The Caribbean oceanic crust is the complex product of the interaction of a mantle plume with the Proto-Caribbean (Farallon) plate 90 Ma ago, forming the second largest plateau after the Ontong Java (Kerr, 2014). Melt modelling of the high-MgO plateau lavas found in Curaçao suggest that the primary magmas contained up to 24 wt% MgO and that they correspond to a 30 %–32 % partial melting of a fertile, heterogeneous peridotite, with a potential temperature of 1564–1614 ∘C (Hastie and Kerr, 2010; Kerr, 2014).

Due to the fact that most of the plateau is currently submerged below large water depths, its structure and geochemistry has been mainly constrained from accreted material along the Caribbean margins, including Colombia, Ecuador, Costa Rica, Jamaica, and Hispaniola (Kerr, 2014; Kerr and Tarney, 2005; Van Der Lelij et al., 2010). In summary, the accreted fragments consist of basaltic and picritic lavas and sills, with gabbros and ultramafic cumulates (Kerr, 2014).

Diebold and Driscoll (1999) and Driscoll and Diebold (1999) presented the most detailed model of the formation of the C-LIP, which they based on seismic reflection profiles. According to their model, the Proto-Caribbean crust formed in the eastern Pacific by seafloor-spreading in the Late Jurassic–Early Cretaceous. This crust had a normal thickness (∼6 km) that later on during the pre-Senonian interacted with a mantle plume. As a result, a widespread eruption of the first basaltic flows, accompanied by extension and thinning of the Proto-Caribbean crust took place. Additionally, the intrusion of igneous material and underplating of residual mantle and ultramafic cumulates contributed to the formation of the thicker portions of the Caribbean plate, which in some places can reach up to 20 km. Following this, during the Senonian additional extension and underplating occurred, causing the uplift and rotation of the Beata Ridge, accompanied by the late stage of basaltic flows. Thus, at least two main pulses of magmatic activity have been identified: at ∼91–88 and ∼76 Ma (Diebold and Driscoll, 1999; Sinton et al., 1998).

Nowadays, it is possible to recognize anomalously thin, extended Proto-Caribbean crust, where crustal thickness ranges between 2.8 and 5 km, in the south-eastern Venezuelan Basin and in some regions of the Colombian Basin. These thin areas are characterized by a rough basement (B′′ horizon), suggesting that no basalt flows covered these domains. The Moho depth also shows a spatial variation: it shallows abruptly below the rough B′′ basement and deepens in areas where the B′′ horizon is smooth (Driscoll and Diebold, 1999). Figure 1 depicts the areas of thinned Proto-Caribbean crust as orange lines in the Colombian Basin (Bowland and Rosencrantz, 1988) (closed polygon) and in the Venezuelan Basin (Kroehler et al., 2011) (eastward of the orange line).

Although there is strong geochemical evidence that supports the origin of the C-LIP as melting of the paleo-Galápagos plume head (Geldmacher et al., 2003; Hastie and Kerr, 2010; Kerr and Tarney, 2005; Thompson et al., 2004), recent kinematic reconstructions back to 90 Ma by Boschman et al. (2014) showed an offset of up to 3000 km between the present-day Galápagos hotspot and the location of the Caribbean plate.

The northern margin of the South American plate is an active zone with two flat-slab subductions that interact at depth: the Nazca (Coiba microplate) plate from the west and the Caribbean plate from the north (some recent references include Bernal-Olaya et al., 2015; Chiarabba et al., 2015; Monsalve et al., 2019; Porritt et al., 2014; Siravo et al., 2019; Syracuse et al., 2016; Vargas and Mann, 2013; Wagner et al., 2017; Yarce et al., 2014). This complex interaction defines a poorly understood tectonic setting. Indeed, the geometry of these subductions is not well constrained yet, although different attempts have been made (e.g. Van Benthem et al., 2013; Bezada et al., 2010; Hayes et al., 2018; Mora et al., 2017).

Since the Late Cretaceous, the oblique collision of the Caribbean plate due to its east–north-east migration from the Pacific Ocean has shaped the South Caribbean margin. As a consequence, several foreland basins have been formed in the north of the South American continent, following the diachronous displacement of the collision front (Escalona and Mann, 2011 and references therein).

Along this margin, the Caribbean plate has subducted beneath the continental South American plate and the Maracaibo block since the Late Cretaceous (Kroehler et al., 2011), forming the South Caribbean Deformed Belt (SCDB; see Fig. 1). However, the lateral extension of this subduction has been actively debated. For instance, the tomographic model of Van Benthem et al. (2013) showed positive P-wave velocity anomalies from 76 to 67∘ W that were only traced in the upper mantle, suggesting that the subduction is ongoing along most of the margin, excluding the southwestern edges with the South American plate and the Panama microplate. According to Kroehler et al. (2011), in the southeast portion of the Caribbean plate an incipient subduction might be developing at present.

From the Pacific, the subduction of the Nazca plate shapes the western margin of the South American continent. In Colombia, north of the Caldas tear (∼5∘ N) the subduction becomes flat (Chiarabba et al., 2015; Vargas and Mann, 2013; Wagner et al., 2017) and is likely associated with the downgoing Coiba microplate (Fig. 1). This flat-slab subduction is characterized by a buoyant oceanic crust, which Chiarabba et al. (2015) associated with a volcanic ridge.

The lithospheric structural starting model includes the thicknesses of eight interfaces: (1) seawater (Fig. 3a), obtained as the difference between sea level and the General Bathymetric Chart of the Oceans (GEBCO) (Weatherall et al., 2015); (2) oceanic and (3) continental sediments (Fig. 3b and c) taken from the CRUST1.0 dataset (Laske et al., 2013); (4) oceanic and (5) continental crystalline crust (Fig. 3d and e) considering the Moho depth from the GEMMA model (Reguzzoni and Sampietro, 2015); (6) the slab shapes of the Nazca (Hayes et al., 2018) and (7) Caribbean (Mora et al., 2017) flat-slab subductions (Fig. 3f); and finally (8) the lithospheric mantle, which was subdivided into seven layers using the SL2013sv S-wave velocity model (Schaeffer and Lebedev, 2013) from 50 to 200 km depth. The integration of the different datasets was made after interpolating to a homogeneous spatial resolution of 25 km, using the minimum curvature algorithm with the software Petrel (© Schlumberger Ltd).

The sediment thickness from CRUST1.0 was selected not only because it includes the continental regions but also because it takes into account information from broadly distributed seismic profiles of EXXON (1985) that allow the recognition of the main sedimentary features without disregarding the spatial connection between both inland and offshore sedimentary systems. The sea level was used as the boundary zone between the oceanic and the continental depocentres.

Additionally, the continent–ocean transition (COT) defined by Gómez-García et al. (2019a, b) was used to separate both oceanic and continental crustal types. The black dashed lines in the oceanic crust map (Fig. 3d) depict regions where the crust has a thickness smaller than 10 km according to Mauffret and Leroy (1997). Most of these regions coincide with places where the calculated crustal thickness is extremely thin, with values lower than 4 km.

Finally, even though the 3D mantle densities published by Gómez-García et al. (2019a, b) were considered (see Sect. 3.2), whose approach takes into account the mineralogical composition of the mantle and the S-wave velocities of Schaeffer and Lebedev (2013), the gravity signal of the Nazca and Caribbean slabs was also tested. The Nazca subduction was taken from the Slab2 dataset (Hayes et al., 2018), which integrates diverse geophysical information into the definition of the slab's 3D structure, including active-source seismic data, receiver functions, seismicity catalogues from local and regional networks, and tomographic models. On the other hand, the Caribbean slab was defined following the hypocentral distribution (Mora et al., 2017). Figure 3f presents the integrated thickness of both flat slabs. In the region where both subducting plates interact at depth the integrated thickness reaches more than 120 km. While the Slab2 dataset includes the Nazca slab thickness, in the Caribbean case a thickness of ∼70 km was assumed a priori (see Fig. S2 in the Supplement).

The lithospheric mantle density is one of the most influential parameters in the modelling of the gravity anomalies because of its massive volume compared to the other lithospheric layers. The mantle density effect on the gravity signal was explored by testing four different mantle configurations (see Sect. S1). The model with the minimum RMSE compared with EIGEN-6C4 was selected as the initial lithospheric configuration. This model includes the 3D mantle densities obtained from the conversion of the S-wave velocity anomalies of the SL2013sv model (Schaeffer and Lebedev, 2013), from 50 to 200 km depth, published by Gómez-García et al. (2019a, b). In this approach, the density conversion is performed following a modified version of Goes et al. (2000) method, using a pressure- and temperature-dependent expansion coefficient (Hacker and Abers, 2004; Meeßen, 2017).

Figures S3 and S4 present the S-wave velocities at different depths (from 50 to 200 km, every 25 km) and their associated densities, respectively. In general, with this approach it is possible to establish a relation between high-velocity anomalies and high mantle densities (blue and dark blue regions in Fig. S3 and S4, respectively) and vice versa with low velocities and low mantle densities. Figure 4 summarizes the conversion results as described hereafter.

The mantle structure shows two different trends. At depths shallower than 75 km, in the Caribbean Sea the mantle is denser compared with the one below the South American continent, with differences up to 40 kgm-3 at 50 km depth. Although the mantle density increases with depth, it is possible to distinguish a subduction-like behaviour from 125 to 200 km below the surface: denser (colder) material below the South American continent, in contrast with a lighter (warmer) mantle below the ocean. These two trends are summarized in Fig. 4a and c for depths shallower than 75 km and in Fig. 4b and d for those deeper than 125 km depth.

Based on the spatial distribution of both parameters (velocity and density), it is not easy to differentiate which portions of the mantle belong to either the Caribbean or the Nazca slabs. Nevertheless, the fastest velocities (up to 4580 ms-1) are concentrated between 68 and 77∘ W, from 125 down to 200 km depth (Figs. 4b and S3).

Because at depths shallower than 125 km the observed patterns do not allow us to make a reliable interpretation about the slabs, it was necessary to explore the gravity effect of both subducting slabs using additional datasets. Figure 5a shows the spatial distribution of the depth to the top of the Nazca (Coiba) slab, as published by Hayes et al. (2018). Two main factors can be highlighted from this geometry: (1) even though it has been published as the Nazca subduction, it is very likely that it also includes an important fraction of the Caribbean slab because of its northward extension, and (2) it is possible to distinguish its flat shape, which widens towards the east, in the region north of the Caldas tear (∼5∘ N), as previously reported by other authors (e.g. Chiarabba et al., 2015; Vargas and Mann, 2013). Similarly, Fig. 5b depicts the depth to the top of the Caribbean slab by Mora et al. (2017), which according to these authors is present westward of ∼72∘ W. This geometry suggests that the Caribbean subduction is not entirely flat, but a sharp increase in its angle is present towards the south and south-east.

Table 1 summarizes the densities assumed in the initial 3D lithospheric model for the different layers considered. This model includes the gravity response of the 3D mantle densities in addition to the integration of the Nazca and Caribbean slabs geometries, as represented in Fig. 3f. When both slabs are integrated into the model, a density of 3163 kgm-3 is obtained with the IGMAS+ inversion algorithm (Sæther, 1997; Schmidt et al., 2011).

After the analysis of the mantle density effect on the modelled gravity was performed, the model which had the minimum RMSE compared with the observed gravity anomalies of the EIGEN-6C4 dataset (Förste et al., 2014) was selected as the initial lithospheric configuration (Sect. 3.2).

The relatively short wavelengths of the still-remaining gravity residuals (observed gravity minus modelled gravity) associated to this model account for rather shallow density heterogeneities. Indeed, the initial model considered the oceanic and continental crusts as homogeneous layers with constant densities (see Table 1). Therefore, in order to have a better representation of the crustal heterogeneities, the forward-modelling of the gravity residuals of the initial model was performed in three main steps. First, the code “Harvester” of the Fatiando a Terra geophysical Python tool (Uieda et al., 2013) was used for defining the boundary between the upper and lower continental and oceanic crusts. Harvester plants “density seeds” at the base of the crust and makes them grow vertically until the gravity residuals are minimized, defining the top of the lower crust. In our case, a density contrast of 300 kgm-3 was assumed between the upper and lower crust for this inversion.

Second, in the regions where the residuals (after the inversion with Harvester) were still considerably large (> 30 and <-30 mGal), bodies in the upper and lower crust within the oceanic and continental domains were defined (see Sect. 4.1). The delimitation of these bodies took into account the geologic evolution of the C-LIP and the tectonics of the Caribbean provinces in general (e.g. well-known volcanic arcs, previously reported underplated material, or extended proto-Caribbean crustal domains). Their densities were assigned depending on whether they matched a positive or negative residual, which suggest a mass deficit or excess in the initial model, respectively.

Finally, after implementing the previously mentioned two steps, large gravity residuals of medium-size wavelengths were still present in the oceanic domain of the Caribbean, which led us to define two high-density bodies in the uppermost oceanic lithospheric mantle (<50 km) of the Colombian and the Venezuelan basins. At shallow depths, trade-offs between the crust and the shallow mantle may occur, affecting the reliability of the tomographic data.

Figure 7a depicts the observed free-air gravity of EIGEN-6C4 at 10 km calculation height. This field has large positive values (>180 mGal) over the topographic highs of the Andes, as well as over the Santa Marta massif and the Panama microplate (Fig. 1). In contrast, negative gravity anomalies lie over the thick depocentres of the South Caribbean margin (Panama Deformed Belt, Magdalena Fan, and South Caribbean Deformed Belt; Fig. 3b), as well as over most of the continental basins (Lower Magdalena, Barinas-Apure, and Maracaibo; Fig. 3b, c).

As described in Sect. 3.3 the first two steps of the forward-modelling process consisted of including crustal heterogeneities that aim to fit the gravity field. Figure 7b shows the residuals associated to a model in which the gravity signal of the water, sediments, heterogeneous crystalline crust, and mantle (deeper than 50 km) were considered. Here, two positive anomalies remain in the Colombian and Venezuelan basins. Thus, the third step on the forward-modelling required the addition of two subcrustal bodies to compensate the gravity misfit in these domains. These bodies are located between the Moho and 50 km depth and have an average density of 3242 kgm-3.

As can be seen in Fig. 7c, after forward-modelling the gravity residuals including the subcrustal bodies, the modelled gravity response resembles the features previously described for the EIGEN-6C4 field. The new residuals have an RMSE of 17.45 mGal and have been minimized for the entire study area (Figs. 7d and S5f) compared with the residuals of the initial model (Figs. 6 and S5d). With the new lithospheric configuration, a decrease of approximately 39.5 % in the RMSE is reached.

This new lithospheric configuration aims to represent the complex tectonic setting of the Caribbean crust by including six different oceanic layers, with low- and high-density bodies (Fig. 8) as summarized in Table 2. The oceanic upper crust has been modelled with an average density of 3000 kgm-3 and its maximum thickness reaches up to 9.55 km (Fig. 8a). Moreover, low- and high-density bodies were defined within this layer. A low-density domain is present below the Aves Ridge (yellow polygon), with an average density of 2900 kgm-3, while a high-density body was found below the Venezuelan Basin (magenta polygon), whose average density is 3300 kgm-3.

The oceanic lower crust (Fig. 8b) has been modelled with an average density of 3100 kgm-3. Within this layer, widely distributed high-density bodies are required to fit the gravity field both for the Colombian and the Venezuelan basins (magenta polygons), with an average density of 3250 kgm-3. The density assigned to the low-density bodies located in the Aves Ridge and the South Nicaraguan Rise is 3000 kgm-3. The maximum thickness of this layer is 24.7 km, towards the west of the Aves Ridge.

As a reference, Fig. 8a includes the places where Mauffret and Leroy (1997) reported a thinner oceanic crust (<10 km, dashed black lines). Similarly, Fig. 8b depicts the areas where extended Proto-Caribbean crust has been reported in the Colombian and the Venezuelan basins (black lines, numbers 1 and 2). Note the relation between these extended domains and the thin (<1 km) or almost non-existent crust present in the structural model.

The continental crust has been split into four different sublayers. The continental upper crust (Fig. 8c) has the maximum thickness (up to 35.3 km) and an average density of 2750 kgm-3. Two low-density bodies were defined within this layer in order to improve the gravity fit. These areas include parts of the Lower Magdalena, Maracaibo, and Falcon basins (yellow polygons). Such bodies have an average density of 2600–2650 kgm-3. Additionally, one high-density body has been integrated into the structure of the upper continental crust (magenta polygon). It has been modelled with an average density of 3000 kgm-3 and is located below the Santa Marta massif, bounded by the Oca and the Santa Marta-Bucaramanga faults.

Finally, the lower continental crust has been defined as a complete unit, with an average density of 3070 kgm-3 and a maximum thickness of 31.8 km reached in the southern part of the eastern cordillera.

Assuming this new lithospheric configuration, the average crustal density highlights high- and low-density domains that correspond with thin and thick crystalline crust, respectively (Fig. 9). In general, it is possible to recognize that lighter crust (Fig. 9a) is associated with the continental domains, as well as with portions of the oceanic arc-like Nicaraguan Rise (Lewis et al., 2011), and the extinct arc of the Aves Ridge (Christeson et al., 2008). On the other hand, the highest densities are concentrated in the Colombian and the Venezuelan basins. Two regions where the calculated thickness of the oceanic crust is “virtually zero” are depicted as white areas in Fig. 9a and as magenta polygons in Fig. 9b and will be discussed in Sect. 5.2.

Within the continental domain, two low-density regions (<2700 kgm-3) are inferred based on the model results: the Lower Magdalena Valley Basin, whose basement has been described as mainly felsic (Mora-Bohórquez et al., 2017), as well as parts of the Falcon and Maracaibo foreland basins. These continental basins are characterized by a thin crust ranging between ∼10 and 20 km (Fig. 9b). In contrast, the crust of the Santa Marta massif has an average density of ∼3000 kgm-3, which is in agreement with a cratonic origin with igneous intrusions, as described by Montes et al. (2019a).

In order to fit the gravity, two subcrustal bodies were additionally required and integrated into the oceanic upper mantle (<50 km depth) (Fig. 10). According to the model results, these bodies might reach an average density of ∼ 3300 kgm-3 if they are placed from the Moho down to 25 km depth (see Fig. S6). Alternatively, an average density of 3242 kgm-3 is required if they extend down to 50 km depth. Due to the fact that a similar high-velocity trend – that would convert to comparable densities – can be followed in the tomographic data integrated into the lithospheric model, down to 75 km depth, the last scenario is selected hereafter as the final lithospheric configuration. These high-density bodies spatially correlate with the areas of thin crust according to Mauffret and Leroy (1997) (dashed lines in Fig. 10) but also include the regions where original Proto-Caribbean crust has been identified by Bowland and Rosencrantz (1988) in the Colombian Basin and by Kroehler et al. (2011) in the Venezuelan Basin (black lines in Fig. 10 – number 1 and 2, respectively). Thus, the high-density bodies are placed in areas where the already shallow crust–mantle boundary is insufficient to account for the excess mass needed to fit the observed gravity.

A summary of the final lithospheric structural and density model is presented in Table 2. As mentioned in Sect. 3.3, the gravity inversion was focused on improving the structural resolution for the oceanic and continental crust and the uppermost oceanic mantle; therefore, the remaining lithospheric layers were preserved as in the initial (reference) model.

In the Caribbean oceanic upper mantle, two high-velocity (and therefore high-density) domains can be followed, at least down to 75 km depth, according to the S-wave tomographic model of Schaeffer and Lebedev (2013) (Figs. 4, S3 and S4) that has been integrated in the initial (reference) density model of the Caribbean. After forward-modelling the gravity residuals to identify density heterogeneities in the crystalline crust, the residuals were still large enough to call for density variations on the uppermost oceanic mantle (<50 km depth), which was initially modelled with a constant density of 3200 kgm-3 (Table 1). Accordingly, the results suggest that two subcrustal bodies with positive density anomalies are present from the Moho down to 50 km depth. These bodies spatially correspond with a fast velocity mantle trend observed in the tomographic data (Fig. 11).

Assuming the thickness and spatial distribution shown in Fig. 10, an average density of ∼3242 kgm-3 is required to fit the gravity field. These bodies, approximately 440 km wide, show a strong spatial correlation with the thinner crust of Mauffret and Leroy (1997) (dashed black polygons) and also with the stretched original Proto-Caribbean crust reported by Bowland and Rosencrantz (1988) in the Colombian Basin (number 1) and by Kroehler et al. (2011) in the Venezuelan Basin (eastward of the black line, number 2).

Nevertheless, it is necessary to point out that the density of these subcrustal domains might not necessarily be exactly 3242 kgm-3. In fact, if their vertical extension is assumed from the Moho down to 25 km depth, the density required to fit the gravity increases to 3300 kgm-3, resulting in similar gravity residuals (see Fig. S6) to the ones shown in Fig. 7d. This illustrates the range of non-uniqueness inherent to gravity modelling but also clearly demonstrates that the need for an excess of mass in these domains is robust.

In both scenarios, the subcrustal bodies might have seismic velocities similar to that of a normal mantle, making their interpretation as mantle plume-related bodies difficult, especially in the context of early seismic experiments carried out in the past. For example, Mauffret and Leroy (1997) interpreted seismic velocities of 8.1 kms-1 as normal mantle velocities, based on isolated sonobuoy measurements both in the Colombian and the Venezuelan basins (see Fig. 22a in their publication).

The presence of shallow subcrustal bodies has been identified for different plateaus and regions where plume interaction with old crust is inferred (see McNutt and Caress, 2007). Some examples include a 4 km thick and 200 km wide subcrustal body found in the Hawaiian–Emperor seamount chain by Watts et al. (1985). According to their interpretation, this body has unusually high seismic velocities for a normal lower oceanic crust, ranging between 7.4 and 7.8 kms-1. More recently, Deng et al. (2014) reconstructed a preserved, lithospheric-scale magmatic intrusion associated with the evolution of the Emeishan Large Igneous Province down to 120 km depth.

In the Caribbean region, underplated material associated with picritic and ultramafic cumulates with seismic velocities of up to 8 kms-1 have been previously identified (e.g. Driscoll and Diebold, 1999; Mauffret and Leroy, 1997). However, the high-density (and thus high-velocity) material as represented in Fig. 11 was never mapped before with such detail and to points so deep beneath the Colombian and Venezuela basins. In the presence of a mantle plume, intrusion of high-density material on the overriding plate is expected. When the plume ceases, the new lithospheric system starts to cool down, and therefore high-density bodies can “freeze” (François et al., 2018) and become part of the mobile lithosphere that moves over the hot asthenospheric mantle, as expected by the plate tectonics theory. Therefore, we interpret these subcrustal bodies as preserved magmatic plume material responsible for the C-LIP formation.

Figure 11 shows a 3D perspective view of the preserved plume material assuming an S-wave velocity of 4540 ms-1 as the boundary of the high-density domains. The velocity contours at depths of 25, 50, and 75 km were taken as main constraints, and the form of the body in between was obtained by morphing of these contours following Brunet (2020) and Brunet and Sills (2015). A total volume of ∼ 12×106 km3 is calculated for what is interpreted as the C-LIP fossil plume conduits. Figure S7 shows the vertical relation between the subcrustal bodies (if a density of 3242 kgm-3 is assumed for them) and the high-density trend at each depth of the tomographic model. The presence of this high-density trend down to 75 km depth suggests that it travelled with the mobile Caribbean lithosphere. Accordingly, Blanco et al. (2017) reported the lithosphere–asthenosphere boundary in the Caribbean region around 75–80 km depth.

Recently, Civiero et al. (2019) used Rayleigh–Taylor models to simulate the 3D mantle convection for Newtonian and non-Newtonian rheologies. Their results predict the development of plumelets in different stages of evolution, creating a complex structure that is sometimes difficult to image using seismic waves. Thus, the observed positive density anomalies in the Caribbean lithospheric mantle might be related with the fossil plume conduits of these type of Rayleigh–Taylor-style instabilities, similar to the ones observed in the northern East African Rift by Civiero et al. (2019).

Previous studies based on geochemical analyses of C-LIP-related magmatic rocks have proposed the present-day Galápagos hotspot as the origin of the thermal anomaly responsible for the development of the C-LIP (e.g. Duncan and Hargraves, 1984; Geldmacher et al., 2003; Kerr and Tarney, 2005; Pindell and Barrett, 1990; Thompson et al., 2004). From a kinematic perspective, Nerlich et al. (2014) obtained a good fit between the paleo-position of the Caribbean plate and the present-day position of the Galápagos hotspot at the time of plateau formation (ca. 90 Ma). In their approach, the Caribbean plateau is assumed to have detached from the Farallon plate and to be fixed relative to South America at 54.5 Ma (“docking” age), while using the finite-rotation poles from Doubrovine et al. (2012) (global moving reference model) for the motion of the Pacific and African plates. However, shortly afterwards Boschman et al. (2014) published an updated kinematic reconstruction of the Caribbean region based on an extensive compilation of geological data from the entire region, which shows a very different motion path of the Caribbean plateau relative to South America (cyan line in Fig. 12) than the model of Nerlich et al. (2014) (green line). A clear offset of the plateau to the east of the modern Galápagos hotspot at the time of C-LIP formation (90 Ma) is observed in the model of Boschman et al. (2014), as previously reported by other authors (e.g. Pindell et al., 2006). More recently, Montes et al. (2019a) reconstructed the tectonic evolution of the northern Andes-Caribbean margin and showed similar results to Boschman et al. (2014) (magenta line in Fig. 12), although a considerable difference in motion and rotation of the C-LIP between 90 and 60 Ma is evident. This difference is due to the fact the Montes et al. (2019a) reconstructions are based on tectonic units in Panama and north-western Andes terrains, while the Caribbean plate moves, bends, and deforms internally to fit those inland proxies.

Most importantly, the absolute motion of the lithospheric plates relative to the Earth's deep interior is crucial to assess whether the C-LIP formed above the Galápagos hotspot 90 Ma ago or not. Several absolute reference models have been proposed over the last few decades, e.g. the moving Indian–Atlantic hotspots model (O'Neill et al., 2005), the fixed Pacific hotspots model (e.g. Wessel and Kroenke, 2008), the true polar wander-corrected paleo-magnetic model (Steinberger and Torsvik, 2008), the subduction reference frame linking surface plate motions to subducted slab remnants mapped from seismic tomography (Van Der Meer et al., 2010), and the global hotspot moving model (Doubrovine et al., 2012). However, the geodynamic consistency of those models has been recently questioned, as they yield unreasonable velocities of slab advance and retreat and of global net rotation of the lithospheric plates (Müller et al., 2016; Schellart et al., 2008; Williams et al., 2015). Tetley et al. (2019) developed a data-optimized global absolute reference frame back to the Triassic that includes global hotspot track observations and estimates of net lithospheric rotation and paleo-trench migration. This geodynamically consistent absolute plate model is implemented in the latest global plate motion model of Müller et al. (2019), which also includes the detailed regional reconstructions of Boschman et al. (2014) for the Caribbean region, and thus corresponds to our “selected model” in the discussion below.

In this section, we aim to re-evaluate the hypothesis that the C-LIP (and interpreted preserved plume conduits) formed above the Galápagos hotspot by testing six different plate motion configurations using the GPlates software (Müller et al., 2018). We used the recently published regional plate kinematic models of Boschman et al. (2014) and Montes et al. (2019a), for which the rotation files are available, and three different absolute reference frames, i.e. the “subduction reference model” of Van Der Meer et al. (2010), the “Global Moving Hotspot model” of Doubrovine et al. (2012), and the most recent “optimized global reference model” of Müller et al. (2019), as summarized in Table 3. The rotation files of all tested models are available as Supplement in Gómez-García et al. (2020) (see the data availability section below).

Figure 13a presents the results of the different kinematic reconstructions back to 90 Ma, using the polygons of the subcrustal bodies in the Colombian and the Venezuelan basins (number 1 and 2, respectively). Here, white circles represent the plausible diameter of the plume head: 2000 km (Campbell, 2005) and 2500 km (Mauffret et al., 2001). Additionally, Fig. 13b depicts the motion paths of the centre of the Caribbean plateau since 90 Ma in absolute reference frame. In both figures, the present location of the Galápagos hotspot is represented by a yellow diamond at 0∘ N, 91∘ W (Nolet et al., 2019). The main outcome of these reconstructions is the clear offset between the C-LIP and the location of the hotspot at time of plateau formation (90 Ma) for all models, as previously reported by Boschman et al. (2014). In our selected model (Müller et al., 2019), the C-LIP centre was located ca. 2500 km to the east of the present-day Galápagos hotspot 90 Ma ago.

To evaluate the reasons behind this misfit, three possible sources of error should be discussed.

The exact paleo-location of the Caribbean plate relative to South America is still not well known, especially prior to ∼67 Ma (age of development of arc magmatism in Panama; Montes et al., 2012), as shown by the differences between the regional reconstructions of Boschman et al. (2014) and Montes et al. (2019a). However, this difference between the regional reconstructions (ca. 750 km) is less significant than the offset between the C-LIP at 90 Ma and the present-day Galápagos hotspot, which ranges from 1264 to 3085 km, depending on the model used (ca. 2500 km for our selected model; Table 3).

To summarize, based on the argumentation above, the major offset between the paleo-position of the C-LIP at 90 Ma and the present-day Galápagos hotspot can be interpreted either as (1) the C-LIP having originated above the paleo-Galápagos hotspot only if this hotspot migrated significantly westward (>1000 km at about 1–2 cmyr-1 in a mantle reference frame) since the establishment of the western Caribbean subduction and had a large plume head during the C-LIP formation (∼2000 km diameter), or (2) the C-LIP having been formed by a different plume, which – if considered fixed – would nowadays be located below the South American continent (Fig. 13).

The complex geologic development of the Caribbean crust involved volcanic mounds, sills, basalt flows, extensional episodes, and underplating of plume material (Driscoll and Diebold, 1999; Mauffret and Leroy, 1997), with heterogeneous, high-MgO rocks (up to 28 wt% MgO, Kerr, 2014). The accreted portions of the C-LIP allowed the reconstruction of the geochemistry and temperature conditions that the mantle plume had at the time of the interaction with the Proto-Caribbean crust (Arndt et al., 1997; Hastie and Kerr, 2010; Kerr, 2005; Van Der Lelij et al., 2010; Neill et al., 2011). Indeed, the high-MgO rocks account for magmas that migrated relatively quickly from the mantle source (Kerr et al., 1998).

Geodynamic models support the evidence found in other large igneous plateaus related with a rapid, kilometre-scale uplift above the plume head. In these regions, the maximum uplift is also associated with lithospheric-scale extension in response to the buoyancy of the plume (e.g. Baes et al., 2016; Deng et al., 2014; François et al., 2018; He et al., 2003), and in most of the cases, the resulting crust is thicker than normal due to the magmatism above the plume head (Ernst, 2014). However, in the C-LIP history, the locations of the plume head and the associated uplift event(s) have been poorly constrained. Nevertheless, Diebold and Driscoll (1999) recognized that the extension and magmatism in the Caribbean were synchronous processes.

As seismic profiles of sufficient quality and spatial coverage are not freely available, 3D gravity models were used to assess the average crustal density configuration on a regional scale (Fig. 9). The average crustal densities highlight high-density areas affected by the plume interaction with the Proto-Caribbean crust, which significantly contributed to the thicker than normal crust that characterizes this plate (Diebold and Driscoll, 1999; Driscoll and Diebold, 1999; Edgar et al., 1971; Ewing et al., 1960; Mauffret and Leroy, 1997). This is in agreement with the early tectonic model of the Caribbean (Diebold and Driscoll, 1999; Driscoll and Diebold, 1999).

However, it is important to note the spatial correlation between the thin crust (<10 km) reported by Mauffret and Leroy (1997) and the extended Proto-Caribbean crust both in the Colombian (Bowland and Rosencrantz, 1988) and the Venezuelan basins (Kroehler et al., 2011) with the subcrustal high-density bodies (Fig. 10). In the conventional tectonic model of the Caribbean of Diebold and Driscoll (1999), the C-LIP plume has been considered to be located below the thicker than normal crustal domains (see Fig. 11 in Driscoll and Diebold, 1999), whereas the relationship between the extended domains of the Colombian and the Venezuelan basins with the C-LIP evolution has not been well understood.

Geodynamic models show that the topographic uplift and subsidence patterns associated with the interaction of a mantle plume with the oceanic lithosphere could be of several kilometres, depending on factors such as the plume size, mantle and lithosphere rheology, and age of the overriding lithosphere (Burov and Cloetingh, 2009; François et al., 2018). Similarly, several studies suggest geological evidence of a rapid (∼3 Ma) crustal doming prior to the eruption of basalts (e.g. Deng et al., 2014; He et al., 2003; Li et al., 2014; Zhang et al., 2016).

The previously described arguments led us to propose that the anomalously thin crust above the plume conduits in the Colombian and the Venezuelan basins corresponds to the centres of uplift. However, the lack of evidence that supports underplated material beneath these basins needs to be explained, and we hypothesize the following two options. (1) The magmatism occurred everywhere, but the uplifted part got eroded, as some geological evidence suggests for C-LIP-related subaerial volcanism (Buchs et al., 2018). Indeed, the erosion of the large igneous plateaus has been widely recognized in other plateaus worldwide, including extreme cases where the preserved structure includes areas with no basalts in the centre of the plume head (Ernst et al., 2005). (2) There was an offset between the plume head and the location of the basalt flows, and the magma paths were probably preferentially located along pre-existing lithospheric weakness zones (Buchan and Ernst, 2018; Ernst et al., 2019).

Based on the spatial correlation between the interpreted plume conduits and the extended crustal domains in the Colombian and the Venezuelan basins, we propose a modification to the tectonic model of the Caribbean, as described in Fig. 14. This cartoon is based on the tectonic model proposed by Driscoll and Diebold (1999) and on the present configuration of the Venezuelan Basin (Figs. 2, 13 and 14 in Diebold and Driscoll, 1999).

The initial configuration of the Proto-Caribbean crust consisted of a normal Farallon plate located in the eastern Pacific (Fig. 14a). In the Early Cretaceous, the plate started to interact with a mantle plume, creating rapid uplift and extension above the plume head (Fig. 14b). In the centre of uplift, extensional structures such as horst and grabens may have formed. Subsequently, the early stage basalts (ca. 90 Ma) spilled over the modified crust, preferentially flowing through predefined weakened regions accompanied by lateral and vertical redistribution of magmas (Ernst et al., 2019). In this stage, underplating of ultramafic cumulates started to contribute to the increase in crustal thickness (Fig. 14c). The lack of evidence for volcanism intruding the Proto-Caribbean domains of the Colombian and the Venezuelan basins today remains enigmatic (see the question marks in Fig. 14c). The thinning of the lithosphere above the plume head should be associated with decompression melting and magmatism, which to our knowledge have never been reported in those areas. Therefore, either the seismic reflection experiments have not been successful in mapping these structures, the basaltic flows were later eroded (Ernst, 2007, 2014), or the episodes of basalt flows took place only in the surrounding areas (with an offset of the plume head) along inherited weak portions of the lithosphere, as Ernst et al. (2019) proposed in their plumbing system model. Finally, a late stage (ca. 76 Ma) of magmatism may have reactivated the former plume conduits and/or formed new ones (Fig. 14d). The frozen plume conduits migrated together with the Caribbean plate from the Pacific, forming the present lithospheric configuration of the Caribbean plate, including the rough–smooth basement morphology described by Driscoll and Diebold (1999). Parts of the C-LIP were accreted along the margins and/or subducted, and thus only the preserved lithosphere forms the Caribbean Sea nowadays (Fig. 14e).

In conclusion, the preserved extended Proto-Caribbean crust in the Colombian and the Venezuelan basins might account for the centres of (kilometre-scale) uplift. The plume-related high-density bodies present in the uppermost 75 km of the Caribbean lithosphere may have contributed significantly to thermal subsidence of these basins, as suggested by the large water depths (up to 5 km) in the Venezuelan Basin (see Figs. 3a and 10, around 14∘ N and 66–68∘ W). Moreover, they might behave as rigid domains nowadays, affecting the overall behaviour of the Caribbean plate in response to regional tectonic stresses.