Subduction Zone Explorer, Arc Geometry and Megathrust Calculator

How to use the tools on this page

Start with the slab profiles. Pick a zone and the section draws the real geometry of that slab, from the trench down to the depth at which earthquakes stop. The three toggles put the locked megathrust, the volcanic arc and the water release depths on the same picture, which is the point: those three things are not separate facts about a subduction zone, they are consequences of the same descending slab. Click anywhere along the slab and you get the depth, the local dip and what is physically happening to the rock there.

Compare two zones before you do anything else. Load Cascadia, then load Tonga. Cascadia is a young warm slab that stops producing earthquakes below about ninety five kilometres. Tonga is old, fast and steep, and its earthquakes reach six hundred and ninety. Same process, same physics, and a factor of seven in how deep the seismicity goes. Everything else on this page is an attempt to explain that difference.

The arc geometry tab does the trigonometry that puts volcanoes where they are. Give it a slab depth beneath the arc and a dip and it returns the distance from the trench to the volcanic front, with the working shown. The input people get wrong is the dip: slabs are curved, so a single dip fitted from trench to arc is an average that matches neither the shallow value nor the deep one. The tool says so rather than pretending the answer is exact.

The megathrust tab sizes a rupture and converts it to a magnitude, and it is deliberately blunt about the fact that every published relation under predicts the largest earthquakes ever recorded. Run the 1960 Chile rupture area through it and the standard interface relation returns 8.84 for an event that was 9.5. That is not a bug in the tool, it is a real limitation of area based scaling, and the reason is explained below.

What a subduction zone is

A subduction zone is a convergent plate boundary where one tectonic plate descends beneath another into the mantle. It is the only place on Earth where lithosphere is destroyed, and it is the return half of the system whose creation half is seafloor spreading.

Three things have to be true for it to work. The descending plate must be oceanic, because continental crust is too buoyant to sink. It must be dense enough to go down, which means old enough to have cooled. And there must be somewhere for it to go, which in practice means a pre existing weakness or an inherited boundary, because starting a new subduction zone from scratch is difficult and remains one of the genuinely open problems in tectonics.

What comes out of that arrangement is the most concentrated set of geological hazards anywhere: every earthquake above magnitude nine in recorded history, most of the world’s explosive volcanoes, and nearly every destructive tsunami.

Why the plate goes down

Oceanic lithosphere is born hot and buoyant at a mid ocean ridge and it cools as it ages. Cooling makes it denser and thicker. By about twenty to thirty million years old it is denser than the asthenosphere beneath it, and from then on it is gravitationally unstable, sitting on top of material it should be sinking through.

What stops it sinking immediately is strength. The plate is a coherent elastic sheet and it needs somewhere to break or bend before it can descend. Once it has started, the descending slab pulls the rest of the plate along behind it, and that slab pull is the dominant driving force in plate tectonics, several times larger than the ridge push acting at the other end.

This is why the ocean basins are young. Nothing on the seafloor is older than about one hundred and eighty million years, because by then it has become dense enough that any available subduction zone will consume it. Continental crust, at about 2.7 grams per cubic centimetre against roughly 3.0 for oceanic crust and more for the mantle beneath it, never becomes dense enough, which is why continents survive for billions of years while ocean floor lasts a few tens of millions.

The three kinds

Ocean under continent produces the Andes: a trench close to the coast, a volcanic arc built on continental crust, and mountains raised by the shortening behind it. The magmas are contaminated by the continental crust they rise through, which is why arc volcanoes on continents erupt andesite and dacite rather than basalt.

Ocean under ocean produces an island arc: Mariana, Tonga, the Aleutians in their western part. There is no continent to contaminate the melt, the arc sits as a chain of volcanic islands, and the trenches are the deepest places on the planet.

Continent meeting continent is the end of the process rather than a kind of it. Neither plate will subduct, convergence is taken up by crustal thickening, and you get the Himalaya. Subduction of the ocean between them continues until the ocean runs out, then stops.

The anatomy, from trench to back arc

Everything in a subduction zone sits at a predictable distance from the trench, and knowing the order is most of the battle.

The outer rise is a broad bulge a hundred or so kilometres seaward of the trench, where the plate flexes upward before bending down. It is elastic, like the far side of a plank you are pushing down on. Normal faults form on it as the top of the plate is stretched, and those faults matter more than their size suggests: they are where seawater gets into the plate, hydrating it and creating the serpentinite that will carry water deep into the mantle.

The trench is the surface expression of the bend. It is deep because the plate is being pulled down faster than sediment can fill it. The Mariana Trench reaches about eleven kilometres. Where sediment supply is high, as off Oregon or in the Makran, the trench is largely filled and shallow, which is a supply effect rather than a difference in the underlying process.

The accretionary wedge is the pile of sediment scraped off the descending plate and stacked against the overriding one. Not every margin has one. Where sediment is thin and convergence fast, the margin is erosive instead, and material is removed from the overriding plate rather than added to it.

The fore arc is the region between the trench and the volcanoes. It sits above the shallow part of the slab, it is cold, and it has no volcanism because the slab beneath it is not yet deep enough to release water into hot mantle.

The volcanic arc is the line of volcanoes, and it sits where it does for a reason covered in detail below.

The back arc is behind the arc, and it can be doing any of three things. It can be quietly riding along, as in most of the Andes. It can be under extension and actively spreading, opening a new ocean basin, as in the Mariana Trough and the Lau Basin. Or it can be under compression, which is what builds mountains behind a margin.

Reading a slab profile

Plot the depths of earthquakes against their distance from the trench and they fall on a dipping plane. That plane is the Wadati Benioff zone, named for Kiyoo Wadati who found it in Japanese data in the nineteen twenties and Hugo Benioff who worked on it independently in the nineteen fifties.

It is worth pausing on what this meant at the time. Wadati identified a dipping sheet of earthquakes extending hundreds of kilometres into the Earth, decades before anyone had a theory that could explain it. It is one of the clearest examples in the subject of an observation arriving long before the framework that makes sense of it.

The seismicity is not the slab. It is the part of the slab that is cold enough and stressed enough to break, which is why a warm slab produces a short Wadati Benioff zone even though tomography shows the slab itself continuing much deeper. When the tool says earthquakes stop at ninety five kilometres under Cascadia, that is a statement about brittleness, not about where the slab ends.

The dip is not one number

Look at any profile in the tool and the slab steepens with depth. Cascadia starts at about twelve degrees and reaches forty. Tonga starts near twenty five and reaches sixty. Northern Chile runs eleven to twenty eight.

That curvature has a practical consequence that catches people out constantly. If you take the arc trench distance and the slab depth beneath the arc and solve for a single dip, you get an average that is steeper than the shallow part and shallower than the deep part, and it will not match either number in a published table. The tool shows you both when you check a real zone against the formula, precisely so the mismatch is visible rather than mysterious.

One number that predicts most of it

If you learn one quantity about subduction zones, learn the thermal parameter. It is the age of the slab in millions of years multiplied by the convergence rate in millimetres per year multiplied by the sine of the dip, and the result is in kilometres.

The logic is direct. Age sets how cold the slab was when it arrived. Rate sets how quickly it descends, and therefore how little time it spends being heated on the way down. Dip sets how much hot mantle it has to push through per kilometre of depth gained. Multiply the three and you have a proxy for how cold the slab still is at any given depth.

The range across real zones is enormous. Cascadia comes out around one hundred and sixty kilometres. Northern Tonga comes out at seventeen thousand eight hundred and forty. That is a factor of over a hundred between two examples of the same process.

And it predicts what you want to know. Cascadia’s earthquakes stop at ninety five kilometres. Tonga’s continue to six hundred and ninety. Sort the database on this page by thermal parameter and then look at the maximum slab depth column, and the correlation is obvious without any statistics.

Where it comes from and what it hides

One more thing to know before you compare numbers on this page. The slab profile tab and the zone database are working at different scales. A profile is fitted to a single transect across the margin, so its age, rate and dip are local. A database row averages a whole segment, hundreds of kilometres of arc along which the age and the rate both change, and its quoted dip range runs to the deepest part of the slab rather than stopping where the section stops. Load the Tonga profile and it returns eleven thousand for a transect at twenty degrees south, where convergence is about a hundred and sixty five millimetres per year. The Northern Tonga database row returns seventeen thousand eight hundred and forty, because up there the rate reaches two hundred and forty. Both are right. The tool shows the two side by side and says why they differ rather than quietly picking one.

Be careful with published values too, because the convention varies. Some authors use the shallow dip, some the deep dip, some an average. Some use the total convergence rate and some the trench normal component. On an oblique margin like the western Aleutians, where convergence is nearly parallel to the trench, using the total rate instead of the trench normal component inflates the answer by a factor of several. The tool states which convention it uses and applies it consistently, and if you are comparing with a paper you should check what that paper did before concluding anything from a difference.

It is also a proxy rather than a calculation. It does not model heat conduction, it does not know about shear heating on the interface, and it ignores the fact that a slab warms from both sides. For predicting the depth extent of seismicity it works remarkably well. For predicting the actual temperature at a point in the slab you need a thermal model, and those disagree with each other by a hundred degrees or more.

Why the volcanoes sit where they do

Look at a map of any subduction zone and the volcanoes form a line, parallel to the trench, at a fairly consistent distance from it. That distance varies between arcs from about a hundred kilometres to over three hundred. The line itself is remarkably sharp, often only a few tens of kilometres wide.

The explanation that survives is that the arc marks where the slab has reached the depth at which it releases water into mantle hot enough to melt. Below that depth the mantle is too cold. The water arrives, the melting point of peridotite drops sharply, and melt is produced in a narrow band.

The observed slab depth beneath arc fronts is often quoted as about one hundred and five kilometres, with a real spread. Syracuse and Abers put it at 105 plus or minus 36 with an observed range of 72 to 173. England, Engdahl and Thatcher, using a different definition based on the top of seismicity rather than a modelled slab surface, get 65 to 130. Those two ranges differ because they are measuring different surfaces, not because either is wrong, and the gap between them is worth up to forty kilometres.

That spread matters, because the tidy version of this story says the depth is constant at about a hundred kilometres and the arc position therefore just follows the dip. The real spread is large enough that the correlation is much weaker than usually taught.

A worked example you can repeat

Use the defaults in the arc geometry tab. Slab depth beneath the arc of 105 kilometres, average dip of 30 degrees, no trench offset.

The slab descends at 30 degrees, so the horizontal distance covered while gaining 105 kilometres of depth is 105 divided by the tangent of 30 degrees. The tangent of 30 is 0.5774, so the distance is 181.9 kilometres.

Measured along the slab surface rather than horizontally you use the sine instead. The sine of 30 is 0.5, so the slab surface distance is 210 kilometres.

Now vary the dip and watch what happens. At 45 degrees the arc trench distance drops to 105 kilometres, because the tangent of 45 is exactly 1. At 15 degrees it rises to 392. At 5 degrees it is 1,200 kilometres, which is far enough inland that in practice the slab never gets there in a state that produces melt, and the arc switches off entirely.

That last case is not hypothetical.

Flat slabs turn the volcanoes off

Three segments in the database are flat slab: central Peru, the Pampean segment of Chile, and central Mexico. In each the slab descends normally to about a hundred kilometres, then flattens and travels horizontally for several hundred kilometres before resuming its descent.

Load the central Peru profile in the tool and you can see it: the slab levels off between roughly three hundred and six hundred and fifty kilometres from the trench, then steepens again.

The consequence is that there is no volcanic arc above a flat slab. With the slab lying directly against the base of the overriding plate there is no wedge of hot mantle between them, so water released from the slab has nothing to melt. Peru has a gap in its volcanic chain that lines up precisely with the flat segment, and so does Chile.

The second consequence is that the flat slab scrapes along the base of the continent, transmitting stress far inland and thickening the crust well away from the margin. The Sierras Pampeanas of Argentina are basement blocks uplifted hundreds of kilometres from the trench, and the leading explanation is the flat slab beneath them. The Laramide orogeny that built the Rocky Mountains, far inland from the western margin of North America, is usually explained the same way.

The slab dehydrates, it does not melt

This is the most commonly taught error in the subject, and it is worth being precise about.

Almost every textbook diagram shows the slab melting and the melt rising to feed the volcanoes. In most arcs that is not what happens. The temperature at the top of the slab at a hundred kilometres depth is roughly 700 to 850 degrees. That is well below the melting point of dry basalt at that pressure, which is over a thousand. The slab is hot, but it is not hot enough to melt.

What it does instead is dehydrate. The oceanic crust arriving at the trench has been altered by seawater for tens of millions of years, and it carries water bound into hydrous minerals rather than as free fluid. As pressure and temperature rise, those minerals become unstable in sequence and release their water.

Clays and zeolites go first, at shallow depths. Chlorite, lawsonite and amphibole break down through the range of roughly sixty to a hundred and thirty kilometres. Antigorite, the serpentine in the hydrated slab mantle, releases a second and larger pulse deeper still, out to about two hundred kilometres.

That water rises into the mantle wedge above the slab, and this is the crucial step. Adding water to peridotite lowers its melting point dramatically, by hundreds of degrees. Mantle at 1,200 degrees that was comfortably solid becomes partially molten the moment water arrives. The wedge melts, not the slab.

The exception, and why it gets misused

Subducted sediment is a different case. Its wet solidus is lower than that of basalt, and at about a hundred kilometres the slab top is genuinely hot enough to melt sediment. That melt exists, and its distinctive isotopic and trace element signature turns up in arc lavas, which is how we know it is there.

But it contributes signature, not volume. The bulk of arc magma is melted mantle wedge. Saying the slab melts because sediment melts is like saying a stew is made of pepper.

If you take one sentence from this section: the slab dehydrates, the wedge melts.

Megathrust earthquakes

The interface between the two plates is the largest fault surface on Earth, and it produces the largest earthquakes. Every event above magnitude nine in the instrumental record has been a subduction megathrust.

The fault is locked over a depth range, not everywhere. Above roughly ten kilometres the sediments are too weak and too wet to store elastic strain, and the interface creeps. Below roughly forty to fifty kilometres, temperature or the transition into mantle rock makes it slide rather than stick. Between those limits it locks, strain accumulates for centuries, and eventually it fails.

A worked example

Use the defaults in the megathrust tab. Locked from ten to forty five kilometres depth, dipping at fifteen degrees, rupturing eight hundred kilometres along strike.

The vertical span of the locked zone is thirty five kilometres. But the fault is not vertical, it dips at fifteen degrees, so the distance measured down the fault plane is thirty five divided by the sine of fifteen degrees. The sine of 15 is 0.2588, giving 135 kilometres of downdip width.

Notice how much bigger that is than the depth range. A shallow dip turns a thin depth interval into a very wide fault, and this is exactly why subduction megathrusts can be so large. A steep fault with the same depth range would be a quarter of the width.

Multiply by the rupture length: 800 times 135.2 gives 108,184 square kilometres of fault surface. That is roughly the area of Iceland, slipping at once.

Convert that area to a magnitude with the standard interface relation and you get 8.70.

Work backward to find the slip that magnitude implies, using a shear modulus of 40 gigapascals, and you get 3.3 metres of average slip across the whole surface. That is a modest number for a magnitude 8.7, and it is modest because the fault is so large: the same moment spread over a smaller surface would need far more slip.

Why every magnitude relation under predicts the great earthquakes

Now do something the textbooks generally do not, and test the relation against the largest earthquakes we have actually recorded.

The 1960 Valdivia earthquake in Chile ruptured about 160,000 square kilometres and was magnitude 9.5, the largest ever instrumentally recorded. Put 160,000 into the standard interface relation and it returns 8.84. That is 0.66 magnitude units low, which is more than twice the stated standard deviation of the relation, and in energy terms it is a factor of about ten.

The 2011 Tohoku earthquake ruptured roughly 100,000 square kilometres and was 9.1. The relation gives 8.67, which is 0.43 low.

The 2004 Sumatra Andaman earthquake ruptured about 221,000 square kilometres and was around 9.15. The relation gives 8.96, which is 0.19 low.

Three events, all under predicted, and the error grows with size. That is not scatter, it is bias, and it has a physical cause.

Rupture width saturates. The locked zone has a downdip limit, so a megathrust can only get so wide, and in practice the maximum is around two hundred kilometres. Once a rupture has used all the width available to it, the only ways left to grow are to extend along strike and to slip harder. Slip does not appear in an area based relation at all. So above roughly magnitude 8.6 the relation compresses, and the biggest earthquakes fall off the bottom of it.

The tool shows this comparison directly rather than hiding it, because a hazard estimate that took 8.84 at face value for the Chilean margin would have been badly, consequentially wrong.

The updip limit is not the barrier it was thought to be

The classic model puts an aseismic zone above about ten kilometres where sediments are too weak to lock, and until 2011 that was treated as reasonably firm. It predicted that megathrust ruptures would stop before reaching the trench.

In March 2011 the Tohoku earthquake ruptured straight through that zone to the trench, with slip near the trench exceeding fifty metres. That shallow slip is what displaced so much water and produced a tsunami far larger than the hazard models had allowed for.

The defensible version now is that the updip zone is where a rupture is unlikely to nucleate, not a barrier that will arrest one that arrives with enough momentum. The practical consequence is direct: a tsunami model that assumes rupture stops at the updip limit will under predict wave heights, which is exactly what happened.

How a megathrust makes a tsunami

The mechanism is simple and worth stating precisely, because it is often described loosely.

During the interseismic period the overriding plate is dragged down and landward by the locked interface, so the fore arc is slowly depressed. When the fault fails, that elastic strain releases and the fore arc springs back up, lifting a slab of seafloor several metres over an area of tens of thousands of square kilometres.

The water above it is lifted with it. That is the tsunami: not a wave generated by shaking, but the entire water column displaced vertically by the seafloor moving.

Two things follow. The tsunami is larger when slip is shallow, because shallow slip produces more vertical seafloor motion than deep slip, which is the Tohoku lesson. And a strike slip earthquake of the same magnitude produces almost no tsunami, because the seafloor moves sideways rather than up.

Tsunami earthquakes, in the technical sense, are events that generate waves much larger than their magnitude suggests. They happen when rupture is very shallow and very slow, in weak sediments near the trench. The 1992 Nicaragua and 1896 Sanriku events are the standard examples. They shake weakly, so people on the coast do not always feel a strong warning, which makes them disproportionately deadly.

Why the earthquakes stop at 660 kilometres

Open the depth tab and look at the histogram. Earthquake numbers are high in the top seventy kilometres, fall away sharply through the intermediate range, reach a minimum around three hundred and fifty kilometres, rise again to a secondary peak near six hundred, then stop almost completely below about six hundred and sixty.

Every part of that shape is a puzzle, because below about fifty kilometres ordinary brittle failure should be impossible. Confining pressure is so high that rock cannot fracture and slide the way it does near the surface. Something else has to be happening.

Intermediate depths: dehydration embrittlement

The favoured explanation for events between about seventy and three hundred kilometres is dehydration embrittlement. A hydrous mineral breaks down and releases water into a rock with essentially no porosity. The water has nowhere to go, pore pressure spikes, and the effective stress holding the fault shut collapses. The rock can then fail in a brittle manner at a pressure that should have prevented it.

The evidence for this is the correlation. Intermediate depth seismicity tracks the depths at which the dehydration reactions run, and it occurs in two distinct planes in many slabs, an upper one in the crust and a lower one in the hydrated mantle, which matches where the water is.

It also explains why the earthquakes fade. The slab runs out of water. Once the hydrous minerals are gone the mechanism has no fuel, and that is the descent into the quiet zone around three hundred to four hundred kilometres.

Deep events: metastable olivine, probably

Below about four hundred and ten kilometres activity picks up again, and dehydration cannot be the answer because the water has already gone.

The leading candidate is transformational faulting. Olivine should convert to a denser structure at about four hundred and ten kilometres, but in the cold core of a fast slab the reaction is kinetically sluggish and a wedge of olivine can survive metastably to greater depth. When it finally does transform it can do so unstably, in a thin shear zone, radiating seismic energy.

This explains the correlation with the thermal parameter neatly: only cold slabs can carry metastable olivine deep enough, and only cold slabs have deep earthquakes.

It is not settled. The metastable wedge has been imaged in some slabs and looked for and not found in others. Anyone who tells you the deep earthquake problem is solved is ahead of the evidence.

The cutoff

Seismicity stops near six hundred and sixty kilometres, which is the depth of the transition from ringwoodite to bridgmanite plus periclase, the boundary between the upper and lower mantle.

Past it, no known mechanism produces a rupture that radiates seismic energy. Slabs do continue below that depth, and tomography images them doing so, some penetrating into the lower mantle and some stalling and buckling at the boundary. They just do it silently.

Cascadia, the quiet one

Cascadia deserves its own section because it is unusual in ways that matter to a lot of people who live on it.

The Juan de Fuca plate going down beneath Oregon, Washington and British Columbia is very young, about nine million years at the trench, and it converges slowly at about forty millimetres per year. That gives a thermal parameter around one hundred and sixty, one of the lowest anywhere, and the consequences follow directly. Earthquakes stop by about ninety five kilometres. There are no deep earthquakes at all. The seismogenic zone is short.

For most of the twentieth century Cascadia was thought to be aseismic, because there is essentially no instrumental record of large megathrust events there. That turned out to be the most dangerous possible misreading.

The resolution came from two directions. Along the Pacific Northwest coast there are drowned forests and abruptly buried marshes recording sudden subsidence, overlain by sand sheets from a tsunami, dated to around 1700. And in Japan there is a written record of an orphan tsunami: a wave that arrived on the evening of 26 January 1700 with no local earthquake to explain it.

Tree ring dating of the drowned forests, combined with the Japanese arrival time and wave height, pins the event to about nine in the evening Pacific time on 26 January 1700, at roughly magnitude nine. It is one of the most satisfying pieces of detective work in the subject, combining palaeoseismology with a foreign historical archive to date a prehistoric earthquake to within hours.

Cascadia is not quiet. It is between events, with an average recurrence of the order of five hundred years and considerable variation, and the last one was over three centuries ago.

What subduction does to the rock

The slab follows a pressure and temperature path unlike anywhere else on Earth: pressure rises very fast while temperature rises slowly, because the slab is descending faster than it can heat up. That produces a high pressure low temperature regime that exists in no other setting.

Blueschist forms in it, at roughly five to twelve kilobars and two hundred to four hundred degrees, and it takes its colour from glaucophane, a sodium amphibole that is only stable under those conditions. Finding blueschist in an ancient terrane is direct evidence that a subduction zone once operated there, and it is one of the cleaner tectonic indicators available.

Push further and you reach eclogite, above about twelve to fourteen kilobars, where plagioclase breaks down entirely and you are left with red garnet in a green omphacite matrix. Eclogite is dense, denser than the mantle around it, and this matters mechanically: the basaltic crust of the slab becomes eclogite as it descends, gains density, and adds to the slab pull driving the whole system. Subduction accelerates itself.

Some eclogites contain coesite, a high pressure form of silica, which requires burial beyond about ninety kilometres. A few contain microdiamonds, requiring more than a hundred and twenty. Those rocks went down to mantle depths and came back to the surface intact, and how they did so without being destroyed is still actively argued.

The budget: what goes down and what comes back

Subduction consumes about three cubic kilometres of oceanic crust a year, which balances what mid ocean ridges produce, because the Earth is not changing size.

Sediment entering trenches runs somewhere between one and a half and two cubic kilometres a year, and that number varies by a factor of two between published compilations, so treat it as an order of magnitude.

Water is the interesting one. The slab carries water down in hydrous minerals, most of it is released and returns through arc volcanoes, but some fraction survives past the arc and goes deeper. Estimates of how much reaches the deep mantle vary widely, and the size of the deep water reservoir is genuinely uncertain, with published figures spanning more than an order of magnitude.

Carbon goes down in carbonate sediment and altered crust, and comes back as volcanic carbon dioxide. Over hundreds of millions of years that loop is part of what regulates atmospheric carbon dioxide and therefore climate, on timescales far longer than anything humans affect.

Not everything returns. Some material stays in the mantle for a billion years or more, which is why the rock cycle is a budget with leaks rather than a closed loop.

Obliquity, and why some margins slide as well as converge

Convergence is rarely exactly perpendicular to the trench. Where it is oblique, the motion splits into a trench normal component that drives subduction and a trench parallel component that has to go somewhere else.

What it usually does is partition onto a strike slip fault in the overriding plate, running parallel to the arc. Sumatra is the standard example: convergence there is about forty five degrees oblique in the north, the perpendicular part is taken up on the megathrust, and the parallel part runs on the Great Sumatran Fault down the spine of the island. That fault produces its own damaging earthquakes, entirely separate from the megathrust offshore.

The western Aleutians take this to its limit. Obliquity climbs to nearly eighty degrees, so the trench normal convergence rate collapses to about sixteen millimetres per year while the total plate motion is much larger. Almost all the motion is trench parallel, and the margin behaves more like a transform boundary that happens to have a trench.

This is a practical trap in the database. If you look up a plate rate and use it as a convergence rate on an oblique margin, you will overestimate the subduction rate substantially, and any thermal parameter you compute from it will be badly inflated. The tool uses trench normal components and says so.

What is taught wrong about subduction zones

That the slab melts and the melt becomes the volcanoes. The slab dehydrates. The water lowers the melting point of the mantle wedge above it, and the wedge melts. The slab itself is several hundred degrees too cold to melt at arc depths, sediment excepted.

That heat and pressure explain intermediate and deep earthquakes. Ordinary brittle failure is impossible at those pressures. Intermediate events need dehydration embrittlement and deep ones probably need a phase transformation, and the deep mechanism is not settled.

That the volcanic arc sits above a slab depth of exactly one hundred kilometres. The observed spread is 72 to 173 kilometres depending on how you define the slab surface, which is wide enough that arc position is a much weaker predictor than usually taught.

That megathrust ruptures stop before reaching the trench. Tohoku went to the trench with over fifty metres of slip, and the resulting tsunami was much larger than the models expected.

That a bigger earthquake means a bigger fault area. Only up to a point. Width saturates near two hundred kilometres, and past about magnitude 8.6 further growth goes into slip. This is why area based magnitude relations under predict the largest events by up to two thirds of a magnitude unit.

That subduction zones are all much the same. Compare Cascadia with Tonga: a factor of over a hundred in thermal parameter, seven in seismicity depth, six in convergence rate. Almost nothing transfers between them except the mechanism.

That the trench is the plate boundary in a simple sense. The boundary is the interface, a surface thousands of kilometres across dipping into the mantle. The trench is only where that surface reaches the seafloor.

Mistakes that catch people out

Using total plate motion as the convergence rate. On an oblique margin that can be several times too large. Use the trench normal component.

Mixing thermal parameter conventions. Shallow dip, deep dip and mean dip give materially different answers, and so do total and trench normal rates. Two published values that disagree may both be right under their own definitions.

Fitting a single dip from trench to arc and then using it as the shallow dip. Slabs are curved. The fitted average is steeper than the shallow part and shallower than the deep part and matches neither.

Taking a magnitude from an area relation at face value at the top end. Above about 8.6 the relations compress. Check the result against the largest event the margin is known to have produced before believing it.

Assuming no historical record means no hazard. Cascadia was thought aseismic for decades. The record was there, in drowned forests and a Japanese tsunami account, it just was not in a seismometer.

Reading the Wadati Benioff zone as the slab. It is the brittle, seismically active part of the slab. Warm slabs go much deeper than their earthquakes do, and tomography sees them doing it.

Slow earthquakes, and what they revealed

For most of the twentieth century a fault did one of two things: it locked, or it crept. In 2001 instruments in Cascadia showed it doing a third thing.

GPS stations on Vancouver Island were moving landward as expected, dragged by the locked megathrust. Then, for a couple of weeks, they briefly reversed direction, moving seaward, before resuming. The fault below had slipped, by a few centimetres, over about two weeks. Fast enough to be a slip event, far too slow to radiate seismic waves. Nobody standing on it felt anything.

At the same time, seismometers recorded a weak, extended, low frequency rumble lasting days rather than the sharp pulse of an earthquake. That is tectonic tremor, and it turned out to be occurring at the same time and place as the slow slip. The pair is called episodic tremor and slip.

In Cascadia these episodes repeat with striking regularity, roughly every fourteen months, in a band just downdip of the locked zone. Other margins show them on their own schedules.

Two reasons this matters. It reveals the transition between locked and creeping behaviour is not a line but a broad zone with its own distinct physics, probably controlled by high fluid pressure from the dehydration reactions running just below. And each slow slip episode transfers a small amount of stress onto the locked zone updip of it, which means the probability of triggering a great earthquake rises very slightly during an episode. Very slightly: the increase is real but small, and no slow slip event has yet been shown to have triggered a megathrust rupture.

It also means a subduction zone that appears quiet on a seismometer may be doing a great deal. Cascadia was called aseismic partly because the instruments of the time could not see any of this.

How a subduction zone starts

This is a genuinely open problem, and it is worth knowing that it is open rather than assuming the textbook has an answer.

The difficulty is mechanical. Oceanic lithosphere older than about thirty million years is denser than the mantle and wants to sink, but it is also a strong plate, and bending it requires more force than plate motions can obviously supply. Once subduction is running, slab pull provides plenty of force, but that is the answer to a different question. Getting from a flat, strong, unbroken plate to a descending one requires an initiation mechanism.

Several are proposed. Nucleation at a pre existing weakness such as a fracture zone or an old passive margin. Forced initiation, where an existing plate boundary is pushed into convergence by a distant collision. Spontaneous collapse at a transform fault where old dense lithosphere sits against young buoyant lithosphere. Polarity reversal after an arc collides with a continent and the subduction flips to the other side.

The Izu Bonin Mariana system is the best studied natural example, because ocean drilling recovered the earliest volcanic rocks and dated the initiation to around fifty two million years ago. What came first was not andesite but a distinctive basalt erupted during the initial sinking, before the mature arc established itself, which tells you the system passed through a stage unlike anything operating today.

The honest summary is that we understand how subduction continues far better than how it begins, and since it must have begun many times, that is a significant gap.

When a ridge goes down the trench

Occasionally a mid ocean ridge is carried into a subduction zone and consumed. Since the ridge is the place new plate is being made, subducting it produces some unusual consequences.

A gap opens in the descending slab, because the two halves of the plate on either side of the ridge continue moving apart as they descend. That gap is a slab window, and through it hot asthenosphere reaches the base of the overriding plate directly.

The effects at the surface are distinctive. Normal arc volcanism stops, because there is no slab there to release water. In its place you get volcanism with a mid ocean ridge chemical signature erupting on a continental margin, which is otherwise nearly impossible. Heat flow rises sharply. And in some cases you get adakites, unusual magmas with a chemistry consistent with melting the young hot edge of the slab itself, one of the few settings where the slab really does melt.

Southern Chile has this happening now at the Chile Triple Junction, where the Chile Rise is being subducted. The Patagonian volcanic gap corresponds to the slab window, and the segment appears in the database with the Antarctic plate rather than Nazca on the downgoing side.

Back arc basins

Behind an arc the overriding plate can be pulled apart, and if that extension goes far enough it becomes genuine seafloor spreading, opening a new small ocean basin behind the arc.

The Mariana Trough, the Lau Basin behind Tonga and the Okinawa Trough behind the Ryukyu arc are all doing this now. The Sea of Japan did it and stopped, which is why Japan is an island arc separated from Asia rather than part of the mainland.

The driving mechanism is usually described as trench rollback. The hinge of the descending slab migrates seaward over time, particularly for old dense slabs that sink steeply, and the overriding plate is stretched to keep up. That is why back arc spreading is most vigorous behind old cold slabs. Mariana, with the oldest slab anywhere at roughly one hundred and fifty five million years, is the classic case.

Not every margin does this. The Andes are under compression behind the arc, not extension, which is why South America has mountains where Mariana has a spreading centre. The difference relates to how fast the overriding plate is advancing toward the trench relative to the rollback of the hinge, and South America is advancing westward fast enough to keep the system in compression.

Where the metals are

There is a practical reason a lot of subduction zone research is funded, and it is copper.

Porphyry copper deposits form in the upper crust above subduction zones, where a water rich arc magma stalls at a few kilometres depth, and the fluid exsolving from it fractures the surrounding rock into a dense stockwork of veins and precipitates copper and molybdenum sulfides. They are low grade and enormous, and they supply the majority of the world’s copper and most of its molybdenum and rhenium.

The Chilean and Peruvian Andes hold the largest concentration on the planet, and the reason connects back to the geometry on this page. The margin has been under compression for a long time, which encourages magmas to stall and fractionate in the crust rather than erupting, and the thick continental crust gives them a long path to interact with. Flat slab episodes appear to help too, by shutting off normal arc volcanism and then reactivating it as the slab resteepens.

Epithermal gold deposits form in the same systems, shallower and cooler, in the volcanic edifice itself. Volcanogenic massive sulfide deposits form in back arc basins where hydrothermal systems on the seafloor precipitate copper, zinc and lead.

So the arrangement of a subduction zone, which plate is going down, how fast, at what angle, and whether the back arc is in compression or extension, determines where a country’s metal endowment sits. That is a long way from an academic diagram.

The Ring of Fire

Something like ninety percent of the world’s earthquakes and roughly seventy five percent of its active volcanoes lie in a band around the Pacific. That is not a coincidence and it is not a mystery: the Pacific is ringed almost entirely by subduction zones.

Going clockwise from New Zealand, the Hikurangi margin runs north into Kermadec and Tonga, then the New Hebrides and Solomon systems, north through the Philippines and Ryukyu to Japan, then the Kurils and Kamchatka, east along the Aleutians to Alaska, and south down the Americas through Middle America to the long Andean margin, ending near Cape Horn.

The Pacific is closing. It is ringed by trenches consuming it, while the Atlantic has almost none and is opening. That asymmetry is why the Pacific plate is being destroyed faster than it is being created, and why the Pacific basin has been shrinking through the Cenozoic.

The main gap in the ring is the western coast of North America between Cascadia and Mexico, where the boundary is the San Andreas transform rather than a trench. That gap exists because a ridge was subducted there, which is the slab window mechanism described above operating in the geological past.

A second worked example: sizing a real event

Work the 2011 Tohoku earthquake from its geometry and see how close you get.

The rupture covered roughly one hundred thousand square kilometres. Average slip was of the order of thirteen metres. The shear modulus of crustal rock at those depths is about forty gigapascals.

Seismic moment is the product of those three: rigidity times area times slip. The area must be converted from square kilometres to square metres, which means multiplying by one million, and that conversion is the single most common unit error in this calculation. Forty gigapascals times one hundred thousand square kilometres times thirteen metres gives 5.2 times ten to the twenty two newton metres.

Convert to magnitude with the standard relation, magnitude equals log of the moment minus 9.1, all divided by 1.5, and you get 9.08. The accepted value is 9.1. Two hundredths of a magnitude unit, from three numbers and a formula.

Now run it the other way. Take the published moment of 5.31 times ten to the twenty two newton metres, divide by rigidity times area, and you recover an average slip of 13.3 metres.

That works because you supplied the slip. The area based relations covered earlier fail on this event precisely because they do not know the slip and have to infer it from the area, and for a rupture this large that inference breaks down. The lesson is not that the physics is uncertain, it is that the shortcut is.

One caution on the magnitude formula itself. There are two conventions in use, differing by whether moment is in newton metres or dyne centimetres, and they differ by exactly one thirtieth of a magnitude unit at every value. That is 0.033, small enough to ignore for most purposes and large enough to be confusing if you are chasing a discrepancy in the third decimal place.

How we know any of this

Nobody has been to a hundred kilometres depth. Every statement on this page about the shape of a slab, the depth of the locked zone or the rate a plate is converging comes from an inference, and the inferences are worth knowing because they have different error bars.

Earthquakes draw the slab

The oldest and still the sharpest tool is the seismicity itself. Locate enough earthquakes accurately and their hypocentres trace out the cold interior of the slab, which is the Wadati Benioff zone the profile tab plots. Modern relocation methods that compare arrival times between nearby events sharpen those clouds enormously, and in well instrumented margins such as Japan and Chile the double seismic zone resolves into two distinct planes separated by twenty to forty kilometres.

The limitation is the one already noted: seismicity shows you the brittle part, not the slab. Below the seismic cutoff the method goes blind.

Tomography images the cold rock

Seismic waves travel faster through cold rock than warm rock. Collect millions of arrival times from earthquakes all over the world, invert them for the three dimensional velocity structure of the mantle, and slabs appear as fast anomalies, typically one to two percent faster than the surrounding mantle.

That is a small percentage doing a lot of work. It is enough to trace slabs continuously from the trench to the base of the mantle in some regions. Resolution is uneven, because it depends on how many ray paths cross a given volume, so slabs beneath dense seismic networks are far better imaged than slabs beneath the southern oceans. Tomographic images should be read as smoothed, not literal: a slab imaged as fifty kilometres thick may well be thinner than that with the image blurred by the method.

Geodesy measures the locking

Continuous GPS stations on the overriding plate record millimetre level motion. During the interseismic period those stations move landward, dragged by the locked interface, and the pattern of that motion inverted for slip on the fault gives a map of which parts of the megathrust are locked and which are creeping.

This is what makes modern hazard assessment quantitative rather than descriptive. It is also what caught slow slip, since a locked patch that briefly releases shows up as a station reversing direction for a fortnight. Seafloor geodesy, which puts transponders on the ocean floor and measures their positions acoustically, extends this offshore, and it is the only way to constrain locking near the trench where the tsunami risk is decided.

Focal mechanisms give the stress

The pattern of first motions across a seismic network resolves the orientation of the fault that slipped and the sense of motion on it. Compile those through a slab and the stress state changes systematically with depth: thrust mechanisms on the shallow interface, then downdip tension in the upper part of the descending slab where it is being pulled, then downdip compression at greater depth where it meets resistance. That sign flip is direct evidence that the slab is being pulled from below at shallow depth and resisted from below at greater depth.

Palaeoseismology extends the record

Instruments have existed for about a century and great subduction earthquakes recur on a timescale of centuries, so the instrumental record is far too short on its own. Buried marsh soils, tsunami sand sheets, drowned forests, uplifted coral microatolls and turbidites triggered on the continental slope all record past events. Cascadia’s 1700 earthquake is known entirely this way, cross checked against a Japanese written record. Chile and Sumatra have similar reconstructions running back a thousand years or more.

The catch is that the geological record dates events well but sizes them poorly. Knowing an event happened is much easier than knowing whether it was magnitude 8.5 or 9.2.

Where the slabs end up

Slabs do not stop when they stop making earthquakes, and tomography shows several distinct fates.

Some stagnate at the six hundred and sixty kilometre boundary and lie flat there, extending horizontally for a thousand kilometres or more. The slab under the Japan Sea does this, and so does the one under the Izu Bonin margin. Whether the boundary resists penetration because of a density contrast, a viscosity jump or simply because the slab has slowed enough to buckle is still argued.

Others go straight through into the lower mantle. Central America and the Aegean slabs appear continuous well below the transition zone, and the Farallon slab, subducted beneath western North America over the last hundred and fifty million years, is imaged as a large fast structure sitting in the lower mantle beneath the eastern United States, thousands of kilometres from where it went down.

That is worth pausing on. The plate that built the Rocky Mountains no longer exists at the surface anywhere, and the only place to look at it is a tomographic image of the mantle under Kentucky.

The eventual destination for material that keeps going is the core mantle boundary, where two enormous slow velocity provinces sit beneath Africa and the Pacific. Old slab material appears to pile up around their margins, and plumes appear to rise from those margins, which would close a very long loop indeed: crust made at a ridge, consumed at a trench, stored for a couple of hundred million years at the base of the mantle, and returned as ocean island basalt in Hawaii or Iceland.

Recurrence, seismic gaps and what overdue means

Because great earthquakes repeat, it is tempting to treat them as scheduled. They are not, and the language around this causes real confusion.

A recurrence interval is an average, usually from a small number of dated prehistoric events. Cascadia’s is often quoted as roughly five hundred years, but the intervals behind that average range from about two hundred to over a thousand. An average drawn from perhaps a dozen events has wide uncertainty, and the underlying process is not periodic anyway. Faults do not accumulate stress and release it on a clock, because how much slip an event produces varies, what the neighbouring segments did affects it, and the strength of the interface itself changes.

The seismic gap idea, that a segment which has not ruptured recently is more likely to rupture next, is intuitive and has performed poorly when tested formally. Several of the largest recent earthquakes occurred where the gap method did not point. Tohoku in 2011 struck a segment expected to produce magnitude 7.5 to 8 events, not a magnitude 9, because segment boundaries thought to be persistent turned out not to be.

That is the deeper lesson. Segmentation is a description of what has happened, not a rule about what can. A margin that has historically produced separate magnitude 8 ruptures on adjacent patches can, on some cycle, rupture them together and produce something much larger. Sumatra in 2004 did this across more than a thousand kilometres of margin.

Overdue is therefore the wrong word. A margin with a five hundred year average that last ruptured three hundred years ago is not late, it is inside a broad distribution. The defensible statement is a probability over a stated window, which is how modern hazard assessments phrase it, and those probabilities are appreciable rather than alarming: figures of the order of ten to fifteen percent in fifty years for a great Cascadia event are typical, with wide uncertainty.

None of that is a reason to relax. A ten percent chance of a magnitude nine in the next fifty years is a very large hazard, because the consequence is so severe.

The numbers worth remembering

If you keep only a handful of figures from this page, these are the ones that do the most work.

Oceanic lithosphere becomes denser than the asthenosphere at roughly twenty to thirty million years old, and nothing on the seafloor survives past about one hundred and eighty.

Volcanic arcs sit above a slab depth of about one hundred kilometres, with the usual range being eighty to one hundred and forty, and the trench to arc distance follows from that depth and the dip.

The locked seismogenic zone runs from about ten kilometres to about forty, so its downdip width is capped near two hundred kilometres, and that cap is why area based magnitude relations fail on the largest events.

Earthquakes stop near six hundred and sixty kilometres everywhere, because that is where the upper mantle ends.

The thermal parameter, age times rate times the sine of the dip, spans about one hundred and sixty for Cascadia to nearly eighteen thousand for Tonga, a factor of over a hundred, and it predicts the depth of the seismicity better than any single one of its ingredients.

And the moment relation, magnitude equals log of the moment minus 9.1 over 1.5, with moment as rigidity times area times slip, will get you within a tenth of a magnitude unit of any well studied event, provided you convert square kilometres to square metres.

Common questions about subduction zones

What is a subduction zone in simple terms?

It is where one tectonic plate slides beneath another and descends into the mantle. It is the only place lithosphere is destroyed, which balances the new crust made at mid ocean ridges. The surface expression is a deep ocean trench, with a line of volcanoes a few hundred kilometres behind it.

What happens at a subduction zone?

An oceanic plate bends and descends beneath another plate. Sediment is either scraped off or carried down. The interface between the plates locks and periodically fails in great earthquakes. As the slab descends it releases water, which melts the mantle above it and feeds a volcanic arc. Deeper still, the slab produces earthquakes to about 660 kilometres and then continues silently into the mantle.

What type of plate boundary is a subduction zone?

A convergent boundary. Convergent boundaries between two oceanic plates or between an oceanic and a continental plate produce subduction. Where two continental plates converge, neither will subduct and you get collision and mountain building instead.

Why does one plate go under the other?

Density. Oceanic lithosphere cools and thickens as it ages, and by about twenty to thirty million years old it is denser than the mantle beneath it. Whichever plate is older, colder and denser goes down. Continental crust is about 2.7 grams per cubic centimetre against roughly 3.0 for oceanic crust, so continental crust essentially never subducts.

What is the Wadati Benioff zone?

The dipping plane of earthquakes that marks the descending slab, named for Kiyoo Wadati who identified it in Japanese data in the 1920s and Hugo Benioff who worked on it in the 1950s. It reaches 660 kilometres in cold slabs and stops by 100 kilometres in warm ones. It marks the brittle part of the slab, not the slab itself.

Why are subduction zone volcanoes so explosive?

Because their magma is wet and silica rich. The water that triggered the melting stays dissolved under pressure, and as magma rises that water comes out of solution as gas. Silica rich magma is viscous, so the gas cannot escape easily and pressure builds until the rock fails. Mid ocean ridge basalt is dry and runny by comparison, which is why it erupts quietly.

Does the subducting slab melt?

Usually not. At about 100 kilometres the top of the slab is roughly 700 to 850 degrees, well below the melting point of dry basalt at that pressure. It dehydrates instead, and the water it releases melts the mantle wedge above it. Subducted sediment does melt, because its melting point is lower, but it contributes chemical signature rather than volume.

How deep do subduction zone earthquakes go?

To about 660 to 700 kilometres in the coldest slabs, and that is the global limit. Warm slabs stop much shallower: Cascadia has essentially nothing below 95 kilometres. The controlling quantity is the thermal parameter, the product of slab age, convergence rate and the sine of the dip.

Why do earthquakes stop at 660 kilometres?

That is the boundary between the upper and lower mantle, where ringwoodite transforms to bridgmanite. Below it no known mechanism produces a rupture that radiates seismic energy. Slabs continue deeper, and seismic tomography images them doing so, but they do it aseismically.

What is the largest earthquake a subduction zone can produce?

The largest instrumentally recorded is the 1960 Valdivia earthquake in Chile at magnitude 9.5. Every earthquake above magnitude 9 in the instrumental record has been a subduction megathrust, because the megathrust is the largest continuous fault surface available. There is no firm theoretical upper limit, only the limit set by how much fault can rupture at once.

How does a subduction zone cause a tsunami?

The overriding plate is dragged down between earthquakes and springs back when the fault fails, lifting a large area of seafloor by several metres. The water column above is lifted with it, and that displacement is the tsunami. Shallow slip near the trench displaces more water than deep slip, which is why the 2011 Tohoku tsunami was much larger than expected once the rupture reached the trench.

What is the Cascadia subduction zone?

The boundary where the Juan de Fuca plate descends beneath Oregon, Washington and British Columbia. The slab is young and slow, so there are no deep earthquakes and the seismogenic zone is short. It was long thought aseismic until drowned forests and a Japanese record of an orphan tsunami pinned a magnitude nine event to the evening of 26 January 1700.

What is a flat slab?

A segment where the descending plate levels off and travels nearly horizontally for hundreds of kilometres before resuming its descent. Central Peru, the Pampean segment of Chile and central Mexico are the main examples. Because there is no wedge of hot mantle between the flat slab and the overriding plate, the volcanic arc switches off entirely above it.

What is the difference between a trench and a subduction zone?

The trench is the topographic depression where the plate boundary meets the seafloor. The subduction zone is the whole system: the descending slab, the interface, the mantle wedge, the arc and the back arc, extending hundreds of kilometres inland and downward. The trench is the visible edge of something much larger.

What is the accretionary wedge?

Sediment scraped off the descending plate and stacked against the overriding one, building a wedge of deformed material behind the trench. Not every margin has one. Where sediment is thin and convergence fast, material is removed from the overriding plate instead, and the margin is described as erosive.

Why is there a gap between the trench and the volcanoes?

Because the slab has to reach the depth at which it releases water into mantle hot enough to melt, roughly 100 kilometres, and it takes horizontal distance to get there. The gap is that horizontal distance and it depends on the dip: a steeper slab reaches arc depth sooner and puts the volcanoes closer to the trench. The region in between is the fore arc, which is cold and has no volcanism.

Related Calculators and Reference

• Pangea Map and Continental Drift Reconstruction shows where Panthalassa went. Almost none of that ocean floor survives, which is why deep time reconstructions are so much softer than modern ones.
• Seafloor Spreading Calculator and Magnetic Stripe Explorer is the other half of the system. Crust made at a ridge there is the crust consumed at a trench here, and the age it carries in is the first term in the thermal parameter.
• Rock Cycle Diagram and Identification Key places subduction in the full loop and covers the blueschist and eclogite that the descending slab produces.
• Interactive Geological Time Scale is where slab ages, arc initiation dates and the 1700 Cascadia event sit relative to the rest of Earth history.
• Mohs Hardness Scale and Mineral Identification covers glaucophane, garnet, omphacite and the hydrous minerals whose breakdown drives arc volcanism.
• Rock Density Calculator turns a mineral assemblage into a bulk density, which is the quantity that decides whether a plate sinks at all.
• Density of Basalt gives measured values for the rock that becomes eclogite on the way down and adds to slab pull.
• USGS Earthquake Hazards Program publishes the catalogues and slab models behind the geometry and magnitude values used here.