TeslaQuake Evaluator · E004

GEOMAGNETIC STORMS VS GLOBAL EARTHQUAKE RATE

Do geomagnetic storms (Kp ≥ 7-/8-/9-) show any measurable effect on the number of declustered earthquakes in the following 14 days?

Result · NULL · NIK · TeslaQuake AI Evaluator · run 2026-09-09 · seed 369

NULL
Result: 295
n (headline cell): 1.015
Rate ratio: 0.454
two-sided p: 0.977–1.052
95 % range (ratio): 6
Cells tested: 3
Files fingerprinted

01 · The data

What was compared

Both series come straight from the agency that publishes them. Nothing is pre-filtered, nothing is hand-picked; the download is cached and fingerprinted (see 05).

Role	Series	Primary source	Range · catalogs
Factor	GFZ Kp index (definitive)	https://kp.gfz.de	1932 → 2026-07-24
Target	USGS ComCat	https://earthquake.usgs.gov/fdsnws/event/1/	A: M5.5+ 1973→ · B: M7.0+ 1932→

Data provenance · Tier A · agency data. Every input was downloaded by NIK from the publishing agency, cached unchanged and fingerprinted with SHA-256 before the test ran. Nobody hand-typed, trimmed or pre-selected a value.

02 · Method, in plain words

How the test was run

Method, as recorded in the result file

Storm onset after ≥7 quiet days (no day at or above the same Kp threshold in the previous 7); declustered (Gardner–Knopoff 1974) counts on days +1…+14 after onset vs. the expected rate (the mean of the shuffled null), vs. the 14 d before, and a 2000-draw shuffled-onset null (dates shuffled within the same calendar year); two-sided p. NULL is declared only where a ±15 % change would have been detectable.

Null model. The same count is taken at random dates, many times over. If the real dates land where the random ones usually land, the factor did nothing measurable. The p-value is the share of random draws that matched or beat the real count (two-sided).

Declustering. Aftershocks are removed with published time–distance windows so one large earthquake is counted once, not fifty times. Without this step almost any date-based test lights up for the wrong reason.

Fixed seed. The random draws start from seed 369, so a re-run reproduces the same numbers, not merely similar ones.

03 · Results

Every cell, nothing left out

Ratio 1.0 means the window after the factor held exactly as many events as any random window. A p above 0.05 means the random draws produced the observed count routinely. The 95 % range says how big a change the cell could have seen: NULL is only declared where a ±15 % change was detectable; a cell that shows no effect but could not have seen one that small is marked INCONCLUSIVE, not NULL.

catalog	storm class	n	Observed / window	Expected / window	Ratio	95 % range	two-sided p	Result
A · M5.5+ · 1973→	Kp ≥ 7- (G3 and up)	295	9.475	9.337	1.015	0.977–1.052	0.454	NULL
A · M5.5+ · 1973→	Kp ≥ 8- (G4 and up)	113	9.584	9.407	1.019	0.957–1.08	0.549	NULL
A · M5.5+ · 1973→	Kp ≥ 9- (top G4 + G5)	25	9.2	9.373	0.982	0.842–1.121	0.855	NULL
B · M7.0+ · 1932→	Kp ≥ 7- (G3 and up)	575	0.443	0.411	1.08	0.952–1.208	0.235	NULL
B · M7.0+ · 1932→	Kp ≥ 8- (G4 and up)	250	0.436	0.408	1.07	0.872–1.268	0.514	INCONCLUSIVE underpowered · could only see a change of ±20 % or more
B · M7.0+ · 1932→	Kp ≥ 9- (top G4 + G5)	65	0.492	0.417	1.181	0.807–1.554	0.391	INCONCLUSIVE underpowered · could only see a change of ±37 % or more

04 · Chart

The headline cell, drawn

Caveat

Kp begins 1932. The Carrington storm of 1859 is outside instrumental coverage and is reported separately from historical records only.

Prior work

Who asked this before us

An evaluation is not a discovery. These are the published results the question already has; where E-pages agree or disagree with them, the row says so.

Love, J. J. & Thomas, J. N. (2013). Insignificant solar-terrestrial triggering of earthquakes. Geophysical Research Letters 40, 1165–1170. doi:10.1002/grl.50211 — Same question, independent method (sunspots, solar wind, geomagnetic indices vs global earthquake occurrence): no measurable effect. E004 agrees.
Gardner, J. K. & Knopoff, L. (1974). Is the sequence of earthquakes in Southern California, with aftershocks removed, Poissonian? Bull. Seismol. Soc. Am. 64, 1363–1367. — Declustering windows used here.

Have a question of your own? One question, one dataset — NIK tests at least one reader submission a week and publishes it either way. Submit an idea →

05 · Re-run it

Reproduce every number above

The script downloads both series from the agencies, declusters, draws the null and writes E004_result.json. Change --window, --draws or --seed and see whether anything moves.

python3 tq_eval_004_storms.py --seed 369 --draws 2000 --window 14 --out . --analysis-end 2026-07-24

Item	Value
Script	tq_eval_004_storms.py v1.3.1
Script SHA-256	2ca595f36ff8e2e774f2b286c415bc94fb3682f9dd2b2145ee777f1ed3ffe718
Run on	NIK · 2026-09-09 · seed 369
Source copy	E004_reproduce.zip — the script, the three data files it used and its result.json, in one download.

06 · Data fingerprints

SHA-256 of every cached download

SHA-256 identifies the exact files this run used — the same files shipped in E004_reproduce.zip. The analysis window is pinned to 2026-07-24, so the command in the zip's README reproduces these numbers on any date. USGS ComCat is revised continuously and definitive Kp grows monthly, so a fresh download will not match these hashes; compare against the files in the zip.

File	SHA-256
kp_daily_max.json	082507cc361095eaabcb9845cca71f064252aa04ca420575304baeb958c92c69
quakes_M5.5_1973.csv	0ac6797e7e28479968a9947d7f7c0c5399fb47badd0a1edd764005ecf8efa55e
quakes_M7.0_1932.csv	5706fe97a042ad08b54412d04a8fb9e307f4408b5b87f92a770e2300c2bb18c0