The dose that won't resolve
A geomagnetic storm with dB/dt of 100 nT per minute induces a geoelectric field at the ground surface of roughly 0.1 to 3 V/km, depending on local crustal resistivity (model-dependent). That range is the number space-weather engineers use for transformer-saturation and pipeline-corrosion models. It is also the number geomagnetic-physiology papers borrow, usually without re-deriving it, to argue storms deliver a biologically meaningful dose. Converted into tissue current density using typical soft-tissue conductivity (0.1–0.5 S/m), that same field range yields roughly 10 microamps to 1.5 milliamps per square metre (model-dependent) — a span that straddles, rather than clears, the current densities thought to matter for cardiac and neural signalling, conventionally cited near 1 mA/m².
That is the whole controversy compressed into one calculation. Depending on which end of entirely defensible assumptions a paper chooses, a storm delivers either a negligible nudge or a dose in the same ballpark as the body's own bioelectric activity. Neither number is wrong. That is exactly why decades of studies keep producing contradictory results on essentially the same phenomenon — not because the field is credulous, but because it has been trying to settle, with single underpowered studies, a question whose honest answer is a dose estimate that spans the ambiguity zone itself (measured; interpretation contested).
The arithmetic, shown
The uncertainty is not primarily about storm intensity. It is about which physical model of body-field coupling is correct, and the field has not standardized on one.
How the 10 µA/m² to 1.5 mA/m² range is built, and where it could be off by orders of magnitude
Two different coupling models are in circulation, and they disagree by many orders of magnitude before any physiology is introduced. Model one treats the body as galvanically embedded in the same extended conducting medium as the ground: the storm's geoelectric field (0.1–3 V/km, i.e. 0.1–3 mV/m) acts directly across tissue of conductivity 0.1–0.5 S/m, giving J = σE in the 10 µA/m²–1.5 mA/m² range used above. This is the model implicit in most epidemiological papers that cite the ground-level V/km figure as if it transfers unchanged to a standing human. Model two treats the body as a small induction loop responding only to local dB/dt via Faraday's law, using a torso-scale loop area (~0.03 m²) rather than the continental-scale loop area that produces grid-relevant geoelectric fields. That calculation yields induced fields many orders of magnitude smaller than model one — closer to picovolts per metre than millivolts per metre. The two models differ because the V/km figure is derived for extended conductors (transmission lines, pipelines, undersea cables) integrating EMF over tens to hundreds of kilometres; a human body is not such a conductor unless it is in direct, extended galvanic contact with the earth, which it normally is not. Additional variance comes from crustal resistivity, which differs by one to two orders of magnitude between igneous and sedimentary terrain, and from storm dB/dt itself, which varies from tens to several thousand nT/min with latitude and storm class. No published human physiology study reports which coupling model, which local resistivity, or which measured (rather than assumed) tissue conductivity it used. Until one does, the dose is not wrong — it is underspecified (model-dependent).
The MRI calibration — and why it cuts against intuition
Earth's static field is approximately 50 microtesla; clinical MRI systems run at 1.5–3.0 tesla, roughly 30,000–60,000 times stronger (established). This comparison is usually deployed to argue that a field strong enough to be used for decades in millions of scans with no established chronic bioeffect from its static component makes a storm's fractional perturbation of Earth's field — on the order of a percent during a severe event — implausible as a health hazard.
That argument is not quite the right one, and getting it right actually matters for where the mechanism debate should go. Static field magnitude does not induce current; its rate of change does. MRI's gradient coils switch at rates on the order of tens of tesla per second specifically because that is the regime regulators use to define peripheral nerve stimulation thresholds — a real, measured, dose-calibrated bioeffect (established). A storm's dB/dt of 100 nT/min is roughly ten orders of magnitude slower and weaker than the switching rates known to trigger nerve stimulation. Properly used, the MRI comparison argues against a brute-force neurostimulation mechanism for storms, not for dismissing geomagnetic physiology outright. It pushes any plausible mechanism away from direct Ohmic excitation and toward something that does not need to beat classical current-density thresholds at all — a dedicated transducer.
Where weak fields do work: birds, fish, and the transducer problem
Biology demonstrably transduces fields in this range; that is not in dispute (established). Migratory birds and several fish species detect geomagnetic-strength fields for navigation. What makes those systems different from a generic human tissue claim is specific and answerable: they possess dedicated, evolved amplifying structures, not diffuse passive conduction. The avian candidate mechanism is a light-dependent cryptochrome radical-pair system in the retina, where a weak magnetic field biases the lifetime ratio of singlet versus triplet spin states in a photochemically generated radical pair — a quantum effect that a cell can amplify into a neural signal because it alters reaction yield, not because it moves enough charge to compete with thermal noise. The fish candidate mechanism is magnetite crystals mechanically coupled to hair cells, where torque on a permanently magnetized particle gates an ion channel — a mechanical lever, not an electrical one. Both have identified neural pathways carrying the signal onward (ophthalmic/trigeminal routes in fish, for example).
No human tissue has an identified, characterized receptor cell population performing either of these jobs. Magnetite has been found in human tissue, including the brain, but its physiological role is unproven (measured; interpretation contested). Human cryptochromes (CRY1/CRY2) are expressed and functional as circadian repressors, and reconstitution experiments putting human CRY into insect magnetosensitive assays have shown retained magnetic sensitivity in that heterologous system — but whether human cryptochrome does anything magnetosensitive in its native human context, at native expression levels, in native tissue, is unresolved (speculative — flagged). That gap — a transducer exists in principle, is not demonstrated in practice — is the single mechanistic route by which the dose-arithmetic problem described above could stop mattering, because a chemical-kinetics or mechanical-torque transducer does not need to clear a current-density threshold at all.
Melatonin: a single bridge, not yet load-tested
The pineal-melatonin hypothesis is the oldest specific mechanistic claim in this literature, going back to correlational work associated with Persinger and Burch linking geomagnetic activity to pineal/melatonin measures. A recent animal study reports that moderate storms (50 nT) modulate melatonin synthesis pathways in a rat model of depression (Brain Res Bull 2025, 215:111369) (measured; interpretation contested). This is informative but narrow: it is one animal study, in a disease model chosen for a different purpose, with no independent replication reported, and no dose-response curve across storm intensities published alongside it. It is consistent with the transducer hypothesis above — the pineal gland is a biologically plausible site for either a magnetite- or cryptochrome-linked effect on a biosynthetic pathway rather than on electrical excitation — but "consistent with" is not "established by." The honest status of the melatonin record is: a mechanistically coherent, single-study, unreplicated finding sitting on top of a decades-old, methodologically dated correlational literature that was never built to the standard current cardiovascular epidemiology now demands.
Heart rate variability: the epidemiology trap
The cardiovascular literature is where the dose-arithmetic ambiguity collides hardest with statistical method. A large cohort analysis using the Normative Aging Study reported reduced HRV in elderly men associated with intense geomagnetic disturbance (Zilli Vieira et al. 2022, Sci Total Environ 839:156235) (measured; interpretation contested). A separate analysis found that many reported HRV–geomagnetic correlations disappear once time series are corrected for autocorrelation and multiple comparisons (Eur J Appl Physiol 2020, 120(6):1371-1381) (measured; interpretation contested). A 2025 scoping review mapping cardiovascular outcomes found substantial heterogeneity and a near-total reliance on ecological study designs that cannot establish individual-level causality (Scoping review 2025). Mechanistically supportive animal work exists — baroreflex sensitivity changes in rabbits during geomagnetic disturbance (Gmitrov & Gmitrova 2004, Bioelectromagnetics 25(2):92-101) — but it is a small, unreplicated, non-human physiology study, not a bridge to human epidemiology on its own.
Three specific, named hazards recur across this literature and explain why it stays contradictory rather than converging: geophysical and physiological time series are both intrinsically autocorrelated, inflating apparent significance when untreated; studies run enough comparisons across lags, indices, and subgroups that multiple-comparison correction changes conclusions; and storm days covary with season, temperature, barometric pressure, and air pollution, any of which independently affects cardiovascular endpoints and few studies adjust for all of them simultaneously.
The record at a glance
| Claimed effect | Reported magnitude | Typical study size | Replication status | Evidence rung |
|---|---|---|---|---|
| Reduced HRV during intense storms (elderly cohort) | Statistically significant reduction; effect size not large | Large cohort (Normative Aging Study) | Not independently replicated at reported effect size; contradicted after autocorrelation/multiple-comparison correction in other datasets | Ecological/epidemiological, confounded |
| HRV–geomagnetic correlation (general literature) | Variable, inconsistent across studies | Small to moderate | Largely disappears after statistical correction | Weak / contested |
| Melatonin synthesis modulation (moderate storm, 50 nT) | Pathway-level modulation in a rat depression model | Single animal study | No independent replication reported | Preliminary, mechanistic, single study |
| Baroreflex sensitivity shift during disturbance | Measurable change in rabbit model | Small animal study (2004) | No major human follow-up identified | Early mechanistic, unreplicated in humans |
| Cardiovascular hospital admissions vs. storm activity | Inconsistent across datasets | Large administrative/ecological datasets | High heterogeneity per scoping review; no causal design | Ecological, low causal weight |
Operational stakes: why the G-scale matters regardless
None of the above should be confused with the engineering domain where the same geoelectric fields are unambiguous. NOAA's Space Weather Prediction Center G-scale grades geomagnetic storms G1 through G5 by planetary Kp index, from G1 (Kp=5, minor) to G5 (Kp=9, extreme) (established). At G1–G2, effects are weak fluctuations in power systems and minor satellite drag; at G4–G5, the scale describes widespread voltage control problems, transformer protective-system tripping and potential damage, HF radio blackouts, degraded satellite navigation accuracy, and increased radiation exposure risk for high-latitude, high-altitude aviation crew and passengers. These are established, calibrated, operationally actionable risks, governed by the same induced geoelectric fields discussed above acting on long conductors — transmission lines, pipelines, submarine cables — where the continental-scale coupling model is the correct one by construction.
G-scale quick reference
| Scale | Kp index | Representative operational impact |
|---|---|---|
| G1 | 5 | Minor power grid fluctuations; weak impact on satellite operations |
| G2 | 6 | High-latitude power systems may experience voltage alarms; drag on low-orbit satellites increases |
| G3 | 7 | Voltage corrections required; HF radio intermittently degraded; GPS error increases |
| G4 | 8 | Widespread voltage control problems; possible transformer damage; HF radio blackout on sunlit side |
| G5 | 9 | Extensive grid system collapse risk; transformer damage; complete HF blackout; satellite navigation degraded for days |
What would change our mind
- Measured, not modeled, tissue dose. Simultaneous local magnetometer dB/dt recordings paired with calibrated phantom or minimally invasive tissue-field measurements at the same site, across multiple crustal-resistivity regions, would replace the current model-dependent 10 µA/m²–1.5 mA/m² range with an actual number and resolve which coupling model is closer to reality.
- A pre-registered, adequately powered cardiovascular cohort. The effect sizes that survive autocorrelation and multiple-comparison correction in the existing literature are small enough that a study with the power to detect them reliably would need continuous HRV/ECG monitoring across several thousand person-days, with season, temperature, pressure, and pollution as co-variates specified before unblinding — an order of magnitude beyond most published designs.
- A direct magnetosensitivity assay in human pineal or retinal cell lines. Testing whether human cryptochrome, in its native cellular context, shows any radical-pair response to oscillating fields matching real storm dB/dt spectra would settle whether a transducer exists at all, independent of any epidemiology.
- Independent replication of the 2025 rat melatonin finding, with a dose-response curve across storm intensities and a second laboratory, would tell us whether the one mechanistically coherent animal result generalizes or was a single-study artifact.



