Volcanic eruption forecasting relies heavily on interpreting precursory signals like increased deformation and seismicity. While escalating rates of these parameters are common precursors, some eruptions, such as the 2010 Eyjafjallajökull eruption, have shown a decline in activity before onset. This study investigates the precursory activity leading up to the 2021 Fagradalsfjall eruption on the Reykjanes Peninsula, Iceland, an area characterized by oblique plate spreading and a history of infrequent eruptions. The Reykjanes Peninsula's unique geological setting, with its en echelon fissure swarms and north-south oriented strike-slip faults, makes it a crucial location to study the interplay between magma intrusion and tectonic stress release in eruption forecasting. Understanding the complex interplay between these factors is critical for improving eruption warning systems and mitigating risks to life and infrastructure. This research focuses on the period from February 24th, 2021, marked by a Mw 5.64 earthquake, until the eruption onset on March 19th, 2021, examining how the observed changes in deformation and seismicity relate to the underlying geological processes.

The established method for eruption forecasting, based on escalating rates of precursory activity (Voight, 1988), has been widely used but doesn't account for all eruption scenarios. Several studies have documented a decline in seismicity before some eruptions, sometimes attributed to sealing of gas migration pathways (e.g., Eyjafjallajökull in 2010, Redoubt Volcano, Telica Volcano). However, Fagradalsfjall presents a unique case involving a decline in both seismicity and deformation, necessitating a more comprehensive understanding of the pre-eruptive processes. Previous research on the Reykjanes Peninsula has shown stress and strain accumulation along the plate boundary (Vadon & Sigmundsson, 1997; Hreinsdóttir et al., 2001; Keiding et al., 2008), providing a context for interpreting the observed precursory activity.

The study employed a multi-faceted approach combining seismic and geodetic data. Manually reviewed earthquake locations (Mw > 1) from February 24th to March 19th, 2021, were analyzed, with a focus on events Mw > 4.5. Global Navigation Satellite System (GNSS) geodesy and Interferometric Synthetic Aperture Radar (InSAR) data from Sentinel-1 were used to measure ground deformation. The GNSS data provided daily cumulative horizontal displacements at multiple stations, while InSAR revealed changes in line-of-sight (LOS) displacement over time. Waveform cross-correlation was used for seismic event relocation. The source of surface deformation was inferred through geodetic modeling using a modified version of GBIS software (Bagnardi & Hooper, 2018). Co-seismic deformation was modeled as shear movement across rectangular planes, and the dyke was modeled with opening and shearing on embedded planes within an elastic half-space. Daily volume change (magma inflow rate) and mean depth of magma emplacement were determined, constrained by combined GNSS and InSAR data. Finally, the model was used to estimate stress changes at depth.

The Mw 5.64 earthquake on February 24th, 2021, initiated a period of intense seismicity and deformation, consistent with magma intrusion into a ~9-km-long dyke (volume ≈ 34 × 10⁶ m³). However, the rates of both seismicity and deformation gradually declined in the days leading up to the March 19th eruption. GNSS data showed displacement rates decreasing from ~10 mm/day to near zero at eruption onset at the KRIV station southeast of the dyke, and from ~4 mm/day to near zero at the LISK station northwest of the dyke. InSAR interferograms similarly revealed a decline in deformation rate. Relocated seismicity indicated a two-segment dyke with seismicity migrating from the northern segment to the southern segment and westward before finally moving southwestward in the days before the eruption. Geodetic modeling indicated that the majority of the deformation resulted from dyke emplacement. Magma inflow rate (volume change) was highest in the initial days (30–35 m³/s), decreasing to <10 m³/s before the eruption. The mean depth of magma emplacement became progressively shallower after March 11th. Modeling revealed that the combined effect of the dyke and other sources (earthquake slip and plate boundary shear) released significant tectonic stress (tens of megapascals at 3.5 km depth), potentially a large fraction of stress accumulated over the 800 years since the last eruption on the peninsula.

The observed decline in deformation and seismicity before the Fagradalsfjall eruption contrasts with the escalating patterns typically anticipated. The findings suggest that in certain geological settings, particularly where tectonic stress is already high, stress release may precede an eruption. The decreasing magma inflow rate may be a consequence of reduced pressure driving the magma upwards as stress is released at the base of the brittle crust. The relatively calm eruption onset further supports this interpretation. The study's findings suggest a possible three-stage process: (1) dyke formation with associated seismicity and deformation, (2) decline in these rates due to stress release, and (3) eruption with minimal further precursory activity. This model bears similarities to observations at other volcanoes, such as Kilauea and Tolbachik, where passive dyke intrusions and rifting have occurred in pre-stressed crust.

The 2021 Fagradalsfjall eruption demonstrates that declining deformation and seismicity can precede eruptions, particularly in tectonic settings with pre-existing stress. This challenges the traditional view that escalation of these parameters is the sole indicator for imminent eruptions. This study highlights the importance of considering the interplay between magmatic and tectonic processes in eruption forecasting and calls for a broader range of monitoring and interpretation strategies. Future work should investigate this precursory pattern in other volcanic rift zones to determine its prevalence and predictive value.

The study focuses on a specific eruption and geological setting. The generalizability of these findings to other volcanoes needs further investigation. While the modeling provided valuable insights, model uncertainties might affect precise interpretations of the stress changes and magma flow rates. Additionally, the lack of readily detectable signals in the upper 1 km of the crust might limit the detection of precursors immediately before the eruption.