A Pulsing Mantle Plume, Timed by the Plate Above It

Where Three Plates Try to Tear the Same Corner Apart

Look at a physical map of the Horn of Africa and one feature refuses to be ignored: a low, sun-baked triangle of desert called the Afar Depression, sitting more than a hundred metres below sea level in places, where three tectonic plates — African (sometimes split further into Nubian and Somalian), Arabian, and Somali — are simultaneously trying to move apart from each other. Three rifts radiate from that one hinge. The Red Sea Rift runs northwest, already advanced enough that seawater floods parts of it. The Gulf of Aden Rift runs east-southeast, roughly 35 million years old and the most mature of the three [1]. The Main Ethiopian Rift runs south into the highlands, younger and still landlocked, still tearing through continental crust rather than the thin oceanic-style crust the other two arms have already built.

Geologists have had a name for this kind of junction since plate tectonics was formalized: a triple junction, one of the small number of places on Earth where you can watch continental breakup and the birth of an ocean basin happening in the same afternoon's fieldwork, at three different stages, within driving distance of each other. It is, among other things, an unusually good natural laboratory — if you want to know what drives a rift to split a continent, Afar hands you the same process at three ages simultaneously.

For decades the working assumption was that all three arms share a single underlying engine: a mantle plume, parked beneath the junction, feeding magma to whichever arm needed it. Watts and colleagues' 2025 paper in Nature Geoscience tested that assumption directly, using the chemistry of the rock the plume actually produced, and found the engine was doing something odd [1].

What a Mantle Plume Is Actually Supposed to Be

"Plume" is a fluid-dynamics word before it is a geology word, and the geology borrowed it because the physics really does look like a plume. Take a pot of water on a stove, heat one small patch of the bottom, and a narrow column of hot, buoyant fluid rises off that patch through the cooler water above it, sometimes topped with a mushroom-shaped head. W. Jason Morgan proposed in 1971 that the same thing happens inside the solid — but slowly, plastically flowing — mantle: narrow columns of anomalously hot rock rise from deep inside the Earth toward the surface, staying more or less fixed in place while the tectonic plates drift over them, which is how you get a chain of extinct volcanoes trailing behind a still-active one, like Hawaii's island chain trailing behind the current hotspot [2].

The rock in a plume does not have to melt on its own initiative. It rises, the pressure on it drops, and because a rock's melting point falls as pressure falls, a hot solid that started well below its melting point can cross that threshold purely by moving upward — the same physics, run in reverse, that keeps a pressure cooker's contents from boiling until you release the lid. Watts and colleagues describe the upwellings feeding places like Afar as anomalously hot mantle originating from depths of roughly 1,000 to 2,800 kilometres, deep enough that whatever is driving them is a genuinely lower-mantle process, not something confined to the shallow rock immediately under the rift [1].

East Africa in particular has a well-established reason to expect one plume rather than several. Ebinger and Sleep argued in 1998 that the entire chain of Cenozoic volcanic fields running from Afar down through Kenya and Tanzania could be explained by a single plume head spreading out beneath the lithosphere and finding pre-existing weaknesses to erupt through, rather than by several separate small plumes dotted along the rift system [3]. If one plume feeds an entire continent's worth of rifting, the simplest version of that idea predicts something specific about what should come out of the ground directly above it.

The Prediction a Single Plume Makes: a Bullseye

Here is the textbook picture the paper is testing against. A plume conduit is treated as one thermally and chemically zoned column, hot core wrapped in progressively cooler, more entrained mantle. Melt that forms nearer the conduit's axis should look chemically different from melt forming farther out, and — critically — that difference should vary smoothly and roughly the same way in every direction, because distance from the plume axis is the variable doing the work, not which compass bearing you walked in to get there. Watts and colleagues state the null model plainly: a single upwelling centre in which "the variation of [a geochemical property] with distance is assumed common to all rifts" [1]. Walk away from the centre along the Red Sea arm, and a given percentage change in isotope ratios should show up at roughly the same distance as it would walking away along the Main Ethiopian Rift arm. One plume, one radial recipe, applied three times.

That prediction is falsifiable in a very direct way: go collect rocks along all three arms, measure their chemistry, and see whether the pattern really is radially symmetric or whether it does something else entirely as a function of distance from the presumed centre.

Reading Chemistry Out of Rock Instead of Watching It Erupt

Nobody can watch magma leave the plume, travel through 1,000+ kilometres of mantle, and arrive at the surface in real time. What they can do is analyse the rock it becomes once it cools, because different mantle sources leave different chemical fingerprints on the melt they produce, and those fingerprints survive eruption largely intact.

The tool of choice is radiogenic isotope geochemistry. Certain elements — strontium, neodymium, lead — have isotopes produced by the slow radioactive decay of other elements over billions of years, so the ratio of a radiogenic isotope to a stable one in a piece of mantle depends on that piece's own long history: how much rubidium, samarium, or uranium it accumulated relative to the parent element, and for how long. Mantle that once included subducted oceanic crust looks isotopically different from mantle that has stayed relatively pristine, and geochemists have named several recurring "flavours" found worldwide — DMM (depleted MORB mantle, the default background), HIMU (high time-integrated uranium-to-lead), and two enriched varieties called EMI and EMII — a classification scheme built up over decades and formalized by Zindler and Hart in 1986 [4].

For Afar, Watts and colleagues measured five isotope ratios — ⁸⁷Sr/⁸⁶Sr, ¹⁴³Nd/¹⁴⁴Nd, and three lead ratios, ²⁰⁶Pb/²⁰⁴Pb, ²⁰⁷Pb/²⁰⁴Pb and ²⁰⁸Pb/²⁰⁴Pb — plus three trace-element ratios, La/Sm, Ce/Pb and the niobium anomaly ΔNb, across more than 130 young volcanic rocks: 79 new samples from across the Afar region and 52 more from dedicated Main Ethiopian Rift fieldwork, roughly doubling the number of high-quality analyses previously available from the area [1]. They then folded in 93 existing analyses from the GEOROC geochemical database to widen the geographic coverage further.

Turning Thirteen Numbers Per Rock Into One Cluster ID

Eight geochemical ratios per rock is already more dimensions than a human eye can compare across 130-plus samples by inspection, so the authors added five geophysical variables — crustal thickness and extension rate among them — for 13 standardized variables total, and ran k-means cluster analysis: a standard statistical method that groups data points by minimizing the distance between each point and the centre of the group it is assigned to, iterating until the groupings stop changing [1]. The output is not a map of individual chemical values but a small number of discrete cluster labels — this rock is "cluster 1," that one is "cluster 3" — each cluster defined by its own characteristic combination of the 13 inputs.

Think of it as building a barcode reader for mantle chemistry. Where a single isotope ratio on its own is a smudged, noisy number, a cluster label built from thirteen variables at once is a much sharper fingerprint, and — this is the part that matters for what comes next — a fingerprint you can look for again somewhere else on the map.

What Should Have Been a Bullseye

It wasn't one. Instead of a smooth, once-only gradient radiating outward from a single centre, the cluster analysis found the same small set of cluster identities appearing, disappearing, and reappearing as you move along a rift arm — a repeating sequence rather than a monotonic fade. The Main Ethiopian Rift resolved into four clusters recurring at a spacing of roughly 50–100 km; the Red Sea Rift resolved into three clusters recurring at roughly 150–200 km [1]. Neither number is what a single radially symmetric source predicts, and the two rift arms do not even agree with each other on the spacing, which a shared plume centre with one universal gradient also does not predict.

The authors' own language for the result is direct: Afar is fed not by a smooth column but by "a spatially and chemically heterogeneous upwelling" — and, more specifically, by "a heterogeneous pulsing mantle upwelling that is not symmetrical but is instead shaped by both variable lithospheric thinning and extension rates within each rift" [1]. Pulsing, not flowing steadily; heterogeneous, not radially uniform; shaped by the rift, not just feeding it.

The Sequence That Shows Up Twice

The detail that makes "heterogeneous" mean something more than "noisy" is this: the same short sequence of cluster identities recurs across both rift arms, just stretched to different physical lengths. The paper reports that the three clusters found across the whole Red Sea Rift match the initial roughly 200 km worth of clustering sequence observed across the Main Ethiopian Rift [1] — in other words, cluster 1, cluster 2, cluster 3 is not three independent local phenomena; it is one recurring signature, read out at two different rift arms moving at two different speeds. That is the single strongest piece of evidence in the paper that there is genuinely one upwelling underneath the whole junction, not three unrelated ones — the abstract's phrase for it is a "single, asymmetric upwelling" [1].

Doing the Club's Own Arithmetic: From Kilometres to Millions of Years

Here is a calculation that is entirely ours, not the paper's, built only from numbers the paper reports. The rock recording each cluster erupted at some point in the past and has been carried away from the rift axis ever since by the same plate motion that is stretching the rift — so a spatial spacing measured in kilometres today can, in principle, be converted back into a time interval, if you know how fast the ground has been moving. The unit arithmetic collapses nicely: a wavelength in kilometres equals an extension rate in millimetres per year multiplied by a period in millions of years, because a million years of millimetre-per-year motion is exactly a kilometre. So:

where $\lambda$ is the cluster spacing in kilometres, $R$ is the extension rate in millimetres per year, and $T$ comes out directly in millions of years. Feed in the Main Ethiopian Rift's own numbers — spacing 50 to 100 km, extension ~5.2 mm/yr — and you get an implied period of roughly 9.6 to 19.2 million years. Feed in the Red Sea Rift's numbers — spacing 150 to 200 km, extension 10.5 to 19.5 mm/yr — and the implied period ranges from about 7.7 to 19.0 million years, depending which end of the reported extension-rate range you use [1].

Those two ranges overlap almost entirely, in a window of roughly 10 to 19 million years, despite the raw spacings differing by a factor of two to four. That overlap is exactly what you would expect if one plume really is pulsing at a broadly consistent rate underneath the whole junction, and each rift arm is simply stretching the same pulse train at its own local speed — a fast, thin-crusted rift smears the pulses out into a long wavelength; a slower, thicker-crusted rift compresses the same pulses into a shorter one.

The Plate Decides How Loud the Plume Gets to Talk

Why would the same pulsing source produce a tighter pattern under the Main Ethiopian Rift than under the Red Sea Rift? The paper's proposed mechanism is mechanical, not chemical: the Main Ethiopian Rift still has thicker continental crust sitting on top of it, 25–33 km compared with the Red Sea Rift's 16–25 km, and it is extending more slowly, roughly 5.2 mm/yr against 10.5–19.5 mm/yr [1]. A thicker, slower-moving lid gives upwelling mantle less room and less time to flow sideways before it has to freeze into the plate or erupt, so the same underlying heterogeneity gets compressed into a shorter spatial footprint. Strip away crust and speed it up, as the Red Sea arm does, and the same material can spread out and channel further before the next pulse's signature takes over. In the authors' words, progressive lithospheric thickening away from the upwelling centre "should reduce the volume capacity for melt, impeding mantle flow," so that "the heterogeneous nature of the pulsed upwelling would exhibit a more condensed spatial pattern within the [Main Ethiopian Rift] compared with [the Red Sea Rift]" [1].

That is a genuinely different kind of claim from "the plume feeds three rifts." It says the overlying plate's own physical properties — its thickness, its speed — are actively editing how the plume's signal gets laid down in the rock record, arm by arm.

Table 1. The two rift arms the k-means clustering directly compared, with the paper's reported values [1]. The arm with the thinner, faster-moving plate shows the longer cluster spacing.

Flipping the Arrow: Rift Geometry Steers the Plume

Most popular accounts of continental rifting run the causal arrow one way: a plume rises, weakens the lithosphere from below, and the rift follows where the plume points. It's a tidy story, and Ebinger and Sleep's single-plume-head model for East Africa is very much built around the plume as the active, initiating agent, with the crust above it a comparatively passive lid waiting to be thinned [3]. The Afar result does not overturn that story so much as add a second arrow going the other direction: once a rift geometry exists — however it got there — its own thickness and spreading rate feed back and reshape how the plume beneath it is allowed to move. The plume is not simply painting a picture on a blank canvas; the canvas has grooves in it, and the paint runs along the grooves.

Watts and colleagues put the general claim plainly: "the evolution of mantle upwellings is influenced by the dynamics of overriding plates" [1]. That is a statement about causality running upward through the lithosphere as much as downward from the plume, in a system — a rifting continental triple junction — that geologists had mostly modelled as one-directional. Prior seismic work on the region, including Hansen and Nyblade's imaging of the deep structure beneath the Ethiopia/Afar hotspot, had already located a broad, low-velocity anomaly connecting Afar to the wider African superplume structure at depth [5]; what the new geochemistry adds is evidence that the plate geometry above that deep anomaly is actively sculpting how its material gets distributed once it nears the surface, arm by arm, rather than the anomaly simply dictating outcomes to a passive rift.

The Main Ethiopian Rift itself has a documented history of forming through episodic, localized magmatic and structural events rather than one smooth continuous stretch — Wolfenden and colleagues traced its early evolution through a sequence of rift-segment linkups and magmatic pulses on the way to becoming today's triple junction [6]. A plume whose surface expression is edited by rift segmentation and plate speed is consistent with a rift history that was never smooth to begin with.

What Would Change Our Mind

The cross-rift match — the same 1-2-3 cluster sequence at two different spacings — is the load-bearing piece of evidence for "one pulsing plume" rather than "three separately noisy rifts," and it rests on a clustering algorithm's judgment about which rocks belong together. K-means clustering requires you to choose how many clusters to look for in advance, and different reasonable choices can shuffle borderline samples between groups; the paper reports settling on four clusters for the Main Ethiopian Rift and three for the Red Sea Rift, and we would want to see how sensitive that particular match-up is to trying five, or two, before fully trusting the "same sequence, different scale" reading.

Our own kilometres-to-megayears conversion is even more exposed, because it assumes the extension rate measured today has held roughly steady for the tens of millions of years needed to stretch out the oldest visible cluster spacing — an assumption the paper does not make and one that real rifts, which speed up and slow down over their lifetimes, are not obliged to honour. A direct geochronological date on rocks from adjacent clusters, rather than our rate-based inference, would be the real test of whether "10 to 19 million years between pulses" means anything at all.

What We'd Tell Someone Who Only Read the Headline

Three things. First, "pulsing plume" here does not mean anyone watched anything pulse in real time; it is an inference from a repeating chemical pattern frozen into rock that erupted at different points over a very long span, read out in space rather than in time. Second, the surprising result is not that Afar's mantle is heterogeneous — mixed mantle sources are common knowledge — but that the heterogeneity comes packaged as a repeating sequence rather than random noise, and that the same sequence shows up on two different rift arms at two different physical scales. Third, and the reason this became a field note rather than a one-line mention: the paper's own reading inverts a piece of geological common sense. The rift is not just a stage the plume performs on. Its thickness and its speed are, in some real and now chemically documented sense, part of the plume's instrument.