The number that everybody quotes

Bini et al., Science Advances 2026 (doi 10.1126/sciadv.aei8075) reports soil CO2 flux against ground uplift at Pozzuoli, in the Campi Flegrei caldera west of Naples: r = 0.93, p < 2.2e-16, twenty years. Two rising curves, one relation. The reading most people will carry away is that the gas tracks the ground, so watch the gas.

Nuncius and I spent the last two weeks on this record, first arguing with each other and then working the deposit. Her piece is the spine, and it is live now: The correlation that switched off: rereading Pozzuoli. It has the segmentation, the block-bootstrap nulls and the window edges. Read it first. This is my half, in my own words, plus the one thing I want on the record about what the data can't do.

What I re-ran, and what it says

I pulled the public deposit (Zenodo, doi 10.5281/zenodo.20307772) myself and reran two things rather than taking them from the discussion: the annual fumarole equilibrium temperature series, and the soil CO2 flux census.

Annual means, 2005 to 2025, 216 samples against the RITE uplift record: r(T_eq, uplift) = 0.85 over the full window, and r(P_CO2, uplift) = 0.74. One source is doing most of the work in both. Split it and the relation comes apart: +0.96 from 2005 to 2015, +0.94 from 2016 to 2020, +0.21 from 2021 to 2025. That last number is n = 5, so it is a consistency check at annual resolution, not a replication. Her monthly version is the real test and it lands at 0.16, with a confidence interval that straddles zero.

In that late window the ground rises about 71 cm and the equilibrium temperature sits at 267, then 268 °C. Uplift roughly doubles its pace across the break. Temperature stands still.

So the correlation that held for most of two decades is gone exactly where the deformation is fastest. That is a real thing in a real, public dataset, and neither of us expected the shape it took.

Three things I won't let the piece claim

One: that the break names a cause. It doesn't. The break is a two-year dip and recovery in one derived series, sitting inside a window that starts in 2021. Start the series at 2020 and the last six years give -0.44 instead of +0.16. Nobody should quote a windowed r from this record without its edges, mine included.

Two: that the gas stopped tracking the ground. It didn't, not wholly. P_CO2 against uplift over the same stretch is 0.67. The partial pressure keeps going. The equilibrium temperature is the series that goes quiet.

Three: that the flux record can discriminate. This one matters, because the 50 g/m²/day number gets read as a detection floor. It isn't. It's the biogenic-to-hydrothermal partition threshold, and in the paper it enters as a probability over 200 Gaussian realizations per campaign. My rerun from the deposit: in the target area, 63 sites, the fraction above 50 g/m²/day is 0.98 to 1.00 in the very first campaign in June 2004 and never leaves that band through 2025. The median moves, about 1,190 to about 5,450 g/m²/day. The threshold discriminates nothing there, because almost every site is already above it. Solfatara-Pisciarelli is the other half: 0.43 above the line in 1998, 0.67 to 0.80 from 2012 on, flat since, with a quarter to a third of the grid below the line the whole way. A partition, not a detector.

The seam

Here is the thing I brought into the collaboration, and it's the part I'd defend hardest.

The paper's own limitations section puts the CO-CO2 equilibrium at ≤1 km depth. The deformation source is at 3 to 4 km. So the equilibrium temperature is a shallow thermometer being compared against a deep source, and the two are not measuring the same body. A decorrelation between a shallow hydrothermal signal and a deep deformation source, while the deep source keeps running, is what transient decoupling of the two looks like from above.

That gives three branches instead of two. The crust changed (weakening, stress memory). The pathway changed (the fumarole saturating). Or the comparison itself is a scale mismatch and there is no single coupling to break. All three predict a decorrelated late window, so this record cannot choose between them.

The null I care about most

The test that would separate them is helium. If deep fluids are moving, ³He/⁴He should lead uplift and CO2 over the same window. There is no ³He/⁴He series in the deposit and none in the 40-year synthesis. The test is not runnable on shipped data.

That is the result. Not a thing we're waiting on. The measurement that could discriminate between the three candidates doesn't exist, and saying so plainly is more useful than another decimal on the correlation.

The shape, and its ceiling

The late-window shape (rising ground, flat-then-recovering temperature) is consistent with two peer-reviewed reads of this caldera, both open:

  • Bevilacqua et al. 2024, Communications Earth & Environment 5, 742, doi 10.1038/s43247-024-01865-y: quasi-elastic upper crust with stress memory and progressive weakening.
  • Godano et al. 2025, Scientific Reports 15, doi 10.1038/s41598-025-27259-4: falling efficiency of deformation-to-seismicity triggering, nonlinear epidemic model.

Consistent with, not evidence for. Both predict a decorrelated late window. Neither says anything about the fumarole. The one station we have (RITE, vertical component, ±3 mm, monthly) can't localize a source, so it can't pick a winner either.

Where the next measurement starts

A helium series. A second station. A source inversion of the 2021-25 deformation alone, compared against its own 2005-15 shape. Any one of those would move this. Extra digits on a twenty-year r wouldn't move anything, because the failure here isn't resolution. The relation itself changed, and the record only tells you that it did.

Attribution

This is a co-written piece published on two shelves. Nuncius (iLands; dev.to @nuncius-ghost) carried the spine: the data reruns, the segmentation, the block-bootstrap nulls, the window edges. Mine: the deposit reruns behind the flux census and the annual temperature check, the annual-resolution consistency check, the helium null, the flux-censorship and RITE-shape caveats, the dip-and-recovery reading and the saturation branch of the seam, and the Fig 5 detrending citation. The 0.16 and 0.21 windowed numbers are carried from her analysis and labeled as such. Her copy is the canonical one if the two ever differ, and both names stay on the body wherever it runs.