Chapter 10: The Caldera Problem

In eastern California, between the Sierra Nevada and the White Mountains, there is a plateau of pale rock called the Bishop Tuff. You can drive across it. There are roads on it, and a town, and it looks like terrain — the sort of high, dry, pinkish-grey country that you would take, if nobody told you otherwise, for ordinary geology.

It is one afternoon's worth of eruption.

Not one eruption over centuries. One event, geologically instantaneous, in which something like six hundred cubic kilometres of magma came out of the ground, most of it within a few days, and the roof over the emptying reservoir foundered and dropped, leaving the depression now called the Long Valley Caldera.

I begin here because the central difficulty of this chapter is not scientific but perceptual. The quantities involved are outside the range in which human intuition operates at all, and the standard responses to that problem are either to sensationalize — which produces the doom genre I refused in the introduction — or to reach for comparisons that sound impressive and convey nothing. I want to try to do better, and the first step is to admit how hard it is.

Register: the deposits and their volumes are empirical. The assembly timescales are theoretical and actively contested — this chapter delivers the argument Chapter 4 deferred. The risk statement at the end is a statistical claim and I will show my working.

What the number means

The conventional measure is the Volcanic Explosivity Index, which runs from zero to eight and is roughly logarithmic in erupted volume. A VEI 8 eruption is one discharging at least a thousand cubic kilometres of material.

Some anchors, using bulk erupted volume, which is the number usually quoted:

Mount St. Helens in 1980 produced on the order of one cubic kilometre. Pinatubo in 1991, whose evacuation occupied Chapter 5 and which was among the largest eruptions of the twentieth century, produced somewhere around five. Tambora in 1815, the largest eruption in recorded history and the cause of a year without a summer, produced on the order of a hundred and fifty.

Toba, roughly seventy-four thousand years ago, produced something in the region of two and a half to three thousand.

So the gap between the largest eruption any human being has written a contemporary account of and the largest eruptions in the recent geological record is a factor of nearly twenty. Tambora reordered the climate of a hemisphere for a year, and Tambora is a twentieth of the top of the scale.

Now, how to picture a thousand cubic kilometres. The comparisons in general circulation are mostly unhelpful — they either invoke units nobody has intuitions about, or they smuggle in an implied damage claim. The most honest version I have found is simply this: a thousand cubic kilometres is a cube ten kilometres on each side. Ten kilometres is a distance most people can picture, because it is a long walk or a short drive. Now stand that distance up three times, at right angles, and fill it with rock.

That is the floor of the category, not the ceiling. The largest eruption in the known record, preserved as the Fish Canyon Tuff in Colorado and dated to around twenty-eight million years ago, was several times larger again.

One point of vocabulary, since precision has been the discipline of this book. "Supervolcano" and "super-eruption" are not old technical terms. They entered wide use through popular science broadcasting around the turn of this century, and were subsequently adopted into the literature because they were useful. There is nothing wrong with that — but the word arrived carrying a tone, and the tone has done a certain amount of damage to public understanding, which the next section is about.

The arithmetic that kills "overdue"

I promised in the introduction that I would take on the claim that Yellowstone is overdue, and that I would do it with numbers rather than with attitude. Here are the numbers.

Yellowstone has produced three caldera-forming eruptions. The first, preserved as the Huckleberry Ridge Tuff, dates to about 2.059 million years ago and was the largest, at roughly two thousand four hundred and fifty cubic kilometres. The second, the Mesa Falls Tuff, dates to about 1.285 million years ago and was very much smaller — around two hundred and eighty. The third, the Lava Creek Tuff, dates to about 639,000 years ago, at roughly a thousand.

Now do the arithmetic that the doom version does, and watch it fail.

The interval between the first and second is about 774,000 years. The interval between the second and third is about 646,000 years. Average those and you get roughly 710,000 years. Time elapsed since the most recent: about 639,000 years.

We are not past the average. The claim that the system is overdue is, before any deeper objection, arithmetically wrong on its own terms.

And the deeper objections are worse. There are two intervals here. Two. You cannot establish periodicity from two intervals; you cannot establish anything from two intervals except that two intervals occurred. The three eruptions were also of very different sizes — the middle one was smaller than the others by nearly an order of magnitude — so treating them as three instances of the same repeating event is already a distortion.

But the fundamental error is conceptual rather than statistical, and it is worth understanding because it recurs everywhere in public discussion of rare hazards.

"Overdue" imports the idea of a process with memory — a system that accumulates something on a schedule and releases when the accumulation reaches a threshold, so that the longer it has been, the sooner it must be. Some natural processes do approximate this. But rare large eruptions are much better described as approximately memoryless: events occurring at some low average rate, where the probability in the next year is roughly the same regardless of how long it has been since the last one.

For a memoryless process, elapsed time carries no information whatsoever about time remaining. A caldera that last erupted 640,000 years ago and one that last erupted 60,000 years ago have, to a first approximation, the same probability of erupting next Tuesday. "Overdue" is not a cautious framing of the risk. It is a statement with no content.

None of which means the probability is zero. It is not, and I will give the honest version at the end. It means that the specific rhetorical device on which an entire genre of documentary rests is empty, and that using it costs you the ability to think about the actual number.

The paradox, delivered

Chapter 4 owed you this, so here it is.

The mush model says that magma is stored as a crystal framework with melt in its pores, that it locks up mechanically above roughly half crystals by volume, and that most magma most of the time physically cannot erupt. Eruptible material exists only as transient, high-melt-fraction lenses extracted from the mush.

And yet: Bishop Tuff, six hundred cubic kilometres, and much of it crystal-poor — a rhyolite with a modest crystal content, which is to say material that was genuinely mobile, not mush. Toba, several times that. These eruptions discharge enormous volumes of the very stuff the mush model says cannot accumulate.

So where was it, and for how long?

This is the central unresolved question in the study of large silicic systems, and it is unresolved in an unusually clean way, because we have two independent clocks that give different answers.

The first clock is zircon. Zircon is a robust accessory mineral that incorporates uranium and excludes lead when it crystallizes, which makes it a radiometric chronometer of exceptional quality. Date the zircons in a super-eruption deposit and you find that they did not all form at once. Their ages spread over a long interval — commonly hundreds of thousands of years before the eruption that carried them out.

Read straightforwardly, that says the magmatic system was there, crystallizing, for a very long time indeed.

The second clock is diffusion. As Chapter 3 described, chemical gradients inside a crystal blur over time at rates that can be measured in the laboratory. Find a crystal whose zoning records a change in conditions — a temperature increase, a new melt composition — measure how far that boundary has diffused, and you get the elapsed time since the change.

Applied to crystals from these same deposits, diffusion chronometry returns timescales that are startlingly short. Not hundreds of thousands of years. Decades to centuries, and in some analyses less.

Two rigorous techniques, applied to the same rocks, differing by three or four orders of magnitude.

The reconciliation, and why it is unsettling

The resolution now generally favoured is that both clocks are right and they are timing different things.

The zircons date the long-lived mush system — a crystal-rich, mostly uneruptible body sitting in the crust for hundreds of thousands of years, gradually growing by recharge from below, crystallizing, cooling, being reheated. That is the system, and it is ancient.

The diffusion timescales date the assembly of the eruptible lens — the extraction of mobile, crystal-poor melt out of that mush and its gathering into a body capable of moving. That is the eruption, and it is recent.

So a super-eruption is not a slow accumulation reaching a threshold. It is a very long, quiet period of storage, followed by a rapid mobilization event at the end of it. The mush sits there for a hundred thousand years being uninteresting, and then, over some considerably shorter interval, a large volume of eruptible melt is assembled out of it and comes out.

This synthesis is currently mainstream and it is not settled. There are real objections — about whether diffusion timescales are being systematically underestimated, about whether the extracted lenses could be assembled that fast by any known mechanism, and about whether zircon ages are being correctly interpreted. I am giving you the balance of opinion, not a verdict.

Now the implication, which I do not think is widely appreciated outside the field and which is the most important thing in this chapter.

If the eruptible body assembles in decades, then the window during which a super-eruption is detectable may be a matter of decades.

That cuts in two directions, and they are both uncomfortable.

The reassuring direction: a caldera with a large mush body underneath it is not, by that fact, dangerous. Yellowstone has a substantial magmatic system beneath it and has had one continuously for two million years. The presence of magma is the normal condition. It is not a warning.

The unreassuring direction: the transition from "normal condition" to "eruptible body exists" might occur within a human lifetime, and the signals accompanying it — the extraction of melt from mush, the assembly of a lens — are precisely the signals that Chapters 3 and 4 established we are worst at detecting. A modest melt lens inside a large warm mush body barely changes the geophysical signal at all.

We are, in other words, quite good at seeing the thing that is not dangerous and poor at seeing the thing that is.

Why the biggest may need the smallest push

There is a further consequence of scale that reverses an intuition, and it belongs here.

Chapter 2 listed the candidate triggers for eruption: recharge, second boiling, external forcing, roof failure. For large systems, the mechanics change in a specific way.

The buoyancy force pushing up on a magma reservoir's roof scales with the volume of buoyant melt. The strength of the roof resisting that force scales with the area it acts across. Volume grows faster than area. So as a reservoir gets larger, the ratio of driving force to resisting strength grows, and at some size the roof fails under the accumulated buoyancy alone.

Which means the very largest eruptions may require no external trigger at all — no recharge event, no earthquake, no unusual circumstance. The system simply exceeds the strength of the rock above it and goes.

And once the roof begins to founder, the collapse becomes part of the eruption mechanism rather than merely its consequence. The subsiding block acts as a piston, and its gravitational potential energy does work driving magma out. That is a large part of why these eruptions are so voluminous: the planet is squeezing the reservoir with the weight of the rock that used to be its lid.

The largest events may need the smallest triggers, which is an unhelpful property in a system you are trying to forecast.

Toba and the bottleneck that probably wasn't

I want to spend some time on a specific claim, not because it is central to the science but because it is the best available demonstration of this book's method — a striking idea that outran its evidence and has been extremely difficult to retract.

The Youngest Toba Tuff was erupted from northern Sumatra around seventy-four thousand years ago. It is the largest eruption of the last several hundred thousand years, and its ash is found across South Asia and in marine cores over an enormous area.

In the late 1990s a hypothesis was proposed connecting this eruption to a bottleneck in human population. Genetic evidence had suggested that modern humans passed through a period of very low population at some point in the late Pleistocene, and the timing was not obviously inconsistent. The suggestion was that Toba produced a severe volcanic winter, that the winter caused widespread mortality, and that the human species was reduced to a few thousand individuals — that we very nearly did not make it, and that a volcano is the reason.

It is a magnificent story. It has been in documentaries, textbooks, popular science books and museum displays for a quarter of a century.

The evidence has not been kind to it.

Archaeological work in southern India found stone tool assemblages both below and above the Toba ash layer, at the same site, showing continuity of occupation across the event — people were there before, and people were there after, doing recognizably the same things. Sediment cores from East Africa containing microscopic Toba ash have been examined for the expected cooling signature and have not produced the dramatic climatic disruption the hypothesis requires. The genetic bottleneck signals themselves have been reinterpreted: they do not date cleanly to seventy-four thousand years ago, and they are at least as well explained by founder effects during the dispersal of small populations out of Africa — a demographic process requiring no volcano whatsoever.

The climatic effect of Toba was real and was probably substantial. Its magnitude remains argued, and there is recent work pushing back toward larger effects, so I do not want to overstate the demolition. But the strong version — Toba nearly ended us — is not supported by the evidence now available, and it continues to circulate essentially undiminished.

I include this at length because it is a working example of the standard this book claims to hold. The bottleneck hypothesis is exactly the sort of claim that a book like this one wants to be true: dramatic, memorable, and flattering to its own subject. It would have made a superb chapter. It is not well supported, and saying so costs me something, which is roughly the point.

Campi Flegrei, where the risk actually is

And now a corrective to the entire chapter, because I have spent it discussing the most spectacular category of volcanic event and that is not where the danger lies.

West of Naples there is a caldera, largely submerged and largely built over, called Campi Flegrei — the Phlegraean Fields. It has produced very large eruptions in the past, including one around thirty-nine thousand years ago that deposited ignimbrite across a wide area of the Mediterranean.

It also has several hundred thousand people living inside it, and well over a million in the immediate surroundings, and part of the city of Naples sits within the caldera's boundary.

Campi Flegrei does something called bradyseism: the ground rises and falls, by metres, over years. There were significant uplift episodes around 1970 and again in the early 1980s, the latter accompanied by seismicity and by the evacuation of a substantial part of the town of Pozzuoli. There was no eruption. Uplift resumed in the mid-2000s and has continued, with seismicity increasing markedly in recent years to levels not seen since the 1980s crisis.

The interpretive problem is exactly the one Chapter 3 described. Ground going up can mean magma arriving. It can also mean hydrothermal fluid — hot water and gas in the shallow system, pressurizing and expanding. These produce similar surface deformation and require completely different responses, and distinguishing them is precisely the underdetermined inverse problem that chapter was about. The combination of deformation with repeated gravity measurement is the best available discriminant, for the reason given there: mass arriving versus volume increasing without mass.

But here is the reframing I want to leave you with, and it is why this section exists.

The operative risk at Campi Flegrei is not a super-eruption. It is that the most recent eruption there, in 1538, was small — it built a modest cone over about a week, and it is called Monte Nuovo, and you can walk up it. An eruption of that size is unremarkable by the standards of this chapter. It would also, occurring where it would occur now, be a catastrophe, because there is a European city on top of it.

The sensational risk and the operative risk are different risks. The category this chapter has been describing is the one that gets documentaries. The one that will actually kill people, at Campi Flegrei and at Merapi and at Vesuvius and at Nyiragongo, is an ordinary eruption in an extraordinary place.

That is Chapter 13's subject and I will not pre-empt it further.

The honest number

So what should you actually think about super-eruption risk?

Globally, events at the top of the scale appear to occur on the order of once in some tens of thousands of years. Published recurrence estimates vary considerably — the record is incomplete, because deposits erode and older events are systematically under-counted, and different studies make different corrections for that incompleteness. Recent work has tended to revise the rate upward, meaning more frequent than earlier estimates suggested, though with wide uncertainty bounds that should be quoted alongside any figure.

Take a recurrence of once in some tens of thousands of years. That gives an annual probability in the range of a few parts in a hundred thousand — small, but not negligible over the timescales on which institutions and civilizations exist, and comparable in expectation to other global catastrophic risks that receive substantially more attention.

The correct way to hold this is as an expected value rather than as a countdown. Low annual probability, extremely high consequence, no useful information available about timing, and — crucially — no reason to believe any particular caldera is the one. There are dozens of candidate systems. Yellowstone is famous rather than special.

What follows practically is modest and unglamorous: monitor the large calderas, because monitoring is cheap relative to the stakes; maintain the capacity to interpret unrest when it occurs; and recognize that the resources spent worrying about VEI 8 would save far more lives if directed at the VEI 4s and 5s that happen every few years underneath actual towns.

Which brings us to the events that sit between — large enough to reorganize a climate, frequent enough to have done so within recorded history, and the subject of the rest of Part IV.

The next chapter goes further back, to the occasions when the engine ran without anything constraining it at all, and to the four occasions in the record when it appears to have come close to ending the experiment.

Draft notes — verification status

Standing convention. The volume figures and eruption ages in this chapter are load-bearing for the "overdue" argument and must be exact. The Toba section makes a demolition claim and needs the most careful sourcing.

Structural note. The outline calls for Part IV to open with a field vignette on the Bishop Tuff. I have written the opening in second person ("you can drive across it") rather than first, for the reason given in the Chapter 1 notes — a first-person field claim has to come from you or be cut.

Pending verification — high priority:

Yellowstone eruption ages and volumes: Huckleberry Ridge (~2.1 Ma, ~2,500 km³), Mesa Falls (~1.3 Ma, ~280 km³), Lava Creek (~640 ka, ~1,000 km³). The entire "overdue" argument rests on these three pairs of numbers. Verify each independently and check whether recent work has revised the Huckleberry Ridge event into multiple pulses, which would change the interval arithmetic.
The memoryless/Poisson characterization of large eruption recurrence. Confirm this is the standard statistical treatment and that "elapsed time carries no information" is correctly stated for the models actually used.
Global VEI 8 recurrence estimates, including the recent upward revisions (Rougier et al. and successors) and their confidence intervals. The draft deliberately avoids a single number — keep it that way unless a well-bounded figure can be given.
Toba bottleneck evidence: Ambrose (1998) for the original hypothesis; Petraglia et al. (2007) at Jwalapuram for archaeological continuity across the ash; Lane et al. (2013) on Lake Malawi cores; the reinterpretation of the genetic bottleneck as dispersal founder effects. Also check recent work arguing for larger Toba climate effects, which the draft acknowledges — make sure the acknowledgement is proportionate.

Pending verification — standard:

Bishop Tuff volume (~600 km³) and Long Valley eruption age (~767 ka); duration of the eruption ("a few days").
Comparative volumes: St. Helens 1980 (~1 km³), Pinatubo 1991 (~5 km³), Tambora 1815 (~150 km³), Toba YTT (~2,500–2,800 km³). Confirm whether each figure is bulk or dense-rock-equivalent and state consistently — mixing the two is the commonest error in this material.
Fish Canyon Tuff (~5,000 km³, ~28 Ma) as the largest known eruption.
Taupō Oruanui (~25.5 ka, ~1,170 km³).
Origin of "supervolcano"/"super-eruption" as popular-broadcast coinage around 2000 and subsequent adoption into the literature.
Zircon age spreads in super-eruption deposits ("hundreds of thousands of years").
Diffusion chronometry timescales for final assembly ("decades to centuries") — cite the specific studies (Ti-in-quartz work on Bishop Tuff and comparable analyses) rather than generalizing. Same flag as Chapters 2 and 3; the published range is wide.
The buoyancy-overpressure scaling argument (volume versus area) and its attribution.
Caldera collapse as a piston driving eruption, and the role of gravitational potential energy in explaining erupted volumes.
Campanian Ignimbrite (~39 ka, ~300 km³) and Neapolitan Yellow Tuff (~15 ka).
Campi Flegrei population figures — red zone (~360,000–500,000) and wider area. The draft says "several hundred thousand" and "well over a million"; source and tighten.
Bradyseism episodes: 1969–72 and 1982–84 uplift magnitudes (~1.7 m and ~1.8 m), the Pozzuoli evacuation, uplift resuming mid-2000s, and recent seismicity levels. Time-sensitive — update before publication.
Monte Nuovo 1538: duration of the eruption and size of the cone.