Sven Erik Matzen

Software Architect | Cloud & Security Expert | AI-enabled Solutions

The Solar Storm That Dated the Vikings: 1021 and the Proof Before Columbus

🎧 Listen to this article

History · 2026-10-18

EU label: fully AI-generated content Fully AI-generated article (no prior review).

The Hook: One Year, Not One Century

Historical dates sound precise and usually are not. "Around 1000," "in the early eleventh century," "between 990 and 1050" — such statements are not estimates in the colloquial sense. They are properly declared uncertainty intervals. Read casually, they look like caution. In fact they are the hard ceiling of what a method can deliver.

For what is arguably the best-known open question in North Atlantic history — when, demonstrably, did Europeans stand in the Americas before Columbus? — that ceiling read, for decades: sometime in the window between about 990 and 1050. Half a century wide. Wide enough to contain two generations, and therefore too wide for almost any historically interesting follow-up question.

In October 2021, a paper appeared in Nature that collapsed this window to a single point. Margot Kuitems, Michael W. Dee, and a team spanning Groningen, Parks Canada, Mannheim, and Memorial University presented an exact felling year for three pieces of wood from the Norse layers at L'Anse aux Meadows in Newfoundland: AD 1021. Not an interval. A calendar year.

What is remarkable is not the number. What is remarkable is where it comes from. It does not come from a new find, a new excavation, or a better instrument in any obvious sense. It comes from the Sun. In late 992, our star threw a shower of particles at Earth's atmosphere that produced a sharp, globally simultaneous spike in radiocarbon. Every tree that grew in the following year wrote that event into exactly one of its rings. A thousand years later, that ring can be located in a piece of wood — and counted forward from.

This article is about that count. But it is mostly about the question underneath it, which reaches far beyond Vikings: what exactly does it mean to have a date? What does a number carry, and what does it not? We will see that the beautiful, sharp result "1021" rests on a surprisingly narrow foundation — not because it is uncertain, but because it says one thing very precisely and is silent about almost everything else. And we will look at three counter-cases — a forgery, a false alarm, and a contested gene — in which it becomes visible how differently load-bearing evidence can be that looks, at first glance, much the same.

Part 1: What the Sagas Tell Us — and What They Do Not Prove

Three Names for a Coast

The literary basis is quickly told. Two Old Norse prose texts, the Grœnlendinga saga (Saga of the Greenlanders) and Eiríks saga rauða (Saga of Erik the Red), report voyages westward. Both were written down in the thirteenth and fourteenth centuries — roughly three centuries after the events they describe.

The sagas name three lands beyond Greenland: Helluland ("land of flat stones," today usually identified with Baffin Island), Markland ("forest land," generally placed in Labrador), and Vinland, which on the current majority view includes Newfoundland and possibly further areas around the Gulf of Saint Lawrence. The figures are familiar: Bjarni Herjólfsson, blown off course in the 980s and sighting land; Leif Erikson, who according to the Saga of the Greenlanders established a small settlement some fifteen years later; later Thorfinn Karlsefni and Freydís. The texts also report encounters with people the Norse called Skrælingjar — a blanket term that scholarship has applied both to Thule Inuit and to Beothuk.

The earliest written mention of Vinland, however, comes not from Iceland but from northern Germany. Around 1075, Adam of Bremen writes in the Descriptio insularum Aquilonis, the fourth part of his Gesta Hammaburgensis ecclesiae pontificum, of an island called Winland, "for the reason that grapevines grow there by themselves, producing the best wine." His source, by his own account, was the Danish king Svend Estridsen. That is, less than a hundred years after the voyages, independent evidence that the name was circulating in northern Europe.

The Fight Over a Vowel

Hanging on that viticulture is an instructive philological dispute. In 1898 the Swedish philologist Sven Söderberg argued (published 1910) that the name should not be read as vín-land ("wine") with a long vowel, but as vin-land with a short one — Old Norse vin in the sense of "meadow, pasture." Vinland would then be pasture-land, not wine-land, and the whole grape question would dissolve. Väinö Tanner (1941) and Valter Jansson (1951) followed this reading.

The other side has prevailed, and with an argument worth understanding. In 1977 Einar Haugen showed that the element vin had already shifted in meaning to "farm, homestead" before the Old Norse stage of the language — so at the point when the naming must have happened, the sense "pasture" was no longer available. Erik Wahlgren called Söderberg's thesis "quite untenable." Kirsten Seaver writes that the short-vowel reading has been "so thoroughly discarded" that only the long-vowelled vin, that is "wine," remains. Birgitta Wallace, for decades the leading archaeologist of the site, considers attempts to substitute other berries for the grapes "insupportable."

Worth noting: this is a linguistic argument, not an archaeological one. It says something about what the Norse meant, not about where they were.

Memory as a Source

Here lies the real difficulty with the sagas, and it is not the difficulty laypeople expect. The objection is rarely "the sagas lie." It is this: the sagas are the product of three hundred years of oral transmission, written down in a Christian literary culture with its own conventions, narrative patterns, and political interests. What stands in them is memory of memory of memory.

Memory research has studied very precisely what happens in that process, and the findings are uncomfortable for historians: remembering is not retrieval from storage but reconstruction, and each reconstruction takes up material from the present. A saga that has run three centuries through the mouths of storytellers can preserve core information — directions, names, striking particulars — while also containing episodes that never happened but told well. The difficulty is not that everything is false. The difficulty is that from the inside, there is no way to distinguish the stable core from what accrued along the way.

This is exactly why the archaeological record is not simply a confirmation of the sagas. It is a second archive, independent of them — and its value lies entirely in that independence.

Part 2: The Place — L'Anse aux Meadows

A Tip from the Village

In 1960 the Norwegian lawyer and explorer Helge Ingstad followed a lead from George Decker, a resident of the village of L'Anse aux Meadows at the northern tip of Newfoundland, pointing to grassy mounds the locals called the "old Indian camp." Between 1961 and 1968, Helge Ingstad and his wife, the archaeologist Anne Stine Ingstad, carried out seven excavation campaigns.

It is worth noting how this find came about: not through a measurement technique but through local knowledge that had been available for decades and that nobody acted on, because nobody was asking in the right category. The mounds were known. Their interpretation was wrong.

What Lay in the Ground

Eight buildings were found — labelled in the literature A through J — built of sod over a wooden frame, in a style known from Iceland and Greenland. The largest dwelling (F) measured roughly 28.8 by 15.6 metres. In total, over 800 Norse objects came to light.

Two findings carry particular weight. First: building J was an iron smithy with a forge and slag, and there was a smelting hut that produced iron from bog ore. On current assessment, only a single smelt took place, yielding enough iron for perhaps 100 to 200 nails. That is very little — and precisely for that reason it is informative. This is not an industrial site; it is the minimum you need in order to repair ships. Second: part of a spindle and a bone needle suggest women were present; stone loom weights came from building G. A bronze fastening pin completes the picture of a settlement rather than a bivouac.

The size of the buildings implies capacity for 30 to 160 people. That has to be set against Greenland's total population of roughly 2,500: an undertaking on this scale was a considerable effort for the colony. The anthropologist John Steinberg considers occupation "at least sporadically for perhaps 20 years" plausible; Eleanor Barraclough reads the place as a temporary boat-repair facility.

The Walnuts That Do Not Grow There

The most elegant single finding at the site is butternuts (Juglans cinerea). Butternut trees do not grow naturally north of New Brunswick — on another formulation, not north of the St. Lawrence. Yet they were lying in the Norse layers at L'Anse aux Meadows.

The inference is more compelling than most archaeological inferences: either the Norse themselves, or material that travelled with them, came from the south. And because butternuts occur in the same range as wild grapes, the finding additionally supports Birgitta Wallace's position that L'Anse aux Meadows is not Vinland but — in her phrasing — a place "described in the sagas as part of Vinland": a northern base from which operations ran further south.

One can see here how a strong argument is built. It uses not the presence of an object but the impossibility of its natural presence. Negative evidence — "this does not grow here" — often carries more than positive evidence.

The southern extent of Vinland nonetheless remains disputed. Proposals run from Cape Breton (Storm 1887; Fischer 1902) through southern Newfoundland (Morison 1971) and Miramichi Bay in New Brunswick (Wahlgren 1986) to present-day New York City (Bergthorsson 1997). Wahlgren argued that L'Anse aux Meadows cannot be Vinland because the saga location has both salmon and wild grape, and placed Vinland in Maine or New Brunswick. And to this day: there is no second unambiguously Norse settlement in Canada or on the North American mainland.

Part 3: The Limit of Radiocarbon Dating

Why 14C Alone Does Not Give a Year

For dating organic material, radiocarbon is the standard tool, and it is excellent. It has only one structural property that many underestimate: by its nature it yields not a year but a probability distribution over years.

Three reasons overlap here. First, radioactive decay is a statistical process; every measurement has counting statistics and therefore an error bar. Second — and this matters more — the atmosphere's 14C concentration was not constant. One therefore needs a calibration curve to translate measurements into calendar years. That curve is not monotonic: it has plateaus on which several centuries return the same measured value, and wiggles that can assign a single measurement to several separate time windows. Third, one never dates the event but the material: a beam cut from the heartwood of a three-hundred-year-old tree is three hundred years older than the moment the tree was felled — the so-called old-wood effect.

For L'Anse aux Meadows this produced the familiar result: a dating range of 990 to 1050 AD, with a mean around 1014. Earlier work had placed the Norse presence between 970 and 1030 AD. That is solid science and at the same time an answer too coarse for questions about sequence: you cannot infer from it whether one voyage took place or five, whether the settlement lasted one summer or twenty years, whether Leif or Karlsefni is meant.

The Dispute of 2019

How tight that limit is became visible in an exchange two years before the 1021 paper, which I consider the most instructive piece of this whole story.

In August 2018 a team around Paul M. Ledger, Linus Girdland-Flink, and Véronique Forbes set out to sample a peat bog about 30 metres east of the Norse ruins for a multiproxy palaeoenvironmental reconstruction of the Norse settlement — pollen, insect remains, plant macrofossils. Instead, at a depth of around 40 centimetres, they struck a previously unknown cultural horizon with trampled surfaces, charcoal, and wood-working debris, but no culturally diagnostic artifacts. Laboratory analysis recovered insect remains, among them early records of beetle species that had been assumed to be post-Columbian additions to the Canadian fauna.

To date their find, the authors combined existing ("legacy") radiocarbon data in a Bayesian model. Their result, in the words of the abstract: the analysis "nuance[s] previous conclusions and suggest[s] Norse activity at LAM may have endured for a century." That sentence made headlines.

Michael W. Dee and Margot Kuitems — the same researchers who would present 1021 two years later — objected in PNAS with a letter whose title contains the entire argument: Duration of activity inestimable due to imprecision of the data. Their position: the data do not support a statement about duration; the modelling manufactures a precision that is not present in the input information. Ledger and colleagues answered with Reply to Dee and Kuitems: Our model is an expression of the uncertainties inherent in the radiocarbon data — their model, they argued, is precisely an expression of those uncertainties, not a concealment of them.

This dispute cannot be settled from the outside, and that is not the point. The point is that both sides had the same data. The difference lay entirely in the question of how much structure a statistical model may legitimately extract from blurred information. And both sides agreed that the problem lay in the imprecision of the inputs — not in the statistics.

What the debate needed, then, was not a better analysis. It needed a better measurement.

Part 4: The Cosmic Anchor

Miyake Events

In 2012 Fusa Miyake, Kentaro Nagaya, Kimiaki Masuda, and Toshio Nakamura published in Nature a finding that changed dating research: in Japanese cedars they detected, for the years AD 774/775, a sharp rise in radiocarbon confined to a single year ("A signature of cosmic-ray increase in AD 774–775 from tree rings in Japan"). In 2013, Nature Communications carried their detection of a second such spike for 993/994 ("Another rapid event in the carbon-14 content of tree rings"). Such single-year spikes are today called Miyake events, after the first author of their discovery.

The physical mechanism is, with high probability, an extreme solar particle event: a burst of energetic protons strikes the atmosphere and briefly raises the production of cosmogenic nuclides. What matters for the evidential case is that 14C is not the only product. In 2015 Fadwa Mekhaldi and colleagues presented the multiradionuclide confirmation in Nature Communications — "Multiradionuclide evidence for the solar origin of the cosmic-ray events of AD 774/5 and 993/4" — that is, simultaneous detection in further nuclides, among them beryllium-10 in Antarctic ice. To this are added historical observations: for late autumn and winter 992, reports of red skies survive from Korea, from the Duchy of Saxony, and from Ulster. F. Richard Stephenson (2015) and Hisashi Hayakawa and colleagues (2017) have evaluated these testimonies systematically and classified them as indications of great magnetic storms in the 990s.

The chain of evidence thus has three mutually independent strands: trees (14C), ice (10Be), and texts (aurorae). None alone would be compelling. Together they are.

The Event of 993

The 993/994 event produced a rise in atmospheric 14C on the order of about 0.91 percent — roughly 9 per mil in the usual Δ14C notation. The 774/775 event was about 1.7 times as strong. For dating applications, however, the decisive property is not amplitude but steepness: the rise completes within a single year. That turns an otherwise slowly varying signal into an edge — and edges can be located.

Here is the conceptual leap that is easy to read past. Ordinary radiocarbon dating compares a measured value against a curve and inherits that curve's blur. A Miyake event inverts the logic: you are no longer looking for the year that fits the measurement, but for the ring in which the known edge sits. That is no longer a statistical inference; it is a count. Statistics yields intervals; counting yields whole numbers.

The side effect is that the method is bound to a new kind of effort. You need wood with an intact ring sequence, you need the outermost growth layer (the waney edge or bark), and you need enough material to measure every single ring separately. Without the bark, you do not know the distance to the felling year — and the precision collapses again.

From Marker to Calendar Year: 29 Rings

Three pieces of wood from L'Anse aux Meadows met exactly these conditions. The Kuitems team chose pieces from contexts that, in the University of Groningen's wording, are "archaeologically attributable to the Vikings," and each of the three displayed "clear evidence of cutting and slicing by blades made of metal." That is an independent argument, not a footnote: metal blades were not a product of the resident Indigenous population. The tool marks identify the wood as Norse irrespective of its age.

The three pieces came from three different trees. In each of them, the 993 signal lay 29 growth rings before the bark edge. Margot Kuitems states the result in the University of Groningen's release as follows: "Finding the signal from the solar storm 29 growth rings in from the bark allowed us to conclude that the cutting activity took place in the year 1021 AD." Michael Dee adds, on the marker itself: "The distinct uplift in radiocarbon production that occurred between 992 and 993 AD has been detected in tree-ring archives from all over the world."

The arithmetic could hardly be simpler: 992/993 plus 29 rings gives 1021. That all three pieces of wood return the same result is the real point. Three independent trees, three independent ring sequences, one felling year — that is an internal replication, and it all but eliminates the likeliest sources of error (miscounting, missing rings, misidentification of the marker).

The paper's abstract states the claim precisely and with restraint at the same time: "Here we provide evidence that the Vikings were present in Newfoundland in AD 1021. We overcome the imprecision of previous age estimates by making use of the cosmic-ray-induced upsurge in atmospheric radiocarbon concentrations in AD 993." And further: the date lays down "a marker for European cognisance of the Americas" and represents "the first known point at which humans encircled the globe."

What the Finding Shows — and What It Does Not

Now comes the part that regularly disappears from the coverage, and that I consider the most important in this whole article.

What does the paper show? That in one particular year, 1021, trees were felled at this place with metal tools. That is all. It is a great deal, but it is precisely bounded.

What does it not show?

First, it says nothing about duration. NASA's account puts this cleanly: "Whether the Vikings were in Newfoundland before or after 1021 is not certain, but the evidence says they were there for at least that year, cutting trees and building things." 1021 is a minimum statement, not an interval. The Groningen release states explicitly that "the number of Viking expeditions to the Americas, and the duration of their stay over the Atlantic, remain unknown," and adds that all current data suggest the whole endeavour was "somewhat short lived." The same release's formulation that 1021 is "the earliest — and only — known year in which Europeans were present in the Americas prior to the arrival of Columbus" is correct — but you have to read known. It is a statement about the state of our knowledge, not about history.

Second, it says nothing about the sagas. No tree ring confirms Leif Erikson. The date is compatible with saga chronology, and compatibility is not proof.

Third, it says nothing about the sequence of events. Who was there, how many, how often, how long, why they left — all open.

And fourth, most importantly: it does not settle the dispute of 2019. The new cultural horizon that Ledger and colleagues found was, on the press release's account, radiocarbon-dated to the late twelfth and thirteenth centuries, that is, after the Norse occupation, and was therefore attributed rather to Indigenous activity. The question of whether Norse activity spanned one year or one century is not answered by 1021; it is merely pinned at one point with an exact nail. A nail is not a line.

I am of the opinion that precisely this restraint is what makes the paper strong. With the same data, it could have been sold as a confirmation of the sagas. It was not.

Part 5: Three Counter-Cases

To judge the quality of evidence, it helps to look at cases where something looked like evidence and was not. This story has three of them, and they fail for three different reasons.

The Vinland Map: A Forgery

In 1957 a world map surfaced showing a land mass west of Greenland labelled Vinland. Paul Mellon bought it and gave it to Yale. In 1965 — according to Yale, "the day before Columbus Day" — it was published with considerable fanfare. It was taken to be the oldest map of America.

The resolution came through chemistry, and it came in stages. In 1972 the forensic specialist Walter McCrone examined the ink and found in the yellowish lines anatase — titanium dioxide in a rounded crystalline form manufactured industrially for pale pigments since the 1920s. That was published in 1974 (McCrone & McCrone, The Vinland Map Ink, Geographical Journal 140). Decades of dispute followed: in the 1980s Thomas Cahill and colleagues found, by PIXE, titanium only in traces (below 0.0062 percent by weight); in 2008 Kenneth Towe argued these figures were at least a thousand times too low. In 2002 Katherine Brown and Robin Clark confirmed significant anatase content by Raman spectroscopy (Analytical Chemistry 74). The parchment itself was radiocarbon-dated in 2002 to 1423–1445 — it is genuine.

On 1–2 September 2021, Yale declared the map definitively a forgery. The analysis covered, in the library's account, "the entire document's elemental composition using state-of-the-art tools and techniques that were previously unavailable"; NBC's reporting names X-ray fluorescence spectroscopy as the technique. The finding: "a titanium compound used in inks first produced in the 1920s pervades the map's lines and text." To this was added a clue that convinces without any instrument at all: a Latin note on the reverse, probably a bookbinder's instruction for the Speculum Historiale, had been overwritten in modern ink so as to look like an instruction for binding the map into that genuine fifteenth-century manuscript. Raymond Clemens, curator of early books and manuscripts at the Beinecke Library, commented: "The altered inscription certainly seems like an attempt to make people believe the map was created at the same time as the Speculum Historiale." And on the whole matter: "The Vinland Map is a fake. There is no reasonable doubt here. This new analysis should put the matter to rest."

Clemens's driest remark is the best one: "Objects like the Vinland Map soak up a lot of intellectual air space." A forgery costs not only truth; it costs attention — for decades.

The lesson: a genuine carrier material (parchment of 1423–1445) is no proof of authenticity for the content. Dating the substrate and dating the information are two different questions. That is exactly the old-wood effect in another guise.

Point Rosee: A False Alarm

In 2014 the satellite archaeologist Sarah Parcak identified, on near-infrared satellite imagery and high-resolution aerial photographs, features at Point Rosee in southwestern Newfoundland that she interpreted as a possible Norse longhouse of roughly 22 by 7 metres. Media interest was considerable.

Excavations followed in 2015 and 2016 under Parcak and Gregory Mumford (with Frederick Schwarz as a further co-director in 2015). The final report of 8 November 2017 to the Provincial Archaeology Office in St. John's was as unambiguous as archaeological reports rarely are: "no evidence whatsoever for either a Norse presence or human activity at Point Rosee prior to the historic period." The supposed turf and wall features were "not man-made"; the bog-ore accumulation and a cracked boulder were classified as natural phenomena. Two dated berries came from the seventeenth to nineteenth centuries.

This case is not a scandal but a success. A hypothesis was generated from remote-sensing data, tested at considerable cost, and discarded by its own originators. The lesson: remote sensing produces candidates, not findings. The passage from signal to fact runs through the shovel — and the shovel is allowed to say no.

The Gene from Iceland: An Open Case

In 2010 a team around Carles Lalueza-Fox and Agnar Helgason reported in the American Journal of Physical Anthropology on a new subclade of mitochondrial haplogroup C1 among Icelanders. The title already carries the caution: "A new subclade of mtDNA haplogroup C1 found in Icelanders: Evidence of pre-Columbian contact?" At least 80 living Icelanders carry the C1e lineage; genealogically it traces back to four families whose ancestors lived between 1710 and 1740 in the same region of southern Iceland.

C1 is, so far as is known, an Indigenous American maternal lineage. The authors' hypothesis is accordingly that the lineage reached Iceland around the year 1000 through at least one Indigenous American woman. The authors themselves consider this the most likely hypothesis — and write in the same breath: "The mystery surrounding the geographical origin of the Icelandic C1e lineage will remain until additional members are found in other populations — ancient or contemporary."

That is the most honest formulation of any that appears in this article. C1e has not so far been found in the sampled Indigenous populations; an Asian origin or a later introduction is not excluded, even though the genealogy rules out the obvious explanation "recently arrived Asian families" for these four families. The lesson: a hypothesis with a question mark in its title is a hypothesis. It does not become a finding by fitting a familiar story well.

Part 6: The Methodological Lesson

Set the four strands of evidence side by side and a ranking emerges that transfers to almost any argument.

Evidence What it carries What it depends on Load-bearing capacity
Ring count from the Miyake marker (1021) one exact felling year intact bark edge, complete ring sequence, global marker very high, very narrow
Butternuts at the site contact with a more southern region botanical range high
Bayesian model over legacy 14C a span, not a duration quality and independence of the input data disputed
mtDNA lineage C1e a hypothesis finding further carriers in other populations open

Four observations carry the transfer into your own work.

First: the strongest finding is the narrowest. The ring count is the most load-bearing of all the strands — and says the least of all of them. That is not coincidence but the rule. Precision and reach stand in tension. If you have a number with four significant digits, you usually have it for a very small question.

Second: a better measurement beats a better analysis. The dispute of 2019 was not decidable by statistical means, because both sides were processing the same blurred information. What resolved it was not a model but a new physical anchor. Every time a debate bogs down in methodology, the question worth asking is: which new measurement — not which new analysis — would end it?

Third: the most useful anchor comes from outside. The solar storm of 992 has nothing to do with Vikings, Newfoundland, or archaeology. That is exactly why it is usable: it is entirely independent of the question in dispute and therefore cannot be contaminated by the assumptions under test. In statistics this principle is called an instrumental variable; in software engineering, an out-of-band check; in metrology, traceability to an independent standard. It is always the same principle.

Fourth: dating the carrier is not dating the content. The Vinland Map is drawn on genuine fifteenth-century parchment. A beam of heartwood is older than its felling year. A file has a creation date that says nothing about the age of its contents. A commit timestamp says nothing about the age of the line. This confusion is so widespread that it deserves a name of its own.

The Central Takeaway

The central lesson of this case is not "the Vikings were in America in 1021." It is this: every precise figure has a domain of validity, and that domain is usually far smaller than the sentence the figure appears in.

"1021" is exact for exactly one claim — in that year, trees were felled here with metal — and for no other. In popular retelling, the exactness immediately migrates: onto the settlement, onto the duration, onto the sagas, onto "the discovery of America." That migration is not dishonesty but a gradient: a sharp number in a blurred environment colours its surroundings.

Precisely this gradient exists in any work that operates with measurements. A p99 latency figure holds for a particular load profile, a particular period, and a particular code path — and is quoted as though it held for the system. A test coverage of 94 percent holds for executed lines — and is read as a statement about correctness. A benchmark result holds for the configuration measured — and appears in the next document as a property of the technology. A compliance certification holds for an audit date and a defined scope — and is quoted as a security claim about the product.

The concrete next step: this week, take the most precise number circulating in your working environment — a latency, an availability figure, a cost forecast, a model accuracy — and write down its domain of validity in one sentence: measured under which conditions, over which period, by which procedure, for which object. Then write down in a second sentence which claims are not supported by it, even though in practice they are rested on it. If both sentences are hard to write, that is the answer to the question of how well the number is understood.

And at the next deadlocked discussion on your team, check whether you are really arguing about interpretation — or whether you are simply missing a measurement that nobody has taken, because taking it would be inconvenient. The solar storm of 992 lay for a thousand years in every tree trunk on Earth. All it needed was someone to count.

Reflection Question

Which number in your work enjoys more authority than its domain of validity warrants — and which independent measurement, untouched by the dispute itself, could either confirm or refute it?

Cross-References in the Vault

Sources

← All articles