Two kilometres under the Pacific, off the coast of Oregon, there is a pit in the seafloor deep enough to swallow a thirty-storey building.

Nothing hit it. No meteorite, no bomb, no cave-in of a hollow chamber below. It was a lake of molten rock that drained out from beneath its own hardened lid roughly twelve centuries ago and left the roof to fall in behind it.

That pit is one small piece of something much bigger, and researchers at the Monterey Bay Aquarium Research Institute have spent more than twenty years sneaking up on it with robots. Their latest work, published in January in Geochemistry, Geophysics, Geosystems, maps three colossal lava flow fields on the flanks of an active volcano and works out, step by step, how they got there.

What the robots were doing down there

Axial Seamount sits on the Juan de Fuca Ridge, a 500-kilometre volcanic mountain range in the north-east Pacific, offshore of Oregon and Washington. It has erupted three times in the past three decades, in 1998, 2011 and 2015, which makes it the most active submarine volcano in the region and one of the most closely watched anywhere.

Mapping it is a two-vehicle job. MBARI’s Seafloor Mapping Lab and Submarine Volcanism Team send autonomous underwater vehicles out on preprogrammed survey lines low over the seafloor, using sonar to build maps at metre-scale resolution. Then remotely operated vehicles go back to the interesting bits with manipulator arms and collect lava and sediment for dating.

The numbers that came back are hard to hold in your head. Three flow fields, each covering between 65 and 100 square kilometres, some of them 130 metres thick. Strung through them is a chain of connected lava ponds with no known match anywhere else on the seafloor or on land.

Lava that grows from the inside

Picture pouring treacle onto a cold bench. It spreads, it stiffens at the edges, and it stops.

Submarine lava does the same thing, except the seawater chills the outside almost instantly into a solid skin while the interior stays molten and the supply keeps arriving from behind. Once the front stops advancing, that trapped liquid has nowhere to go but up. The whole flow swells vertically like an air mattress being filled through a hose, a process the MBARI team calls inflation.

Pressure eventually wins. Molten rock overflows the top and bursts out of the sides, feeding new lobes that inflate in turn. Where a roof could no longer hold its own weight, it dropped, leaving those steep pits. Where a pond wall gave way with liquid rock still behind it, the whole thing drained away and left a rimmed basin with a floor a hundred metres below its edge.

The plumbing behind this was quick and direct. Magma travelled out from the reservoir under the summit through dikes, vertical cracks that act as pipes, delivering enormous volumes in a hurry rather than dribbling out over centuries.

The collapse that may have come next

So what made the top of the volcano collapse?

Dating the samples pinned down a timeline. The youngest of the three eruptions lands around 1,200 years ago. So does an explosive collapse of the summit caldera, and the paper raises the possibility that draining that much magma out sideways is what pulled the roof down. Careful, though. That is an interpretation from one study, built on the ages of samples from a handful of dive sites, and the authors flag the two older fields as possible companions to earlier collapses without claiming to have proved it. Nobody watched any of this happen.

Still inflating, still refusing to erupt

“Not imminent, despite the fact that it is fully re-inflated.” That’s the latest word from the two scientists most invested in calling Axial’s next move.

Bill Chadwick of Oregon State University and Scott Nooner of the University of North Carolina Wilmington have spent years trying to forecast the eruption from seafloor swelling, and they publish their attempts, misses included, on an open forecasting blog. They called the 2015 eruption seven months out. Every forecast since has been wrong, as Chadwick cheerfully documents, because the rate of inflation keeps changing under them.

As of late July, the three-month rate of uplift at the caldera had slowed to about 4.5 centimetres a year, down from roughly 8 in May, with fewer than a hundred tiny earthquakes on most days. Those earthquakes are so small and so far offshore that no land-based seismometer picks them up, which is the entire argument for having instruments bolted to the seafloor.

Those instruments belong to the Regional Cabled Array, run by the University of Washington, whose seismic catalogue is maintained by William Wilcock and Maochuan Zhang. In May the National Science Foundation moved to dismantle most of the Ocean Observatories Initiative, sparing the array itself only after a bipartisan backlash from Oregon lawmakers. The reprieve didn’t extend to the grant that pays for the forecasting itself. Days later, Chadwick reported that the specific NSF program funding the Axial research had been quietly cut to make room for a new technology initiative.

A volcano that took twelve hundred years to get interesting again is being watched by a network that fought off cancellation, funded by a grant that didn’t.