Seismic Images of Pressurized Sources and Fluid Migration Driving Uplift at the Campi Flegrei Caldera During 2020–2024

Abstract

After the subsidence phase that followed the 1982–1984 bradyseismic crisis, a gradual ground uplift at Campi Flegrei caldera resumed in 2005, while volcanic-tectonic earthquakes have steadily increased in frequency and intensity since 2018, with a significant intensification observed since 2023. This rise in seismic activity enabled a new tomographic study using data collected from 2020 to June 2024. In this work, 4161 local earthquakes (41,272 P-phases and 14,683 S-phases) were processed with the tomoDDPS code, considering 388,166 P and 107,281 S differential times to improve earthquake locations and velocity models. Compared to previous tomographic studies, the 3D velocity models provided higher-resolution images of the central caldera’s structure down to ~4 km depth. Additionally, separate inversions of the two 2020–2022 (moderate seismicity) and 2023–2024 (intense seismicity) datasets identified velocity variations ranging from 5% to 10% between these periods. These changes observed in 2023–2024 support the existence of two pressurized sources at different depths. The first, located at 3.0–4.0 km depth beneath Pozzuoli and offshore, may represent either a magma intrusion enriched in supercritical fluids or an accumulation of pressurized, high-density fluids—a finding that aligns with recent ground deformation studies and modeled source depths. Additionally, the upward migration of magmatic fluids interacting with the geothermal system generated a secondary, shallower pressurized source at approximately 2.0 km depth beneath the Solfatara-Pisciarelli area. Overall, these processes are responsible for the recent acceleration in uplift, increased seismicity and gases from the fumarolic field, and changes in crustal elastic properties through stress variations and fluid/gas migration.

1. Introduction

Campi Flegrei caldera (CFc), located near Naples (Italy), is a volcanic field that hosts a nested caldera complex on the Tyrrhenian margin of the southern Apennines (Figure 1). The structural framework of the CFc remains a topic of debate (e.g., [1,2]). Many researchers propose the existence of a nested caldera structure, primarily shaped by two major caldera-forming eruptions (Figure 1a) [3,4]). Since the Roman era, the CFc has experienced a gradual subsidence of about 1–2 cm per year, occasionally interrupted by periods of rapid ground uplift [5]. The most recent eruption occurred in 1538 AD, following a prolonged uplift episode of more than 17 m that began approximately 100 years earlier, leading to the formation of the Monte Nuovo phonolitic cone [6,7]. In the past century, the caldera has undergone several periods of instability, characterized by alternating phases of uplift and subsidence. The first three recent uplift events that disturbed caldera subsidence occurred in 1950–1952 (74 cm), 1970–1972 (159 cm), and 1982–1984 (178 cm) [8,9]. While the 1970–1972 unrest had minimal seismic activity, the 1982–1984 crisis included nearly 16,000 low-magnitude earthquakes, the largest reaching magnitude 4.0, causing structural damage and the evacuation of over 40,000 Pozzuoli residents [10]. Following a 20-year period of subsidence, a partial recovery of the 1982–1984 uplift began in 2005.

In 2005, ground uplift resumed, and in the area of maximum uplift, the Rione Terra (R in Figure 1a) is now 34 cm higher than its peak in 1984 and 138 cm higher than in 2005 (inset in Figure 1b). Seismic activity during this uplift phase has progressively intensified, especially since 2018. Between 2023 and 2024, over 35 earthquakes with duration magnitudes (Md) greater than 3.0 were recorded, including a Md 4.4 event on 20 May 2024. This phase of uplift is also associated with increased degassing in the Solfatara-Pisciarelli fumarolic area, which is the site of most degassing activity (Figure 1a).

History and present state of activity, therefore, make the volcanic hazard of the CFc high, with real potential for medium-high impact eruptions [11,12]. Given the high urbanization—approximately 500,000 inhabitants residing in the so-called red zone within the caldera—as well as its proximity to Naples, the region’s vulnerability to both seismic and volcanic hazards is further exacerbated. The volcanic area is officially monitored by the Osservatorio Vesuviano of the Istituto Nazionale di Geofisica e Vulcanologia (INGV). A combination of seismological, geodetic, volcanological, and geochemical monitoring is employed to assess volcanic activity and associated hazards [13].

The recent increase in the number and magnitude of earthquakes over time has raised public concern, prompting the issuance of a national government decree [14] that allocated resources for seismic risk mitigation measures, such as the retrofitting of buildings. Recently, Iervolino et al. [15] indicated that the reference moment magnitudes for the largest earthquakes in the caldera range from 4.4 to 5.1, with the upper bound considered a conservative estimate. Consequently, concerns regarding public safety in Campi Flegrei have markedly increased.

As part of the above initiatives, the PLINIVS Study Centre (Centre of Competence for Volcanic Risk, Italian Civil Protection Department) conducted a speedy vulnerability survey of 12,700 structures in the area of intervention, including 3700 non-residential buildings. Preliminary results indicate that among the 9000 residential structures surveyed, approximately 10% are classified as highly vulnerable—potentially at risk of collapse—while an additional 35% require further investigation (in https://mappe.protezionecivile.gov.it/it/mappa/mappa-ricognizione-areale-della-vulnerabilita-nella-zona-di-intervento-dei-campi-flegrei/ is reported the synthesis map for the vulnerability of 9078 buildings; accessed on 1 March 2024).

Figure 1. (a) Digital elevation model (DEM) of the Campi Flegrei caldera, illustrating the caldera rims. The outer caldera margin is represented by an orange dotted line, while the inner caldera margin is marked with a yellow dotted line. The morphological boundary of the resurgent dome is delineated by a red dotted line (refer to Figure 2 in [16] for more details). The map also shows the location of Mofete (MF1), San Vito (SV1, SV3), and Agnano (CF23) deep boreholes (yellow circles), seismic stations (red triangles), and earthquake epicenters (white circles), which have been re-located using a 1D velocity model. The inset in the upper left shows the regional setting of the Phlegraean Volcanic District, including its main tectonic and volcanic structures, modified from [17]. (b) Histogram displaying the number of daily earthquakes (black) from 2020 to 30 June 2024, with grey-shaded areas highlighting two distinct periods (I from 2020 to 2022 and II from 2023 to June 2024) during which separate tomographic inversions were conducted for comparison. The inset includes (1) a histogram of daily earthquakes since 1983 (black), with the count displayed on the left axis, (2) a cumulative plot of released energy (blue line), and (3) the vertical daily displacement at RITE (red line, Rione Terra, Pozzuoli). The values for (2) and (3) are provided on the right axis.

Figure 1. (a) Digital elevation model (DEM) of the Campi Flegrei caldera, illustrating the caldera rims. The outer caldera margin is represented by an orange dotted line, while the inner caldera margin is marked with a yellow dotted line. The morphological boundary of the resurgent dome is delineated by a red dotted line (refer to Figure 2 in [16] for more details). The map also shows the location of Mofete (MF1), San Vito (SV1, SV3), and Agnano (CF23) deep boreholes (yellow circles), seismic stations (red triangles), and earthquake epicenters (white circles), which have been re-located using a 1D velocity model. The inset in the upper left shows the regional setting of the Phlegraean Volcanic District, including its main tectonic and volcanic structures, modified from [17]. (b) Histogram displaying the number of daily earthquakes (black) from 2020 to 30 June 2024, with grey-shaded areas highlighting two distinct periods (I from 2020 to 2022 and II from 2023 to June 2024) during which separate tomographic inversions were conducted for comparison. The inset includes (1) a histogram of daily earthquakes since 1983 (black), with the count displayed on the left axis, (2) a cumulative plot of released energy (blue line), and (3) the vertical daily displacement at RITE (red line, Rione Terra, Pozzuoli). The values for (2) and (3) are provided on the right axis.

This is not surprising, as, in many urban areas of southern Italy, an estimated 60–80% of buildings do not comply with the seismic design standards introduced in the 1980s and refined in subsequent decades.

At present, the ever-changing conditions indicate that more in-depth investigations are essential for mitigating seismic risk, protecting the population from hazards linked to the rising frequency and energy of volcanic earthquakes, and determining whether seismicity and other parameters can serve as precursors to an eruption. At the same time, a deeper understanding of the caldera’s structural framework and subsurface is essential, as these factors significantly influence both eruption dynamics and seismic behavior.

Numerous geological and geophysical studies, including various tomographic investigations, have been conducted to improve the understanding of the CFc subsurface structure. Aster and Meyer [18] analyzed 228 seismic events recorded by a temporary network from the University of Wisconsin-Madison during the 1982–1984 bradyseismic crisis, deriving both P-wave (Vp) and S-wave (Vs) velocity models of the caldera.

Subsequent analysis mainly utilized datasets from the 1982–1984 seismic activity and from the 2001 SERAPIS active seismic experiment to refine velocity models and investigate the subsurface structure [19,20,21,22,23,24,25]. The most recent tomographic studies analyzed velocity changes, also incorporating recent data and highlighting time-dependent variations in the velocity structure [26,27]. The repeated application of three-dimensional (3D) seismic tomography, or 4D tomography (in space and time), can be considered a powerful tool to detect and track changes in the velocity structure under an active volcanic system.

This study presents new three-dimensional (3D) Vp, Vs, Vp/Vs, and Vp*Vs models for Campi Flegrei during the 2020–June 2024 period, which was marked by a progressive and significant increase in seismicity during the ongoing uplift phase that began in 2005.

First, a tomographic inversion was performed for the entire period of 2020–2024, and then the resulting seismic images were compared with those from Battaglia et al. [23], which used data from the 1982–1984 seismic crisis and the 2001 SERAPIS active experiment. Additionally, given the significant increase in both the number and intensity of seismic events since 2023, tomographic images from the 2023–2024 and 2020–2022 periods were compared to analyze velocity variations. This comparison aims to identify potential changes associated with the ascent of fluids and/or magma in recent years.

2. Materials and Methods

2.1. Tomographic Inversion for the Vp, Vs, and Vp/Vs Structures

It is well known how the accuracy and reliability of results obtained from tomographic inversion depend on several factors, including the inversion code used and the initial velocity model. The choice of starting velocity model is a critical aspect of the inversion process, as it significantly influences the accuracy and reliability of the final results. Selecting an appropriate initial model helps minimize potential biases and artifacts that could arise during the inversion [28], allowing for a more precise depiction of the subsurface structure.

In this study, various 3D starting models were utilized to thoroughly assess their impact on the inversion results. These included the 3D velocity models derived from the 1D velocity model from Vanorio et al. [21], with different Vp/Vs ratios ranging between 1.7 and 1.78, as well as the 3D velocity models obtained in Battaglia et al. [23]. This approach enabled exploring how different initial conditions could influence the final imaging of the subsurface structure, ensuring that the detected features are robust.

Moreover, the choice of the tomographic inversion code plays a pivotal role in accurately imaging the subsurface structure, as different algorithms have distinct approaches to solving the inversion problem and handling data uncertainties [29]. To enhance the reliability of results, two well-established inversion codes of the SimulPS [30] and tomoDD [31] families were used and compared. The SimulPS14 code [32,33] is particularly effective for generating initial velocity models, thanks to its robust handling of local earthquake datasets and ability to make preliminary adjustments to the velocity structure. Conversely, the tomoDD code excels at refining these models through its double-difference approach, which reduces residual errors and enhances the resolution of subsurface features. Double-difference (DD) algorithms minimize the travel time differences between pairs of earthquakes observed at common stations. This method focuses on relative locations of earthquakes, which improves resolution in clustered seismicity. Moreover, the double-difference approach reduces the sensitivity to errors in the initial velocity model, making it computationally efficient for resolving small-scale features. The tomoDD version used in this study, named tomoDDPS [34], has advanced features for adjusting S-P times to account for different propagation velocities of P and S waves through various materials. Then, tomoDDPS is able to provide a more robust Vp/Vs model than that from the original DD tomography code (tomoDD), obtained simply by dividing Vp by Vs [35].

The use of both codes enabled cross-validation of the results, evaluation of the influence of the selected starting models, and the attainment of a detailed and reliable representation of the Campi Flegrei caldera’s complex subsurface. This combined approach ensured that the final inversion outcomes were not overly dependent on a single method, thereby enhancing their robustness.

2.2. Dataset and Parameterization of Inversions

The majority of earthquakes in the Campi Flegrei caldera take place at depths less than 3 km and exhibit a nearly elliptical distribution at the large caldera scale (Figure 1), as shown in the reference catalog of the Istituto Nazionale di Geofisica e Vulcanologia—Osservatorio Vesuviano [36] (INGV-OV, https://doi.org/10.13127/gossip). The seismic activity mostly occurs inland, particularly under the Solfatara-Pisciarelli area and to the north of Pozzuoli, while it deepens offshore.

Typically, earthquakes have a duration magnitude of Md ≤ 1. However, since 2020, and particularly from early 2023 onward, there has been a significant rise in the average monthly magnitude. This increase includes over 35 earthquakes with Md ≥ 3.0, the largest being an Md 4.4 earthquake recorded on May 20, 2024.

In this study, only located events and the associated P- and S-wave arrival times from the INGV-OV earthquake database [36] were considered.

Typical lower magnitude events (Md < 2) present 6–10 P and 2–4 S arrival times. The arrival times were extracted from 28 stations of the INGV-OV network (triangles in Figure 1), located within 15 km from the catalog epicenters. Since the data were collected by different operators at various times, the uncertainties in phase picking are not always reported and therefore not well known. Generally, the precision to which arrival times could be measured is ~0.01 s for P phases and ~0.2 s for S phases. Therefore, two different inversions were performed, applying a uniform picking uncertainty across the entire dataset: one with a weight of 0 for P-phases and 1 for S-phases, and the other with a weight of 1 for P-phases and 2 for S-phases. From the initial dataset, earthquakes were selected that had at least 6 P and 3 S phase reads, azimuthal gaps smaller than 200, and RMS time residuals smaller than 0.2 s. This selection provided a refined dataset consisting of 4161 earthquakes from the original 4710 events of the INGV-OV database, with a final number of 41,272 P-wave readings and 14,683 S-wave readings.

As previously mentioned, the SimulPS14 code was initially applied using the 1D Vp model from Vanorio et al. [21] as the starting model (hereafter referred to as the Vanorio model). After a thorough evaluation of the 3D results considering Vp/Vs ratios ranging from 1.7 to 1.78, a constant Vp/Vs ratio of 1.7 was selected to minimize the residual. In the second step, the resulting Vp, Vs, and Vp/Vs models were used as starting points for further refinement with the tomoDDPS code (hereafter referred to as the Vanorio scheme). Furthermore, an inversion was also performed using tomoDDPS with the 3D velocity models from Battaglia et al. [23] (hereafter referred to as the Battaglia scheme), and the results from the two final velocity models were compared.

To resolve subsurface velocity structures, the tomographic inversions with the Vanorio model and Vanorio scheme employed a grid of nodes spaced 1 km horizontally and 0.5 km vertically, extending to a depth of 8 km. Conversely, the Battaglia scheme inversion used a finer grid spacing of 0.5 km in the longitude, latitude, and depth directions, providing a more detailed resolution. Considering Battaglia’s model [23], Figures S1 and S2 (Supplementary Materials) illustrate the distribution and density of seismic rays passing through different regions of the study area for Vp and Vs, respectively.

To avoid overfitting and ensure stable inversion results with SimulPS14, appropriate regularization techniques and smoothing constraints were applied to control the trade-off between data fit and model smoothness. The most suitable parameterization was selected after several trial inversions, based on the resulting model, the final data misfit, and the resolution matrix. To determine the optimal damping parameter, single-iteration inversions were performed while varying the damping until an appropriate balance between misfit reduction and model complexity was achieved. Ultimately, a damping value of 30 for Vp and 25 for the Vp/Vs ratio was chosen.

In the tomoDDPS inversion process, there are several important parameters that need to be determined: the weights for each phase, the weighting scheme for the absolute and differential catalog data, and damping. The phase weights used in this study were 1 for the P phase (0 in phase picking) structures and 0.5 (1 in phase picking) for the S phase structures. The weight of the P phase was set higher than that of the S phase because determining the S arrival time is generally more challenging and involves greater uncertainty compared to the P arrival time. For the weighting of the catalog and differential data, a greater weight is assigned to the absolute data at the beginning of the inversion to obtain the global velocity structure. Conversely, at the end of the inversion, a greater weight is assigned to the differential data to refine the velocity structure in more detail [31]. Initially, a weighting factor of 100 was applied to the absolute catalog data relative to the differential data during the inversion process. The optimal damping parameter was assessed based on the condition number (CND)—defined as the ratio of the largest to the smallest eigenvalue—which was maintained between 80 and 100.

Using the Battaglia scheme, the 2020–2024 dataset was inverted using 41,272 P-phases and 14,683 S-phases, along with 388,166 P and 107,281 S differential times. After 11 iterations, the P and S velocity structures were obtained, and earthquake locations were determined. For the unweighted catalog phases, the starting RMS value of 0.094 s decreased to 0.064 s, representing an approximate reduction of 32%. Conversely, for the weighted catalog, the starting RMS was 0.232 s, which was reduced to 0.043 s, resulting in an 82% decrease. Comparable results were achieved with the Vanorio scheme. For the unweighted catalog phases, the starting RMS of 0.082 s decreased to 0.064 s, yielding an approximate reduction of 22%. For the weighted catalog, the starting RMS value of 0.142 s was reduced to 0.043 s, representing a decrease of 70%. Considering the final velocity model obtained using the Battaglia scheme, hypocenters were localized with uncertainties of less than 0.3 km in both epicentral coordinates and depth, and the average root-mean-square (RMS) traveltime residual was less than 0.032 s.

3. Results

3.1. The 2020–2024 Inversion

For the entire dataset spanning 2020–2024, the results from the Vanorio and Battaglia schemes were initially compared. Although some details differed, the main features of the tomographic images were similar in the well-resolved volume. Given the similarities between the two velocity models, subsequent analysis of results focused specifically on the seismic images derived from the Battaglia scheme. This choice was further justified by the following two reasons: (1) the better accuracy of Battaglia’s models as starting models, particularly in terms of the Vp structure, which benefited from denser sampling due to the SERAPIS active experiment, and (2) the greater flexibility of tomoDDPS in selecting initial velocity models, including the option to start from complex 3D models derived from other studies, which can lead to more accurate final models [34]. Additionally, employing Battaglia’s models [23] enabled the integration in the external regions (not well-resolved) of the new velocity models, resulting in a more complete and accurate representation of the Campi Flegrei caldera’s subsurface structure.

The tomographic models for Vp, Vs, and Vp/Vs are presented at various horizontal depths in Figure 2a–c and along two cross-sections (WE and SW–NE) in Figure 3a–c. Instead, the panel in Figure 3d displays the Vp*Vs model mapped at a depth of 2.5 km, along with the SW–NE cross-section.

Figure 2. Maps showing the final solutions for Vp (a), Vs (b), and Vp/Vs (c) models. Relocated 3D earthquakes (black circles) occurring within ±0.25 km from each layer are depicted. S indicates the Solfatara area. The blue continuous lines delineate the well-resolved areas. Outside of these regions, the velocity models from Battaglia et al. [23] are included.

Figure 2. Maps showing the final solutions for Vp (a), Vs (b), and Vp/Vs (c) models. Relocated 3D earthquakes (black circles) occurring within ±0.25 km from each layer are depicted. S indicates the Solfatara area. The blue continuous lines delineate the well-resolved areas. Outside of these regions, the velocity models from Battaglia et al. [23] are included.

The reliability of these models has been validated by synthetic tests and resolution criteria (Figures S3–S7; see Supplementary Materials for further details). These indicated that the 3D Vp, Vs, and Vp/Vs structures of models are well resolved in the central part of the study area down to approximately 4.0 km depth. In Figure 2a–c, the boundary of the reliably resolved region is delineated by a thick blue line. In areas outside this boundary, where model resolution is poor, the displayed velocity models are those of Battaglia et al. [23]. To facilitate direct comparison with the tomographic images of Vp models presented by Vanorio et al. [21] and Battaglia et al. [23], the original color scales were maintained with only minor modifications.

The tomographic images shown in Figure 2a–c and Figure 3a–d, and the 3D representation of the Vp/Vs ratio in Figure 4 were obtained through 3D interpolation using Voxler 4 software (Golden Software) with the Inverse Distance to a Power method.

The 3D approach, compared to the 2D layer interpolation typically presented in some tomographic studies, generally provides a more accurate and detailed representation of subsurface structures. Indeed, by accounting for variations in all three spatial dimensions, 3D interpolation can better capture the complexity of the subsurface, including vertical changes and subtle structural features that may be missed or not visible with 2D interpolation methods.

Key features of the Vp model include the following:

• A wide relative low P-velocity anomaly (2.0 < Vp ≤ 4.0 km/s) in the central part of the caldera, mainly extending inland and down from approximately 0.5 to 3.0 km depth (L1 in Figure 2a). Between 2.5 and 3.0 km depth, this anomaly takes on an almost L-shaped aspect.
• A relative low P-velocity (4.0 < Vp ≤ 5.0 km/s) region located at a depth of 3.0 and 4.0 km (L2 in Figure 2a), showing an almost elliptical shape and encircling a small high Vp anomaly (5.5 < Vp ≤ 6.0 km/s) at 4.0 km depth. This elliptical anomaly appears to delimit the morphological boundary of the internal resurgent dome [16], within which deeper earthquakes are mainly concentrated.
• Some portions of the arc-like high P-velocity anomaly (3.0 < Vp ≤ 5.0 km/s), previously identified in earlier tomographic studies [19,20,21,23] (H1 in Figure 2a). This anomaly, attributed to the buried rim of the Campi Flegrei outer caldera, is primarily located along the southern and western borders of the Gulf of Pozzuoli; it extends from about 0.5 km to approximately 2.0 km in depth, with seismic velocities increasing with depth. It has been interpreted as consolidated lava and/or tuffs and interbedded lava sequences that signify the buried rim of the caldera [19].

Regarding the Vs structure, the most notable features of the model include the following:

• A wide relative high S-velocity anomaly (2.5 < Vs ≤ 3.0 km/s) located inland at about 2.0 km depth, between Pozzuoli and Campi Flegrei, coinciding with the region where a great part of earthquakes occurred (HS in Figure 2b).
• A relatively low S-velocity anomaly (2.0 < Vs ≤ 3.0 km/s), located at a depth of 3.0 to 4.0 km (marked as LS in Figure 2b and Figure 3c), is primarily situated offshore and south of Pozzuoli.

Analyzing the Vp/Vs structure reveals slightly more conservative absolute values of Vp/Vs compared to the previous studies. They range from 1.6 to 2.5, suggesting that direct inversion of Vp/Vs from S-P data is probably more stable than deriving Vp/Vs from separate Vp and Vs models.

Horizontal and vertical cross-sections of the Vp/Vs ratio (Figure 2c and Figure 3c) reveal:

• Two high Vp/Vs (≥1.9) volumes (H2 in Figure 2c and Figure 3c) are located in the shallower central part of the caldera, at depths ranging from approximately 0 to 1.5 km, surrounded by regions with normal to low Vp/Vs. The larger of these volumes lies beneath the inland area of Pozzuoli and partially extends offshore.
• An extended, thick high Vp/Vs layer (H3 in Figure 2c and Figure 3c), with values ranging from 1.9 to 2.5, located beneath a large portion of the caldera at depths of approximately 1.0 to 2.5 km. In the caldera’s center, this high Vp/Vs layer becomes discontinuous.
• A volume with a low Vp/Vs ratio, with values ranging from 1.6 to 1.66, that rises from depth in the central part of the caldera and reaches approximately 1.5 km b.s.l. (L3 in Figure 2c and Figure 3c). This structure widens as it extends downward, merging into an almost continuous layer at 3.5 km, which likely extends beyond 4.0 km depth.

The 3D representation of these low- and high-Vp/Vs volumes (Figure 4) in the central part of the caldera provides a clearer view of their shape and distribution at depth. Figure 4 displays the isovalues for both low Vp/Vs (≤1.66) and the shallower high Vp/Vs (≥2.1) structures. The low Vp/Vs structure appears as a complex, branching volume extending from approximately 1.5 to 4.0 km depth. The two high Vp/Vs volumes in the central part of the caldera, situated between 0.2 and 1.5 km depth, are positioned just above the branched low Vp/Vs region.

It is important to note that Figure 4 does not display the very shallow low Vp/Vs anomalies at depths ≤ 0.5 km, nor does it include the thick layer with high Vp/Vs values (between 1.0 and 2.5 km depth) lacking in the central part of the caldera and surrounding the two central high Vp/Vs volumes.

The branching low Vp/Vs volumes can be interpreted as regions where fluid and gas phases ascend toward the surface, with a portion contributing to the Solfatara-Pisciarelli shallow hydrothermal system (Figure 3c). These findings are consistent with the interpretations of Vanorio et al. [21] and Battaglia et al. [23], who associate the low Vp/Vs anomaly observed beneath the caldera center, between 2.0 and 4.0 km depth, and below 3.5 km in the surrounding areas, with overpressured water and/or gas-bearing rocks (thermo-metamorphic, Mesozoic carbonate and crystalline rocks; see Figure 3 in [37]) at supercritical conditions. In contrast, the high Vp/Vs thick layer between ca. 1.0 and 2.5 km, and the two shallower high Vp/Vs anomalies at around 0–1.5 km depth in the central part of the caldera, likely indicate the presence of brine-saturated rocks pre- and post-caldera volcano and marine deposits (Figure 3 in [37]), possibly resulting from steam condensation due to lower temperatures measured at shallow depths [38].

It is widely acknowledged that the Vp/Vs ratio is particularly sensitive to changes in rock fluid saturation, making it an excellent indicator of variations in pore fluid content.

Instead, the Vp*Vs product primarily highlights variations in lithology, porosity, and crack density while minimizing the influence of fluid saturation on seismic properties [21,39,40]. The Vp*Vs product has been used as an indicator of porosity in sedimentary rocks and a measure of rock stiffness and elastic properties [41]. Under constant conditions, Vp*Vs decreases with increasing porosity [42]. However, at sufficient depths where cracks close, porosity approaches zero, weakening this correlation. Additionally, as rocks heat and near their melting point, both Vp and Vs decrease simultaneously, significantly reducing their product. Even when Vp*Vs is high, indicating a relatively intact and stiff rock matrix, a decrease in Vp/Vs may suggest increased fracturing or variations in fluid and gas saturation that preferentially affect P-wave velocities. This suggests that while high Vp*Vs values point to a competent rock framework, localized reductions in Vp/Vs can signal damage or alterations in pore structure due to microcracking or the presence of fluid/gas. Then, to identify potential lithological variations at depth or detect zones with varying porosity and crack density, images of the Vp*Vs product (Figure 3d) were derived from the obtained P-wave and S-wave velocity models.

The dominant structure in the Vp*Vs model features a medium-to-high Vp*Vs in the upper layers that overlies a deeper zone with very low Vp*Vs values (Figure 3d). This deeper very low Vp*Vs region possibly indicates the presence of hot rocks at depths greater than 4.0 km and is potentially associated with the strong geothermal gradient or heating from deeper magma. However, it lies outside the well-resolved part of the models, remaining uncertain. Additionally, the eastern part of the model, which exhibits high Vp*Vs values, is also not well resolved. These limitations in resolution suggest that these zones of the Vp*Vs model must be interpreted with caution. However, the most significant feature in Figure 3d is a high Vp* Vs anomaly, delineating a volume that extends from the surface to approximately 3.0 km in depth beneath the Solfatara-Pisciarelli fumarolic area. This zone is marked by the most significant concentration of seismic activity. Moreover, a broad NNE–SSE elongated low Vp*Vs anomaly is also evident, extending from Pozzuoli to the central part of the caldera. This likely represents another notable structural feature, possibly indicating a zone of enhanced fracturing or altered porosity.

A more detailed discussion of all these observations is provided in Section 4.

3.2. Temporal Variation of Vp, Vs and Vp/Vs (2023–2024 vs. 2020–2022)

Four-dimensional tomographic inversion aims to detect changes in subsurface properties over time by comparing seismic data from different periods. In time-lapse tomography, 4D velocity changes are calculated by comparing differences between 3D images generated from data collected over different time intervals. Four-dimensional seismic tomography is routinely used in mature fields to monitor the effectiveness of enhanced oil recovery (EOR) techniques and in different tectonic contexts, enhancing the understanding of subsurface dynamics. For example, it can track the movement of injected fluids, such as CO₂ or water, and their impact on oil displacement. In volcanic environments, changes in both Vp and Vs and especially temporal variations in the P-to-S wave velocity ratio (Vp/Vs) have been documented, such as at Etna during its activity between 2002 and 2003 [43]. The authors interpreted the observed lower Vp/Vs ratios as evidence of the intrusion of volatile-rich basaltic magma feeding the eruption. Similarly, significant changes in seismic velocities were detected beneath Mammoth Mountain [44], where the variations were consistent with CO₂ migration into the upper 2.0 km, with corresponding depletion in surrounding areas, correlating with surface venting zones. Furthermore, in the Geysers geothermal field, between 2008 and 2014, the strengthening of a low Vp/Vs anomaly after 2010 was attributed to alterations in the water injection regime [45]. Furthermore, at the Nevado Del Ruiz, the temporal velocity changes from 2000 to 2016 indicated that the volcanic system underwent changes related to the ascent of new magma, which interacted with preexisting magma bodies, leading to the recent emplacement of a small dome at the bottom of the active crater [46]. Variations in seismic properties are typically linked to changes in stress and deformation, fluid and gas migration, and magma accumulation beneath volcanoes. However, the application of time-repeated tomography can be limited by differences in ray coverage between time periods, which may result from the not always optimal distribution of seismic sources (local earthquakes) and receivers. To address this, careful data selection and inversion procedures are necessary to ensure that the time resolution of the images is consistent across periods, thereby preventing spatial anomalies from being misinterpreted as temporal changes.

By consistently using the same starting velocity model across multiple inversions, the differences observed in the final velocity models are more likely to reflect true temporal variations in the subsurface rather than artifacts introduced by variations in the initial conditions. This approach significantly reduces but does not eliminate the bias that might arise if different starting models are used, as variations due to the choice of different initial models could overshadow or distort the actual changes occurring in the subsurface [28]. Thus, maintaining a consistent starting velocity model improves the stability and comparability of the inversion results, ensuring that the detected changes can likely be primarily due to physical changes in the subsurface.

Given the significant increase in seismicity observed since 2023 (Figure 1b), the 2020–2024 dataset was divided into two sub-periods and inverted separately. The resulting Battaglia scheme models for 2020–2022 and 2023–2024 were compared to identify potential velocity changes. To assess the resolution of the two seismic tomographic models, checkerboard tests were conducted for each period (Figures S8 and S9, Supplementary Materials), following the same approach as for the 2020–2024 dataset.

The images obtained from the two inversions were then compared, revealing that, despite some differences in detail, the main features of the tomographic images remain consistent. Figure 5a,b illustrates the temporal variations in velocity, derived by subtracting the velocity values of the 2023–2024 model from those of the 2020–2022 model.

The observed differences highlight the evolution of velocity structures during 2023–2024, revealing areas where seismic velocities have increased or decreased. Specifically, both Vp and Vs show a broad, ring-shaped decrease, generally ranging from −2.5% to −10%. This decrease is particularly prominent at depths between approximately 2.0 and 4.0 km, becoming more evident below 3.0 km, with some zones inside the ring exhibiting reductions also exceeding −10%.

For Vp, the region of decreased velocity surrounds, at depths of 3.0–4.0 km, a well-defined volume oriented roughly NW–SE, where seismic velocities increased by +2.5% to +10% (U1 in Figure 5). This increase is located between Pozzuoli and the adjacent offshore region, with the maximum increment (+10%) occurring between 3.5 and 4.0 km depth. In contrast, the zone with increased Vs (+2.5% to +10%) (U2 in Figure 5) lies mainly at a depth of 2.0–3.0 km, with the maximum increment (+10%) occurring at ca. 2.0 km, primarily concentrated near and beneath the Solfatara area.

Considering the Vp/Vs ratio, the most significant variation (ranging from −2.5% to −10%) is observed between ca. 1.0 and 3.0 km depth, with a maximum reduction of −10% occurring at approximately 2.0 km depth, almost corresponding with the Vs increase (D1 in Figure 5). This observation might align well with seismic changes expected during the depletion of liquid pore fluids and their replacement by steam, a process observed to be progressive and extreme in other geothermal systems such as The Geysers and Larderello-Travale (e.g., [40,47]). Instead, the observed deeper patterns of Vt and Vs increase at 3.0–4.0 km may suggest the presence of an inflating source at depth, due to a small intrusion of magma enriched in supercritical fluids. This intrusion likely influences the surrounding crustal structure through stress changes and fluid migration that reduce seismic velocities.

A more detailed discussion of all these observations is provided in Section 4.

4. Discussion

In this study, seismic imaging of the shallow volcanic structure beneath the central part of the Campi Flegrei caldera (CFc) for the period 2020–2024 yields more refined Vp, Vs, Vp/Vs, and Vp*Vs models compared to previous tomographic investigations, reaching depths of approximately 4.0 km. The following discussion integrates these findings with recent geochemical and geodetic studies and compares the observed velocity variations in 2023–2024 to assess whether the ongoing unrest—accelerating since 2023—is primarily driven by magma intrusion or hydrothermal processes, a key challenge in understanding the current dynamics of CFc. The volcanological literature on CFc presents contrasting interpretations, with several studies attributing the unrest to shallow magma intrusions and many others to hydrothermal perturbations associated with the injection of deep magmatic gases (e.g., [48] and references therein).

The Vp and Vs velocity models presented here enhance the understanding of the subsurface structures (Figure 2a,b and Figure 3a,b), though they do not provide clear evidence of low-velocity perturbations indicative of a distinct magma intrusion. The only exception is a relatively not well-confined low-Vs anomaly, located at a depth of 3.0–4.0 km (marked as LS in Figure 2b and Figure 3c), offshore and south of Pozzuoli.

It is well known that seismic velocities generally increase with depth due to the progressive compaction and densification of rocks under increasing overburden pressures. They are also influenced by multiple factors—including rock composition, porosity, fracturing, minerals, fluid saturation, temperature, and pressure—that can alter elastic properties and complicate the interpretation of seismic anomalies. Additionally, it is important to consider that ray-based seismic imaging methods may underestimate low-velocity anomalies and melt fractions, because seismic waves preferentially travel through high-velocity regions, potentially bypassing low-velocity zones [49]. This limitation is particularly significant when dealing with small magma intrusions (<1 km³) or when melt is distributed in thin, interconnected pockets rather than forming a large, coherent body [50].

Differently, the Vp/Vs ratio and the Vp*Vs product have provided additional constraints for the identification of the causes of some observed seismic velocity variations (Figure 2c, Figure 3c, and Figure 4). The analysis of the compressional-to-shear wave velocity ratio (Vp/Vs) has proven particularly useful for constraining the distribution of fluids in volcanic and geothermal environments. In saturated or partially saturated rocks, fluid content and phase significantly influence P-wave velocities while having little effect on S-wave velocities [21]. Fluid phase transitions alter fluid compressibility and, consequently, the bulk modulus [51,52], while shear moduli remain largely unchanged. As a result, low Vp/Vs ratios are generally associated with gas-bearing rocks, which exhibit high fluid compressibility, whereas high Vp/Vs ratios indicate liquid-bearing rocks [21].

One of the most significant features of the tomographic results is the 3D reconstruction of the low Vp/Vs volumes (<1.66) beneath the central caldera (Figure 4). This 3D body appears as a broad, interconnected network of gas-bearing rocks extending from depths of 3.5–4.0 km up to approximately 1.5 km below the fumarolic areas of Campi Flegrei (Solfatara-Pisciarelli). It can be inferred that this complex structure may represent the degassing pathway linking a deeper source to the surface fumarolic areas [53].

Another notable feature is the thick, extensive high Vp/Vs layer (values ranging from 1.9 to 2.5) located between 1.0 and 2.5 km in depth beneath much of the caldera (Figure 2c and Figure 3c). This layer is attributed to liquid-bearing rocks and brine-saturated zones resulting from steam condensation. In the central caldera, this high Vp/Vs layer becomes discontinuous, and at shallower depths (0–1.5 km), two distinct high Vp/Vs zones emerge. The larger of these is situated beneath the inland area of Pozzuoli and extends partially offshore [23].

Geothermal drilling at Mofete and San Vito (Figure 1; [17,38]) confirms that volcanic sediments infiltrated by meteoric water form a porous, fluid-saturated matrix. These wells reveal a multi-reservoir geothermal system extending from 550 to 2700 m in depth, characterized by high-salinity fluids, temperatures of 250–390 °C, and distinct fluid origins—seawater-derived at depths <2000 m and meteoric-magmatic below. These high temperatures of approximately 400 °C favor the decarbonation of deep limestone rocks, releasing additional CO₂ and contributing to the intense fumarolic activity observed at the surface in areas such as Solfatara and Pisciarelli.

Deep hot brines ascend through faults and fractures, mixing with shallow groundwater [54], while volcano-tectonic discontinuities facilitate the upward migration of fluids, including deep CO₂-rich mineralized fluids. Exceptionally high geothermal gradients measured in these wells [17,55] suggest a direct link between geothermal anomalies and hydrothermal activity. The interaction between cooler water and a deeper, hotter gas/steam reservoir may trigger rapid phase transitions (flashing), generating steam.

In the sketch model presented in Figure 6, the temperature distribution from the 3D model by Petrillo et al. (Figure 10 in [53]) is overlaid onto the W–E cross-section of the Vp/Vs structure. A strong correlation is observed between the temperature boundaries and both high and low Vp/Vs regions, particularly in the western and central parts of the caldera. Additionally, the sketch model reports the most significant features obtained from comparing P- and S-wave velocity and Vp/Vs ratio (Figure 6) between the 2023–2024 and 2020–2022 periods.

The observed velocity variations at 3.0–4.0 km depth beneath the Pozzuoli area and its offshore (a ca. NW–SE elongated volume) indicate a significant increase in Vp (+2.5% to +10%), along with a minor increase in Vs, and a slight decrease in the Vp/Vs ratio. Extensive interdisciplinary research supports that tectonic structures oriented NW–SE and WNW–ESE may significantly influence recent unrest at Campi Flegrei (e.g., [48] and references therein).

Two possible scenarios may explain these velocity variations:

(1)

A small magma intrusion enriched in supercritical fluids. In this case, the intrusion would stiffen the surrounding rock, leading to a significant increase in Vp, a modest increase in Vs, and a slight decrease in Vp/Vs. This behavior aligns with observations that high-pressure, high-density supercritical fluids reduce pore space and enhance rock stiffness [21,50].

(2)

The accumulation of pressurized fluids or high-density, mineral-saturated brines. These fluids similarly stiffen the rock matrix, increasing Vp, though their geochemical signatures differ from those associated with magma intrusions [41].

In both scenarios, the pressurized source influences the surrounding crust through stress changes and fluid migration, creating favorable conditions for rock fracturing and faulting. This is evidenced by the surrounding zones of Vp decrease (−2.5% to −10%), which can be linked to fluid-induced weakening and microfracturing in the adjacent rocks.

At shallower depths (2.0–3.0 km), a significant increase (+10%) in Vs is observed, encircled also in this case by broader zones of Vs reduction (−2.5% to −10%). A decrease in Vs is typically associated with microfracturing, as the formation of microcracks lowers the shear modulus and, consequently, S-wave velocity. Additionally, fluid saturation reduces effective stress and acts as a lubricant within the rock matrix, further decreasing the shear modulus and Vs [41,58,59]. These observations support the interpretation that regions of reduced Vp and Vs during the 2023–2024 period are more damaged and less rigid, likely due to increased fracturing and fluid infiltration.

In particular, beneath the Solfatara-Pisciarelli area, where an increase in S-wave velocity (Vs) is observed, a reduction of approximately 10% in the Vp/Vs ratio has been detected. This may be linked to a complex local structural setting, indicating the presence of relatively intact stiff rock in which steam has gradually replaced liquid water under high-pressure and high-temperature conditions, thereby enhancing surface emissions. The resulting steam condensates can form brines and migrate upward along fracture networks, potentially explaining the intensification of fumarolic activity observed at the surface. Furthermore, the increased seismic activity occurring since 2023 in this region may have further promoted fracturing and the migration of fluids and gases within the rocks. Consequently, the localized reduction in the Vp/Vs ratio observed during 2023–2024 (Figure 5a,b) likely occurred in a region where the rock matrix remains largely intact, as suggested by the high Vp*Vs product. These seismic changes are reminiscent of those documented in other geothermal systems. For instance, at The Geysers and Larderello-Travale [40,47], the gradual replacement of liquid pore fluids by steam is linked to pronounced seismic anomalies. Similarly, in the Coso geothermal area, a high Vp*Vs channel is observed in conjunction with a low Vp/Vs ratio, which has been interpreted as evidence of a rising or fossil magmatic body, likely representing the hot, low-porosity, and unfractured core of an intrusion [39].

Another dominant feature observed in the Vp*Vs structure is the elongated NNE–SSW region with small Vp*Vs values located beneath Pozzuoli and the central part of the caldera between about 1.5–4.0 km of depth, surrounded by medium Vp*Vs values. The presence of the youngest normal faults in the central sector of the caldera, which indicate a local NNE–SSW extension likely associated with resurgence phenomena (Figure 5 in [1,60]), could account for this observed low Vp*Vs anomaly. This anomaly, therefore, may reflect the intense fracturing and mechanical weakening in the affected rocks, reducing the Vp*Vs product relative to surrounding areas.

Overall, these findings highlight the complex interplay between fluid migration, phase transitions, and the mechanical and physical properties of rocks in the highly heterogeneous geothermal and volcanic settings, such as in the Campi Flegrei caldera, providing some critical insights into the subsurface processes that drive unrest in the area.

Recent studies focusing on the ongoing unrest highlight several aspects that warrant consideration.

Caliro et al. [61] refined some previous geochemical models [62,63,64] by integrating hydrothermal and magmatic contributions. They propose that the increased H₂S emissions observed from 2018 to 2022 may result from decompression-driven degassing of mafic magma at ~6 km depth, coupled with sulfur remobilization within the hydrothermal system. However, they caution that geochemical data alone cannot precisely constrain the pressure or depth of the magma source. Most recent geochemical studies consistently indicate that the hydrothermal system is responding to a deep pressurization source, with magmatic fluids migrating upward and reaching the hydrothermal system. This process induces substantial geochemical and geophysical modifications, including alterations in the gas compositions of fumarolic emissions and progressive changes in the mechanical properties of the crust as detected by seismic and geodetic measurements.

Ground deformation studies [9,56,57,65] consistently identify a primary magmatic pressure source between 3.0 and 4.0 km depth. Tizzani et al. [57] use multi-platform, multi-frequency InSAR data (Cosmo-SkyMed and Sentinel-1) from 2011–2022 to identify an inflating source at 3.0–4.0 km depth, interpreted as a pressurized magmatic intrusion. Giudicepietro et al. [56] analyze GNSS and Multi-Temporal DInSAR data from 2015 to 2023, highlighting a broad, bell-shaped deformation affecting the entire caldera (evident since 2021) interpreted as either a sill intrusion or a magmatic fluid accumulation zone inflating at ~3.8 km depth. Both Tizzani et al. [57] and Giudicepietro et al. [56] also identified a very shallow secondary source at 400–500 m of depth located beneath the Campi Flegrei that may be feeding fumaroles in the area. Astort et al. [65] analyze surface deformation at CF from 2007 to 2023, suggesting the action of both a shallow and a deep source. They identify a deformation source that progressively widens and migrates upward from 5.9 to 3.9 km depth, with inflation linked to the ascent of 0.06–0.22 km³ of magma from the deeper source (≥8 km), which undergoes minor deflation.

The tomographic results presented here align well with ground deformation studies and modeled source depths. Although the 2023–2024 temporal variations in Vp, Vs, and Vp/Vs indicate a deeper primary pressurized source at 3.0–4.0 km depth beneath Pozzuoli (including its offshore area) and a secondary pressurized source at approximately 2.0 km depth beneath the Solfatara-Pisciarelli fumarolic field—likely associated with the increase in seismicity—there is no clear evidence of magma intrusion. However, if a deeper magma intrusion enriched in supercritical fluids were present, the expected low Vp/Vs anomaly would be indistinguishable from that of the surrounding rock, which already exhibits values below 1.66.

5. Conclusions

Distinguishing between magma intrusions and hydrothermal effects remains a critical challenge for understanding unrest at Campi Flegrei. Past interpretations have alternately attributed unrest to shallow magma intrusions or to thermal-fluid disturbances in the shallow geothermal system. The analysis of seismic data from 2020–June 2024 produced updated 3D models of Vp, Vs, Vp/Vs, and Vp*Vs down to approximately 4.0 km depth, almost consistent with previous tomographic studies. The tomoDDPS code used in the study, based on the double-difference approach, reduces the sensitivity to errors in the initial velocity model, making it computationally efficient for resolving small-scale features, and has advanced to provide a more robust Vp/Vs model by adjusting S-P times to account for different propagation velocities of P and S waves through various materials. It allowed for the inversion of a dataset of 4161 seismic events recorded during the 2020–2024 period using both 41,272 P-phases and 14,683 S-phases, and 388,166 P and 107,281 S differential times.

Despite the obtained models not revealing clear low-velocity anomalies typical of magma intrusions—except for a localized, poorly confined low-Vs anomaly at 3.0–4.0 km depth offshore and south of Pozzuoli—the comparison of tomographic images from 2023–2024 and 2020–2022 shows significant velocity variations. In particular, a notable increase in Vp at 3.0–4.0 km depth beneath Pozzuoli and its offshore area was observed, accompanied by a substantial increase in Vs at shallower depths (2.0–3.0 km) and a decrease in the Vp/Vs ratio at around 2.0 km beneath the Solfatara-Pisciarelli area. Both the zones of increased Vp and Vs are surrounded by broad regions of velocity decrease.

One plausible interpretation is that the Vp increase at around 3.0–4.0 km depth reflects the effect of a deeper, roughly NW–SE-elongated pressurized source that stiffens the rock—either a magma intrusion enriched in supercritical fluids or an accumulation of pressurized, high-density fluids. This stiffening results in higher P-wave velocities relative to the surrounding, more porous, fluid-saturated rock.

These findings corroborate recent ground deformation analyses, which suggest the presence of a pressurized source between 3.0 and 4.0 km depth. In particular, Astort et al. [65] indicate that the inflation source at 3.9 km depth is linked to the ascent of 0.06–0.22 km³ of magma. This represents a very small volume (<1 km³) that seismic tomography might fail to detect due to its resolution limitations at present, being influenced by factors such as the spacing between seismic sources and receivers, as well as the adequacy of ray coverage. Moreover, if the magma intrusion is enriched in supercritical fluids, the expected low Vp/Vs anomaly may be indistinguishable from that of the volume located at approximately 1.5 km b.s.l. in the central part of CFc, which merges into an almost continuous layer at 3.5–4.0 km depth (Figure 4) and exhibits a low Vp/Vs ratio (1.6–1.66).

The upward migration of fluids from the deeper source may have caused the observed increase in Vs at around 2.0 km depth, indicating a distinct, pressurized process within the shallow hydrothermal system. This shallower zone—characterized by increased Vs and a decreased Vp/Vs ratio—may represent a volume of relatively intact, stiff rock where, under high-pressure and high-temperature conditions, steam has gradually replaced liquid water. A process that further influenced the active feeding system of the Solfatara-Pisciarelli fumarolic field through the increase in degassing. Meanwhile, the resulting overpressure in this shallow region and associated stress transfer have weakened the surrounding rock through microfracturing, thereby promoting the increase in seismicity in this zone.

These depth-related differences likely reflect the interplay between deep pressurization effects, shallow fluid migration, and phase transitions, which together produce distinct seismic signatures in Vp and Vs. Overall, the findings corroborate recent ground deformation studies and modeled source depths associated with the ongoing unrest, underscoring the complex interaction between deep magmatic processes and shallow hydrothermal dynamics.

Fully resolving the origin of this pressurized source—and understanding its impact on the geothermal system—requires a robust, integrated approach that combines geophysical, geodetic, geochemical, and petrological data into a unified model for a more precise interpretation of these subsurface processes. Furthermore, additional multidisciplinary monitoring efforts and targeted investigations, such as a new active seismic experiment, are believed necessary. Given the ongoing unrest, concerns are growing among the scientific community, local authorities, and residents, as a potential eruptive episode could require the evacuation of hundreds of thousands of people. However, even in the absence of an eruption, the increasing seismicity poses significant risks in this densely populated area.