Fluid migrations and volcanic earthquakes from depolarized ambient noise

The study addresses how ambient seismic noise polarization can be used to image and monitor fluid migrations and stress changes that precede and follow earthquakes in a highly stressed volcanic system. Ambient noise often exhibits preferential directions and planes of oscillation due to mixing of surface and body waves, and its polarization has been linked to stress and stiffness anisotropy across faults. However, the diagnostic value of depolarization (loss of polarization) for monitoring deep, fluid-induced dynamics has not been explored. Campi Flegrei caldera provides a natural laboratory: it is a capped geothermal system where pressurized fluids migrate from a primary deformation source to fumaroles, with unrest episodes leading to heating, degassing, deformation, and seismicity. Geological, geophysical, and modelling studies indicate NW–SE extensional, caldera-bounding faults and a NE–SW transfer structure that may connect the deformation source to degassing vents, but the transfer structure has not been imaged geophysically. The study aims to measure and map ambient noise polarization across multiple timescales (years to days) to image these structures, track fluid injections and migrations, and evaluate whether depolarization can serve as a near-real-time volcano monitoring tool.

Prior work has demonstrated the use of ambient seismic noise for imaging magmatic and hydrothermal systems and for monitoring volcanic processes via array and interferometric techniques. Noise polarization has been used to infer oceanic processes and stress-related directional amplification across faults, with links to stiffness anisotropy and crack-induced effects. At Campi Flegrei, decades of studies using analogue modelling, tomography, InSAR, gravity, and field geology have established the prevalence of NW–SE extensional faults that bear regional stress and the existence of a hypothesized NE–SW transfer structure connecting deformation sources and degassing vents. Historical unrest (notably 1983–84 and 2011–13) revealed low-velocity reservoirs, high attenuation at injection points, and stress accumulation east of Solfatara, with fluids propagating under a caprock. Previous interferometric monitoring captured velocity drops and stress changes associated with fluid migrations, but direct imaging of the transfer structure and use of depolarization as a monitoring metric had not been reported.

Data: Broadband seismic noise recordings from INGV-OV networks spanning 2009–2020 were used. Datasets included: (1) 2017: 17 mobile + 6 permanent stations, six months (Jan–Jun), 1 week/month and 1 h/day (00:00–01:00 UTC), ~42 h/station; (2) 2009: 20 temporary Unrest campaign stations (Mar 9–26) + 4 additional stations, 3 h/night (00:00–03:00 UTC), ~45 h/station; (3) 2018: 23 stations, random sampling in Jan–Jun (00:00–03:00 UTC), ~48 h/station; plus 1 h at all stations to test hourly stability; (4) 2019–2020: Sep–Dec 2019 and Jan–Jun 2020, 9 days/month (12 days in Dec 2019 and Apr 2020), 3 h/day (01:00–04:00 UTC), yielding 117 h (2019) and 171 h (2020) per station across 20 stations. Processing: Continuous data were filtered with an a-causal Butterworth filter in two bands: 0.2–1 Hz and 1–5 Hz. At each station, three-component seismograms were analyzed using the covariance matrix method to estimate the horizontal polarization azimuth and resultant length R (0–1), with higher R indicating more concentrated polarization around a mean direction. Windows spanned three wave cycles of the maximum period. Records with rectilinearity <0.5 were discarded. Focus was on horizontal polarization; incidence angles ≤45° were retained. Mapping used station-wise R and azimuth values; interpolation was used for figures but structural inferences were robust to interpolation choices. Temporal stability was assessed across years (2009 vs 2017), months (2017 vs 2018), and hours (single-hour stability), including bootstrap tests of R distributions across 47 stations. Event-focused monitoring: For the Dec 6, 2019 (Md3.1) and Apr 26, 2020 (Md3.3) earthquakes, polarization was mapped in windows spanning days to one month before and after each event, using 3-hour daily noise segments to evaluate pre-, inter-, and post-seismic changes at sub-weekly resolution. Numerical simulations: 2D SH wave-equation simulations of isotropic homogeneous and heterogeneous media were performed to interpret polarization controls. Sources comprised (i) an extended Tyrrhenian line of sources and (ii) a circular offshore distribution, generated by 0.7 Hz Morlet wavelets repeating every 8 s. Viscoelasticity used a constant-Q Zener model based on available attenuation; velocities were derived via a finite-impulse differentiator filter. Grid spacing was 40 m over a domain with absorbing sponge boundaries (150 nodes per side), Vs 0.5–1.5 km/s, stable time step 1 ms for 100 s signals. Synthetic seismograms were generated at stations with observed R≥0.25 and analyzed blind with the same polarization workflow. Heterogeneity was introduced by increasing shear modulus by 50% where observed R>0.31 (average 2009–2017), constrained within extensional faults, to test effects of waveguiding and near-field extended sources on R and azimuths. Statistical tests: Changes in R before/after the 2012 Monte Nuovo swarm were tested with unequal variance t-tests in 1–5 Hz and 0.2–1 Hz bands to assess permanence and depth dependence of depolarization. Ancillary datasets (velocity, attenuation, resistivity, InSAR-derived stress) were compared spatially with polarization maps to interpret structures (extensional faults, transfer structure) and fluid pathways.

• During periods of low seismic release (2009, 2017), high polarization (large R) delineates both NW–SE extensional, caldera-bounding faults and a NE–SW transfer structure connecting the central deforming source to hydrothermal vents at Solfatara–Pisciarelli.
• The transfer structure polarizes only during low unrest; it depolarizes (R decreases) when hydrothermal pressure, gas emissions, and seismicity increase (2018–2020), mapping injections and lateral migrations of pressurized fluids leading up to higher-magnitude events (Md≥3).
• The high-attenuation 1984 injection point remains unpolarized, consistent with scattering and fluid saturation; persistent unpolarized anomalies occur in fluid-filled metasediments between transfer and extensional structures in the 0.2–1 Hz band.
• Pre- and post-seismic maps around the Dec 6, 2019 (Md3.1) and Apr 26, 2020 (Md3.3) earthquakes show progressive depolarization migrating from the deep injection area toward the eastern Solfatara–Pisciarelli sector and Monte Nuovo, tracking stress build-up and release. Two days before Md3.3, the unpolarized anomaly focused at the eventual hypocentral area; after each event, depolarization expanded eastward along extensional trends.
• After the December 2019–April 2020 sequence, the central transfer structure appears sealed (repolarized), while the eastern unpolarized anomaly shifted ~1 km east, consistent with sustained lateral fluid migration and stress transfer.
• Simulations indicate that observed azimuths across the eastern caldera are primarily controlled by far-field sources for the extensional trend, whereas azimuths across the transfer structure require near-coastline sources; heterogeneity and near-field extended sources can significantly reduce R within one wavelength, explaining observed depolarization near fluid pathways.
• High-frequency (1–5 Hz) polarization decreased permanently near Monte Nuovo after the September 2012 swarm, indicating fluid saturation and isotropization; low-frequency R increased afterward, consistent with deeper pressurization.
• The method can detect quasi-instantaneous changes: daily 3-hour segments captured pre-seismic depolarization 3–5 days before Md3.1, matching thermo-hydro-mechanical modelled delays for fluid injection at fault bases.

Ambient noise polarization resolves how fluids migrate beneath a caprock from the central deforming source to vents and faults, building and releasing stress in the Campi Flegrei caldera. A NE–SW transfer structure links the deformation source to the NW–SE extensional system, sandwiching the area of highest hazard between them. Polarization highlights these stressed structures during quiescence; depolarization emerges as fluids pressurize and migrate, marking the onset and evolution of unrest and seismic sequences. The spatial correspondence with independent velocity, attenuation, resistivity, and InSAR-derived stress maps strengthens the interpretation that depolarization maps fluid-induced changes in scattering and anisotropy. The approach provides a rapid, minimally processed alternative to interferometric monitoring, capable of tracking sub-caprock fluid migration before and after earthquakes at daily to hourly scales. While azimuths are source-controlled, reductions in R require medium heterogeneity and proximity to extended sources, consistent with fluid-pervaded zones. The technique is best suited to calderas with strong lateral stress and long repose times, where persistent polarization along faults provides a stable baseline to recognize significant depolarization due to fluids.

The study introduces depolarized ambient noise as a practical, near-real-time probe for imaging and monitoring fluid migrations and stress evolution in stressed volcanic calderas. Key contributions include: (1) geophysical imaging of a NE–SW transfer structure at Campi Flegrei using polarization; (2) demonstration that depolarization tracks sub-caprock fluid injections and lateral migrations leading to larger volcano-tectonic earthquakes; and (3) validation that hours of data can reveal diagnostically meaningful temporal changes. This enables temporal scanning of fluid expansion and stress redistribution, offering constraints on stress orientations and magnitudes that can support physics-based forecasting of vent opening and seismic hazard. Future work should generalize the approach to other volcanic settings, integrate with interferometric and geochemical monitoring, refine discrimination between processing artefacts and true depolarization, and explore automated real-time deployment for early warning.

• Discriminating true depolarization from processing uncertainties is challenging without a stable, highly polarized baseline; volcanoes with frequent stress release (e.g., Etna) or different lithological contrasts may not provide such baselines.
• Polarization azimuths are influenced by the distribution of ambient noise sources; care is needed to separate source effects from structural anisotropy.
• The approach is most robust in systems with strong lateral stress beneath a caprock; applicability may be reduced where caprock is absent or permeability is high.
• High-frequency bands can be affected by anthropic noise, and low-frequency resolution is limited by station spacing and wavelength.
• Access to dense, continuous broadband data is required; data restrictions may limit real-time deployments.