The Carrington Event: The Solar Storm That Set the Wires on Fire

The event opens with an observation, and it is the first of its kind anyone ever made. On September 1, 1859, the British astronomer Richard C. Carrington and, independently, Richard Hodgson observed a white-light solar flare, an eruption of extraordinary brightness from a large sunspot group. It was the first documented observation of a solar flare in history.

Then a gap, and the length of the gap is itself evidence. The flare was followed approximately 17.6 hours later by the onset of a severe geomagnetic storm. A transit time that short implies a coronal mass ejection moving at roughly 2,500 kilometers per second, which is extremely fast. Typical CME transit times run 2 to 4 days.

What arrived is the largest storm on the instrumental record. The Carrington Event of September 1 to 2, 1859 was the most intense geomagnetic storm in recorded history, caused by a coronal mass ejection from the Sun striking Earth's magnetosphere.

One instrument is doing most of the work. The Colaba magnetometer at Mumbai recorded a horizontal magnetic field depression of approximately -1,600 nT. That is a measured local deflection at one observatory. Hold on to that phrasing, because every argument below turns on it.

The famous number is not that measurement. It is a derivation from it. Tsurutani, Gonzalez, Lakhina and Alex (2003) derived a Dst of about -1,760 nT for the 1859 storm as consistent with the Colaba local-noon horizontal-field response of dH = 1,600 plus or minus 10 nT. Dst is a global index; it measures the intensity of the ring current around the whole planet. The Colaba trace is one station's record of what happened above one place. Our own file prints -1,760 nT in its summary and -1,600 nT in its body as though one measurement were being quoted twice at different precisions. They are two different quantities, and the step between them is the entire dispute. Our file also names no source for the -1,760 figure anywhere in its thirteen-entry bibliography.

That step was challenged in print almost at once. Akasofu and Kamide published a Comment on the Tsurutani paper in the Journal of Geophysical Research in 2005, disputing the move from a single observatory's horizontal-field record to a global Dst value. That exchange is the origin of the whole three-way disagreement that follows.

Siscoe, Crooker and Clauer (2006) argued that the Colaba instrument may have been affected by local anomalies, and that the true global Dst was likely -850 to -900 nT. Notice what that concedes and what it does not. It roughly halves the headline figure while leaving 1859 the strongest recorded geomagnetic storm by a significant margin.

One strand of the literature goes considerably further, and it deserves to be put at full strength rather than mentioned in passing. Cid, Palacios, Saiz, Guerrero and Cerrato (2014) argue that the Dst given in the literature for the Carrington storm was obtained from the magnetic record of just one observatory, and question whether a local magnetic disturbance recovers on the same hyperbolic law that a global ring-current signature does. Building on their own 2013 work, they put the storm's minimum Dst at -685 nT, which they describe as comparable to the 1989 Quebec storm. If Cid and colleagues are right, the defining event of this whole subject was not in a class of its own. It was a bad storm of a kind the modern grid has already survived once.

For scale: the strongest storm of the modern era, March 1989, reached a Dst of -589 nT. Set that beside the three estimates above and the shape of the disagreement is plain. Against -1,760 nT, 1859 is a different animal entirely. Against -685 nT, it is the same animal, slightly larger.

There is a definitional trap here, and our own file walks straight into it. In the risk literature, Carrington-class is an operational threshold rather than a description of 1859, and Riley (2012) sets it at a storm with Dst below -850 nT. But -850 nT is also one of the three candidate values for the 1859 storm itself, and it is the low one. So the same number is being made to serve as both the boundary of the class and as an estimate of the event the class is named after. Whether 1859 was twice the threshold or barely reached it is precisely what is unresolved, and any sentence that treats Carrington-class and the Carrington Event as interchangeable has quietly assumed an answer.

Coronal mass ejections are large-scale eruptions of magnetized plasma from the solar corona, carrying roughly 10^12 to 10^13 kg of material at speeds of 300 to over 3,000 km/s. Orientation decides whether one matters to us. When a CME's magnetic field points southward, anti-parallel to Earth's dayside magnetopause field, it enables efficient energy transfer into the magnetosphere through magnetic reconnection, and that is what drives a geomagnetic storm.

The coupling to the ground is straightforward physics. Geomagnetically induced currents arise when rapid changes in Earth's magnetic field during a storm induce electric fields in the Earth's surface, which then drive quasi-DC currents through any grounded conducting network: power grids, pipelines, undersea cables. The storm does not need to touch the hardware. It only needs the hardware to be long, conductive and earthed at both ends.

And that is where a transformer is vulnerable. In transformers, geomagnetically induced currents cause half-cycle saturation, leading to harmonic distortion, overheating, and potentially irreversible damage to high-voltage transformer windings. Irreversible is the operative word. A saturated transformer is not a device that trips and resets.

The March 1989 geomagnetic storm, at Dst -589 nT and triggered by a CME, caused the collapse of Hydro-Quebec's power grid in 92 seconds, leaving 6 million people without electricity for up to 9 hours. That is the whole demonstration: a modern power system taken down faster than a reader can finish this paragraph, by a storm nobody would call historic. Our file describes this as happening at about one third the intensity of the Carrington Event, and that ratio needs the caveat from section 02 attached to it. It holds against the -1,760 nT estimate. Against Cid's -685 nT it does not hold at all, and the 1989 storm becomes something close to a like-for-like preview.

On July 23, 2012 the STEREO-A spacecraft, orbiting the Sun ahead of Earth, detected a CME that passed through Earth's orbital position. Analysis by Baker and colleagues, published in Space Weather in 2013, concluded that the CME's speed of about 2,500 km/s, its magnetic field strength and its duration were all comparable to the 1859 Carrington Event.

Earth had passed through that same orbital position approximately one week earlier. Had the eruption occurred 7 to 9 days sooner, Earth would have been directly in the CME's path.

Two figures from that work are worth carrying with their status marked. Baker's own summary of the event, as reported by NASA, is that in his view the July 2012 storm was in all respects at least as strong as the 1859 Carrington event. NASA's account of the same work gives a Dst for the July 2012 storm of about -1,200 nT had it struck Earth, describing that as comparable to the Carrington Event and twice as severe as the March 1989 Quebec storm. That -1,200 nT is a modelled counterfactual for a storm that did not hit us, not a measurement of one that did, and it reaches this article through a NASA popular-science feature rather than through the peer-reviewed paper. It belongs in the article, labelled, and not in a headline.

The near-miss underscored that Carrington-class events are not merely historical curiosities but occurring solar phenomena, and that one could strike Earth with essentially no warning beyond the roughly 15 to 60 minute transit time from the L1 monitoring point to Earth. Section 09 takes that last clause apart, because it is the single most misquoted sentence in this subject.

Our file leads with $1 to $2 trillion in first-year economic damages from a Carrington-class event, with recovery taking 4 to 10 years, attributed to the National Academy of Sciences in 2008. Three things about that attribution are wrong, and they compound. It is not a National Academy finding: it is a scenario analysis by John Kappenman of Metatech Corporation, reproduced inside a National Research Council workshop report, a format that records what participants presented rather than endorsing it. The modelled storm is not Carrington-class: it is a repeat of the May 1921 geomagnetic storm, which is smaller than 1859. And the figure is United States only, not global. Every popular retelling that says a Carrington event would cost the world one to two trillion dollars has inherited all three errors at once.

Stated properly, the scenario is this. Kappenman and Metatech modelling of a repeat of the May 1921 storm predicts permanent damage to roughly 350 extra-high-voltage transformers in the United States, up to about 130 million people without power, a first-year United States economic impact of $1 to $2 trillion, and full recovery in 4 to 10 years depending on the extent of the damage. That is a real, specific, published scenario by a named modeller, and it is worth taking seriously on those terms. What it is not is a consensus estimate, a measurement, or a finding of the National Academies.

It also did not go unexamined. The 2011 JASON review, commissioned by the Department of Homeland Security to assess exactly this worst-case grid scenario, declined to endorse it. Its stated conclusion, as reported by the Federation of American Scientists, was that the reviewers were not convinced the worst case scenario is plausible, nor that the analysis behind it, built on proprietary algorithms, is suitable for deciding national policy. JASON recommended instead that a rigorous and fully transparent risk analysis of the United States grid be carried out. That review appears nowhere in our own research file, which is the single largest gap in it: the document presents a contested scenario as settled by leaving out the body that contested it.

One detail inside that scenario has been mangled in transmission, and our own file mangles it. The file states that the United States alone has approximately 350 high-voltage transformers of the type most vulnerable to damage. That is a misreading of the scenario's own output. The 350 figure is the number of extra-high-voltage transformers the model predicts would be permanently damaged, not a count of how many exist. The real United States fleet is far larger. The JASON review describes the same scenario as predicting catastrophic damage to more than 300 extra high voltage transformers, which is the same quantity said correctly. Read as a census, the sentence tells a reader that the American grid rests on 350 machines. It does not, and the difference matters in both directions: the loss is smaller than the sentence implies as a share of the fleet, and the sentence is worthless as a description of the fleet.

The reason any of this takes years rather than weeks is manufacturing, not repair. Our file gives high-voltage transformers of this type typical replacement lead times of 12 to 24 months, and notes that most are custom-manufactured overseas. That figure is now out of date, and the update runs against our file rather than for it. Large power transformers are custom-designed and historically carried procurement lead times of one year or longer, stretching beyond 20 months when key parts or materials were hard to obtain; as of the 2024 Department of Energy Large Power Transformer Resilience Report to Congress, 36-month lead times are commonly quoted and maximum lead times reach as much as 60 months. If the recovery argument depends on the replacement bottleneck, and it does, the honest current number makes the point more forcefully than the one our file carries.

There is a second major estimate, from the insurance side, and it must not be blended with the first. Lloyd's of London, in a 2013 report produced with Atmospheric and Environmental Research, estimated that 20 to 40 million Americans could lose power from a Carrington-level storm, with cascading impacts on water supply, food distribution, medical systems, financial systems and communications. Our file gives the outage as weeks to months; the report's own range is wider and franker, 16 days to one or two years depending on replacement transformer availability. It put total economic losses at up to about $2.6 trillion and placed the greatest risk along the Atlantic corridor between New York and Washington DC, with elevated risk in parts of the Midwest and along the Gulf Coast. That $2.6 trillion is a different and larger figure from the $1 to $2 trillion above, produced by different people modelling a different thing. The two are not a range and should never be quoted as one.

On the defensive side the picture is real but modest. Mitigation measures including GIC blocking devices, transformer monitoring, strategic spare transformer stockpiles and improved space weather forecasting have been partially implemented since the 2008 report, but progress is incremental and most national grids remain significantly vulnerable.

Riley, in Space Weather in 2012, estimated the probability of a Carrington-class event, defined as Dst below -850 nT, within a decade at approximately 12 percent, derived from extreme-value statistics applied to the historical record of geomagnetic storm intensities. This is a genuine published estimate, correctly cited in our file, and it is also the highest figure in the literature and the one popular writing reaches for. It should never appear on its own.

Kataoka, in Space Weather in 2013, estimates the probability of another Carrington-type storm over the next decade at 0.04 to 0.06, that is 4 to 6 percent, using cumulative distribution functions built from an 89-year list of magnetic storms recorded at Kakioka. Our own file cites this paper for a figure of roughly 1 to 4 percent per decade. That is not its result.

Chapman, Horne and Watkins, in Geophysical Research Letters in 2020, are not measuring the same thing at all. Using the aa index across the last 14 solar cycles, they report a 14-cycle average of roughly 4 percent chance per year of at least one great storm, meaning Dst below -500 nT, and roughly 28 percent per year of at least one severe storm below -250 nT, and they frame a Dst of about -809 nT as roughly a 1-in-151-year event. They note that the Carrington event's estimated amplitude sits within the same distribution as other extreme activity seen in the aa index since 1868. Co-author Richard Horne's own summary of the result is that a super-storm can happen more often than we thought. Our file cites this paper as producing a lower estimate. It is per year rather than per decade, and its own conclusion points in the opposite direction.

The real floor of the published range belongs to a paper our file never cites. Morina, Serra, Puig and Corral, in Scientific Reports in 2019, estimate the probability of a Carrington-like event with Dst below -850 nT in the next decade at 0.92 percent, with a 95 percent confidence interval of 0.46 to 1.87 percent, using a counting process with Weibull inter-occurrence times. The authors state explicitly that this is substantially lower than the roughly 12 percent previously reported, and they attribute the difference to exponential rather than power-law behaviour in the tail of the Dst distribution. That is the actual disagreement, stated by the people having it: not about the Sun, but about the shape of a tail.

One line from further back, and only one, because that record has its own file in this wing. Geological proxies, beryllium-10 and carbon-14 spikes in ice cores and tree rings, record several solar particle events over the past 10,000 years that may have exceeded the Carrington Event, of which the Miyake events of 774 CE and 993 CE are the strongest identified. The caution attached is important: those were extreme solar energetic particle events rather than necessarily extreme CME-driven geomagnetic storms, so they are evidence that the Sun can do more than 1859, not evidence about how often it does 1859.

Some risk analysts have suggested that a Carrington-class event, combined with modern society's near-total dependence on electricity and digital systems, could trigger cascading infrastructure collapse potentially resulting in large-scale fatalities, from loss of water treatment, medical care and food distribution. Our file states that these catastrophic scenarios are plausible but depend heavily on the duration of the outage, its geographic extent, government response and pre-positioned mitigation. It names no analyst and no study for any of it, which is why this sits where it sits. The dependencies are the honest part of the claim: every one of them is a variable nobody has measured, because the event has not happened to a wired civilization.

Claims that a solar superstorm could strike without warning are partially misleading. Solar flares travel at light speed, giving roughly an 8-minute warning of the flare itself, while CMEs take 15 to 96 or more hours to reach Earth. The DSCOVR satellite at L1 provides approximately 15 to 60 minutes of advance warning once the CME passes L1, and ACE and DSCOVR data enable real-time forecasting that did not exist in 1859. The truthful version of the sentence is much less dramatic and much more useful: the warning is short, it is not zero, and what it is short for is protective action on a grid, not evacuation.

That window is confirmed by the agency operating the satellite. NOAA's Space Weather Prediction Center states that from its position at Lagrange point 1, about one million miles from Earth, DSCOVR can typically provide 15 to 60 minutes of warning before a CME's particles and magnetic field reach Earth, by observing sudden increases in density, total interplanetary magnetic field strength and solar wind speed.