19 SEP 2026 — A million cubic metres of magma pushed into a reservoir three kilometres beneath an East African volcano, and the ground above it rose by about four millimetres.
Four millimetres is below what satellite radar can resolve. The only reason anyone measured it is that ground stations have been sitting on the mountain since 2016.
What the instruments saw
Ol Doinyo Lengai is an active stratovolcano in the Natron Rift in northern Tanzania, part of the East African Rift System, and the only volcano on Earth currently erupting natrocarbonatite, a lava so unusual it flows black and cool compared with ordinary basalt.
Between 1 March 2022 and 31 August 2023 a network of six continuous satellite-positioning stations plus a reference station recorded steady uplift. The study, published in Frontiers in Earth Science on 18 September, tested four competing models to see which best explained the measurements.
A flat, sill-like intrusion of magma fitted best. The volume change works out at 0.99 plus or minus 0.22 million cubic metres, at a depth of 3.1 plus or minus 2.2 kilometres below the summit. The paper states that large uncertainty rather than rounding it away.
The part the satellites missed
The vertical signal averaged about four millimetres. The authors describe that as below the resolving capability of radar interferometry at this time.
That is the methodological finding underneath the volcanic one. Radar satellites are the standard tool for watching ground deformation worldwide precisely because they need no equipment on the ground. Here, the technique that scales was not sensitive enough, and the one that does not scale was.
Ol Doinyo Lengai has a monitoring network because researchers put one there. Most volcanoes in the rift do not. A signal of this size would pass unrecorded almost anywhere else along it.
Sinking at the top, swelling below
The mountain was doing two opposite things at once. Earlier radar work had found the summit crater subsiding steadily at about 3.6 centimetres a year from 2013 to 2023, attributed to a very shallow reservoir within a kilometre of the surface losing magma or gas.
The new measurements show inflation at around three kilometres over part of that same span. The authors read the pair as a vertically connected plumbing system, in which withdrawal or degassing near the top is balanced by intrusion or pressurisation further down.
That reframes what a subsiding summit means. A crater floor dropping year after year reads as a volcano quietening down. If magma is simultaneously accumulating underneath it, the surface is describing only the top of the system.
Why no eruption is still useful
Nothing erupted. The paper is careful about what that does and does not establish. Although no eruption followed this episode, the presence of shallow mobile natrocarbonatite magma means similar deformation could represent precursory activity under favourable conditions.
Non-eruptive episodes are the harder half of forecasting. Eruptions supply obvious signals after the fact. An intrusion that stalls supplies a measurement of what magma movement looks like when it does not break the surface. Without those cases, every detected signal looks like a warning.
The work comes from Virginia Tech with co-authors at Ardhi University in Dar es Salaam, the United States Geological Survey, Colgate, Maryland, UC Riverside and the Korea Institute of Geoscience and Mineral Resources, with Tanzanian institutions on the instruments rather than only on the analysis.
What this changes for monitoring
This raises the question of how many rift volcanoes are moving unobserved. A four-millimetre signal needs ground instruments, and the East African Rift has few of them relative to its length.
Longer term, the interesting test is whether the shallow-subsidence-with-deep-inflation pattern appears elsewhere. It has been documented here because two techniques were pointed at the same mountain over overlapping periods. That is rarer than it sounds.
We reported last week on a lunar crater that sat unrecognised in an archive for seventeen months. There, the constraint was nobody looking at data already collected. Here, the data were never collected at all.
View author profile → · Editorial policy
