Chapter 5: Two Warnings

The eruption began at nine minutes past nine in the evening. The first lahar reached the town at half past eleven.

In between there were two hours and twenty minutes, a functioning telephone network, a local radio station, a Red Cross that had already called for evacuation, and a hazard map — printed, distributed, and delivered to the authorities the previous month — which showed, correctly, that the town was in the path.

Around twenty-three thousand people died at Armero on the night of the thirteenth of November, 1985, and almost none of the reasons were geological.

I have been building toward this chapter for four chapters, and I want to be careful about what I am using it for. It would be easy to write it as an indictment, and there are people in it whose decisions were poor. But the useful reading is not that somebody was wicked. It is that a chain with many links failed at nearly every one of them, and that the links which held were the scientific ones. Armero is the case that separates two things this book keeps insisting are different: an epistemic problem — we cannot see inside the mountain — and a communication problem, which is everything downstream of the science. Conflating them is how the second one keeps happening.

Register: the monitoring record and the physical events are empirical. The account of institutional behaviour is drawn from subsequent reconstructions and inquiries and is interpretive; reasonable people have emphasized different links in the chain.

The year before

Nevado del Ruiz is a glaciated volcano in the Colombian Andes, and the glacier is the important part. It had not produced a major eruption since the nineteenth century, and there was a documented event in the eighteenth, and in both cases the mechanism had been the same: a comparatively modest eruption melts a portion of the ice cap, the meltwater mobilizes loose volcanic debris, and the resulting lahar runs down the river valleys that drain the mountain.

One of those valleys is the Lagunillas. At its mouth, on a fan of older mudflow deposits — that is, on the evidence of the last time this happened — sat the town of Armero, some forty-odd kilometres from the summit.

Unrest began roughly a year before, in late 1984: increased seismicity, fumarolic activity, sulfur smell. By the middle of 1985 it was obvious enough that Colombian scientists were pressing for monitoring equipment, and instruments began to arrive — slowly, in small numbers, some of them not operational until very late. On the eleventh of September 1985 the volcano produced a phreatic eruption, a steam-driven blast that generated a small lahar and demonstrated the mechanism in miniature, two months in advance, as clearly as anyone could have asked.

By this point the science was not in serious doubt. Scientists from Colombia, working with visiting delegations, understood the hazard well: an eruption of quite ordinary size could melt enough ice to send lahars down the drainages, and the drainages had towns on them.

The map

In the first half of October 1985 a hazard map was produced. It was preliminary, made under time pressure, and it was — this is the crucial and verified fact — substantially correct. It identified the valleys that would channel lahars. It showed Armero within the zone of expected inundation. One assessment made at the time put the probability of the town being hit, in the event of a significant eruption, at something close to certainty.

The map was presented to national, regional and local authorities. It reached the press. It was discussed publicly in the weeks before the eruption.

And then very little happened.

The reasons that emerge from later reconstructions are ordinary and, I think, more disturbing for being ordinary. Evacuating a town of that size is expensive, and the coffee harvest was in progress. There was no clear line of authority: the science sat with one set of institutions, civil defence with another, and local government with a third, and no one of them could act alone or was obliged to. There was scepticism about whether the scientists were exaggerating — one strand of press commentary treated the warnings as alarmism, and the phrase used in some accounts was "volcanic terrorism." And there was the reasonable-sounding objection that unrest had already been going on for a year without an eruption, which is the objection Chapter 4 predicted would be raised, because it is usually correct.

Above all there was the assumption, never quite stated and never quite examined, that there would be time — that if things became serious, someone would say so, and then people would leave.

The night

On the evening of the thirteenth of November the volcano erupted. It was not, by volcanic standards, a large eruption. It would not be remembered at all if the mountain had not been wearing ice.

Hot pyroclastic material moved across the ice cap and melted a fraction of it — a small fraction; the glacier was not destroyed. The meltwater ran into the drainages, picked up loose debris, and became what the map had said it would become: lahars, moving fast, confined by valley walls, heading for the towns at the valley mouths.

The gap between the eruption and the arrival of the mud at Armero was two hours and twenty-one minutes.

What happened in that window is the part that should be taught in every risk-communication course in the world. Ash began falling on the town in the evening and there was alarm; there were also reassurances. Accounts describe local officials and a radio broadcast urging residents to stay calm and remain in their homes — advice which is correct for ashfall, which is exactly wrong for a lahar, and which was given by people who had not internalized the distinction. The regional Red Cross had ordered an evacuation earlier in the evening. It was not carried out — the power failed during a violent thunderstorm, and the storm's rain and thunder may also have masked the sound of the mountain. When the ash appeared to ease, the alarm eased with it.

At half past eleven the first lahar reached the town. It was tens of metres wide, moving at speed, carrying boulders and trees and the remains of everything upstream. It went through Armero in a matter of minutes and buried most of it.

Armero had about twenty-eight thousand seven hundred inhabitants. Three-quarters of them died.

What they knew, and when

Now the verdict, because it matters for the whole of the rest of this book.

The science was right. The mechanism was correctly identified over a year in advance and demonstrated in miniature two months in advance. The hazard map was produced before the eruption, correctly delineated the inundation zones, and put Armero inside one. The monitoring, though thin and late and under-resourced, was adequate to establish that the volcano was in unrest and dangerous.

There was no failure of volcanology at Nevado del Ruiz. There was a failure of everything that has to happen after volcanology.

That distinction is not a defence of scientists — and I want to be careful here, because it would be self-serving for a book like this one to conclude that the only blameless people in the story are the ones who produced the data. The uncomfortable corollary is that a discipline whose product only saves lives if institutions act on it has an obligation to think about institutions. Getting the map right and handing it over is not the end of the job. It was treated as the end of the job, and more than twenty-three thousand people died some forty-odd kilometres from a mountain that had been telling them for a year.

The discipline did draw that conclusion, and it drew it quickly, which is the redeeming part of this story and the hinge into the second half of the chapter.

The photograph

I want to say something about Omayra Sánchez, and then move on, because there is a way of writing about her that I do not want to engage in.

She was thirteen. She was trapped in the ruins of her house, pinned in debris and water, with her legs held by something underneath that rescuers could not reach without equipment nobody had brought. She remained conscious and lucid for around three days while people who could not free her stayed with her, and journalists filmed her, and then she died.

The photograph taken in her final hours went around the world and became the image of Armero, and it did something no hazard map had managed: it made the disaster impossible to look away from. There is an argument, which I think is correct, that the international response to Armero — and the institutional changes that followed — owed more to that photograph than to any technical report.

That should sit uneasily, and I do not want to resolve it for you. The same observation appeared in Chapter 3, from the other direction: the Kraffts' footage worked at Pinatubo because images do something to people that numbers do not. Here is the same mechanism operating after the fact, at the cost of a child's death being watched by the world.

If you want a single sentence for why the discipline is so preoccupied with communication, it is that the most effective piece of volcanic risk communication in the twentieth century was a photograph of a dying girl, and it arrived three days too late to be of any use to her.

What Armero built

Within about a year of Armero, the United States Geological Survey established a standing rapid-response capability — a small team, with portable instruments, whose purpose was to deploy to volcanoes in unrest anywhere in the world, at the request of the host country, and work alongside local scientists.

It exists because of Armero. That is the explicit rationale: the recognition that the equipment, the experience and the institutional weight needed to handle a volcanic crisis were unevenly distributed around the world, that they were often not where the volcano was, and that the gap could be closed by putting a team on an aeroplane.

Five years later, that team went to the Philippines.

Pinatubo

Mount Pinatubo had not erupted in around five centuries. It was forested. It was not being monitored. Locally it was regarded as a mountain rather than as a hazard, and a substantial population lived around it, along with a large American air base.

In the first half of 1991 it began to show unrest — steam explosions, seismicity, sulfur. From that point the Philippine Institute of Volcanology and Seismology, joined by the USGS team, had roughly two months to do everything: install a monitoring network from nothing, work out what a volcano nobody had studied was likely to do, design an alert system, and persuade several hundred thousand people to leave.

What they built, under that pressure, is the template.

They installed seismometers and worked out where the earthquakes were coming from. They measured sulfur dioxide emission and watched it rise and then drop — the drop being read, correctly and against the intuitive interpretation, as a sign the system was sealing itself and pressurizing rather than calming down. They mapped the deposits of previous eruptions and discovered that this quiet mountain had a history of very large ones.

And they designed a five-level alert scheme, wrote down in advance what each level meant and what action it required, and published it before it was needed. That last point is not a detail. An alert scheme agreed in advance converts a judgement call under pressure into the execution of a pre-existing plan, and it means that raising the level is not a fresh act of individual courage every time.

The evacuation was staged, with the radius extended progressively as the alert level rose — ten kilometres, then twenty, then thirty. The air base was evacuated days before the climax. Communities were reached by an information campaign that included the Krafft film.

On the fifteenth of June the volcano produced one of the largest eruptions of the twentieth century. Pyroclastic flows filled the valleys. And a typhoon arrived at the same time, saturating the ash and bringing it down on roofs.

Roughly four hundred people died, most of them from structural collapse under wet ash rather than from the eruption's direct hazards. The USGS and PHIVOLCS estimate that the forecasts and evacuations saved somewhere between five thousand and twenty thousand lives.

What actually made the difference

The instruments at Pinatubo in 1991 were not fundamentally better than those available at Ruiz in 1985. Seismometers, tiltmeters, gas spectrometers, and a great deal of fieldwork. There was no technological revolution in the intervening six years.

The differences were institutional, and I want to list them plainly, because this is the operational core of the chapter.

A pre-agreed alert scheme. Defined in advance, published, with each level tied to specified actions. No one had to invent a threshold in the middle of a crisis.

A single credible chain of authority. It was clear who was issuing the warning and who was responsible for acting on it. At Ruiz, responsibility was distributed among institutions in a way that let every one of them reasonably believe someone else would act.

Communication designed for the audience. Not reports. Briefings, repeated, with film, to the specific people who would have to make and obey the decision — including military commanders, whose calculus was different from the civilian one and who were addressed accordingly.

A willingness to be wrong, loudly and early. The team accepted a high rate of false alarm days deliberately. They raised alert levels on evidence that might have come to nothing, and they said so. This is the expensive part, and I will come to what it costs.

Local scientists in the lead. PHIVOLCS ran it. The visiting team assisted. That matters practically — local scientists know the terrain, the language, the institutions and the communities — and it matters for credibility, which is the currency the last mile runs on.

None of that is technology. All of it is design.

The costs nobody puts in the ledger

I am wary of how well the Pinatubo story reads, so let me put the other things in.

The evacuation displaced tens of thousands of people, and among them were the Aeta, an indigenous population who had lived on the mountain and whose communities did not survive the displacement intact. They lost land, subsistence, and a good deal of a way of life, in evacuation centres and resettlement schemes that were not designed with them in mind. The forecasting saved their lives and the response damaged their existence, and both of those statements are true simultaneously. A version of this chapter that celebrates Pinatubo without that paragraph is doing propaganda.

And the eruption did not stop killing when it stopped erupting. Enormous volumes of loose ash sat on the slopes, and for years afterward the rainy season remobilized it into lahars that destroyed towns which had survived the eruption itself. The death toll accumulated slowly, in the aftermath, in places that had already been evacuated once and had been allowed to return.

The other side of the ledger

Now the harder half, which the Pinatubo story tends to obscure: most volcanic unrest does not produce an eruption. That is not a caveat. It is the base rate. An observatory that treats every episode of unrest as an imminent eruption will be wrong most of the time.

In 1976, on Guadeloupe, La Soufrière entered a period of vigorous unrest. Roughly seventy thousand people were evacuated, for months. The scientific community split publicly and bitterly over whether the activity was phreatic — steam-driven, alarming, and not a precursor to magmatic eruption — or the beginning of something worse. The dispute was conducted in the press, with reputations attached.

There was no magmatic eruption. The evacuation lasted months, wrecked the local economy for a season, and the faction that had argued the activity was merely phreatic was vindicated. The episode did real damage to the discipline's standing, and it is still cited, nearly fifty years on, whenever someone wants to argue that volcanologists over-warn.

At Campi Flegrei, in the early 1980s, ground uplift and seismicity led to the evacuation of a large part of Pozzuoli. There was no eruption. There has still been no eruption. There is a population living there now, inside a caldera that continues to inflate and subside, who have been through this before and drawn their own conclusions about what it means.

And in 2019 at Whakaari, in New Zealand, an eruption occurred at a volcano that was on a raised but not restrictive alert level, killing twenty-two people who were on the island as tourists. The aftermath was legal rather than scientific — prosecutions, questions of liability, and a chilling of the willingness of anyone to be the person who said a place was safe enough to visit.

The pattern across these is not that scientists are unreliable. It is that the same honest assessment produces a costly non-event most of the time and a saved city occasionally, and the institutional and political consequences of those two outcomes are wildly asymmetric. Nobody is promoted for an evacuation that turned out to be unnecessary. The mayor who empties a town for nothing has ended a career; the mayor who does not empty a town and is unlucky has ended more than that.

I should mention what happened at L'Aquila, because it is the extreme case even though it is seismological rather than volcanic. After the 2009 earthquake, Italian scientists and a public official were prosecuted over reassurances given before the event; the scientists' convictions were substantially overturned on appeal, but the process took years. Whatever one concludes about the specific facts, the effect on scientific communication across Europe was immediate and predictable: it made reassurance legally dangerous, which does not make anybody safer. It makes advice vaguer.

Probability is the honest product

All of which points at a reframing that the discipline has largely adopted and that the public has largely not.

The question "will it erupt?" has no answer. The question that can be answered, and answered increasingly well, is: what is the probability, over what time window, of an event of what size, affecting what area?

The formal machinery for this is the event tree. You lay out the sequence of things that would have to happen — unrest continues; magma is involved rather than just hydrothermal fluid; magma reaches the surface; the eruption is of such-and-such a size; the hazard follows this drainage — and attach a probability to each branch, drawing on the monitoring data, the geological record of that volcano's past behaviour, and the structured judgement of experts. Multiply along the branches and you get an estimate for each outcome, with an uncertainty attached.

This is genuinely better than what came before, and it has one great virtue: it forces disagreement to become explicit. Two scientists who feel differently about a volcano can be made to say where on the tree they differ and by how much, which is enormously more productive than two people saying "I'm worried" and "I'm not."

But it hands an unsolved problem to somebody else. An event tree returns a distribution. A mayor requires a decision. The translation from one to the other is not a scientific operation — it involves valuing lives against livelihoods, weighing a certain economic loss against a probabilistic catastrophe, and deciding who bears which. Those are political and ethical judgements, and scientists have no special standing to make them.

The honest division of labour is that the scientist owns the distribution and the official owns the decision. It works only if the official understands what a distribution is, and if the scientist does not quietly smuggle a recommendation into the framing — and, most importantly, if both of them agreed on the arrangement before the crisis, which is precisely what the Pinatubo alert scheme accomplished and what did not exist at Ruiz.

What a warning is for

Let me finish Part II where it has been heading since Chapter 3.

We cannot see inside the mountain. Our images are underdetermined. Our timing is, for the structural reasons set out in Chapter 2, probably not knowable with precision. All of that is true and none of it is what killed Armero.

Armero died of a two-hour-and-twenty-minute gap that nobody used, because no one had agreed in advance who would say the word, and because the advice that reached people was correct for the wrong hazard, and because a year of unrest without an eruption had taught everyone that unrest without an eruption was what happened. The science had done its part. The map was right. The map was in the room.

Pinatubo worked because a set of unglamorous arrangements existed beforehand: a scale, a chain of command, a communications plan, a film, and a group of people who had decided in advance that they would rather be embarrassed than sorry.

Which yields the conclusion this book keeps returning to, and which I will state as flatly as I can:

A volcanic forecast is not a scientific product. It is a social one. It is manufactured out of science, but it fails or succeeds on the last mile — on whether the person who has to act trusts the person who is telling them, and on whether they agreed beforehand what would happen.

That is why an appendix at the back of this book, in among the seismic and gas and deformation protocols, contains a protocol for the communication plan, and why I have put it last and argued it should be built first.

It is also why the question of who is standing at the end of that last mile — who is believed, who is warned, who is in a position to leave — is not a footnote to volcanic risk but very close to the whole of it.

That is Chapter 13, and we will get there.

But first, having spent five chapters on how the engine works and how badly we can see it, I want to make the case that this engine built the world. Part III begins at the bottom of the ocean.

Draft notes — verification status

Standing convention. Several load-bearing facts here were resolved during the thesis-paper verification pass. The institutional narrative is drawn from post-event reconstructions and needs the most careful checking, because it assigns responsibility.

Already verified — do not re-check:

Nevado del Ruiz hazard map completed in preliminary form in early October 1985 and presented to national, regional and local authorities; correctly showed lahar inundation of Armero.
Armero death toll ~23,000–25,000 (contested range; the draft uses "around twenty-three thousand" and "three-quarters of the town's population" — reconcile these two figures against each other and against the Introduction and Chapter 2 before print).
Pinatubo: USGS/PHIVOLCS estimate of 5,000–20,000 lives saved; approximately 400 deaths; the Krafft film used in briefings.

Pending verification:

RESOLVED. Eruption at 21:09 on 13 November 1985; first lahar reached Armero at 23:30 — an interval of two hours and twenty-one minutes. Draft corrected from "two and a half hours."
Distance from summit to Armero (draft: "forty-odd" / "forty-eight kilometres" — pick one and source it).
Onset of unrest (late 1984) and the 11 September 1985 phreatic eruption and its associated lahar.
The date the hazard map was distributed, and the reported probability assessment for Armero's inundation.
The "volcanic terrorism" characterization in the Colombian press — verify the phrase and its source before quoting it even indirectly.
PARTLY RESOLVED, AND CORRECTED. Local officials including the town priest did instruct residents to stay calm and go inside, and local radio carried similar messages. But the Red Cross evacuation order was not countermanded — it failed to be executed because electrical power was lost during a thunderstorm, whose rain and thunder may also have masked the sound of the volcano. The draft has been corrected; the earlier version implied a deliberate reversal that the record does not support. Still worth checking against primary reconstructions rather than encyclopedic summary.
RESOLVED. Armero had ~28,700 inhabitants; roughly three-quarters were killed. Total across all valleys >23,000, including ~1,800 at Chinchiná.
Omayra Sánchez: age (13), duration trapped (~60 hours / "around three days"), circumstances, and the photographer and provenance of the photograph.
USGS Volcano Disaster Assistance Program: founding year (1986) and whether the causal attribution to Armero is explicit in its own founding rationale, not merely commonly asserted.
Pinatubo repose interval (~500 years); 2 April 1991 phreatic onset; the five-level alert scheme; evacuation radii (10/20/30 km); Clark Air Base evacuation date and personnel numbers.
The declining SO₂ flux interpreted as sealing/pressurization — confirm this was actually part of the 1991 reasoning and is not retrospective tidying.
Typhoon Yunya's coincidence and the attribution of most deaths to ash-loaded roof collapse.
Aeta displacement: numbers, resettlement outcomes, and long-term consequences. Handle with care and cite properly.
Post-eruption lahar fatalities in subsequent years.
La Soufrière, Guadeloupe 1976: evacuation numbers (~70,000), duration, the Tazieff/Allègre dispute, and the outcome. The draft avoids naming the disputants — decide whether to name them.
Campi Flegrei 1982–84 bradyseism and the Pozzuoli evacuation numbers.
Whakaari/White Island 2019: alert level at the time, fatalities (22), and the legal aftermath.
L'Aquila 2009: the charges, the convictions, and the appeal outcome for the scientists versus the official. Get this right — it is frequently misreported in both directions.

The case that the engine made the living world