Deformation and seismicity decline before the 2021 Fagradalsfjall eruption

Volcano observatories rely on interpreting precursory signals—typically accelerating deformation and seismicity—to issue timely eruption warnings. Classical approaches such as Voight’s material failure forecast method interpret escalating rates in strain or seismicity to forecast eruption timing. However, some eruptions exhibit different behaviour, including reductions in precursory activity immediately before onset. Prior examples include the 2010 Eyjafjallajökull flank eruption, and declines noted before events at Redoubt and Telica. In some cases, reduced seismicity has been linked to sealing of gas pathways and increased pressurization with uplift. The study investigates the 2021 Fagradalsfjall eruption to document and explain an unusual pattern: concurrent declines in both deformation and seismicity in the days before eruption, and explores how interactions between tectonic stress and crustal properties can produce such precursors, with implications for eruption forecasting on oblique rift systems.

The material failure forecast method (Voight, 1988) and its Bayesian/stochastic enhancements have been widely applied to accelerating precursory signals. Yet, counterexamples exist where seismicity and/or deformation decline before eruption: Eyjafjallajökull 2010 (decline after months of elevated activity), Redoubt (1989–1990 and 2009) and Telica (1999), sometimes attributed to sealing of gas pathways leading to pressurization and increased surface deformation. Broader context on the Reykjanes Peninsula includes long-term strain accumulation and oblique plate motion, with arrays of strike-slip faults accommodating shear during non-eruptive periods. Previous unrest since late 2019 involved repeated intrusions and inflation episodes. Comparisons are later made to passive rifting and dyke intrusions at Kīlauea and in East Africa, and to Tolbachik (1975–1976), where seismicity declined 1–2 days before fissure opening, highlighting that precursor declines can be characteristic in certain tectonic settings.

The study integrates seismology, geodesy (GNSS and InSAR), and geodetic modelling. Seismic data include manually reviewed events (Mw > 1), detection of >53,000 earthquakes (24 Feb–19 Mar 2021), and waveform cross-correlation relocations to map dyke geometry and migration. GNSS stations (e.g., KRIV, LISK) provided daily horizontal displacement time series showing co-seismic steps and post-seismic/deformation trends. Sentinel-1A/B InSAR (descending track T155) produced 6-day interferograms spanning late February to March 2021, revealing evolving line-of-sight (LOS) deformation; positive/negative LOS volumes were computed by summing unwrapped LOS changes exceeding ±28 mm thresholds times pixel area. For geodetic modelling, a modified GBIS Bayesian inversion framework was used with sources embedded in a uniform elastic half-space: (a) a vertical dyke with opening and potential shearing, (b) shear on a subvertical plate-boundary-aligned plane representing distributed slip on mapped N–S strike-slip faults, (c) rectangular shear planes for the two largest earthquakes, and (d) a localized contracting point-pressure source near the dyke centre. A one-segment dyke model with uniform opening/shear was first tested, then refined to a two-segment dyke consistent with relocated seismicity: a northern segment striking N45°E and a southern segment N23.5°E with distributed opening and slip. Joint inversion of GNSS and InSAR constrained daily dyke volume change (magma inflow rate) and the corresponding mean depth of magma emplacement. Stress changes (dilatational and shear) were estimated, including Coulomb failure stress changes on regional faults. Temporal evolution of seismicity was separated between dyke and plate-boundary areas to assess relative timing of declines.

• A dyke approximately 9 km long was emplaced between the surface and ~8 km depth with a total volume of ~34 × 10^6 m³ starting on 24 February 2021. This emplacement and associated shear triggered intense seismicity, including a Mw 5.64 event on 24 February and numerous Mw ≥ 4 events (64 total Mw ≥ 4.0 events detected during the interval).
• GNSS and InSAR show initially high deformation rates after 24 February that gradually decreased, reaching near-zero deformation at eruption onset (19 March 2021). For example, KRIV moved ~10 mm/day (declining), LISK increased to ~14 mm/day by 3 March then declined; InSAR 6-day interferograms show decreasing LOS deformation with the slowest rates in the last days pre-eruption.
• Seismicity and seismic moment release decreased toward eruption onset. After 14 March (Mw 5.33), only one event above Mw 4.0 occurred (Mw 4.20 on 15 March). Two days before eruption, activity concentrated in small swarms, including at the eruption site. In the last ~15 hours, only shallow (<1 km), small (<M2) events with depleted high frequencies were recorded near the vent.
• Seismic relocations delineate a vertical two-segment dyke (north: N45°E; south: N25°E/N23.5°E). Seismicity migrated from the northern segment (24 Feb–3 Mar) to the southern segment and westward along the plate boundary, then along the southern segment after 7 March. The seismicity rate decline began ~3 days earlier in the plate boundary area than within the dyke.
• Inferred magma inflow (dyke volume change) decreased over time: ~30–35 m³/s (24 Feb–3 Mar), ~10–20 m³/s (3–15 Mar), and <10 m³/s (15–19 Mar). The mean depth of emplacement shallowed after 11 March. A near-linear relationship between inflow rate and emplacement depth indicates control by flow up a deep conduit with minimal pressure needed for lateral dyke intrusion. This relationship predicts an initial eruption rate ~7 m³/s and suggests a feeding source depth ~19 km; the conduit cross-sectional area is of order a few m².
• Modelled stress changes at ~3.5 km depth reached tens of MPa, vastly exceeding annual tectonic dilatational stress accumulation (tens of kPa/yr), implying a substantial fraction of centuries of accumulated tectonic stress was released during the sequence. Combined dyke opening and simple shear on the plate-boundary plane effectively relieved tectonic stress in this oblique rift setting.
• Overall, deformation and seismicity declined over several days before eruption onset, demonstrating that tectonic stress release and shallow weak crust can lead to reduced precursory signals immediately prior to eruption.

The findings demonstrate that in oblique rift settings like the Reykjanes Peninsula, dyke emplacement and associated shear can release accumulated tectonic stress, leading to declines in both deformation and seismicity prior to eruption. As stress at the base of the brittle crust is relieved, magma must rise higher before intruding, increasing the pressure drop and reducing inflow rates, which in turn diminishes surface deformation and seismicity. This pattern contrasts with classical escalating precursors predicted by the material failure forecast method but aligns with behaviours observed in passive rifting contexts (e.g., Kīlauea, East Africa) and Tolbachik, where seismicity declined before fissure opening. Practically, a lull or decline in geophysical signals—rather than acceleration—can herald imminent eruption in pre-stressed rifted crust. The results underscore the need for volcano observatories to consider tectonic stress interactions and crustal layering when interpreting precursors and issuing warnings, as an eruption may commence quietly once dyke-related stress has been released.

This study documents a rare but important precursor pattern: concurrent declines in deformation and seismicity in the days before the 19 March 2021 Fagradalsfjall eruption. Joint seismic and geodetic observations, coupled with Bayesian geodetic modelling, show dyke emplacement, plate-boundary shear, and localized deflation releasing a substantial fraction of tectonic stress accumulated over centuries. Magma inflow rates decreased and emplacement shallowed, consistent with flow through a deep conduit, culminating in a quiet surface breach with minimal seismic energy release. The work advances understanding of how tectonics and crustal structure govern pre-eruptive signals in volcanic rifts and cautions against relying solely on accelerating precursors. Future research should assess the prevalence of this behaviour across rift systems, refine coupled stress–flow models, and integrate real-time multi-sensor data to improve forecasting and risk mitigation.