Introduction

Between 5.97 and 5.33 million years ago (Ma), the Mediterranean Sea underwent a period of isolation from the global ocean - the Messinian salinity crisis (MSC) - allegedly leading to a km-scale sea-level drawdown and widespread salt deposition1,2,3,4,5. The ≈1 km3 volume of resulting evaporitic rocks6,7 sequestered ≈5% of the global ocean salts, which implies that during the MSC there was an intermittent influx of Atlantic waters supplying the necessary ions for salt deposition, in addition to the ions contributed by river runoff8. The return to fully stable marine conditions at the end of the MSC was initially seen as a gradual process taking place through a waterfall across the Strait of Gibraltar. Although a “geologically instantaneous” flood was foreseen9, the duration of the refill was estimated at as much as 10 ka10. The cataclysmic scenario for the end of the crisis gained support from seismic reflection data revealing an erosion channel stretching from the Gulf of Cadiz (Atlantic Ocean) to the Alboran Sea, which suggested that a single-stage flooding event - known as the Zanclean megaflood—refilled the Mediterranean11,12. This megaflood is thought to have been caused by overtopping Atlantic waters through a Late Messinian isthmus near the modern Gibraltar Strait, which initially filled the western Mediterranean, and subsequently spilled over the intrabasinal Sicily Sill, filling the eastern Mediterranean. Estimates suggest that the megaflood’s discharge and duration ranged from 68 to 100 Sverdrups (Sv = 1 million m3 s–1), and between 2 and 16 years, respectively; the peak discharge and water velocity likely occurred during the overspill across the Sicily Sill11,13,14. The dimensions of the inferred Zanclean megaflood would make it the largest known flooding event on Earth, one to two orders of magnitude larger than well-documented Pleistocene events induced by the northern hemisphere ice-sheet disintegration15,16, and one of the largest known in the solar system11.

The megaflood hypothesis is supported by a number of observations. In both outcrop and drill samples, a sharp lithological and palaeontological boundary marks the transition from shallow, non-marine sediments to overlying deep, open marine water deposits17,18, whereas sapropels point to a substantial transfer of salt from the western to the eastern Mediterranean19. Erosional and depositional features associated with the megaflood include a buried 390 km long channel and 35 km long megabars near the Strait of Gibraltar20,21. In the western Ionian Basin, a 1500 km³ chaotic sedimentary deposit lies buried to the east of Noto Canyon, an exposed theatre-shaped canyon carved into the Malta Escarpment22,23. Additionally, in southwest Sicily, a 5–7 m thick sandy contouritic deposit is interpreted to have formed by bottom currents during the reconnection of isolated water bodies24. Direct access to these locations has been limited to just a few drilling sites, and the megaflood hypothesis remains subject to debate25,26,27. Several of the cited studies encountered limitations, including lack of a robust age model, ambiguous geophysical data interpretation, and modelling that relies on low-resolution topographic data and poorly constrained palaeogeography. Alternative scenarios have also been proposed for the termination of the MSC. These scenarios include infilling by an overspill of Paratethyan (former Black Sea) water followed by Atlantic inflow once the Mediterranean Basin was refilled28, or by Atlantic seawater restoring normal open marine conditions in an evaporitic basin that was not subject to substantial sea-level drawdown4. Validation of the Zanclean megaflood hypothesis requires directly characterising the evidence left by the passage of flooding waters in terms of geomorphology, sedimentology, and stratigraphy. Here we report direct onshore evidence of megaflood erosional and depositional indicators, preserved as subaerial rock outcrops in SE Sicily, along with geophysical imaging of a buried erosional channel in the continental shelf offshore. Supported by numerical modelling of the flow, our shoreline-crossing field evidence confirms that a cataclysmic megaflood overtopped the Sicily Sill at the termination of the MSC. We are thereby able to constrain the evolution of the flow direction, depth, velocity, and shear stress of the Zanclean megaflood in the central Mediterranean region.

The intrabasinal Sicily Sill in SE Sicily (Fig. 1b) is part of the Hyblean foreland domain, the northernmost promontory of the African continental plate29. Positioned about 60 km SE from the front of the Sicilian Fold and Thrust Belt (Fig. 1a), the geology and geomorphology of the sill remain poorly characterised. The sill comprises a NE-SW trending, 13 km-wide depressed region that underwent tectonic extension and subsidence during the Miocene. Since the Pliocene, the sill has been uplifted together with the Hyblean Plateau29, and presently lies at an elevation of no more than 120 m above sea-level (Fig. 1b). The lithological framework of the sill consists of a thick, middle-upper Miocene, open shelf to shallow water sedimentary sequence, which is unconformably covered by pelagic chalk deposits of Zanclean age30 and breccia. Messinian evaporites crop out locally. The continental margin east of the Sicily Sill features a 12 km wide shelf. The shelf leads eastward to the 3 km high carbonate cliff of the Malta Escarpment, which hosts the 6 km wide submarine Noto Canyon31. The theatre-shaped head of the Noto Canyon had been suggested to reflect erosion of the Malta Escarpment by the Zanclean megaflood22.

Fig. 1: Location map.
Fig. 1: Location map.
Full size image

a Location map and digital elevation model of SE Sicily and adjacent seafloor (sources: SITR, EMODnet). The front of the Sicilian Fold and Thrust Belt, the Malta Escarpment, Noto Canyon, and the inferred buried megaflood deposit are labelled. b Digital elevation model of the Sicily Sill with outlines of the bases of the ridges. Sampled sites are denoted by red dots.

Results

Ridge morphometry

The central part of the sill is dominated by >300 ridges spread across an 8–10 km wide corridor (Fig. 1b). At their base they have a mean length (L) of 640 m (σ = 554 m) and a mean width (W) of 360 m (σ = 250 m). Their mean relief is 40 m (σ = 28 m). The overall circular mean direction of the long axes of ridge tops and bases is 40° (Fig. 2a). With respect to the mean direction, the majority of ridges are asymmetric in vertical sections along each long axis, with lee sides steeper than the stoss sides (Fig. 2d). In plan view, many ridge tops display aspect ratios (W/L) of 0.27 (Fig. 2c), which is commensurate with streamlining32,33,34,35. Although some ridge bases do have aspect ratios approaching 0.27, the majority are less-well streamlined and elongated (aspect ratio of 0.49; Supplementary Fig. 3). Ridges tend to become more streamlined for shorter lengths and lower heights (Supplementary Fig. 4). Shorter, lower and more streamlined ridges occur in the SE-half of the study area, whilst longer, higher and less streamlined ridges are located in the NW (Fig. 2b).

Fig. 2: Ridge morphology.
Fig. 2: Ridge morphology.
Full size image

a Rose plot of the circular orientation (in °) of the long axes of the ridge tops. Aspect ratio is denoted by the radial axis. Red arrow denotes mean orientation. b Map of bounding areas around ridges, with colour indicative of ridge relief. Lower W/L ratios imply a greater degree of streamlining. c Plot of aspect ratio of ridge tops. Aspect ratios of >0.27 for some of the ridge tops point to the occurrence of blockier, poorly streamlined or longer shapes. d Profile photograph (taken from the south) of a ridge. Location in figure b.

Ridge outcrops, age and deformation structures

A breccia unit is preserved along the top of most elongated ridges (Fig. 3a–c). The outcrops of the unit are up to 8 m thick and their lateral continuity never exceeds a few tens of metres. Breccia is primarily poorly sorted and clast-supported. Rarely, it is a well sorted, clast-supported breccia, or a poorly sorted, matrix-supported breccia (block-in-matrix). Clasts are angular to sub-rounded, pebble to boulder size. Clast composition varies from calcarenites and calcirudites, to marls and marly-calcarenites, all of which crop out in the nearby Hyblean Plateau. The matrix is largely sandy, with a minor silt fraction containing frequent mollusc shells (pectinids and oyster remains). Breccia displays loose (mostly) to moderate (rarely) calcareous cementation, which is occasionally hard. It is mostly structureless, with coarse sub-horizontal to gently dipping (up to 20–30°) parallel bedding and lamination with lateral terminations on erosional surfaces, which outline lenses of coarser clasts in the matrix. Rare gross coarsening-up and clasts imbrications are observed. The sedimentary fabric and metre- to decametre-scale structure of the breccia is highly variable across the study area, and we do not observe trends or preferential distribution of the fabric.

Fig. 3: Ridge outcrops.
Fig. 3: Ridge outcrops.
Full size image

a Poorly sorted and clast-supported breccia (Br) unit with polygenic sand matrix and boulder size sub-rounded clasts. b Clast-supported breccia with angular monogenic clasts. c Erosional surface defining the stratal break-up in the Tripoli Formation overlain by the breccia unit. d 3D photomosaic of a ridge with outcrops of the breccia unit, the overlying Trubi and underlying Tripoli Formations (Fm) (Source: Google Earth Image © 2024 Airbus). Locations in Fig. 1.

Below the breccia, three units were observed: (i) White-grey biogenic calcarenites rich in quartz, feldspars, molluscs and echinoids. Among foraminifera, benthic species reach values between 80% and 95%, with abundant individuals of Elphidium, Ammonia, Gyroidina, Asterigerinata, Bolivina. Planktonic foraminifera oscillate between 5% and 20%, with few specimens of Neogloboquadrina acostaensis, Globigerina bulloides and Dentoglobigerina spp. (Supplementary Fig. 5). Applying the regression formula of van der Zwaan et al.36, a palaeodepth for neritic environments (between 40 and 70 m) was inferred. This unit is upper Tortonian (9.4–7.2 Ma) in age and coeval with the uppermost part of the Tellaro Formation (cfr. Siracusa Member30). (ii) Yellow-brown calcarenites rich in fragments of molluscs (Pecten vigolensis, Anadara turonica, Parvamussium sp. and Ostrea sp.), echinoids, and bryozoans, and clasts of quartz and feldspar (Supplementary Fig. 5). Among the microfossils, Ammonia is the dominant genus, whilst ostracods associated with brackish environments are also present. This unit is upper Tortonian in age and interpreted as similar to the Carrubba Formation described in the eastern part of the Hyblean Plateau, which is a heteropic facies of the Siracusa Member37. (iii) White sandy marlstone, displaying cyclic alternation with grey marls, red laminates and whitish pseudo-diatomitic laminates, passing upwards to carbonatic laminated limestones. The marls are rich in planktonic foraminifera with values higher than 85–90% (Globigerina bulloides, Turborotalita quinqueloba, T. multiloba, Neogloboquadrina acostaensis, Orbulina universa, Globigerinoides quadrilobatus), whereas the benthic foraminiferal assemblages include stress tolerant species of dysoxic conditions as Bulimina echinata, B. aculeata, B. elongate and Bolivina dentellata (Supplementary Fig. 5). In the red laminates and in the whitish pseudo-diatomitic beds, planktonic microfossils are very abundant and benthic species are only sporadically presents. This is due to the anoxic conditions in the deep water column. These are characteristic of the uppermost part of the Messinian Tripoli Formation (7.05–5.97 Ma)38,39 (Fig. 3c, d). A unit comprising brecciated whitish marly limestones, rich in oogons of Chara sp., lies discordantly above the Tripoli Formation. These lithologies are truncated at the top by the breccia unit. The oogons belong to Charophyta that are typical algae of brackish environment and that have already been described in other Late Messinian Lago-Mare units3 (Fig. 3d). The unit overlying the breccia is only locally visible. It comprises white marly/silty limestones containing Sphaeroidinellopsis dehiscens, Neogloboquadrina acostaensis, Globigerinoides quadrilobatus and Globigerina bulloides, with Siphonina reticulata and Globorotalia margaritae in the upper section (Supplementary Fig. 5). This unit is interpreted as the Trubi Formation (Fig. 3d). These microfossil assemblages confirm the MPl 1and MPl 2 biozones, which correspond to the earliest part of the Zanclean40 (MPl1 biozone: 5.33 – 5.13 Ma; MPl 2 biozone: 5.1–5.0 Ma) (Supplementary Fig. 5). In this unit, planktonic foraminifera reach values higher than 60%, suggesting an upper part of the bathyal environment between 300 and 400 m depth. Our observations contrast with the previous interpretation of the breccia as Quaternary in age37.

Two types of soft sediment deformation structure are observed at metre- to decametre-scales in some of the outcrops. In the breccia and the underlying Messinian basement there are common soft-sediment deformation structures comprising metre-scale buckle folding, mostly asymmetric and occasionally overturned. The folding is disharmonic in strata with different competence, and deformation bands can be observed in yellow Miocene calcarenites (Fig. 4c, d). Exclusively in the breccia, sedimentary injections occur as sub-vertical or high angle fractures filled with a higher proportion of fine-grained matrix and clasts of the same composition as the surrounding unit (Fig. 4a, b). The fracture fill does not display flow structures. In addition, there are shear bands in the matrix surrounding larger boulders and broken intraclasts embedded in the breccia unit.

Fig. 4: Soft-sediment deformation structures and clastic injections induced by the deposition of the breccia (Br) unit.
Fig. 4: Soft-sediment deformation structures and clastic injections induced by the deposition of the breccia (Br) unit.
Full size image

a Vertical clastic (sandy) injection in the breccia unit. b Inclined clastic (gravel) injections in the breccia unit. (c) Soft-sediment deformation in the Tripoli Formation overlain by the breccia unit (with enlarged boundary between breccia unit and Tripoli Formation shown in d). Locations in Fig. 1.

Buried erosional shelf channel

Shore parallel and perpendicular seismic reflection profiles from the continental shelf east of the Sicily Sill and west of the upper sections of Noto Canyon reveal a prominent high-amplitude reflector, which represents an erosional unconformity over large sections (Fig. 5a–c). The age of the overlying seismic units is not constrained by physical age dating. However, the low-reflectivity seismic facies directly above the high-amplitude reflection resemble well-known facies of the lower Pliocene from the eastern and western Mediterranean41,42,43. Strata above the low-reflective unit comprise a lower aggrading and upper prograding sedimentary unit with channel-fill geometries (Fig. 5a, b). These geometries are observed below the inner shelf and the contourite drift geometries below the shelf edge. Following the analogous Plio-Pleistocene sequence described at the base of the Malta Escarpment, we interpret the unconformity between the lower aggrading and upper prograding units as a consequence of the mid-Pleistocene Transition44,45, when the 100 ka ice age cycles were established and global sea- and base-level dropped by more than 100 m during glacial maxima. Consequentially, the unconformity is named the mid-Pleistocene unconformity (MPU) (Fig. 5a, b). These lines of evidence suggest that the high-amplitude reflector dates back to the Zanclean at the very least, which leads us to interpret it as the Margin Erosional Surface (MES)1.

Fig. 5: Buried erosional shelf channels upslope of Noto Canyon.
Fig. 5: Buried erosional shelf channels upslope of Noto Canyon.
Full size image

a Part of seismic reflection profile HH24-03 illustrating a cross-section of the larger erosional shelf channel. b Part of seismic reflection profile HH24-29 showing a dip section along the larger erosional shelf channel. Both (a) and (b) display the location of the mid-Pleistocene unconformity (MPU). Inset (c) shows truncated reflectors in seismic reflection profile HH24-05. TWT = two-way travel time; VE = vertical exaggeration. Locations of profiles in Figure (d). d Interpolated sub-seafloor depth of the MES across the continental shelf east of the Sicily Sill and upslope of Noto Canyon, outlining the two erosional channels. White arrows denote inferred direction of water flow. Digital elevation model of SE Sicily and hillshade map of the adjacent seafloor are also shown.

The depth of the Margin Erosional Surface varies from a few metres below the seafloor close to the coast, to approximately 700 m below the flanks of Noto Canyon (Fig. 5d). Upslope of Noto Canyon, the Margin Erosional Surface features a 8–20 km wide and up to 270 m deep W-shaped channel, which gently dips from the coastline east of the ridges onshore, towards the lower slope into the head of Noto Canyon (Fig. 5a, b, d). A shorter 5 km wide and 100 m deep U-shaped channel is located 5 km to the north and dips into the northern tributary of Noto Canyon. The base of these channels is highly irregular and shows truncations of the underlying strata (Fig. 5c). Localised, sharp vertical offsets in both dip and strike directions point to uplift along faults at the centre of the larger channel (Fig. 5a), and accounts for the transformation from a presumed original U-shaped to the current W-shaped cross-section.

Hydrodynamic modelling

We perform 2D planform, vertically-averaged, hydrodynamic modelling of the Zanclean megaflood across the Sicily Sill for two scenarios—(i) prior to the formation of the shelf channel and the Noto Canyon on the Malta Escarpment (pre-Noto scenario), and (ii) after their formation (post-Noto scenario)—to estimate flow properties of the megaflood and their evolution. We set different levels for the overspilling western Mediterranean (zw). Because the sill is at an elevation of 25 m, there is no water flow for zw < 25 m. The flow of water over the sill increases from 0.76 Sv for zw = 50 m (pre-Noto scenario) to 20.8 Sv for zw = 100 m (post-Noto scenario) (Supplementary Fig. 6). The calculated Froude number is sub-critical in most of the model domain and supercritical in the upper shelf channel.

Figure 6 shows two representative stages of the Zanclean megaflood - before (zw = 70 m) and after (zw = 90 m) the formation of the shelf channel and Noto Canyon. Prior to the formation of these two features, the maximum velocity, discharge and shear stress in the study area are estimated at 20 m s–1, 4.1 Sv, and 2.3 kPa, respectively. After the formation of the canyon, these increase to 32 m s–1, 13.0 Sv and 6.5 kPa, and a clockwise change in flow direction is observed.

Fig. 6: Results of the 2D hydrodynamic model.
Fig. 6: Results of the 2D hydrodynamic model.
Full size image

Modulus (colour-bar scale) and vector field of water velocity, and free surface elevation (contours in m) (left panels), and bottom shear stress (colour-bar scale) and water depth (contours in m) (right panels) for the pre-Noto (a, b) and post-Noto (c, d) scenarios. The stated water level in the open western boundary (zw) is indicated in each panel. The resulting E-W water flow is 4.1 Sv for (a, b) and 13 Sv for (c, d) (1 Sv = 106 m3 s–1). The red rectangle corresponds to the area in Fig. 1b.

Discussion

Based on the field observations, we propose that the elongated ridges, the breccia unit and associated deformation structures, and the buried erosional shelf channel were formed by the passage of the Zanclean megaflood from the western to the eastern Mediterranean Basin across a shallow-water marine corridor in SE Sicily at the Messinian-Zanclean boundary (Fig. 7b). The ridges (Fig. 1b) are streamlined landforms compatible with large-scale, turbulent water flow with a predominant NE direction46,47,48, which inundated and eroded a planar surface of lower to upper Messinian formations (Fig. 7c, e). The ridges are primarily erosional remnants. The upper portions of the ridges responded more rapidly to both initial erosion and subsequent deposition along the ridge crestline and are thus more readily streamlined by the flow. In contrast, the greater bulk off the base of the ridge is affected by more two-dimensional flows of the same duration running along each flank, giving rise to erosion but with no opportunity for deposition. Such incision of a planar surface to form grooves separating ridges is common across a range of geophysical flow scales, resulting in spiralling erosive flow within the grooves and deposition on the ridge tops49,50 (Fig. 7c). Streamlined ridges have been documented in other proven Pleistocene catastrophic flood terrains51,52,53. The poor degree of streamlining of some ridges in our study area indicates a relatively short duration of inundation, inadequate to fully streamline the eroding ridges, with dissection of some ridges by cross-flows also resulting in poorly-streamlined landforms34,35. The breccia unit consists of material that has been eroded from the ridge flanks and the Hyblean foreland domain by the megaflood, transported as bedload and suspended load, and deposited rapidly along the ridge crests49,50 (Fig. 7c, e). We attribute the absence of MSC deposits between the Tripoli Formation and the breccia unit to erosion by the megaflood. The emplacement of the breccia unit gave rise to soft-sediment deformation of the unconsolidated Messinian formations. Sedimentary injections, typical of other megaflood deposits54,55,56, suggest a marked increase in pore water pressure, which reaches and exceeds the lithostatic load of the recently emplaced breccia. Localised under-consolidation of basal, unlithified formations by the load of rapidly deposited breccia favoured a decrease in effective stress, allowing plastic, soft-sediment deformation in basal layers (e.g57.; Supplementary Discussion, Supplementary Fig. 1). The alignment of the elongated ridges with the buried erosional shelf channels leading to the Noto Canyon (Fig. 5) confirms that the shelf channels were eroded by the flood and acted as conduits funnelling the flood water and its sediment load from the SE Sicily Sill to the Malta Escarpment, where the Noto Canyon was carved into the lower continental slope. The end-product of this process was the deposition of the ≈1500 km3 chaotic sedimentary deposit at the foot of the slope in the western Ionian Basin22. At the end of the megaflood, stable marine conditions favoured pelagic sedimentation during the Zanclean (Trubi Formation) (Fig. 7d).

Fig. 7: Idealised schematic conceptual model.
Fig. 7: Idealised schematic conceptual model.
Full size image

Model for the formation of the shelf channel, elongated ridges, breccia unit and associated deformation structures by the Zanclean megaflood across the Sicily Sill. a–d 2D cross-section of the Sicily Sill at different stages. Figure c and e represent the same stage. The flow of the water in figure (c) is away from the reader. The tilt in figure d is a result of the uplift of the Hyblean Plateau.

The field observations also provide insights into the characteristics and evolution of the megaflood. The asymmetry observed in our ridges is characteristic of relatively shallow inundation58,59, in contrast to the asymmetry of emergent or deeply submerged ridges60. By assuming that the highest ridges were submerged, we estimate a conservative value of 20–40 m for the depth of flood inundation above the ridge tops (Fig. 6). A number of observations suggest that the megaflood underwent a change in flow direction (from SW-NE to WSW-ENE) and an increase in velocity. First, the progressive dissection of a planar surface would be expected to initially produce flat-topped mesas, then elongated ridges, and ultimately streamlined, whaleback landforms32,61. The spatial variability in ridge morphology (Fig. 2b)—with shorter, lower and more streamlined ridges in the SE and longer, higher and less streamlined ridges in the NW—is not a result of geological variations37. Instead, it indicates that water flow in the SE of the study area, likely situated away from the flood margin, was deeper, stronger and/or more sustained in a WSW-ENE direction. The general change in ridge alignment (Fig. 2b), with the higher ridges oriented N-NNE in the NW and the smaller ridges exhibiting ENE-E orientation in the SE, suggests a clockwise change in flow direction from SW-NE to WSW-ENE as the flood progressed. Second, the top-down or lateral clastic injections observed in the breccia unit can be explained by a process that occurred after the bulk of the volume of the breccia had been deposited. We deduce that this process can be related to an increase in water flow velocity following the deposition of the breccia (Supplementary Discussion, Supplementary Fig. 2). The associated increase in shear stress would have simultaneously produced erosion and clastic injections, where the hydrodynamic component of the flow exceeded the lithostatic pressure in the recently deposited breccias. We infer that changes in megaflood direction and velocity were driven by the formation of the larger shelf channel and Noto Canyon on the Malta Escarpment further downstream, the orientations of which are aligned with the most eastern ridges in SE Sicily (Figs. 1, 2b, 5d). This inference is validated by our hydrodynamic model results (Fig. 6), which show that the excavation of the shelf channel and Noto Canyon, and the increase in the water level, had a noticeable impact in turning the flow direction clockwise and increasing the maximum velocity, discharge and shear stress during the megaflood. The estimated shear stresses far exceed reported critical values for erosion of soft rocks62 similar to the Tellaro and Tripoli-like formations. Boulders eroded from the deepening troughs between ridges and the shelf channel would have been suspended and comminuted63 to form a substantial sediment load, which would have lowered erosion thresholds even further.

The volume of rock eroded from the sill (the difference between the pre-Noto and post-Noto topography surfaces) is 2.6 km3. Even if combined with the volume of the larger shelf channel (20 km3) and Noto Canyon (45 km3), the number is too small to explain the megaflood deposit identified in the western Ionian Basin (1430–1620 km3[,22). This observation suggests that most of the deep basin megaflood deposit was eroded and transported from further upstream by the same flood rather than originating from erosion in the modelled area.

Alternative hypotheses to the megaflood hypothesis for the formation of the ridges, breccia unit, and deformation structures include tectonic deformation and fluvial subaerial erosion. Tectonic shortening is an unlikely cause, given the absence of reverse faults at the Sicily Sill, which is situated away from the collisional front (Fig. 1a)29. In fact, since the Neogene, this region has been subject to extensional strain, leading to large subsiding corridors in the hanging-wall blocks64. The hypothesis of post-MSC subaerial fluvial erosion also appears unlikely; the absence of a distinct fluvial channelled network, the ridge morphologies and the small, localised nature of the catchments argue against this explanation. We also reject the possibility of subaerial fluvial erosion as the cause for the larger shelf channel, in view of its considerable size relative to the largest potential catchment area onshore.

Our study presents limitations regarding the estimation of maximum discharge values from the hydrodynamic modelling. For a water depth (zw) of 100 m, we observe an additional flow of ≈4 Sv exiting through the southwestern boundary of the computational domain. Obtaining discharges of 100 Sv11,65 would require modelling a larger domain extended towards the south. We did not consider this because of uncertainties surrounding the palaeogeography of the central Mediterranean region at the end of the MSC, and the lack of field evidence indicating the presence of additional southern gateways.

This study provides the most compelling evidence yet for the largest megaflood postulated to have occurred on Earth, which led to the cataclysmic refilling of the Mediterranean at the termination of the MSC. We present the first large-scale evidence, in the form of erosional ridges topped by breccia deposits, for the passage of the Zanclean megaflood across the Sicily Sill during the overflow from the western to the eastern Mediterranean. This evidence is reinforced by geophysical evidence of a buried shelf channel that connected these onshore, upstream erosional and depositional indicators to the Noto Canyon and the buried chaotic sedimentary body in the western Ionian Basin. Numerical models reveal the crucial role played by the erosion of shelf channel and Noto Canyon in increasing the velocity and stabilising the direction of water flow as the megaflood overtopped the Sicily Sill. Our findings imply substantial (km-scale) lowering of the Mediterranean sea-level prior to the Zanclean megaflood11. Furthermore, they indicate that the geologic and geomorphic indicators of the megaflood have persisted in the subaerial and marine record for more than 5 Ma. This conclusion implies that it may be worthwhile to search for additional evidence of the megaflood along the margins of the Mediterranean.

Methods

Field data

Field observations, photographs and samples were collected during fieldwork carried out in SE Sicily in March and September 2023. These data were complemented by a digital elevation model (LiDAR, 2 m grid size, collected in 2008) and aerial photographs (15 cm grid size, collected in 2018) available from the Geoportal Regione Sicilia (https://www.sitr.regione.sicilia.it/).

Ridge mapping and geomorphometric analyses

The publicly available datasets were uploaded in a Geographic Information System. The bases of the ridges were delineated by extracting profile curvature and identifying continuous concave breaks of slope. A polygon was drawn along lines of high profile curvature where the adjacent slope gradient was >5°. The same was done for the top using a convex break of slope. Bounding boxes were generated for each ridge, from which the ridge length, width and the bearing of the long axis were measured. A trough depth map was extracted using hydrology tools66. The deepest values between ridges were extracted and interpolated to generate a smooth surface across the deepest sections of the troughs. The original digital elevation model was subtracted from this interpolated surface, and the highest value of relief was extracted for each ridge.

Geological sample analyses

To determine the stratigraphic ages of the lithological units present in the studied area, 73 samples were collected above and below the breccia unit (Fig. 1b). In the outcrops where the stratigraphic units were thicker than 2 m, a sampling interval of 50 cm was used. From each sample, 30 g of dried sediment were washed through a 63 μm sieve and dried to 40 °C. All samples were subject to micropaleontological analysis by identifying the foraminiferal assemblages. Thin sections were prepared for analysing microfacies and to determine the mineralogical composition to discriminate the lithologies. The foraminiferal identification was based on a qualitative analysis carried out on the >63 and >125 μm fractions. For this study, the planktonic foraminiferal biostratigraphic scheme proposed by ref. 40 was used. For benthic foraminifera, the classifications by refs. 67,68 were used.

Marine geophysical survey

The seismic profiles were acquired during the M199 (DELUGE) cruise69 on board the R/V Meteor in February-March 2024. The data set consists of over 1400 profile kilometres. The seismic source was a 6-tip Applied Acoustic Delta Sparker with a shot interval of 7 s. The streamer was a 500 m, 120 channel digital streamer (4.2 m channel spacing, near-offset of 110 m). Despite having a dominant frequency of around 150 Hz (dominant wavelength of ≈10 m at the seafloor), the maximum vertical resolution of the data is ≈20 m due to the double pulse generated by the source. The processing sequence included: (1) resampling from 0.5 ms to 1 ms, (2) crooked line common mid-point binning (10 m bin spacing), (3) semblance velocity analysis (2.5 km spacing), (4) spherical divergence correction, (5) shot domain threshold median noise attenuation, (6) normal move-out correction using 2–30° angle mute and fold-compensated stacking, (7) finite difference migration, (8) post-stack targeted F-K de-multiple sequence (to suppress the first and second water bottom residual multiple energy), and (9) F-X deconvolution, followed by a 35-45-300-400 Hz Ormsby bandpass filter (zero-phase) and a top mute above the water bottom. Depth conversion of the Margin Erosional Surface was performed by assuming a seismic velocity of 1.8 km s−1 in the overlying sediments and 1.5 km s-1 in the water column.

Numerical modelling

The topographic reconstructions used as pre-Noto and post-Noto scenarios in the numerical model have been made based on present topography/bathymetry grids with a cell size of 100 m. For the post-Noto scenario, the grids were modified to remove the overlying Plio-Pleistocene sediments (as interpreted in ref. 22), expose the larger shelf channel upslope of Noto Canyon, and infill the northern tributary of Noto Canyon (which is inferred to have been formed by slope failure during the Pleistocene23) by linearly interpolating across the canyon tributary rim. To restore the southwards tilting of the southern flank of the Hyblean Plateau related to flexural forebulge uplifting70, we have tilted the topographic model northwards by 0.3°, which causes the study area to be the lowest pass between the western and eastern Mediterranean within this model domain. For the pre-Noto scenario, starting from the post-Noto grid, we have fully infilled the shelf channel and Noto Canyon, and raised the topography of the ridges area by following a smooth envelope following the ridge tops, using linear interpolation in both cases. The latter was intended to reconstruct relief prior to the megaflood erosion. These reconstructions are somewhat speculative due to the lack of constraints on the 3D continuity of the buried part of the Noto Canyon or the amount of erosion that occurred above the present ridge topography. Therefore, it must be kept in mind that their purpose is primarily to test the different flow conditions expected before and after the formation of the shelf channel and Noto Canyon, and compare them with the ridge morphometry.

The numerical model is based on 2D depth-averaged barotropic hydrodynamic equations and includes a wet/dry algorithm that continuously updates the computational domain. The numerical code, developed at the University of Seville71, has been shown to be a robust computational tool to deal with cataclysmic floods (e.g., the Zanclean megaflood65; the late Pleistocene Lake Bonneville flood72). It has been adapted for the present study with the pre-Noto and post-Noto topographic grids, with a domain covering 14.7617°-15.5844°E in geographical longitude and 36.4345°-36.9516°N in latitude.

For the friction coefficient, \({k}_{f}\), we adopted the formulation of72,73 as a function of the water depth, \(d\), and the Manning coefficient \(n\), as \({k}_{f}=g{n}^{2}{d}^{-1/3}\), with \(g\) being the acceleration of the gravity. A uniform value of \(n\) = 0.04 has been selected for this study, similar to the other applications of 2D-hydrodynamic models for outburst floods: \(n\) = 0.05 (Lake Kuray–Chuja74), \(n\) = 0.04–0.05 (for valley floor and step margins, Missoula floods75), or \(n\) = 0.03–0.05 (Bonneville Lake72). The sensitivity of our model output to the selection of the Manning coefficient has been assessed.

The boundary conditions were set as follows: (i) no flow through the emerged terrain; (ii) a gravity wave radiation condition in the open north and east boundaries; (iii) uniform and constant water elevation in the open western boundary, \({z}_{W}\) (which forces a W-E flow); (iv) a null gradient condition in the open south boundary.

As initial conditions, we used uniform water levels in the Ionian Basin, \({Z}_{{IB}}\), and allowed the inundation of the domain west of the sill with the water level \({z}_{W}\). The numerical simulation was initialised with still waters and allowed the flow to find its path through the dry terrain to reach a quasi-steady state for the specified boundary condition (which provides a picture of the stage of the flood). It is worth noting that \({Z}_{{IB}}\) had no effect on the water flow when it remained lower than the Malta Escarpment. Thus, the ascribed values (500 m and 100 m below the grid datum) for the pre-Noto and post-Noto scenarios only provided qualitative information on the stage of the Zanclean megaflood.

A series of runs was performed for the pre-Noto scenario with \({z}_{W}\) values of 50, 60 and 70 m, while we used \({z}_{W}\) values of 70, 80, 90 and 100 m for the post-Noto scenario. This approach provides a gross picture of the temporal evolution of the flood, as shown in Supplementary Fig. 6. To enhance convergence and speed up the process, the fields of water elevation and water current from a steady state were used as initial conditions for a subsequent stage with a higher water level at the western boundary.

The model generates spatial maps with the elevation of the free surface of the water, water depths, current vectors, bottom shear stress, friction coefficient, and Froude number, as well as time series of the W-E water flow and of any of the above parameters at selected stations within the domain.

Using a Manning coefficient n = 0.05 instead of 0.04 results in a reduction of ≈15% in water velocity, shear stress and discharge without changing the main features of the flow reported in Supplementary Fig. 6. The erosive power of the flow progressively diminishes as the water level in the eastern Mediterranean exceeds the depth of the sill.