sign in sign up
<<
Home , Azores hotspot, Dynamic topography, Greenland Plate, Iceland hotspot, Mantle plume, Seismic tomography

Two deep- sources for doming and in the North Atlantic

Petar Glišovic(Петар Глишовић)a,1 and Alessandro M. Forteb

aGeotop, Université du Québec à Montréal, Montréal, QC, Canada H3C 3P8; and bDepartment of Geological Sciences, University of Florida, Gainesville, FL 32603

Edited by Barbara A. Romanowicz, University of California, Berkeley, CA, and approved May 20, 2019 (received for review September 19, 2018) The North Atlantic Igneous Province (NAIP) erupted in two major extent of the mantle and thus allows us to obtain explicit links pulses that coincide with the continental breakup and the opening between mantle dynamics and emplacement of the NAIP. To of the North over a period from 62 to 54 Ma. The date, it has been mostly assumed that this mantle dynamic con- unknown mantle structure under the North Atlantic during the nection could be entirely modeled in terms of a Paleocene represents a major missing link in deciphering the geo- under the ancestral . As we show below, there is an dynamic causes of this event. To address this outstanding challenge, equally important upwelling below the ancestral hotspot we use a back-and-forth iterative method for time-reversed global that is a major contributor to North Atlantic surface topography modeling over the Cenozoic Era which incorporates and mantle melting in the Paleocene and . models of present-day tomography-based mantle heterogeneity. We find that the Paleocene mantle under the North Atlantic is Validating Backward Convection Solutions characterized by two major low-density plumes in the : Before embarking on a detailed consideration of the recon- one beneath and another beneath the Azores. These structed evolution of mantle heterogeneity under the North strong lower-mantle upwellings generate small-scale hot upwellings Atlantic, it is important to verify its reliability and geodynamic and cold downwellings in the . The upwellings are consistency. A detailed discussion of this verification is presented dispersed sources of and topographic uplift that were in SI Appendix and we summarize here the main points. active on the rifted margins of the North Atlantic during the In the first test, when the reconstructed mantle heterogeneity formation of the NAIP. While most studies of the Paleocene evolution at 70 Ma is used as the starting model for a convection simula- , ATMOSPHERIC, of the North Atlantic have focused on the proto-Icelandic plume, our tion integrated forward to the present day, we obtain strong AND PLANETARY SCIENCES Cenozoic reconstructions reveal the equally important dynamics of a correlations to present-day heterogeneity given by the global hot, buoyant, mantle-wide upwelling below the Azores. tomography models (SI Appendix, Fig. S1). This test also dem- onstrates that mantle evolution is sensitive to the differences in mantle plumes | North Atlantic Igneous Province | time-reversed mantle radial viscosity variations (SI Appendix, Fig. S2A), and the V2 convection viscosity model (12) yields the highest global correlation to the tomography models. he Paleocene separation of Greenland from northwest In the second test, we verify whether the present-day hetero- TEurope and from North America is closely linked to the geneities predicted by the (forwardly integrated) reconstructions at massive outburst of igneous activity (the total crustal volume is 70 Ma yield a fit to a wide range of convection-related datasets that ∼6.6 × 106 km3; ref. 1) that is now deposited in a region that includes Baffin Island, the western and eastern margins of Significance Greenland, the mid-Norwegian margin, the Faeroe Islands, and the British Isles (2, 3). It has been hypothesized that this volca- nism played a role in triggering the Paleocene–Eocene Thermal A vigorous debate has raged over the past decades on the Maximum (4, 5). The emplacement of the North Atlantic Igne- mantle plume origin of the North Atlantic Igneous Province ous Province (NAIP) occurred in two main events (5, 6), phase 1 (NAIP). This debate has persisted for such a long time because magmatism (∼62 to 59 Ma) and phase 2 magmatism (∼56.5 to the 3D structure of the mantle under the North Atlantic during 54 Ma). These phases are most often thought to be linked to the the Cenozoic was, until now, unknown. This paper presents emergence of anomalously hot structure inside the stretched and thermodynamically consistent tomography-based reconstruc- tions of two deeply seated hot mantle upwellings (plumes) rifted produced by a plume that is now centered under Greenland and the Azores whose evolution is tracked beneath the present-day position of the (2, 3, 6). A from 70 Ma to the present day. We show how these dual up- recent appraisal of NAIP geochronological data suggests that re- wellings gave rise to the small-scale mantle melt sources and gional volcanism progressed without significant pauses and that topographic changes of the NAIP. We believe that this discov- associated magmatic activity continued until about 20 Ma (7). It ery will contribute to resolving the long-standing debate on has been also suggested that the dispersed distribution of NAIP the origin of the NAIP. created in several discrete events requires the presence of either

another mantle plume (8), south of the Iceland hotspot (9), or Author contributions: P.G. and A.M.F. designed research; P.G. performed research; P.G. multiple phases of mantle upwelling (6). While the mantle-plume and A.M.F. analyzed data; and P.G. and A.M.F. wrote the paper. hypothesis is considered a viable candidate for explaining the The authors declare no conflict of interest. NAIP petrogenesis (10), there are a number of competing theories This article is a PNAS Direct Submission. and models, including , meteorite impact, small-scale Published under the PNAS license. -related convection, and melting of a fertile mantle (a review Data deposition: All codes and data reported in this paper have been deposited in the may be found in ref. 11). Here we apply a recently developed data Geotop, Research Centre on the Dynamics of the Earth System, Université du Québec à assimilation method that yields a quantitative geophysical re- Montréal (https://www.geotop.ca/en/recherche/donnees/geophysique/manteau). construction of the Cenozoic evolution of 3D thermal structure 1To whom correspondence may be addressed. Email: [email protected]. below the North Atlantic employing a tomography-based mantle This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. convection model. This modeling provides a detailed spatiotem- 1073/pnas.1816188116/-/DCSupplemental. poral mapping of 3D thermal anomalies that span the entire depth Published online June 13, 2019.

www.pnas.org/cgi/doi/10.1073/pnas.1816188116 PNAS | July 2, 2019 | vol. 116 | no. 27 | 13227–13232 Downloaded by guest on October 1, 2021 constrain the 3D density structure and dynamics of the mantle (SI plumes is resolved in all tomography-based flow predictions tested Appendix,TableS1). This test shows (perhaps not surprisingly) that here and that they remain active today (SI Appendix,Fig.S5). mantle reconstructions based on the joint (seismic–geodynamic) At 55 Ma, strong low-density anomalies (δρ/ρ < −0.2%) and tomography models yield the best overall fits to the present-day high vertical flow rates spanned the upper half of the mantle geodynamic constraints. (depth < 1,445 km) below the Azores hotspot and below a region On the basis of these tests, in the following discussion we will extending from South Greenland (SG) to the Iceland hotspot mainly focus on mantle reconstructions obtained with the V2 (Fig. 1B). The upwelling below the injects a large viscosity and GyPSuM tomography model (13). The predictions volume of hot material in the upper mantle that is directed to- obtained with the other tomography models are nonetheless im- ward the north and west (SI Appendix, Fig. S4 A and B). This portant and for this reason we included them in cluster analyses of finding suggests the “Azores plume” was a major contributor to predicted vertical flow (SI Appendix) and mantle melting predic- the NAIP, along with the SG upwelling (Fig. 1B). About 10° east – C tions (presented below). Furthermore, the reconstructed mantle from the Iceland hotspot, the west east cross-section (Fig. 1 ) evolution under the North Atlantic obtained with the S40RTS shows an additional low-density anomaly and localized upwelling tomography model (14) is shown in SI Appendix,Fig.S3. in the upper mantle beneath the nascent North Atlantic ridge. This same cross-section shows that the deeply rooted Greenland Reconstruction of Paleocene Mantle Flow Under the North plume under the ancestral Iceland hotspot appears to be located Atlantic at the confluence of two large-scale convection rolls below the Maps of the reconstructed Paleocene mantle flow show an up- North American and Eurasian plates. We note that our re- welling below Greenland characterized by a locus of maximum construction of mantle structure and flow at 55 Ma also reveals a ′′ localized upper-mantle upwelling beneath Ireland and Scotland vertical flow that emerges from the D layer below the northern D portion of Greenland and shifts southward at shallower depths (Fig. 1 ) located in the eastern margin of the NAIP (4). A comparison of the Iceland plume reconstructions obtained before impacting the lithosphere mostly below the Greenland SI Appendix margins (Fig. 1 and SI Appendix,Fig.S4A and B). The north– with the GyPSuM and S40RTS tomography models ( , Fig. S3) shows that even though the latter provides a less strong southaxisofthiselongated“Greenland upwelling” or “plume” is expression in the deep mantle, the Cenozoic evolution of this plume centered under the ancestral Iceland hotspot (Fig. 1C) and it is remains centered below the Iceland hotspot. This result provides characterized by long wavelengths in the high-viscosity lower added confidence that our back-and-forth iterative technique for mantle which evolve into a number of shorter-wavelength up- SI Appendix A performing time-reversed convection simulations is robust and wellings in the low-viscosity upper mantle ( ,Fig.S4 performs well under less-than-optimal starting conditions. and B). We also find that this mantle-wide Greenland plume is a positively buoyant feature that persisted from the Paleocene to the Implications for Melt Production and Surface Volcanism present day, where it now resides below Iceland (SI Appendix, Fig. C The reconstructed upwellings of hot mantle material predicted S4 ), thus providing a stable source for the injection of hot ma- by our back-and-forth convection simulations (Fig. 1 and SI Ap- terial into the below the North Atlantic basin. We pendix,Fig.S4) are potentially important sources of Paleocene melt similarly find that a nearly vertical upwelling persists beneath the production under the North Atlantic. We estimate the locations of B ancestral Azores hotspot throughout the Paleocene (Fig. 1 and melting by superimposing the reconstructed lateral temperature SI Appendix A B ,Fig.S4 and ), although its present-day expression variations on a reference temperature profile that satisfies a half- (SI Appendix,Fig.S4C)isasanupper-mantlebranchofthe space cooling model for 50-My-old lithosphere (SI Appendix,Fig. “West-African Superplume” (12). A cluster analysis (SI Appendix) S6) and then map out all regions that exceed the solidus temper- shows that buoyancy-driven vertical flow of these two mantle ature of dry (15). This mapping of regions of mantle melting does not necessarily translate into corresponding maps of subaerial (surface) volcanism. To establish this connection we need a model of the thermomechanical processes involved in the trans- A B port of from (sub)lithospheric depths to the surface via dike formation and propagation (16). This modeling is beyond the scope of our current simulations. With this important caveat in mind, in the following we consider four potential source regions for magma production (Fig. 2) below the Azores, Iceland, SG, and the British Tertiary Igneous Province (BTIP). CD The spatial disposition of the reconstructed melt zones (Fig. 2A) reveals candidate sources for the BTIP, the Faeroe Islands, and the Rockall Plateau . The predicted melt zones also de- lineate potential sources for the onshore basalts in west and east Greenland, as well as the offshore areas of thickened crust along the coast of Greenland. Further south, our reconstruction suggests a large melt zone related to the Azores plume that includes the Newfoundland and extends northward toward the SG melt zone, including the offshore margin of Labrador. A cluster Fig. 1. A convection prediction of mantle heterogeneity below the North analysis (Fig. 2B) demonstrates that the occurrence of mantle Atlantic at 55 Ma. (A) The 2D map shows the geographical orientation of melting near SG and the Azores is a common feature in all mantle cross-sections (B–D). The blue and black lines show the reconstruction tomography-based convection models tested here. We note, position of the plate boundaries and coastlines (39), respectively. The green however, that the GyPSuM simulation is the only one that predicts line and gray shading represent the present-day position of plate boundaries melting regions beneath the British Isles, the Faeroe Islands, and and coastlines, respectively. (B–D) The cross-sections show the pattern of density anomalies that also yield a match to the geological inference of plateaus in the North Atlantic (e.g., Rockall Plateau). We find the depth interval over which the mantle melting surface plate velocities at this time. The distance between yellow dots on SI Appendix cross-sections is 10°. The cyan and magenta circles show the present-day mostly occurred ranges between 50 and 100 km ( , Fig. location of Azores and Iceland hotspots. (C) The double-head magenta ar- S7), and this finding agrees with geochemical modeling of rare- row marks the location of the paleo mid-Atlantic ridge. earth elements in flood basalts (17). The predicted volume of

13228 | www.pnas.org/cgi/doi/10.1073/pnas.1816188116 Glišovic and Forte Downloaded by guest on October 1, 2021 A geochronological evidence for multiple pulses). While the melting beneath SG shows a behavior similar to that for Iceland during the Cenozoic (Fig. 2C), our results suggest substantially greater melt volumes produced by SG. Melting below the Azores and SG po- tentially provided the largest magmatic reservoirs for the NAIP. The possible connections between surface volcanism and re- gions of mantle melting discussed above can also be visualized by considering particle paths, shown in Fig. 3, traced in the recon- structed spatiotemporal evolution of the mantle flow field. The predicted paths further support the connection between North Atlantic volcanism and the reconstructed regions of melting B below central Greenland (e.g., Faeroe Islands path) and eastern (e.g., Skye and Bergen paths) rift margin of Greenland. They also highlight the connection between volcanic rocks in the Ca- nadian (e.g., Clyde River) and the Canadian Atlantic shelf (e.g., Newfoundland Seamounts) to mantle melting under the ancestral Iceland and Azores hotspots, respectively. Finally, we found that the S40RTS-based reconstruction pre- dicts melting beneath Iceland even for a reference boundary- layer geotherm corresponding to lithospheric age of 150 My (SI Appendix, Fig. S9B) because S40RTS generates higher temper- ature anomalies under these regions than GyPSuM. As noted above, this melting signal peaks during phase 2 magmatism. Both C (GyPSuM and S40RTS) reconstructions also predict present-day melting beneath Iceland (SI Appendix, Fig. S9) if we assume a thinned lithosphere whose age is 25 My (SI Appendix, Fig. S6). Dynamic Surface Topography EARTH, ATMOSPHERIC,

Observations (or geological inferences) of current (or past) to- AND PLANETARY SCIENCES pographic uplift produced by hot mantle upwellings provide strong constraints on the location, dimension, and integrated buoyancy (and hence temperature anomalies) of these upwell- ings. It is therefore of considerable interest to obtain geologic Fig. 2. A prediction of mantle melt production. (A) Map of mantle melting estimates or reconstructions of the topography of the North at 70 km depth during the Paleocene. The red shading represents the region Atlantic region in a time window that encompasses the intense in which the solidus temperature of dry peridotite (15) is exceeded. The Paleocene–Eocene volcanic activity that led to the formation of yellow contour lines represent areas with a high vertical velocity exceeding the NAIP. Following decades of hydrocarbon exploration and one-half of the maximum velocity in the mapped region (SI Appendix, Fig. S4 seismic stratigraphic mapping, the sedimentary basins along the A and B). The mantle melt production is predicted from the GyPSuM to- East Greenland–northwest (NW) European margin have yielded mography model (13). (B) Cluster analysis of mantle melting at 70 km depth valuable spatial constrains on the transient uplifts this region predicted from three global tomography models (13, 14, 31) at 62.5 Ma and experienced at the Paleocene–Eocene boundary (19, 20). 55 Ma (given by columns). Maps are color-coded according to how many models assign high-temperature regions to the melting cluster (green = 1; yellow = 2; red = 3). (C) The total volume of mantle melting as a function of time predicted from both GyPSuM (13) and S40RTS (14) tomography models (given by columns). The curves are color-coded with reference to the disks in the map. These volumes are calculated by integrating all areas (see A) where the melting temperature is exceeded by 100 K. In the maps shown in A and B the blue and black lines represent the reconstruction position of the plate boundaries and coastlines (39), respectively.

mantle melting varies depending on the source region, with the Azores and SG yielding the largest volume, followed by Iceland (Fig. 2C). The reconstructed time evolution of mantle melting varies with the tomography model, with S40RTS yielding a peak production between 50 and 60 Ma, while GyPSuM yields a progressive increase that peaks between 55 and 30 Ma. These findings suggest that a mantle melt supply was available for a protracted period of surface Fig. 3. Particle trajectories predicted with time-dependent flow recon- volcanic activity (7), while a local peak in melting near 55 Ma structions. The flow lines shown above track the trajectories followed by (predicted with S40RTS) below the ancestral Iceland hotspot is in passive particles (tracers) in the mantle flow field that has evolved over the accord with previous inferences of increased (phase 2) magmatism past 70 My in response to the reconstructed changes in mantle buoyancy at this time on the northeast Atlantic margin, which followed the distribution predicted using the GyPSuM tomography model (13). The col- earlier phase 1 activity in West Greenland and Baffin Bay (5, 6) (Fig. ored circles on each flow line are spaced apart at 10-My intervals and their 2 A and C). Considering the northwestward drift of Greenland colors correspond to the depth of the particle (see color bar on the right). Blue and black lines show the reconstruction position of the plate bound- relative to the (assumed) fixed position of the Iceland hotspot, the aries and coastlines (39), respectively at 60 Ma, while the green line and gray underlying mantle melt source may have also contributed to the shading represent the present-day position of plate boundaries and coast- third (minor) magmatic pulse at ∼49 Ma along the East Greenland lines, respectively. The downward extension of the flow lines into the lower rift margin (18) (see, however, ref. 7 for a recent appraisal of the mantle is shown in SI Appendix, Fig. S8.

Glišovic and Forte PNAS | July 2, 2019 | vol. 116 | no. 27 | 13229 Downloaded by guest on October 1, 2021 Previous efforts to model the spatiotemporal evolution of this requires the highest possible resolution in the convection convection-driven topography in the North Atlantic include modeling. We address this challenge in the following. plate-tectonic reconstructions of the descent of subducted slabs We begin by noting that the geological compilation by ref. 19 and the corresponding changes in surface topography driven by suggests that most of the interpreted Paleocene–Eocene uplifts their negative buoyancy (21). The predictions along the NW European margin range between ∼0.1 and ∼1 km, show continuous subsidence across Greenland and its margins with the notable exception of the NW Rockall Trough where the ∼ ∼ since the Late Jurassic and a small amount of differential uplift in interpreted uplift ranges between 1.5 and 2 km. Furthermore, the Barents Sea over the past ∼30 My. It is thus clear that dif- the interpretation of uplift associatedwiththeBTIPwasnomore ferential changes in -driven dynamic topography cannot than 200 m during the phase 1 magmatism that preceded more explain uplift in the North Atlantic during the Paleocene–Eocene widespread, higher-amplitude uplift during phase 2 (6). With these geologic constraints in mind, we now turn to a consideration of the transition. The most recent modeling of mid- dynamic pattern of dynamic topography change predicted by our time- topography changes in the North Atlantic has involved the ap- reversed reconstructions of the buoyancy distribution in the mantle. plication of adjoint mantle convection methods using global seis- The Late Paleocene process that occurred during the final mic tomography models as an initial condition (22). This adjoint stages of the North Atlantic rifting involved extensive stretching convection modeling, which was carried back to 40 Ma, yielded and thinning of the crust–lithosphere system that is theoretically predictions of uplifted dynamic topography driven by hot mantle controlled by complex (brittle and ductile) rheological structure upwelling that are characterized by long wavelengths. The out- and corresponding topography variations that may present shorter standing challenge then is to obtain a geologically acceptable wavelength features than in the underlying mantle plume structure model of the convection-driven evolution of North Atlantic to- (23). In view of the possibly complex deformation of the lithosphere pography that extends back to the time of the formation of the in this rifting environment, we focus here on the time-dependent NAIP. A key ingredient in resolving this challenge is to reproduce vertical displacement of the surface that is generated by sublitho- topography changes on sufficiently short spatial wavelengths and spheric changes in mantle buoyancy (Fig. 4 A–E). (Corresponding

A F K

B G L

Fig. 4. Paleogene predictions of dynamic surface C H M topography. (A–E) The evolution of dynamic topog- raphy obtained with the GyPSuM tomography model (13), which excludes the contribution from changes in lithospheric density, down to a depth of 125 km. A Lagrangian perspective is employed, where the to- pography is viewed by an observer fixed to each of the tectonic plates (North America, Greenland, and Eurasia) that occupy the North Atlantic. The dynamic D I N topography at each instant is therefore mapped in present-day geographic coordinates and we use the Euler rotation poles of ref. 33 to calculate the needed rotations of the predicted topography fields. (All rotations are with respect to the No-Net Rotation frame employed in the mantle convection simula- tions.) The gray shaded regions depict whose age is younger than the time at which the topography is mapped (and hence has no definition). E J O (F–J) Changes in dynamic topography, calculated relative to the dynamic topography predicted at 70 Ma (frame A). The gray shaded regions depict oceanic crust whose age is younger than 70 Ma. (K–O) Changes in dynamic topography, calculated relative to the present-day dynamic topography (frame E). The gray shaded regions depict oceanic crust whose age is younger than each time at which the differential to- pography is calculated. The predicted topography fields are in all cases calculated from a spherical har- monic expansion to degree and order 100.

13230 | www.pnas.org/cgi/doi/10.1073/pnas.1816188116 Glišovic and Forte Downloaded by guest on October 1, 2021 topography maps that include the contributions from changes in edges of a sublithospheric plume head (25). Such shallow, small- lithospheric buoyancy predicted by our back-and-forth convection scale thermal heterogeneity may provide an additional magma modeling are shown in SI Appendix,Fig.S10.) source that could contribute to the formation of flood We present two complementary perspectives in quantifying provinces (26) and their associated uplift. the predicted variations of the convection-induced topography: Past interpretations of North Atlantic uplift and magmatism (i) changes relative to the preexisting dynamic topography at preceding and during the Paleocene–Eocene transition have been 70 Ma (Fig. 4 F–J) and (ii) changes relative to present-day dy- based on the popular conceptual and theoretical model of a single namic topography (Fig. 4 K–O). Relative to the 70-Ma dynamic symmetric upwelling plume that spreads horizontally outward when topography, the dominant feature (as seen by an observer fixed it impacts the base of the lithosphere (19, 27). However, our Pa- to the Greenland plate) is a dome of uplift that propagates from leocene predictions of mantle heterogeneity provide a markedly the southwestern margin of Greenland at 65 Ma (Fig. 4F) to the different geodynamic setting to explain the development of the southeastern margin at 55 Ma (Fig. 4H), at which time most of NAIP. We find that two major mantle plumes are implicated in the the eastern margin of Greenland experienced between ∼200 to Paleocene and Eocene evolution of North Atlantic topography and ∼600 m of differential uplift. Between 55 and 50 Ma, the Green- magmatism, namely the Azores and Iceland plumes, where the land dome continued to grow and shifted northward, where it latter continues to be very active in the present day. This setting achieved its peak expression (Fig. 4I). In the -up to the Pa- includes two large-scale, deeply seated hot upwellings originating in leocene–Eocene transition, the NW European margin also expe- the lower mantle that create short-wavelength, dispersed upwellings rienced differential uplift in a number of localities extending from within the upper mantle (Fig. 1 and SI Appendix,Fig.S4). These (off- and onshore) to offshore sites located northwest of upper-mantle upwellings are closely associated with the predicted the British Isles and Ireland. The differential uplift produced by loci of magmatism and topographic uplift on the spreading margins sublithospheric buoyancy changes at these NW European loca- that surrounded Greenland during the Paleocene (Figs. 2 and 4). tions ranges between ∼200 and ∼400 m, which are within the range We found that this complex dynamic evolution of multiscale up- of the geologic estimates compiled by ref. 19. These regions of wellings continuously injected hot material into the lithosphere uplift also appear to be correlated to recent maps of geological during the Paleogene, thus providing a long-lived, deep-mantle hiatuses in the North Atlantic presented in ref. 24. source for magmatic pulses of NAIP from ∼62 to ∼50 Ma (9, 18) Relative to the present-day topography, the predicted evolu- (but see ref. 7) and a possible Iceland hotspot track (4, 28). tion in dynamic topography shows that large portions of the crust surrounding the North Atlantic Ocean were uplifted by up to Materials and Methods ∼ – EARTH, ATMOSPHERIC, 800 m at the time of the Paleocene Eocene transition (Fig. To test hypotheses for the formation of the NAIP, we use a recently developed AND PLANETARY SCIENCES 4N). We note, in particular, that Paleocene dynamic topography data assimilation technique that involves a back-and-forth iterative nudging method highs (Fig. 4 L–N) are predicted at the following locations: SG, for reconstructing the past evolution of mantle convection using, as an initial the British Isles, Norway (notably around the Lofoten archipel- condition, the present-day 3D mantle structure derived from ago), and a very large fraction of the present-day Atlantic shelf (29, 30). As illustrated in ref. 30, seismic tomography models contain a record of past states of mantle structure and thus, using an appropriate data assimilation east of Newfoundland (e.g., Grand Banks and Flemish Cap), procedure, we can successfully track the evolution of mantle heterogeneity from extending further east to an elongated topographic high that the present day into the geologic past. While the incorporation of present-day encompasses the Newfoundland Seamounts. These Paleocene tomography is a key ingredient in backward convection modeling, the less-than- and Eocene topography highs offshore southeastern New- perfect resolution of mantle structure by seismic tomography may be a source of foundland are associated with mantle buoyancy generated by the substantial uncertainty in time-reversed reconstructions of past mantle heteroge- ancestral Azores plume (Fig. 1B). neity. Tomographic inversions provide an inherently nonunique inference of mantle heterogeneity that is further compounded by the application of damping Synthetic Plume Simulations and smoothing in the inversions. We therefore tested the following three global We compared our tomography-based mantle reconstructions with tomography models: (i)TX2008(31)and(ii) GyPSuM (13), both derived from the joint inversion of seismic, geodynamic, and mineral physics data, and (iii)S40RTS theoretical forward-in-time numerical simulations (described in SI Appendix (14) derived from the inversion of seismic (S-wave velocity) data alone. The main ) of a single mantle plume rising beneath a hypothetical difference between TX2008 and GyPSuM is that the latter also incorporates hotspot (here assumed to represent Iceland). We find that this globally distributed P-wave travel time measurement and an enhanced iterative deep-seated hot upwelling generates small-scale thermal instabil- procedure for the joint resolution of 3D density and seismic velocity anomalies in ities in the upper mantle, extending radially tens of degrees from the mantle. Furthermore, both TX2008 and GyPSuM incorporate compositional the central plume axis, long before the main plume body enters (i.e., chemical) contributions to mantle density anomalies. the upper mantle (SI Appendix,Fig.S11A). These small-scale In view of the uncertainties inherent in the seismic tomographic imaging shallow anomalies produce a complex dynamic topography pat- of 3D mantle structure (13, 14, 31), we also considered purely theoretical tern characterized by elongated topography highs arranged convection models of a single thermal plume rising below the Icelandic around the axis of the main upwelling. The dispersed of hotspot (see SI Appendix for details). This synthetic plume model provides a reference for interpreting our reconstructions of the Cenozoic evolution of these localized upper-mantle anomalies (SI Appendix,Fig.S11C D mantle heterogeneity below the North Atlantic and the corresponding and ) resembles that obtained with the tomography-based re- changes in dynamic surface topography. constructions (Fig. 1). Once the main buoyant upwelling passes ∼ Another important feature of our backward convection modeling is the across the upper mantle (requiring 10 My), it generates a central inclusion of a plate-like boundary condition (32) that involves the use of geo- dome of topographic uplift (e.g., SI Appendix,Fig.S11C). logic reconstructions of tectonic plate geometries during the Cenozoic (33). This surface boundary condition is based on an explicit coupling of effectively rigid Discussion and Conclusions tectonic plates to the underlying mantle flow, generating predictions of both As shown by ref. 23, it is possible that complex rheological poloidal (divergent and convergent) and toroidal (strike-slip) flow (34). We stratification of the crust–lithosphere system will modulate the emphasize that the surface plate velocities are not imposed but rather are surface uplift produced by a mantle upwelling, generating com- predicted at any given time through viscous coupling to the buoyancy-driven mantle flow (32). Geological reconstructions of plate motion histories provide plex topography with scale lengths that can be significantly less the only direct constraints on the distribution of mantle buoyancy in the than the scale length of the underlying mantle plume. In addi- geologic past. We thus implemented a mathematical inverse procedure that tion, time-reversed convection modeling is still unable to incor- determines the minimal perturbation of the reconstructed density field in the porate a strong thermal boundary layer at the top of the mantle mantle that yields a match to the geologically determined plate motions (29). that is responsible for the appearance of small-scale thermal Our back-and-forth data assimilation method for time-reversed convection instabilities (with lateral dimensions of ∼100 to 250 km) on the modeling incorporates geodynamically constrained depth variations of mantle

Glišovic and Forte PNAS | July 2, 2019 | vol. 116 | no. 27 | 13231 Downloaded by guest on October 1, 2021 viscosity. As discussed in ref. 29, this fundamental input represents another plate rotations and that the latter are a consequence of extreme lateral source of uncertainty in the time-reversed convection simulations. We con- variability of the effective viscosity of the lithosphere (34). (The implications sidered two mantle viscosity profiles, V1 and V2 (SI Appendix,Fig.S2A), that are and dynamic feedback of the plate-like character of the lithosphere are both inferred from joint inversions of the global convection-related observables further discussed in SI Appendix.) (plate velocities, gravity anomalies, crust-corrected dynamic topography, and The values we employ for reference properties of the mantle (e.g., vis- core-mantle boundary ellipticity) and a suite of ice-age geodynamic data as- cosity, mean density and gravity, thermal conductivity, thermal expansion, heat sociated with glacial isostatic adjustment (specifically, the Fennoscandian re- capacity, and internal heating) are given in SI Appendix, where we present laxation spectrum and a set of decay times determined from the postglacial sea- more details on the pseudospectral modeling of global thermal convection. level history in Hudson Bay and Sweden) (12, 35). Compared with the V1 profile, V2 has a thicker, higher-viscosity lithosphere, a low-viscosity asthenospheric Code Availability. Mantle-flow kernels employed to calculate mantle tem- layer that is displaced to greater depth (220 km), a continuous variation of perature structure and the dynamic topography predictions have been de- viscosity from the upper to the lower mantle, and a higher viscosity in the posited in the Geotop, Research Centre on the Dynamics of the Earth System bottomhalfofthemantle. (https://www.geotop.ca/fr/recherche/donnees/geophysique) (38). The use of a 1D (depth-dependent) viscosity, rather than a 3D viscosity, is naturally open to question. We note that recent efforts to reconcile both Data Availability. All data analyzed in this study were previously published in gravity anomalies and topography in mantle flow calculations suggest rather the cited references. Additional data related to this paper have been de- modest lateral variations in viscosity (36), indicating that several opposing posited in the Geotop, Research Centre on the Dynamics of the Earth System microphysical controls may be important in the mantle (37). The in- (https://www.geotop.ca/fr/recherche/donnees/geophysique) (38). corporation of lateral variations in viscosity that are compatible with both mineral physics and geodynamic constraints is an outstanding issue in con- ACKNOWLEDGMENTS. We acknowledge the support of this work provided vection modeling. We therefore opted to work with depth-dependent vis- by the Natural Sciences and Engineering Research Council of Canada (Grant cosity profiles that have been directly verified against a wide suite of 217272-2013-RGPIN). A.M.F. also acknowledges the support for this work geodynamic surface constraints (35) and independent mineral–physical provided by the University of Florida (UF). The convection simulations in this modeling (37). We note, moreover, that our backward convection models study were carried out thanks to supercomputing facilities of Calcul Québec include a significant toroidal component of flow generated by the surface consortia at Université de Montréal and on the HiPerGator at UF.

1. O. Eldholm, K. Grue, North Atlantic volcanic margins: Dimensions and production 20. M. E. S. Champion, N. J. White, S. M. Jones, J. P. B. Lovell, Quantifying transient rates. J. Geophys. Res. Solid Earth 99, 2955–2968 (1994). mantle convective uplift: An example from the Faroe-Shetland basin. Tectonics 27, 2. A. D. Saunders, J. G. Fitton, A. C. Kerr, M. J. Norry, R. W. Kent, “The North Atlantic TC1002 (2008). Igneous Province” in Large Igneous Provinces: Continental, Oceanic, and Planetary 21. G. E. Shephard et al., Circum Arctic mantle structure and long wavelength topogra- Flood Volcanism, J. J. Mahoney, M. F. Coffin, Eds. (American Geophysical Union, phy since the Jurassic. J. Geophys. Res. Solid Earth 119, 7889–7908 (2014). Washington, DC, 1997). 22. L. Colli, S. Ghelichkhan, H. P. Bunge, J. Oeser, Retrodictions of Mid Paleogene mantle 3. R. White, D. McKenzie, Magmatism at rift zones: The generation of volcanic conti- flow and dynamic topography in the Atlantic region from compressible high reso- nental margins and flood basalts. J. Geophys. Res. 94, 7685–7729 (1989). lution adjoint mantle convection models: Sensitivity to deep mantle viscosity and – 4. M. Storey, R. A. Duncan, C. C. Swisher III, Paleocene-Eocene thermal maximum and tomographic input model. Res. 53, 252 272 (2017). the opening of the Northeast Atlantic. Science 316, 587–589 (2007). 23. E. Burov, T. Gerya, Asymmetric three-dimensional topography over mantle plumes. – 5. A. Saunders, Two LIPs and two Earth-system crises: The impact of the North Atlantic Nature 513,85 89 (2014). 24. Y. Vibe, A. M. Friedrich, H. P. Bunge, S. R. Clark, Correlations of oceanic spreading Igneous Province and the on the Earth-surface . Geol. Mag. rates and hiatus surface area in the North Atlantic realm. Lithosphere 10, 677–684 153, 201–222 (2016). (2018). 6. A. D. Saunders et al., Regional uplift associated with continental large igneous 25. P. Glišovic, A. M. Forte, Importance of initial buoyancy field on evolution of mantle provinces: The roles of mantle plumes and the lithosphere. Chem. Geol. 241, 282–318 thermal structure: Implications of surface boundary conditions. Geosci. Front. 6,3–22 (2007). (2015). 7. C. M. Wilkinson, M. Ganerød, B. W. H. Hendriks, E. A. Eide, Compilation and appraisal 26. D. L. Anderson, “Scoring hotspots: The plume and plate paradigms” in Plates, Plumes of geochronological data from the North Atlantic Igneous Province (NAIP). Geol. Soc. and Paradigms, G. R. Foulger, J. H. Natland, D. C. Presnall, D. L. Anderson, Eds. – Lond. Spec. Publ. 447,69 103 (2017). (Geological Society of America Special Papers, Geological Society of America, Boulder, 8. W. J. Morgan, Hotspot tracks and the early rifting of the Atlantic. Tectonophysics 94, CO, 2005), vol. 388, pp. 31–54. – 123 139 (1983). 27. R. A. Hartley, G. G. Roberts, N. White, C. Richardson, Transient convective uplift of an 9. M. Ganerød et al., The North Atlantic Igneous Province reconstructed and its relation ancient buried landscape. Nat. Geosci. 4, 562–565 (2011). to the plume generation zone: The Antrim Group revisited. Geophys. J. Int. 182, 28. L. A. Lawver, R. D. Müller, Iceland hotspot track. 22, 311–314 (1994). 183–202 (2010). 29. P. Glišovic, A. M. Forte, A new back-and-forth iterative method for time-reversed 10. E. L. Brown, C. E. Lesher, North Atlantic magmatism controlled by temperature, convection modelling: Implications for the Cenozoic evolution of 3-D structure and mantle composition and buoyancy. Nat. Geosci. 7, 820–824 (2014). dynamics of the mantle. J. Geophys. Res. Solid Earth 121,1–18 (2016). 11. R. Meyer, J. van Wijk, L. Gernigon, “The North Atlantic Igneous Province: A review of 30. P. Glišovic, A. M. Forte, On the deep-mantle origin of the . Science 355, models for its formation” in Plates, Plumes and Planetary Processes, G. R. Foulger, D. 613–616 (2017). M. Jurdy, Eds. (Geological Society of America Special Papers, Geological Society of 31. N. A. Simmons, A. M. Forte, S. P. Grand, Joint seismic, geodynamic and mineral America, Boulder, CO, 2007), vol. 430, pp. 525–552. physical constraints on three-dimensional mantle heterogeneity: Implications for the 12. A. M. Forte et al., Joint seismic–geodynamic–mineral physical modelling of African relative importance of thermal versus compositional heterogeneity. Geophys. J. Int. – : A reconciliation of deep-mantle convection with surface geophysical 177, 1284 1304 (2009). š  constraints. Earth Planet. Sci. Lett. 295, 329–341 (2010). 32. P. Gli ovic, A. M. Forte, R. Moucha, Time-dependent convection models of mantle 13. N. A. Simmons, A. M. Forte, L. Boschi, S. P. Grand, GyPSuM: A joint tomographic thermal structure constrained by seismic tomography and geodynamics: Implications – model of mantle density and speeds. J. Geophys. Res. Solid Earth 115, for mantle plume dynamics and CMB heat flux. Geophys. J. Int. 190, 785 815 (2012). 33. D. B. Rowley et al., Kinematics and dynamics of the East Pacific Rise linked to a stable, B12310 (2010). deep-mantle upwelling. Sci. Adv. 2, e1601107 (2016). 14. J. Ritsema, H. J. van Heijst, A. Deuss, J. H. Woodhouse, S40RTS: A degree-40 shear- 34. A. M. Forte, W. R. Peltier, The kinematics and dynamics of poloidal-toroidal coupling velocity model for the mantle from new Rayleigh wave dispersion, teleseismic trav- in mantle flow: The importance of surface plates and lateral viscosity variations. Adv. eltimes, and normal-mode splitting function measurements. Geophys. J. Int. 184, Geophys. 36,1–119 (1994). 1223–1236 (2011). 35. J. X. Mitrovica, A. M. Forte, A new inference of mantle viscosity based upon joint 15. M. M. Hirschmann, Mantle solidus: Experimental constraints and the effects of peri- inversion of convection and glacial isostatic adjustment data. Earth Planet. Sci. Lett. dotite composition. Geochem. Geophys. Geosyst. 1, 1042 (2000). 225, 177–189 (2004). 16. E. Rivalta, B. Taisne, A. P. Bunger, R. F. Katz, A review of mechanical models of dike 36. T. Yang, M. Gurnis, Dynamic topography, gravity and the role of lateral viscosity vari- – propagation: Schools of thought, results and future directions. Tectonophysics 638,1 ations from inversion of global mantle flow. Geophys. J. Int. 207,1186–1202 (2016). 42 (2015). 37. P. Glišovic, A. M. Forte, M. W. Ammann, Variations in grain size and viscosity based on 17. R. S. White, D. McKenzie, Mantle plumes and flood basalts. J. Geophys. Res. 10, vacancy diffusion in minerals, seismic tomography, and geodynamically inferred 17543–17585 (1995). mantle rheology. Geophys. Res. Lett. 42, 6278–6286 (2015). 40 39 18. C. Tegner et al., Ar- Ar geochronology of Tertiary mafic intrusions along the East 38. P. Glišovic, A. M. Forte, Two deep mantle sources for Paleocene doming and volcanism in Greenland rifted margin: Relation to flood basalts and the Iceland hotspot track. the North Atlantic. Geotop, Research Centre on the Dynamics of the Earth System. Earth Planet. Sci. Lett. 156,75–88 (1998). https://www.geotop.ca/fr/recherche/donnees/geophysique. Deposited 5 June 2019. 19. J. Maclennan, S. M. Jones, Regional uplift, gas hydrate dissociation and the origins of 39. T. H. Torsvik, R. D. Müller, R. Van der Voo, B. Steinberger, C. Gaina, Global plate the Paleocene-Eocene Thermal Maximum. Earth Planet. Sci. Lett. 245,65–80 (2006). motion frames: Toward a unified model. Rev. Geophys. 46, RG3004 (2008).

13232 | www.pnas.org/cgi/doi/10.1073/pnas.1816188116 Glišovic and Forte Downloaded by guest on October 1, 2021