t)) diagnostic were more common during periods of sustained southward IMF (B_z).
That effect was actually larger than the sector transition effect:
p = 0.039
Using a simpler instantaneous southward-(B_z) definition gave:
p = 0.044
I definitely want to come back to this...
So what did I actually find?
The simplest way I can put it is this:
The lowest stability states in DRIFT occurred about 10% more often shortly after an IMF sector transition than they did during the rest of the record.
And, when I deliberately scrambled the timing between the two datasets while preserving their internal time structure, the real result landed in the:
of 1,000 shifted null tests
That’s enough for me to think there is something here worth testing...
What I want to test next
The obvious next step is to replace my proxy with a vetted, independent HCS or sector boundary catalog and run the test again without changing the DRIFT definition.
I’d also like to stop treating every low stability hour as its own observation and instead group neighboring hours into independent low stability episodes.
Then I want to test:
actual cataloged HCS crossings
whether one sector polarity behaves differently from the other
whether the result changes around CIRs and high-speed streams
whether Solar Cycle 24 really is special
whether the relationship survives controls for geomagnetic activity
and the separate (R(t)) versus southward-(B_z) result
I want to know whether the changing heliospheric environment contains repeatable information about when Earth’s polar motion enters an unusual state.
This first pass says maybe….And, I think that’s interesting enough to keep going.
References
Polar motion / DRIFT: Craig Stone (@nobulart), DRIFT Dashboard. DRIFT calculates diagnostics from Earth Orientation Parameter and polar motion observations, including IERS based polar motion data.
Solar wind: NASA OMNI / CDAWeb, including interplanetary magnetic field measurements, solar wind plasma variables, Kp, ap, Dst and SYM/H.
ICME context: Richardson & Cane ICME catalog.
Sector boundary test: The intended WSO/Svalgaard boundary catalog could not be retrieved during this run because of an SSL connection failure. The main result reported here therefore uses an IMF phi angle sector transition proxy derived from OMNI observations.
Analysis period: January 1, 1996 to June 9, 2026.
