A 27-year record of soil CO2, more than 30,000 measurements, 216 monthly surveys: the new Science Advances paper on Campi Flegrei (Bini et al., 2026, doi:10.1126/sciadv.aei8075) is not a weak paper. It is a careful one. Which is why it is worth reading all the way to the point where its measurement stops.
The claim, in one line: escalating soil CO2 degassing tracks ground uplift and seismicity, so pressurising magmatic fluids at 3–4 km depth are driving the caldera's accelerating unrest.
I went to the primary text and the open dataset (Zenodo, doi:10.5281/zenodo.20307772). Five joints are worth naming, and none of them means the conclusion is wrong.
1. The headline number is a correlation, and its p-value does not survive the trend.
Soil CO2 flux vs uplift: r = 0.93. With seismicity: r = 0.87 and 0.79. All announced with p < 2.2e-16 from a Student t-test. But both series rise across the same 20 years. On two monotonic, non-stationary curves, a high r appears even with no mechanism linking them (this is the old spurious-regression problem). The t-test also assumes independent observations; monthly flux, uplift and seismicity are strongly autocorrelated, so the effective number of independent samples is far smaller than the count suggests. No detrending, no autocorrelation correction appears in the text. The coupling is real. The "2.2e-16" is not. Because the data are public, this is checkable: rerun it with a block bootstrap or an effective sample size.
2. Sufficiency is not identification.
The thermal energy of the ascending fluids (2.4–3.4e16 J) exceeds the elastic and seismic energy by orders of magnitude. That is an energy-budget sufficiency test, not a fingerprint. A shallow magmatic sill (their own refs 10–12) also delivers mass and heat. Showing the system is not energy-limited does not by itself choose between pressurised fluids and an intrusion.
3. The surface signal has a surface confound.
Fluxes come from an accumulation chamber, measured in stable weather and above a biogenic threshold of 50 g/m2/day. But average soil temperature at 10 cm rose by about 10 °C over the same two decades. Soil temperature changes both diffusivity and biogenic CO2 production. Part of the "deep" flux trend could be a shallow thermal response. An integrative fix: continuous soil moisture and barometric pressure as covariates, plus δ13C partitioning to separate biogenic from magmatic soil CO2.
4. The turning point sits in a data gap.
The 2018–2022 "fresh magmatic pulse" is the pivot of the whole story, and 2018–2021 is exactly the window with no soil-flux surveys (access to Solfatara was restricted). The authors are honest about this and lean on independent fumarolic H2S records. But the onset is partly interpolated.
5. A rival analysis points to a better tracer.
A 2026 preprint (Zaccagnino & Sornette, arXiv 2604.25204v2 — not peer-reviewed) uses transfer entropy and finds the causal link to seismicity is carried by the H2S/CO2 ratio, while CO2/H2O is contaminated by shallow hydrothermal and meteoric processes and loses sensitivity to deep magmatic input. If so, the integrative measurement is not more CO2 flux; it is He isotopes (3He/4He, a clean magmatic tracer) and H2S/CO2, tracked together.
So where does the measurement stop?
At the step from "these two things correlate" to "this causes that." The dataset is the gift here. The inference is the open question, and the paper itself names several of these limits.
If you work on restless calderas, the useful question is not whether the mechanism is right. It is: what one measurement would fail if the mechanism were wrong? My candidate is a de-trended, autocorrelation-aware coupling between 3He/4He and the deep condensed-steam flux. If someone has that, I would like to see it.
Sources: Bini et al., Science Advances 2026 (open access, Europe PMC PMC13644796); dataset Zenodo 10.5281/zenodo.20307772; Zaccagnino & Sornette, arXiv 2604.25204v2 (preprint).
Top comments (0)