Magma storage depths and crustal-upper mantle structure of Campi Flegrei caldera (Southern Italy) unveiled through receiver functions analysis Authors: V. Ortega-Ramos, L. D’Auria, J. L. Granja-Bruña, I. Cabrera-Pérez, L. Pappalardo, G. Buono, N. M. Pérez Published: 2026 ABSTRACT Campi Flegrei is an active caldera in Southern Italy and one of the most hazardous volcanoes in the world. Here, we provide the first comprehensive 3D geophysical image of its plumbing system down to 50 km using the seismic receiver function technique. We imaged major crustal and upper mantle structures and located seismic discontinuities, identifying two significant low-velocity zones. A shallow LVZ, in the eastern part of Campi Flegrei caldera at 10 km of depth, has a vertical prolate ellipsoid shape (6 × 5 × 4 km), suggesting a localized crustal magma transfer or intrusion zone. A deeper northward-dipping LVZ is 17 km thick, extending from 16 to 33 km depth and containing up to 30 wt% melt, indicating a subcrustal primitive magma source. These findings agree with previous petrological studies and enhance our understanding of the Campi Flegrei magmatic system, its ongoing activity, unrest, and associated hazards.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-51786-3. Introduction Determining the depths at which magmas are stored is critical in understanding volcano dynamics and assessing associated hazards. This knowledge enhances the understanding of volcanic unrest signals from monitoring data, offering a more reliable framework for evaluating potential future episodes of activity. Identifying key signals indicative of unrest ultimately contributes to more accurate eruption forecasting1,2. The Campi Flegrei caldera (CFc) is a compelling case study due to its dynamic and complex volcanic system. Over the last 70 years, the caldera has exhibited significant variations in as gravity values, ground deformation, hydrothermal system, and seismic activity3–5,8,17,18. The subsurface processes driving these changes remain a topic of active debate and controversy, underscoring the challenges in unravelling the behaviour of such a multifaceted system3–5. The CFc is located in a densely populated area along the west coast of southern Italy, near Naples (Fig. 1). It is part of the Campanian volcanic district, which also includes the islands of Procida and Ischia, as well as the Somma-Vesuvius volcano. Its volcanic activity started ca. 250 ka and has been characterized by at least two large-scale caldera-forming eruptions: the Campanian Ignimbrite (Dense Rock Equivalent, DRE: ~300 km3, 40 ka)6 and the Neapolitan Yellow Tuff (DRE: ~30 km3, 14.9 ka)7. In the last 15 ka, volcanism has been concentrated in three stages of activity (~ 15-10.5.5, 9.6–9.1, 5.5–3.8 ka; total DRE: ~7.5 km3), producing at least 70 eruptions with a Volcanic Explosivity Index (VEI) reaching up to 58–10. During this period of activity, the central part of the caldera experienced a long-term uplift of ca. 100 m, partly recorded in the present-day elevation of La Starza marine terrace (ca. 40 m above sea level), that emerged ca. 5 ka11–15. The last volcanic activity occurred after ~ 3000 years of quiescence, with the Monte Nuovo eruption in 1538 (VEI = 2), which was preceded by intense and prolonged precursory phenomena14,16,17. The end of this eruption marked the beginning of a long rest period of significant subsidence, which lasted until 1950-52 when the first ground uplift of 0.5 m occurred18. This was followed by two main bradyseismic crises of 1970–1972 and 1982–1984, which caused ground elevation of 1.7 and 1.8 m respectively, and were accompanied by intense shallow seismicity18. A new ongoing unrest phase began in 2005 and continues today. It is characterized by ground uplift, hydrothermal activity and increased seismicity3–5. This unrest phase produced two of the largest recorded earthquakes in the area, with a magnitude of 4.6 on 13th March and 30th June 2025, approaching the potential maximum event that could occur in Campi Flegrei19. Based on petrological, geochemical and geophysical investigations, a comprehensive understanding of the first kilometres of the volcanic system has been developed18,20–35. However, despite numerous studies, the location of the main magma chamber(s) is still a matter of debate, and the reconstruction of the CFc magmatic system down to its roots in the crustal-upper mantle structure remains largely unexplored. Fig. 1a) CFc shaded relief map from Shuttle Radar Topography Mission (SRTM; https://portal.opentopography.org/), showing the locations of the seismic stations used in the study (green triangles), the locations of the vertical seismic cross-sections (black dashed lines) and the boreholes (red circles). Key geological structures of the caldera, such as crater rims (black solid lines) and caldera rims (red and yellow lines), are also indicated. b) Global map showing the locations of the earthquakes (Mw > 5.5) used in the analysis, with concentric circles representing the epicentral distances (30º and 90º) from the seismic stations. a) CFc shaded relief map from Shuttle Radar Topography Mission (SRTM; https://portal.opentopography.org/), showing the locations of the seismic stations used in the study (green triangles), the locations of the vertical seismic cross-sections (black dashed lines) and the boreholes (red circles). Key geological structures of the caldera, such as crater rims (black solid lines) and caldera rims (red and yellow lines), are also indicated. b) Global map showing the locations of the earthquakes (Mw > 5.5) used in the analysis, with concentric circles representing the epicentral distances (30º and 90º) from the seismic stations. Based on available petrological data on volcanic products from the Campi Flegrei region over the entire caldera history, it was suggested that a long-lived, multi-depth magmatic system exists22,26,27. This system consists of a shallow (1.5–2.0.5.0 kbar, corresponding to 6–8 km of depth) zoned, up to felsic (trachyte-phonolite) storage area, recharged by a deeper (> 2.0 kbar, corresponding to > 8–10 km of depth) mafic (trachybasalt-shoshonite) source. At shallower depths (< 4 km), ephemeral magma storage zones can form at the transition between caldera-filling tuffs and underlying carbonate bedrock, where magma resides briefly before either erupting or cooling and crystallizing in the case of aborted eruptions26,27. The petrological evidence is partially coherent with geophysical findings. In particular, Zollo et al.36 located a magma layer at a depth of 7.5 km, based on seismic velocity anomalies from multichannel seismic reflection profiles and seismic amplitude attributes (i.e., AVO, Amplitude vs. Offset analysis). Furthermore, Fedi et al.23 interpreted a wide gravity low in the Campanian Volcanic area as a vast source of partially molten, low-density material extending from approximately 8 to 30 km of depth. Additionally, Nunziata et al.37 identified a low S-wave velocity (Vs) layer with a 5% velocity reduction that begins at about 14–15 km of depth and extends to the Moho, beneath Roccamonfina in the north, Campi Flegrei, the Bay of Naples, and Mt. Vesuvius in the south. Isaia et al.38 identified a low-resistivity anomaly at 8–20 km of depth (total volume 915 km³) under CFc through magnetotelluric imaging, interpreted as a crustal mush zone (melt fraction of 10%). Although the presence and temporal stability of small magmatic bodies at shallower depths are not entirely understood3, based on seismic attenuation images derived from the 1982–1984 earthquake dataset, de Lorenzo et al.39, and De Siena et al.40 proposed the presence of small amounts of melt at a depth of approximately 3–4 km during this unrest episode. Furthermore, Patanè et al41. reveal notable temporal variations in seismic velocities with a significant increase in Vp at 3–4 km depth beneath Pozzuoli, accompanied by variations in Vs and a decrease in Vp/Vs. These observations are in agreement with previous geophysical models of D’Auria et al.42. Calò & Tramelli35,43 using P-wave seismic tomography with data from 1982 to 1984 and 2005–2023, identified an anomalous body at a depth of 3.5–4 km, consistent with molten rocks. More recently, Giacomuzzi et al.31 reported a low Vp and high Vp/Vs anomaly located at a depth of 4–5 km beneath the central part of the caldera, indicative of small volumes of melt during the 1982–1984 unrest and after 2019, in agreement with earlier P- and S-wave microearthquake tomography and rock physics constraints that also point to fluid and melt-rich regions at comparable depths44. Given the complexity of the proposed models, it is crucial to investigate the depth of the primary magma storage zones and obtain a comprehensive 3D image of the CFc magmatic system to understand the geophysical and geochemical variations observed during periods of volcanic unrest and to define potential future eruptive scenarios. The present study aims to investigate the geological structure of the continental crust and the upper mantle of the Campi Flegrei region, as well as the relative distribution of magma bodies in the current plumbing system using the receiver function (RF) technique. This technique is widely used and has been validated across different geodynamic settings to investigate subsurface crustal and upper-mantle structures45–51. The RF technique is particularly effective in resolving vertical velocity contrasts and detecting seismic discontinuities at greater depths below the seismic stations52, allowing us to investigate the crustal and upper mantle structures down, in this case up to 50 km of depth. By exploiting its sensitivity to variations in S-wave velocity (Vs) within the lithosphere, the method enables the construction of Vs profiles beneath individual seismic stations using teleseismic recordings53,55, which can then be interpolated to generate 3D models. We first present the results of the RF technique, which yield a 3D model of seismic velocities and discontinuities in the upper mantle and crust of Campi Flegrei, with particular emphasis on identifying low-velocity zones (LVZs) indicative of regions containing molten magma. Furthermore, we quantify the melt fraction using the 3D Vs model and realistic petrological compositions of the crustal and mantle rocks beneath CFc. We discuss our results in the framework of over 40 years of petrological and geochemical knowledge of the caldera to get an integrated model of the internal structure of CFc. These data serve as a basis for understanding the dynamics driving unrest episodes and for anticipating potential future eruption scenarios. Results By applying RF analysis to data from 13 seismic stations of the INGV monitoring network (Fig. 1), and to more than 900 teleseismic events recorded between January 2016 and November 2022, we derived, for the first time, a comprehensive 3D seismic image of the crustal and upper mantle structure beneath the 15 km wide CFc. Figure 2 presents the relocated seismicity in Campi Flegrei from January 2022 to March 202555, along with the Vs model, using four maps at different depths (horizontal slices) (Fig. 2a) and four vertical cross-sections (Fig. 2b) to highlight the main structures beneath the CFc, extending to 50 km of depth. In the model, we identify five distinct layers (A to E) based on the Vs values (Table S1): Layer A (Fig. 2b, vertical sections) is characterized by a relatively low Vs, ranging from 3.0 to 4.6 km/s. Beneath the seismic stations CPOZ (Pozzuoli city center) and CPIS (Pisciarelli hydrothermal site), located near the caldera center, we observe a particularly low Vs value of 2.5 km/s. On average, this layer is 3 km thick and hosts most of the earthquakes recorded in recent decades35,43(Fig. S1). Layer B (Fig. 2b, vertical sections) shows higher Vs values ranging from 4.6 to 5.5 km/s. This layer has an average thickness of 5 km and extends to a depth of around 8 km. Layer C (Fig. 2b, vertical sections) shows a Vs decrease from 4.6 to 4.0 km/s with a significant drop to about 2 km/s beneath the CNIS (Nisida crater, SE part of the caldera) station. Layer C starts at about 8 km in depth, extends to 16 km in the southern part of Campi Flegrei, and reaches to 20 km in the northern. The Layer D (Fig. 2b, vertical sections) appears from 16 km to an average of 33 km (Fig. 2) and is marked by a broad low Vs zone, ranging from 2.3 to 3.3 km/s. This layer extends across the whole model, smoothly deepening towards the north, while maintaining a consistent thickness of about 17 km. At the bottom of the model, we observe Layer E (Fig. 2b, vertical sections), characterised by an increase of Vs from 3.3 to 4.0 km/s. Fig. 23D Vs profiles of CFc. (a) Map view showing relocated seismic events across the CFc and horizontal Vs slices at depths of 0, −10, −15, and − 30 km. Shallower events (0 to 2.5 km) are represented by black dots and events occurring between 2.5 and 5 km depth are highlighted in red. (b) Vertical cross-sections along transects A-A′, B-B′, C-C′, and D-D′ illustrate the distribution of Vs. The region labelled R1 corresponds to a pronounced low-Vs anomaly within layer C, whereas M1 denotes a LVZ in layer D. Blue contours delineate regions where the inferred melt fraction is 30 wt%. The topographic base map is derived from Shuttle Radar Topography Mission (SRTM) data obtained from the OpenTopography (https://portal.opentopography.org/). 3D Vs profiles of CFc. (a) Map view showing relocated seismic events across the CFc and horizontal Vs slices at depths of 0, −10, −15, and − 30 km. Shallower events (0 to 2.5 km) are represented by black dots and events occurring between 2.5 and 5 km depth are highlighted in red. (b) Vertical cross-sections along transects A-A′, B-B′, C-C′, and D-D′ illustrate the distribution of Vs. The region labelled R1 corresponds to a pronounced low-Vs anomaly within layer C, whereas M1 denotes a LVZ in layer D. Blue contours delineate regions where the inferred melt fraction is 30 wt%. The topographic base map is derived from Shuttle Radar Topography Mission (SRTM) data obtained from the OpenTopography (https://portal.opentopography.org/). Layer C shows a pronounced decrease in Vs in the southeastern sector of the caldera, particularly beneath the CNIS station, where there is an anomaly with a Vs value of 3.6 km/s, indicating an average decrease of 1 km/s in Vs at 10 km of depth, contrasting with models obtained from other stations that show relatively higher Vs values. This anomaly is also visible in the cross-sections B-B’ and C-C’ and is marked as R1 in Figs. 2 and 3 and S2. We represent the relative Vs variation in Fig. 3 to enhance this anomaly. This model reveals an anomalous region measuring 6 km in height, 5 km in width and 4 km in length (R1 in Figs. 2 and 3 and S18), with a total volume of approximately 120 km3. At the same time in layer C, we observed a positive relative Vs variation (marked as A in Fig. 3f-g) influenced by the surroundings at this depth, with no significant changes in Vs (Fig. 3g). Fig. 3Horizontal and vertical cross-sections of the relative Vs variation extracted from the shear-wave velocity model for the CFc. V.H. on the bottom of the figures indicates the vertical exaggeration (vertical to horizontal scale ratio), set to 0.5. Panels (a-b) show horizontal slices of the relative Vs variation (%) with respect to the mean velocity model at depths of 0 and − 15 km, respectively. In comparison, panel (c) displays the 3-D Vs distribution at a depth of −15 km. Black triangles mark the locations of the seismic stations, and the black lines in panel (a) indicate the positions of the vertical cross-sections shown below. Panels (d) and (f) present vertical sections along transects B-B′ and C-C′, respectively, illustrating relative Vs variations down to 22 km depth. Warm colours (red to yellow) denote negative anomalies, whereas cold colours (blue to cyan) indicate positive anomalies. The dashed black contour indicates at least a −25% of relative Vs variation. Panels (e) and (g) show the corresponding Vs distributions along the same transects, highlighting the Vs distribution across the model. Warm colours (red to white) indicate low Vs, while cool colours (blue to purple) correspond to high Vs values. The region labelled R1 marks the lowest Vs zone beneath the Nisida area. M1 represents the lowest Vs values beneath CFc. Dashed grey contour (A) in panels (f) and (g) shows a positive relative Vs variation. The topographic base map is derived from Shuttle Radar Topography Mission (SRTM) data obtained from the OpenTopography (https://portal.opentopography.org/). Horizontal and vertical cross-sections of the relative Vs variation extracted from the shear-wave velocity model for the CFc. V.H. on the bottom of the figures indicates the vertical exaggeration (vertical to horizontal scale ratio), set to 0.5. Panels (a-b) show horizontal slices of the relative Vs variation (%) with respect to the mean velocity model at depths of 0 and − 15 km, respectively. In comparison, panel (c) displays the 3-D Vs distribution at a depth of −15 km. Black triangles mark the locations of the seismic stations, and the black lines in panel (a) indicate the positions of the vertical cross-sections shown below. Panels (d) and (f) present vertical sections along transects B-B′ and C-C′, respectively, illustrating relative Vs variations down to 22 km depth. Warm colours (red to yellow) denote negative anomalies, whereas cold colours (blue to cyan) indicate positive anomalies. The dashed black contour indicates at least a −25% of relative Vs variation. Panels (e) and (g) show the corresponding Vs distributions along the same transects, highlighting the Vs distribution across the model. Warm colours (red to white) indicate low Vs, while cool colours (blue to purple) correspond to high Vs values. The region labelled R1 marks the lowest Vs zone beneath the Nisida area. M1 represents the lowest Vs values beneath CFc. Dashed grey contour (A) in panels (f) and (g) shows a positive relative Vs variation. The topographic base map is derived from Shuttle Radar Topography Mission (SRTM) data obtained from the OpenTopography (https://portal.opentopography.org/). To investigate the possible content of partial melting, we quantified the liquid fraction across the investigated area using a thermodynamic equilibrium model for a given mineral assemblage composition56. This analysis reveals a heterogeneous percentage distribution of partial melt, particularly between 18 and 28 km of depth (Fig. S2). The liquid fraction, shown in the horizontal sections of Figure S2, indicates that significant partial melting begins at a depth of about 20 km, where the liquid fraction reaches up to 30 wt%. Cross-sections A-A’, B-B’, C-C’, and D-D’ (Fig. S2) reveal the spatial distribution of melting. The B-B’ profile shows a broad zone of partial melting between 20 and 30 km, while C-C’ and D-D’ show lateral variations in melt extension and the smoothly deepening of this layer toward the north. Discussion The 3D Vs model derived from RFs provides a full view of the crust and upper mantle beneath the CFc, extending to 50 km in depth, highlighting multiple layers characterised by distinct seismic velocities. In this section, we interpret and discuss the geological significance of each layer identified in our model based on available geological and petrological information (Figs. 2 and 3 and S2). Layer A, extending to a depth of 3 km, is associated with the most seismically active zone (Figs. 2 and 3 and Fig. S1). It shows low Vs values (2.5–4.6 km/s), indicating low consolidation and/or highly fractured materials likely of volcanic and sedimentary origin. At this depth, the stratigraphy beneath the caldera is well-documented by deep geothermal boreholes drilled by Agip-ENEL57. These boreholes are located near the northwestern margin of the caldera (in the Mofete and Licola areas; Mf1, Mf5, L1, respectively, in Fig. 1) and its central portion (San Vito area; SV1 and SV3 in Fig. 1). The stratigraphic sequence predominantly comprises volcaniclastic rocks, mainly tuffs, interbedded with minor lava flows and marine to fluvial sedimentary rocks of Pleistocene age58,59. These original lithologies have undergone significant alteration due to interactions with water and gases at progressively higher temperatures with increasing depth. This process has altered primary rock-forming minerals, leading to the formation of secondary hydrothermal minerals and decreasing the rock substrate’s porosity and permeability. Notably, at depths of 2–3 km, these rocks host an extensive geothermal system that spreads from the Solfatara-Pisciarelli area to the east to the Secca delle Fumose to the west, adjacent to the Monte Nuovo cone25,57,58. Layer B, between 3 and 8 km of depth, is characterised by the highest Vs (4.6–5.5 km/s) in the model (Figs. 2 and 3 and Fig. S1At this depth, the subsurface is known only through geophysical, geochemical and petrological data3,59. According to various studies, Mesozoic carbonate sequences are present between 4 and 8–11 km of depth, consistent with the composition of the upper crust exposed in the surrounding Campanian region33,58. However, the absence of carbonate lithics (rock fragments) in the volcanic successions of the CFc has led some researchers to propose the presence of intrusive syenitic rocks61,62. Further beneath, the Palaeozoic metamorphic basement begins and starts the first deep magma reservoir62. At this first deep magma reservoir between 7.5 and 8.5 km, Zollo et al36. identified a low-velocity zone, which they interpret as containing up to 80% melt. However, due to our limited vertical resolution model, we cannot resolve this zone clearly and only observe a decrease in Vs. The vertical resolution of the model is estimated to be on the order of 1 km (see Sect. 5. Methods), which supports the reliability of the main crustal features identified in this study. This level likely represents a zone where ascending magma accumulates, forming thin magmatic sills (ponding zone). This process may occur at the transition between carbonate and metamorphic rocks22,37, driven by differences in their rheological and mechanical properties. Layer C, up to a depth of 16–20 km, reveals an LVZ from 4.6 to 4.0 km/s, compared to the overlying layer B. This decrease in Vs may be influenced by increased fracturing, the presence of fluid layers, and small zones of partial melt. Within this layer, we highlight a relative Vs variation in the southeastern part of the caldera (R1) in Figs. 2 and 3. This layer is primarily interpreted as a magma transfer region linking the underlying vast magmatic source (Layer D) with the shallower crustal levels. Based on the relative Vs variation model, we interpret this layer as a magma transfer region linking the deeper magmatic source (Layer D-M1) with shallower crustal levels, where it may locally preserve remnants of past magma accumulation. Related to this, R1 anomaly suggests a focused pathway within this transfer zone, where magma and/or fluids may ascend. Layer D, averaging 16 to 33 km in depth, deepens further toward the north of the investigated area, reaching 36 km. In this layer, our model indicates the lowest Vs values between 2.3 and 3.3 km/s (Fig. 2) and a higher percentage of melt ranging from 15 to 30 wt% (Fig. S2), with an average thickness of 17 km (M1 in Figs. 2 and S2). This zone likely corresponds to the primary source of magma in the area. From seismic data, little is known about the lower crust beneath the caldera. Ferrucci et al.63 identified the Moho depth at approximately 25 km beneath the CFc and 35 km beneath the nearby Somma-Vesuvio. However, the structure of the lower crust has been inferred from gravity data, who presents a low-density anomaly centred beneath the Gulf of Naples with causative source(s) at depths of 8–30 km, attributed to variations in density distribution23. Our results are consistent with this interpretation, as in the model the Vs increase at this depth, indicating a sharp seismic structural discontinuity as the Moho. Finally, layer (E), from 33 km of depth up to the bottom of our model (50 km), presents values of Vs of 4.0–4.5 km/s, typically of the upper mantle (Fig. 2). The top of this layer marks the lower limit of the magma source volume of CFc, consistent with petrological data26,27. Previous geophysical models have typically inferred a single, vertically extensive low-density or low-velocity volume between about 8 and 20–24 km depth, with melt fractions ranging from 30 to 10% (e.g., Fedi et al.²³, Isaia et al.³⁸). By contrast, petrological constraints have long suggested discrete magma storage zones at 6–8 km and > 12–16 km depth22,26,27. Our RF-based model, which images a sill-like upper-crustal ponding level near 7–8 km and a deeper, partially molten mush zone extending from 16 to 33 km, is therefore more consistent with this latter picture of vertically separated storage levels than with a single, contiguous reservoir. In this contest, the anomaly R1 observed within layer C may represent a transitional zone linking these levels, acting as a pathway for magma and fluids between the deep source and shallower regions. Our multilayer image of the feeding system suggests that the ongoing unrest since 2005 arises from the interplay between a deep, volumetrically extensive source (M1) in layer D and a mid- to lower-crustal transfer/intrusion zone represented by R1 in layer C, with cascading effects within the shallow hydrothermal domain within layer A, where the seismicity is concentrated. In particular, the lower layer (M1), 17 km thick and spanning 16–33 km in depth, exhibits the lowest seismic velocities and the highest melt fractions (15–30 wt%). Thus, it constitutes the long-term magma supply. Upward transfer occurs through body R1 in the SE sector of the caldera, between 10 and 16–20 km of depth (Fig. 2); this volume, coincident with the most remarkable heterogeneity in melt content at 18–28 km, acts as a staging and routing region that channels flux from M1 toward shallower levels. The size of R1 (≈ 6 × 5 × 4 km; ~120 km³) implies that even modest increases in input from M1 can drive appreciable local pressure and changes in elastic properties, with measurable deformation and seismicity. As we based our interpretation on the geometrical and seismic characteristics of the velocity layers, rather than on explicit dynamic modelling, we suggest that magma transfer of magma, fluids or volatiles may occur more broadly throughout layer C, where R1 likely represents a more localized and efficient pathway where such transfer is focused. Near the top of this crustal pathway, the rheological contrast at the carbonate-metamorphic basement transition (7.5–8.5 km) provides an efficient sill-like ponding level, consistent with our observed reductions in Vs at those depths, where episodic injections and temporary storage can modulate bradyseismic uplift and upper-crustal seismicity. In the shallowest 2–3 km, the extensive hydrothermal system pervading caldera-fill tuffs provides an additional amplifier: increased input and degassing from below, transmitted through R1/sill levels and the hydrothermal column, can elevate fluid pressures and temperatures, fostering accelerations in deformation and clusters of seismicity up to the most energetic events recorded in the current sequence (M4.6 on 30 June 2025). In summary, we propose that the persistence of M1 and its pulsed contributions, funnelled through R1 and paced by sill-level storage at 10 km with occasional shallow pockets, drives the pressurization-depressurization cycle that characterizes recent unrest at Campi Flegrei, integrating seismological, petrological, and geochemical evidence into a single, coherent model. Conclusions In conclusion, our data allow us to provide a comprehensive 3D image of the CFc plumbing system down to 50 km, including main seismic discontinuities, Vs variations, and partial melting. For the first time, we identified a prominent Vs anomaly (R1) in the southeastern sector of the caldera, roughly beneath the CNIS (Nisida) station, with a vertical prolate ellipsoid shape (6 × 5 × 4 km), revealing structural heterogeneities in the lower crust, with an local remnants of previous magma accumulation between 10 and 16 km of depth. We believe this volume represents focused pathway in the magma transfer zone (Layer C) linking the deep magmatic roots with the shallow ponding region, where ascending magma could have triggered eruptions or cooling and crystallizing during aborted events. The most significant and extensive low-Vs anomaly is observed at depths between 16 and 33 km (Layer D; M1), where melt fractions range from 15 to 30%wt. This zone is therefore interpreted as the primary deep magma source feeding the system. These findings are consistent with petrological models that propose a shallow ponding level at 6–8 km (Layer B) which persisted throughout the caldera history26,65–69, recharged by a deeper mafic source (M1), from which melt is transferred upward through Layer C and locally through R1. Finally, integrating these findings with previous petrological and geophysical data provides, for the first time, a complete image of the CFc magmatic system. This model could help understand the signals of volcanic unrest in this high-risk, densely populated area. Furthermore, we note the robustness of the RF technique for investigating volcanic structures and identifying the presence of magmatic reservoirs67. Methods Data pre-processing For this study, we used the data recorded in thirteen broadband seismic stations belonging to the Istituto Nazionale di Geofisica e Vulcanologia (INGV) seismic network (FDSN code: INGV; Fig. 1) operated by INGV, Osservatorio Vesuviano. We selected teleseismic events with an epicentral distance between 30º and 90º from CFc and magnitudes higher than 5.5 (Mw), obtaining a catalogue of 913 events from January 2016 to November 2022 (Fig. 1). To prepare the seismograms for deconvolution, we cut them between 20 s before and 120 s after the theoretical P-wave arrival time. We applied a 0.2–0.5 Hz bandpass filter to maximize sensitivity to deeper crustal and upper mantle structures, which are the main target of this study. While this frequency band reduces vertical resolution in the shallow crust (0–8 km), it effectively suppresses noise and provides adequate resolution at depth, where our primary interest lies. A total of 8630 RFs were computed using the multiple-taper correlation technique of Park & Levin70(Fig. S4 to Fig. S16). To improve data quality, RFs were stacked using a median filter, thereby minimizing the impact of noisy traces. The final dataset consists of RFs from the thirteen broadband seismic stations, each processed individually to obtain 1D velocity models. Back-azimuthal stacking was employed to enhance signal coherence, and the resulting RFs served as the basis for subsequent transdimensional inversion. Rather than focusing on individual traces, we prioritized extracting consistent velocity structures across the network, a widely validated approach in both volcanic and tectonic settings50,53,71. For each station, RF datasets were selected based on signal-to-noise ratio and teleseismic coverage to ensure robust velocity estimates. Representative RF stacks for all the stations used are shown in the supplementary materials (Fig. S4-S16). Transdimensional inversion model and resolution check We derived a 1D Vs profile beneath each station from the transdimensional inversion of RFs, later combined into a 3D Vs model of CFc. The Vs and radial component were chosen for their sensitivity to vertical velocity contrasts, making them particularly effective for imaging crustal and upper mantle discontinuities in volcanic settings. S-waves were used, given their sensitivity to fluids and their ability to highlight potential magma accumulation zones64. To obtain the 1D Vs profiles beneath each station, we employed the Bayesian transdimensional approach of Bodin et al.72 to get a posterior probability distribution of seismic velocities (Fig. S4-S16), with layer thickness and velocity treated as free parameters. By including a maximum number of layers in the model space, the method adaptively balances resolution and stability while capturing significant discontinuities without overfitting. Uncertainties in both velocity and thickness are quantified from the ensemble of models sampled via Markov Chain Monte Carlo, inherently accounting for trade-offs between these parameters and supporting the robustness of the inversion. At each depth, we selected the statistical expectation of the Vs probability density as the representative velocity, while uncertainties were quantified from the 25th and 75th percentiles of the distribution. A diagonal covariance matrix was assumed for RF errors, with noise considered to be uncorrelated across time samples. However, this simplification may narrow uncertainty bounds, improve computational efficiency, and yield results consistent with geological and petrological constraints. Forward modelling of RFs follows the method of Levin & Park73 which calculates synthetic RFs for a stratified model with uniform velocity and density layers. Synthetic RFs were generated for each model sampled during the inversion and compared with observed RFs to evaluate misfit (Fig. S4-S16). While this approach does not explicitly account for anisotropy or lateral heterogeneity, it reliably reproduces the timing and amplitude of primary converted phases, allowing robust resolution of vertical velocity contrasts and discontinuities in the crust and upper mantle. The inverse model extends to 50 km depth, considering the deeper parts of the plumbing system beneath the caldera. We explored 10⁶ models to obtain Vs profiles for each station, which we subsequently integrated to construct a 3D Vs model. We assessed the vertical resolution achievable by our method through synthetic resolution tests. These tests, based on layered velocity models and realistic noise conditions, indicate that the transdimensional inversion can resolve vertical structures of approximately 1 km. Although the S-wave wavelengths in the 0.2–0.5 Hz frequency band suggest a theoretical resolution of 4–6 km, stacking RFs from multiple teleseismic events and the probabilistic nature of the inversion enhance effective resolution. We performed a synthetic test to assess the vertical resolution of our models. Using a known velocity model, we generated synthetic RFs and applied the transdimensional inversion, successfully retrieving a Vs model down to 50 km depth. Comparison of the retrieved and true models shows excellent agreement in both the velocity profiles and the corresponding synthetic RFs (Fig. S17), supporting a vertical resolution of approximately 1 km. Then we obtain the 3D Vs model by interpolating the Vs profiles. Determining melt fraction To quantify the melt percentage within our Vs model, we employed MAGEMin, a software designed to compute mineral phase equilibria under specified pressure, temperature, and chemical compositions (Riel et al.56). MAGEMin allows an understanding of how temperature variations influence mineral stability and melt production, providing critical insights into the relationship between seismic velocity and partial melting in heterogeneous crustal and mantle environments. The resulting liquid fraction estimates from this iterative process are fundamental to determining the melt distribution within the study region. In our methodology, we established compositions for the crust, the mantle, and the magma, along with a lithostatic pressure gradient. The remaining parameters, the temperature and the melt fraction, are determined through simple functional minimization. In practice, for each point in the model, we fix the composition and pressure and search, using thermodynamic modelling, for the temperature whose corresponding Vs matches the seismological model. To mitigate uncertainties in seismic velocity, mineral composition, pressure, and temperature within the melt fraction estimations, we used representative compositions derived from petrological data. We applied a lithostatic pressure gradient consistent with regional crustal structure. The temperature and melt fraction were obtained via functional minimization, thereby ensuring internal consistency between seismic and thermodynamic models. Although a complete sensitivity analysis is beyond the scope of this study, the melt fraction values reported are robust within the expected range of physical parameters. They are consistent with previous geophysical and petrological constraints. To establish the rock composition over the whole model, we used the published petrological data on the volcanic products from past eruptions representative of the entire volcanic history of the caldera (Fig. S3)26 to define: Magma composition. Throughout the history of the Campi Flegrei volcanic district, magma composition has primarily consisted of slightly silica-undersaturated rocks, ranging from trachybasalt to trachyphonolite26. Moreover, trachybasaltic-shoshonitic compositions are also represented in the Phlegraean volcanic district by the products of the Ischia and Procida islands. In the following discussion, these mafic compositions will be treated together, as common mantle sources and parental mafic magmas have been indicated for alkaline Campanian districts74. The mafic reservoir is characterized by a liquidus temperature of approximately 1200 °C, corresponding to a viscosity of 100−101 Pa·s under saturation conditions with water content between 5% and 9% by weight. This reservoir is located at a pressure ranging > 3–4 kbar, which corresponds to a depth from 12 to 16 km, extending down to the Moho discontinuity at approximately 25–30 km of depth. Felsic magma level exhibits a liquidus temperature of about 1100 °C. The viscosity of the felsic magma under these conditions is 102−103 Pa·s under saturation conditions with a water content ranging from 4.5% to 6% by weight. It is located at a depth between 6 and 8 km, corresponding to a pressure of about 1.5–2.0.5.0 kbar, at the transition between carbonate and metamorphic rocks.Mantle source. Spinel peridotite xenoliths found in the Monte Vulture volcano have been derived from the subcontinental lithospheric mantle beneath central-southern Italy75. Clinopyroxene-poor lherzolites and harzburgites are the most common rock types, with subordinate wehrlites and dunites.Crustal composition. The starting material analogous to the lower continental crust is a migmatite (CT54B sample) belonging to the Gennargentu Igneous Complex (Sardinia, Italy). The migmatite is a leucosome-depleted diatomite characterized by a fine-grained (200 μm), granoblastic and foliated texture. It hosts a mineral assemblage of quartz, K-feldspar, plagioclase, biotite, cordierite, ilmenite, hercynite and corundum76. The lower continental crust beneath Campi Flegrei is believed to be composed of rocks that have undergone multiple episodes of partial melting, leaving behind a restitic (residual) composition similar to that of the CT54B migmatite. The Gennargentu Complex, where the sample was obtained, is an ancient orogenic belt that, like the crust beneath Campi Flegrei, has undergone high-grade metamorphism and partial melting. This makes it a good analogue for the crustal rocks expected beneath the Campi Flegrei region74. Magma composition. Throughout the history of the Campi Flegrei volcanic district, magma composition has primarily consisted of slightly silica-undersaturated rocks, ranging from trachybasalt to trachyphonolite26. Moreover, trachybasaltic-shoshonitic compositions are also represented in the Phlegraean volcanic district by the products of the Ischia and Procida islands. In the following discussion, these mafic compositions will be treated together, as common mantle sources and parental mafic magmas have been indicated for alkaline Campanian districts74. The mafic reservoir is characterized by a liquidus temperature of approximately 1200 °C, corresponding to a viscosity of 100−101 Pa·s under saturation conditions with water content between 5% and 9% by weight. This reservoir is located at a pressure ranging > 3–4 kbar, which corresponds to a depth from 12 to 16 km, extending down to the Moho discontinuity at approximately 25–30 km of depth. Felsic magma level exhibits a liquidus temperature of about 1100 °C. The viscosity of the felsic magma under these conditions is 102−103 Pa·s under saturation conditions with a water content ranging from 4.5% to 6% by weight. It is located at a depth between 6 and 8 km, corresponding to a pressure of about 1.5–2.0.5.0 kbar, at the transition between carbonate and metamorphic rocks. Magma composition. Throughout the history of the Campi Flegrei volcanic district, magma composition has primarily consisted of slightly silica-undersaturated rocks, ranging from trachybasalt to trachyphonolite26. Moreover, trachybasaltic-shoshonitic compositions are also represented in the Phlegraean volcanic district by the products of the Ischia and Procida islands. In the following discussion, these mafic compositions will be treated together, as common mantle sources and parental mafic magmas have been indicated for alkaline Campanian districts74. The mafic reservoir is characterized by a liquidus temperature of approximately 1200 °C, corresponding to a viscosity of 100−101 Pa·s under saturation conditions with water content between 5% and 9% by weight. This reservoir is located at a pressure ranging > 3–4 kbar, which corresponds to a depth from 12 to 16 km, extending down to the Moho discontinuity at approximately 25–30 km of depth. Felsic magma level exhibits a liquidus temperature of about 1100 °C. The viscosity of the felsic magma under these conditions is 102−103 Pa·s under saturation conditions with a water content ranging from 4.5% to 6% by weight. It is located at a depth between 6 and 8 km, corresponding to a pressure of about 1.5–2.0.5.0 kbar, at the transition between carbonate and metamorphic rocks. Mantle source. Spinel peridotite xenoliths found in the Monte Vulture volcano have been derived from the subcontinental lithospheric mantle beneath central-southern Italy75. Clinopyroxene-poor lherzolites and harzburgites are the most common rock types, with subordinate wehrlites and dunites. Mantle source. Spinel peridotite xenoliths found in the Monte Vulture volcano have been derived from the subcontinental lithospheric mantle beneath central-southern Italy75. Clinopyroxene-poor lherzolites and harzburgites are the most common rock types, with subordinate wehrlites and dunites. Crustal composition. The starting material analogous to the lower continental crust is a migmatite (CT54B sample) belonging to the Gennargentu Igneous Complex (Sardinia, Italy). The migmatite is a leucosome-depleted diatomite characterized by a fine-grained (200 μm), granoblastic and foliated texture. It hosts a mineral assemblage of quartz, K-feldspar, plagioclase, biotite, cordierite, ilmenite, hercynite and corundum76. The lower continental crust beneath Campi Flegrei is believed to be composed of rocks that have undergone multiple episodes of partial melting, leaving behind a restitic (residual) composition similar to that of the CT54B migmatite. The Gennargentu Complex, where the sample was obtained, is an ancient orogenic belt that, like the crust beneath Campi Flegrei, has undergone high-grade metamorphism and partial melting. This makes it a good analogue for the crustal rocks expected beneath the Campi Flegrei region74. Crustal composition. The starting material analogous to the lower continental crust is a migmatite (CT54B sample) belonging to the Gennargentu Igneous Complex (Sardinia, Italy). The migmatite is a leucosome-depleted diatomite characterized by a fine-grained (200 μm), granoblastic and foliated texture. It hosts a mineral assemblage of quartz, K-feldspar, plagioclase, biotite, cordierite, ilmenite, hercynite and corundum76. The lower continental crust beneath Campi Flegrei is believed to be composed of rocks that have undergone multiple episodes of partial melting, leaving behind a restitic (residual) composition similar to that of the CT54B migmatite. The Gennargentu Complex, where the sample was obtained, is an ancient orogenic belt that, like the crust beneath Campi Flegrei, has undergone high-grade metamorphism and partial melting. This makes it a good analogue for the crustal rocks expected beneath the Campi Flegrei region74. Supplementary Information Below is the link to the electronic supplementary material.Supplementary material 1 (DOCX 25264.6 kb) Supplementary material 1 (DOCX 25264.6 kb)