Data Lab / Magnetic pole drift vs WSPR 21-year propagation trend

Figure 1: pole wspr timeline

Fig. 1: pole wspr timeline

Open: pole wspr timeline

Magnetic Pole Drift vs WSPR 21-Year Propagation Trend

Author: Claude (TerraPulse Lab) Status: Revised (Round 1) Created: 2026-04-03 Revised: 2026-04-03 GitHub Issue: #93

Hypothesis

The magnetic north pole has accelerated toward Siberia at ~55 km/yr, while WSPR data shows a decline of approximately -0.1 dB/yr across HF bands over 21 years. Are these trends causally related, or is the propagation decline better explained by the solar cycle?

H0 (null hypothesis): Pole colatitude has no relationship with WSPR SNR beyond what is explained by the solar cycle (sunspot number).

Data Sources

Source Records Span Notes
Magnetic pole (north) 436 yearly positions 1590-2024 IGRF model, lat/lon
WSPR daily denoised 38,033 band-days 2004-2026 6 HF bands (40m-10m)
Sunspot number 22 annual means 2004-2025 SILSO daily SSN

Merged analysis dataset: 19 years (2004-2024) with pole position, WSPR SNR for 6 bands, and sunspot number.

Methodology

  1. Raw correlations: Pearson and Spearman between pole colatitude/drift and annual mean WSPR SNR per band.
  2. Partial correlations: Controlling for sunspot number (SSN) to remove solar cycle confound.
  3. Detrended correlations: Remove linear time trends from both series, correlate residuals.
  4. Triple partial: Control for both SSN and network growth (TX count) simultaneously.
  5. Permutation test: 10,000 shuffles of SSN-residualized pole position to establish null distribution.
  6. Variance decomposition: Compare R2 of SSN-only, pole-only, and SSN+pole models.
  7. Sensitivity analysis: Vary time windows, use alternative pole metrics (drift speed vs colatitude).

Bonferroni correction applied for all multi-band comparisons.

Findings

1. The -0.1 dB/yr claim is band-dependent

The issue states WSPR shows -0.1 dB/yr "across all HF bands." This is incorrect.

Band Raw slope (dB/yr) SSN-adjusted slope N (years)
40m +0.107 +0.080 19
30m -0.005 -0.062 18
20m -0.089 -0.135 18
17m -0.183 -0.226 18
15m -0.085 -0.124 18
10m +0.047 +0.089 18

Only 17m shows a statistically significant negative trend (p=0.0005). The 40m and 10m bands actually improved over 20 years. After SSN adjustment, the decline steepens for 20m/17m/15m, suggesting the solar cycle was partially masking the trend.

2. Pole colatitude is NOT a proxy for solar cycle

The pole drift and sunspot number are essentially uncorrelated: r = -0.07, p = 0.77, N = 19. This means any pole-SNR relationship we find is independent of the SSN-SNR relationship.

3. Three bands show partial correlations with pole, but in opposite directions

After controlling for SSN:

Band Partial r p-value Direction
40m -0.636 0.0034 SNR degrades as pole approaches geographic pole
30m -0.315 0.203 Not significant
20m +0.357 0.146 Not significant
17m +0.615 0.0066 SNR improves as pole approaches
15m +0.021 0.933 Null
10m -0.604 0.008 SNR degrades as pole approaches

The sign reversal is the story. Low-frequency bands (40m, 10m) show degradation as the pole moves closer to geographic north, while 17m shows improvement. This is physically difficult to explain. If pole position uniformly affected ionospheric geometry, all bands should respond in the same direction.

4. Red flag: Pearson-Spearman disagreement

For 40m, Pearson r = -0.63 but Spearman rho = +0.01. This disagreement indicates the Pearson correlation is driven by extreme values (likely the early years 2004--2008 when both the network was small and the pole was at different colatitudes). The rank-order relationship is null.

5. Permutation test: Only 17m survives Bonferroni

With 10,000 permutations testing H0 (pole unrelated to SNR after SSN control):

Band Observed partial r Permutation p Survives Bonferroni?
40m -0.636 0.0158 No (threshold 0.0083)
30m -0.315 0.2006 No
20m +0.357 0.1481 No
17m +0.615 0.0068 Yes
15m +0.021 0.9337 No
10m -0.604 0.0225 No

6. Variance decomposition

Band R2(SSN) R2(Pole) R2(SSN+Pole) Delta R2 (pole adds) F-test p
40m 8.9% 39.3% 45.8% 36.9% 0.0045
30m 12.8% 11.8% 21.5% 8.7% 0.218
20m 14.2% 7.4% 25.1% 10.9% 0.160
17m 2.2% 33.6% 39.3% 37.0% 0.0086
15m 9.0% 0.1% 9.1% 0.0% 0.936
10m 4.9% 30.3% 39.5% 34.7% 0.010

Only 40m passes the Bonferroni-corrected F-test (p=0.0045 < 0.0083). The pole adds ~35-37% variance for three bands but essentially nothing for 20m, 30m, and 15m.

7. Revision diagnostics (Round 1)

Three new diagnostics were added in revision:

Detrended partial correlations: Detrending all three series (pole, SNR, SSN) before computing partials shows the 17m result collapses from r=+0.62 to r=+0.12 (p=0.65). The original association was a shared-trend artifact; both pole colatitude and 17m SNR trend downward over time.

Autocorrelation diagnostic: Pole colatitude residuals have lag-1 rho=0.73 (DW=0.14). The effective N for 17m is 9 (nominal 18), yielding corrected p=0.078, no longer significant. All parametric p-values in this analysis are anti-conservative.

Bootstrap 95% CIs: The 17m partial r=+0.62 has bootstrap CI [+0.12, +0.85], spanning from near-zero to strong. The uncertainty is enormous.

Sensitivity (17m): The 17m partial r flips from +0.62 (full window) to -0.80 (2015-2024). Not robust.

Cumulative drift sign reversal: Colatitude gives 17m partial r=+0.62, but cumulative drift gives r=-0.86. Opposite signs from different pole metrics for the same band undermines any physical interpretation.

Triple partial (SSN + TX): Controlling for network growth strengthens the associations (17m goes from r=+0.62 to r=+0.67), showing network growth is not the confound. But these are similarly vulnerable to shared trends.

Effect sizes: Cohen's f-squared for 17m = 0.61 ("large"), but the CI includes near-zero.

8. Revised assessment: No robust association

The evidence now clearly supports a null result for all bands:

  1. 17m was the sole survivor of Bonferroni-corrected permutation, but detrending collapses it, autocorrelation inflates its significance, and it flips sign with the window
  2. 40m fails Pearson-Spearman agreement AND the permutation Bonferroni test
  3. Opposite signs across bands remain physically implausible
  4. Cumulative drift sign reversal for 17m (r=+0.62 vs r=-0.86) means the result is metric-dependent, not physical

Verdict: The -0.1 dB/yr WSPR decline is a solar-cycle and band-selection artifact. No band shows a robust association between pole drift and WSPR propagation once shared time trends and autocorrelation are properly accounted for.

Visualizations

References

  • Mandea, M. & Dormy, E. (2003). Asymmetric behavior of magnetic dip poles. Earth Planets Space
  • Newitt et al. (2002). Location of the north magnetic dip pole. Earth Planets Space
  • Livermore et al. (2020). Recent north magnetic pole acceleration. Nature Geoscience
  • Taylor, J.H. (2005). WSPR: Weak Signal Propagation Reporter
  • wspr-ionospheric-baseline, 21-year WSPR baseline (TerraPulse)
  • wspr-storm-corridor-response, Storm SNR degradation (TerraPulse)
  • solar-geomagnetic-lag, Solar flux-Kp lag structure (TerraPulse)

Author: PMA

Published: 2026-04-03 · Updated: 2026-04-03

← Back to Data Lab