A Volcano That Has Been Dead for 250,000 Years Is Still Breathing

Six Thousand Metres of Nothing Happening

Uturuncu stands in the Sur Lípez, in the far southwest corner of Bolivia, about fifty kilometres from the Chilean border and a long way from anything else. Six thousand and eight metres. The rock is dacite, silica-rich and stiff. Pasty stuff, the kind that piles up into steep domes rather than running downhill in sheets. Brown and cold, streaked with sulphur. A rough road climbs to within a few hundred metres of the summit, built decades ago for a sulphur mine, which makes this one of the highest places on Earth you can reach in a truck.

It has not erupted for 250,000 years, plus or minus five thousand [1]. Sit with that number for a second. When lava last came out of Uturuncu, nobody lived in the Americas. Nobody lived outside Africa either. Neanderthals were doing fine.

And yet.

Since 1992, when satellite radar first became good enough to measure ground motion from orbit, the surface around Uturuncu has been going up, and it has not stopped. Not dramatically. About a centimetre a year at the centre, which is roughly the rate at which a fingernail grows. The dome it sits on is tens of kilometres wide [3]. Underneath, small earthquakes go off at an average of around three a day, arriving in swarms of five to sixty a few times a month, and nearly all of them are shallow [8]. The summit vents gas.

Somebody at some point called it a zombie volcano. The name stuck, partly because it is a good name and partly because every popular piece written about Uturuncu since has reached for it as the obvious handle. This one included.

What the name papers over is what the moving is made of.

The cutaway above is the frame we fill in. One layer per section, in the order the evidence arrives, so that by the end you are looking at the same picture the paper's authors are looking at and know where each line in it came from. Right now it holds a mountain and a ruler.

The Shape of the Problem

The measurement that started all of this is called InSAR, short for interferometric synthetic aperture radar, and the idea behind it is simple even if the engineering is not. A satellite bounces radar off the ground and records the strength of the echo together with its phase, meaning where in its cycle the returning wave happens to be when it arrives back. Fly over the same patch a month later, compare the two phases pixel by pixel, and a shift of half a radar wavelength shows up as a fringe. Radar wavelengths are centimetres. So the technique can see the ground move by a few millimetres from six hundred kilometres up, which remains, to us, an outrageous thing to be able to do.

Pritchard and Simons pointed it at the central Andes in 2002, looking at roughly nine hundred volcanoes at once in a survey nobody had the data to attempt before. Four of them were deforming. The largest signal was a 70-kilometre-wide dome rising at about a centimetre a year, centred on a volcano that every catalogue of the day listed as dormant, which is the sort of result that either breaks a catalogue or breaks a model [3].

Ten years later, working with a longer record, Fialko and Pearse found the other half of the pattern. The dome is not alone. It sits inside a broad, shallow bowl of subsidence, a moat, reaching out past seventy kilometres from the summit and dropping at one to a few millimetres a year [4]. Seen in cross-section, uplift in the middle and a dip around the rim, the whole thing looks like a hat, and the literature has called it the sombrero ever since.

A dome you can explain with almost anything. A dome with a moat rules out most of the anythings in one stroke. That single stroke is the reason this article exists.

Why the moat matters takes some machinery to see, so the next section builds it, starting from a single buried sphere and two lines of algebra.

What a Buried Balloon Does to the Ground Above It

Suppose something underground swells. What happens at the surface?

Make the crust a block of elastic material with a flat top, the way an engineer would treat a steel beam: push it and it deforms, let go and it springs back, and the relationship between push and deformation is a straight line. Bury a small sphere at depth \(d\). Now inflate it, which is to say increase the volume of that sphere by \(\Delta V\). The rock has to go somewhere, and where it goes is up and out.

Kiyoo Mogi solved this in 1958 at the Earthquake Research Institute in Tokyo, working from ground-tilt records at Japanese volcanoes, and the answer came out as two lines of algebra [2]. At a horizontal distance \(r\) from the point directly above the sphere, the surface moves vertically by

and horizontally, straight away from the centre, by

where \(\nu\) is Poisson's ratio, the number describing how much a material bulges sideways when you squash it, taken as 0.25 for crustal rock. The model stops there. Two formulas, four inputs, no computer required. Sixty-seven years later it remains the first thing a volcano geodesist fits to a new deformation signal, because it is transparent and because it is usually good enough to get the depth roughly right.

Three things follow from those two lines of algebra, and all three do work later on.

First, and this is the one that matters most: the sign never changes. Look at the vertical formula. Everything in it except \(\Delta V\) is positive. The depth \(d\) is positive because the source is underground; the bracket \((r^2+d^2)^{3/2}\) is positive because it is a sum of squares raised to a power; \(\pi\) and \((1-\nu)\) are positive. So the sign of \(u_z\) follows the sign of \(\Delta V\) and nothing else, at every radius from zero to infinity, which is the single fact this article leans on hardest. An inflating source lifts the entire surface of the planet, forever, by a diminishing amount. A deflating one lowers all of it. Neither can lift the middle and drop the rim. A moat is not a thing one Mogi source fits badly, but a thing one Mogi source cannot do at any depth or any strength whatever.

Second, the horizontal motion tells you the depth for free. Directly above the source the ground goes straight up and not sideways, so \(u_r = 0\). Far away it goes nowhere. In between sits a maximum, and differentiating shows it falls exactly at \(r = d/\sqrt{2}\), about 0.707 times the depth, so measuring where the horizontal motion peaks reads the source depth straight off the ground surface with no inversion at all.

Figure 1 draws the first two of those relations for our own fitted shallow source.

Third, the width of the bump is also a depth gauge. The ratio \(u_z(r)/u_z(0)\) works out to \(d^3/(r^2+d^2)^{3/2}\), which falls to a half at \(r = d\sqrt{2^{2/3}-1} = 0.766\,d\), so the half-width of the dome and the depth of the source are the same measurement twice. A wide dome means a deep source, and a 70-kilometre-wide dome means one around 46 kilometres down, which is deeper than anyone hoping for a tidy shallow chamber would like.

Layer three adds the earthquakes. They are worth a word because they sit so shallow. More than 1,700 events, recorded between April 2009 and October 2012, clustered between the surface and about five kilometres depth in a dome-shaped cloud under the summit [1, 8]. They sit more than fifteen kilometres above anything that anyone would be willing to call a magma chamber. Whatever is cracking rock, it is cracking it near the top.

Bench Notes: The Afternoon We Tried to Make a Moat With One Source

We could not get the real interferograms, and would not have known what to do with them if we had. So we made our own data.

The recipe: put an inflating source at 20 km and a deflating one at 80 km, pick their strengths so the profile comes out looking like the published Uturuncu pattern, sample it every two kilometres out to 120 km, and add Gaussian noise with a standard deviation of 0.20 mm/yr. That last number is roughly what a multi-year stacked radar velocity field achieves in practice. The seed is 20250428, which is the paper's publication date, and it is written into the script, so anyone rerunning it lands on our exact digits.

The synthetic profile peaks at +9.71 mm/yr at the centre, crosses zero at 24.7 km, and bottoms at −1.97 mm/yr at 42 km, which makes the moat about a fifth as deep as the dome is high. Close enough to the real thing to be useful. Far enough from it that nobody should mistake one for the other.

Then we fitted one source to it.

The fitting is easier than it sounds. Fix the depth and the model goes linear in the volume change, because \(u_z\) is \(\Delta V\) multiplied by a fixed shape, so you can sweep the depth on a grid and solve the volume change exactly by least squares at every node. No optimiser. No starting guess. Nothing to get stuck in. From 2 to 200 km in 250-metre steps: 793 fits, every one exact.

Best answer: a source at 12.25 km, inflating at 6.62 × 10⁶ m³/yr. Root-mean-square misfit 1.29 mm/yr. Variance reduction 81.77%, which sounds respectable and is the number a careless person would quote. Variance reduction is the fraction of the squared signal the model accounts for, so 81.77% means it captures most of the energy in the profile, and the energy is nearly all in the big central bump.

The number that gives it away sits elsewhere. Nowhere on the profile does the best single source dip below +0.011 mm/yr, which is eleven microns a year of uplift standing where the data show two millimetres a year of subsidence. Positive, everywhere, by construction. Average the moat annulus from 40 to 100 km. Observations, −1.37 mm/yr. Model, +0.08 mm/yr.

Not a residual. A model being asked to do something it cannot do. Figure 2 shows it happening: the dashed curve flattens out just above zero and stays there while the data dive underneath it into the moat that no single source can reach.

Obvious objection: maybe we picked a bad depth.

We had the whole misfit curve already, so checking this took no extra work at all. The curve bottoms at 12.25 km and climbs monotonically in both directions, reaching 3.00 mm/yr by 200 km and 2.48 mm/yr at 2 km, and no depth anywhere in that sweep does better than six and a half times the noise floor. Figure 3 plots that curve, which is one of the flatter and more boring things we have ever drawn, and its flatness is the entire finding.

Same trick, two dimensions. Grid over the shallow depth and the deep depth, solve both volume rates exactly by least squares at every pair, refine the winner at 50-metre steps.

Result: a shallow source at 20.80 km inflating at +2.544 × 10⁷ m³/yr, and a deep source at 77.40 km deflating at −1.075 × 10⁸ m³/yr. RMS misfit 0.171 mm/yr, which is below the 0.20 mm/yr of noise we put in, meaning the model is now fitting a little of the noise as well as all of the signal. Variance reduction 99.68%.

It also found the hidden truth. We built the data from sources at 20.0 and 80.0 km, and the inversion returned 20.8 and 77.4, errors of 4.0% and 3.2%, with the deep volume rate back to within half a per cent.

One number surprised us. The deep source has to be 4.2 times larger in volume change. We did not arrange that. It falls out of the geometry, because a deep source produces a broad flat signal and needs a lot of volume to make a visible dent at the rim. Fialko and Pearse ran into the same factor when they tried a two-source model on the real Uturuncu data, found it needed the lower source to be about four times larger, judged that physically awkward, and went looking for a different mechanism instead [4].

We take that agreement as a sign our arithmetic is not broken, and as a warning that fitting a shape is not the same as explaining it.

Layer four is ours, and it is labelled as ours wherever it appears from here on. The shallow marker sits at 20.8 km. The deep one sits at 77 km, far below the bottom of the frame, with no obvious physical object to be, which is what a point-source model looks like when it is asked to do a job that needs a distributed one. We would rather print the problem than crop it out of the picture.

The Arithmetic

Plain division from here to the end of the section. Every number below is either an input we chose or a line the script printed.

Our fitted shallow source inflates at 2.544 × 10⁷ m³ per year. Call it 0.0254 km³/yr, or 0.806 cubic metres per second. Sparks and colleagues put the rate at about one cubic metre per second, working from the 1992 to 2006 radar record [5]. The two agree to within a quarter. We claim nothing more than that.

Uturuncu's edifice holds roughly 85 km³ of rock, built between about 890,000 and 250,000 years ago [5]. Divide 85 × 10⁹ m³ by 640,000 years: 1.33 × 10⁵ m³/yr. At that rate, and no faster, magma arrived through the whole stretch when Uturuncu was building a mountain.

Ratio of the two: 192.

Read the volume change as magma and you have to believe magma now arrives 192 times faster than it managed across the 640,000 years when it was, demonstrably, delivering a mountain.

Run the present rate backwards. Multiply 2.544 × 10⁷ m³/yr by 250,000 years: 6,360 km³. Seventy-five times the volume of the mountain. Two and a half times the Atana ignimbrite, which at around 2,500 km³ is the largest eruption the region has ever produced, an ignimbrite being the sheet of welded ash left by a pyroclastic flow. Nothing of that size is under Uturuncu. So the rate is not steady, and has not been running long.

Thirty-three years of measured uplift. Accumulated volume change, 0.84 km³.

Now the other direction. The two-source fit does not describe a system filling up. It describes a shallow source gaining volume while a deeper and much larger one loses it.

Line T7 is the one to sit with. The net volume change of the fitted system is negative: −8.2 × 10⁷ m³ a year, or −2.7 km³ over the 33 years, which is the whole argument in one line. The top swells while the whole loses. Something mobile moving upward out of a deep reservoir leaves that signature, and it is the opposite of what assembling an eruption looks like from the outside.

Last piece. How long would the ground have to keep this up to accumulate enough magma for an eruption? The answer depends entirely on what you assume the volume change is made of. Three assumptions, side by side.

The middle column allows for something the paper measured. The magma body under the region is at most about 25% melt, the rest crystals, and a crystal-rich mush stops behaving like a liquid and starts behaving like a wet sandcastle somewhere near half crystals [1]. Extracting one cubic kilometre of eruptible melt means reorganising roughly four cubic kilometres of mush. One cubic kilometre takes 1.6 centuries on that reading, and 75 centuries at the rate magma actually arrived here across 640,000 years, which is the difference between a human timescale and a geological one.

Three columns, three orders of magnitude apart. The spread is the honest width of an estimate built this way. Figure 4 plots the same rows on a logarithmic timeline, so the spread shows at a glance. We are not going to narrow it, and anyone who does should be asked how.

Layer five is the reason the middle column of that table exists. The Altiplano-Puna Magma Body was found in the 1990s as a zone where seismic waves slow down dramatically, sitting between roughly 15 and 25 kilometres depth and running for something like two hundred kilometres across the Bolivia, Chile and Argentina border country [9]. Later imaging refined its shape and confirmed the size [10]. No larger body of partially molten rock is known anywhere in Earth's continental crust, and the 2025 paper's own modelling puts its melt fraction at up to about 25% [1]. Not a lake of lava. Hot rock with liquid in the gaps.

What Seventeen Hundred Earthquakes Showed

The paper does none of the above. Theirs is a seismology paper, and it answers the question by looking rather than by dividing.

Between April 2009 and October 2012 a temporary network recorded more than 1,700 earthquakes underneath Uturuncu, and Liu and colleagues used them to run a tomography, which works on the same principle as a medical CT scan: every earthquake sends waves through the rock to every station, each path samples a different slice, and if you have enough crossing paths you can solve for the speed of sound at each point in a three-dimensional grid [1]. Out comes a map of two wave speeds: P waves, compressional, the same kind as sound in air, and S waves, shear, a sideways wiggle that no liquid carries at all.

They also inverted for azimuthal anisotropy, meaning the extent to which P waves travel faster in one horizontal direction than another, which happens when the rock is threaded with aligned cracks. Resolution: about 2.5 km for the wave speeds, 5.0 km for the anisotropy [1].

The picture has a shape. A vertical column rising from roughly 10 km below sea level up toward the summit, narrow, distinct from the rock around it, and connecting the top of the deep magma body to the shallow crust where all those earthquakes are happening. Beneath the summit, a few kilometres down, a region where the ratio of the two speeds drops. Around and below it, regions where the same ratio rises.

The column is layer six, and the single most important object in the whole paper. It starts at the top of the magma body and ends near the summit, it is narrow, and it does not look like the rock around it. Something has made a preferred path through fifteen kilometres of crust. Hudson and colleagues had already found earthquakes strung along a similar route and read them as evidence of fluid working its way upward [7], so the column is not a surprise so much as a confirmation drawn properly.

The next section explains why that ratio tells you what fills the pores.

How You Turn a Speed Into a Fluid

One piece of physics carries the whole paper, and it is elegant enough to be worth slowing down for.

The speed of a seismic wave depends on how stiff the rock is and how heavy it is. Stiffness comes in two kinds. Bulk modulus \(K\) resists being squeezed from every side at once. Shear modulus \(\mu\) resists being twisted or slid. The two wave speeds use them differently:

Now the trick. A fluid sitting in the pores of a rock has no shear strength at all. Water does not resist twisting, and neither does gas or magma on seismic timescales, so changing the pore fluid leaves \(\mu\) alone to a very good approximation. It changes only \(K\) and the density \(\rho\).

Fill the pores with brine, which is salty water and nearly incompressible, and \(K\) stays high while \(\rho\) goes up, so the numerator of \(V_s\) is unchanged while its denominator grows and \(V_s\) falls. \(V_p\) barely moves. The ratio \(V_p/V_s\) goes up.

Fill them with gas instead. Gas squashes easily, so \(K\) collapses while \(\rho\) drops a little. \(V_p\) falls hard. \(V_s\) actually rises slightly, because the rock got lighter without getting floppier in shear. The ratio goes down, and a long way down.

So a low \(V_p/V_s\) anomaly is a gas flag and a high one is a brine or melt flag, and the two point in opposite directions, which is what makes the diagnosis possible at all. Liu and colleagues pushed this further than a flag by running a petrophysical inversion: take the measured speeds, assume a rock type, and solve for the porosity and the saturation that reproduce them. Under the summit they recover a gas-bearing region with a porosity of 2.5 to 8.2%. In the fractured zones around it, brine at 2.7 to 4.2%, and where the column meets the magma body, a mixture of melt, brine and gas with gas saturation no higher than about 30% [1].

An independent check confirms where that gas sits. Muir and colleagues took Uturuncu's own dacite into the laboratory, ran it through pressure vessels at volcanic temperatures, and worked out from which mineral assemblages appear at which pressures that the magma was last stored at 100 ± 50 megapascals, meaning 1.9 to 5.7 kilometres below the surface [15]. Those are the same few kilometres in which the seismic image now finds its gas. The last magma to leave this volcano was parked exactly where the exhaust collects today, which is either a coincidence or a sign that old plumbing stays plumbing.

The last number carries the forecast, because gas saturation is what makes a silicic eruption explosive, with dissolved volatiles coming out of solution to drive the fragmentation. 30% saturation in a few per cent of pore space is not a charged system. It is a leaky one.

The Best Case Against

Every result deserves its strongest opponent, and this one has several worth taking seriously, so here they are in descending order of how much they worry us.

The first objection is the one that bites: the paper images a pathway and infers a cause, and those are two entirely different acts of measurement. The seismic data all come from a single three-year window that ended back in 2012. The deformation record runs from 1992 and has changed over that time, slowing from something near 10 mm/yr to a few mm/yr, with continuous GPS at the summit suggesting the rate is not even monotonic [6]. A snapshot of where the fluids sit is not a measurement of how fast they move, and the paper does not close the budget by showing that the fluid flux through the chimney equals the geodetic volume change. It shows that a plumbing system exists, that it is well placed to do the job, and that the alternative leaves no trace where a trace should be. A strong circumstantial case. A circumstantial case.

The second is resolution. Lateral resolution is 2.5 km for the velocity models, and the rays that provide it come from earthquakes that are themselves clustered in the top five kilometres, which means coverage is best where the earthquakes are and worst everywhere else, particularly directly beneath. A dike a few hundred metres wide would be invisible, and dikes that size are a perfectly ordinary way for magma to travel through cold crust. "We did not see fresh magma" is weaker than "there is no fresh magma." The gap between them is exactly the resolution.

The third is that petrophysical inversion is not unique. Turning a velocity ratio into a porosity and a saturation requires a rock-physics model, and every such model rests on an assumed mineralogy and on assumed shapes for the cracks and pores. Flat cracks and round pores can hold the same porosity and give very different velocities. The quoted 2.5 to 8.2% porosity range reflects some of that spread. It does not reflect all of it, because the uncertainty in the model form itself is hard to put a number on and almost nobody tries.

The fourth is that the alternatives have not been eliminated. Fialko and Pearse favoured a ballooning diapir of buoyant melt rising through the middle crust [4], and del Potro and colleagues found independent gravity evidence for a diapir-shaped low-density body [13]. Hickey, Gottsmann and del Potro showed that a viscoelastic crust, which creeps rather than springing back, needs far less volume change to make the same uplift and moves the inferred source [12]. Gottsmann and colleagues modelled the same anomaly as thermomechanical reorganisation of mush [11], and later as transcrustal flow of a compressible fluid [14], which is close to the 2025 picture but not the same thing. This field has not converged. Fitting neatly into one of several live hypotheses settles nothing.

And the fifth is about time. Low eruption risk is a statement about now. We have 33 years of deformation data and three years of seismic data, set against a volcano with a 640,000-year construction history and a 250,000-year pause since the last eruption. Absence of eruptible magma today constrains the next few decades and says very little about the next few thousand years, a limit the paper itself is careful to state. Even the university's own announcement led with the anatomy rather than with a forecast [19]. The headlines downstream were less careful.

Layer eight is the objection drawn as a picture. Those two faint bands are the country rock the inversion has to assume before it can invert anything: a quartz-rich schist above, a metapelitic gneiss below, both of them ordinary metamorphosed sediment [1]. Their assumed mineralogy sets the stiffness of the empty rock frame, and every porosity and saturation in the paper is measured against that frame rather than against anything observed directly. Change the assumed rock and the numbers move. The directions of the anomalies move much less, whichever rock you assume, and the argument rests on the directions rather than on the magnitudes, which is why it survives this objection.

Fourteen Hundred Mountains, and What We Say About Them

Now the part that reaches past Bolivia.

Roughly 1,400 volcanoes on Earth have a Holocene eruption on record, meaning an eruption in the last 11,700 years, and satellites can now measure ground motion at essentially all of them, increasingly without a human in the loop. The number of volcanoes known to be deforming went from 44 in 1997 to more than 220 by 2016, and that rise is a story about instruments rather than about volcanoes [17]. A later catalogue assembled 339 separate deformation episodes at 160 volcanoes and found that the ones which did precede eruptions clustered at 3 to 5 kilometres depth [18], which is a useful fact and also a slightly uncomfortable one, because Uturuncu's signal is nothing like that shallow.

Around 2010 the detections started outrunning the interpretations, and the question of what a deforming volcano actually means stopped being academic and started being a question observatories had to answer on a schedule.

The best answer anyone has is a 2014 study by Biggs and colleagues, who went through 18 years of satellite data covering 198 volcanoes, of which fifty-four deformed. Of those fifty-four, 25 also erupted, which is 46%. Of the ones that did not deform, 94% did not erupt [16].

Read that pair carefully, because it is routinely misread. Deformation is a decent signal, roughly quintupling the odds relative to a quiet mountain, and also a coin flip, because more than half of the volcanoes that visibly swelled over an 18-year window did not erupt during it.

Uturuncu is the extreme case that shows what the coin flip is actually made of. It has the whole checklist: uplift, a moat, daily earthquakes, summit gas, and a magma body underneath that is the largest known anywhere in the continents. And when someone finally looked properly, the answer was plumbing.

The lesson is not that deformation means nothing. Deformation measures volume change, volume change is not the same thing as mass arriving, and a model that cannot tell those apart will keep confusing a leak for a delivery.

That distinction is worth a number. Our fitted 2.5 × 10⁷ m³/yr costs 61 million tonnes a year if it is dacitic magma at 2,400 kg/m³, and 5 to 10 million tonnes a year if it is supercritical water and carbon dioxide at 200 to 400 kg/m³. A factor of six to twelve, and that is still generous to the magma, because a compressible fluid can change the volume of a pore network by expanding where it already is, with no new mass arriving at all. The Mogi model cannot distinguish those cases and never could. Forty years of reading uplift as intrusion walked straight into that blind spot, at one volcano after another, and nothing about the model warned anybody it was happening.

Layer nine crosses out the vent. Every symptom on the monitoring checklist is present at Uturuncu, and the one thing that is missing is a working path for magma to reach the surface. It has been missing for a quarter of a million years.

How to Break This

If you wanted to overturn the fluid explanation, or to confirm it properly, here is what we would go and measure, in rough order of how decisive we think each one is.

1. Weigh it. Repeat microgravity surveys at fixed benchmarks. Gravity responds to mass; deformation responds to volume. If the ground keeps rising and the gravity at the same point falls or holds steady, the volume change is being made by something light or by nothing at all, and the fluid case is made directly rather than by inference. If gravity increases as the ground rises, mass is arriving, and the fluid case is in serious trouble, which is exactly why this is the cleanest test available. It needs no new theory.
2. Measure the horizontal motion densely. For a single source \(u_z/u_r = d/r\) exactly, so a GNSS network with enough stations reads the depth straight off the surface, with no inversion anywhere in the chain. If the horizontal field peaks at a radius implying a source at 5 km rather than 20, the shallow hydrothermal system is doing more of the work than anyone currently thinks.
3. Repeat the tomography. The seismic image dates from 2009 to 2012, so deploy again and difference the two. A change in \(V_p/V_s\) inside the chimney over a decade would mean fluid is moving on a human timescale, which is the time-domain link the current argument lacks.
4. Watch the gas at the surface. If summit fumarole flux and composition track the deformation rate across a decade, the chain from deep body to ground motion is closed at both ends.
5. Listen for the other kind of earthquake. The 1,700 events are brittle-failure earthquakes in the top five kilometres. Long-period events or deep tremor underneath the magma body are the classic signature of magma on the move, and their continued absence is itself evidence, provided somebody keeps listening.

Four of those five a well-funded observatory could do tomorrow. A comfortable position for a scientific claim to be in.

Layer ten puts the instruments on the ground, and those five markers are gravity benchmarks, the only thing in this whole cross-section that measures mass instead of shape.

On the Word Zombie

The metaphor is wrong, and wrong in an instructive direction.

A zombie is a dead thing that moves because something is animating the corpse, and the implication is menace: it looks finished and is not finished, so do not turn your back on it. Applied to a volcano, the word primes exactly the reading this paper argues against.

Under Uturuncu sits a cooling body of crystal mush, enormous and very old, doing what cooling bodies do. As crystals grow out of a melt, the water and carbon dioxide that were dissolved in the liquid have nowhere to go and are forced into the shrinking pockets of remaining fluid until they come out as a separate phase. That gas is buoyant. It leaves. It finds the cracks, follows them up, dissolves minerals on the way, deposits others, warms the rock it passes, and pushes on the pore space hard enough to lift a mountain a centimetre a year and to snap rock often enough to register as three small earthquakes a day.

None of that is a corpse twitching. It is exhaust.

A better image is a fire that went out a long time ago under a very deep pile of ash, still warm, still sending smoke up through the pile, with the smoke finding new channels as old ones clog. The smoke is real. The mountain moves because of it. You would want to know about it if you lived nearby, and the scattered villages of the Sur Lípez are near enough to count, which is why the forecast matters and not only the metaphor. But the fire is not relighting, and nothing in the ground says it is.

The forecasting lesson underneath the metaphor is the part we would keep. A volcano can show every symptom on the list and still be driven entirely by fluids escaping something old, and the symptoms are real measurements worth taking. Two opposite stories fit them equally well. Telling the stories apart takes more than one instrument.

Everything in sections 4 and 5 and in Figures 1 to 4 is the club's own simplified model, run on synthetic data we made up, and it proves nothing about the real volcano on its own. Liu and colleagues did the seismology, the petrophysics and the real error budget. We did the arithmetic a school reader can check, and printed every assumption so that you can go and break it yourself, which is more use than agreement.

References