LETTERS PUBLISHED ONLINE: 9 MAY 2010 | DOI: 10.1038/NGEO855 Mantle upwellings above slab graveyards linked to the global geoid lows Sonja Spasojevic 1 * , Michael Gurnis 1 and Rupert Sutherland 2 The global geoid is characterized by a semi-continuous belt of lows that surround the Pacific Ocean, including isolated minima in the Indian Ocean, Ross Sea and northeast Pacific and west Atlantic oceans. These geoid lows have been attributed to Mesozoic subduction 1,2 . Geodynamic models that include slab graveyards in the lower mantle as inferred from seismic topography or from plate reconstructions correctly predict the general trend of geoid minima 3,4 . However, these models fail to accurately reproduce localized geoid lows in the Indian Ocean, Ross Sea and northeast Pacific Ocean. Here we show that the geoid lows are correlated with high-velocity anomalies near the base of the mantle and low-velocity anomalies in the mid-to-upper mantle. Our mantle flow models reproduce the geoid minima if the mid-to-upper mantle upwellings are positioned above the inferred locations of ancient subducted slabs. We find that the long-wavelength trough in the geoid is linked to high-density slab graveyards in the lower mantle, whereas upwelling regions in the mantle above 1,000 km depth cause discrete lows within the larger trough. We suggest that this mode of upwelling in the mid-to-upper mantle is caused by buoyant hydrated mantle that was created by processes around and above subducted slabs. The shape of the global geoid is characterized by a semi- continuous 10–30-m-amplitude negative anomaly surrounding the Pacific Ocean, with high-amplitude ( - 40 to -90 m) localized minima in the Indian Ocean, Ross Sea, northeast Pacific Ocean and west Atlantic Ocean within this trough (Fig. 1a). The more continuous eastern hemisphere low extends from Siberia through India into the Ross Sea. The more segmented western hemisphere low includes a north–south trending anomaly in the northeast Pacific Ocean and a zone that includes Hudson Bay and the western Atlantic Ocean. Localized geoid lows persist after isostatic corrections for the lithosphere have been applied 5 . Analysis of the gravity field as a simultaneous function of position and spectral content 6 shows that about half of the Hudson Bay anomaly could be explained by incomplete postglacial rebound, but such a mechanism cannot explain any significant component of the other geoid lows, because they are not correlated with positions of past ice sheets. The primary explanation for the shape of the geoid is related to heterogeneity and dynamics within the mantle 1–4 . It has previously been noted that geoid lows are globally correlated with locations of Mesozoic subduction 1,2 , whereas geoid highs are correlated with present-day subduction zones and hotspots 7 . Active subduction zones are characterized by geoid highs resulting from a dominating positive mass anomaly of cold, upper mantle slabs compared with a low from dynamic topography 8,9 . As slabs sink into the higher-viscosity lower mantle, the dynamic response functions switch sign 9 , resulting in an association of lower mantle slabs with negative geoid anomalies. We analysed global 1 Seismological Laboratory, California Institute of Technology, Pasadena, California 91125, USA, 2 GNS Science, PO Box 30368, Lower Hutt 5040, New Zealand. *e-mail: sonja@gps.caltech.edu. tomographic images of seismic velocity to explore the hypothesis that geoid lows are spatially correlated with slab graveyards and high seismic velocities 1,2 . It is clear that geoid lows are indeed underlain by volumes with high seismic velocities near the base of the mantle (Fig. 1b, Supplementary Fig. S1a), but we also find that there are anomalously low seismic velocities in the upper part of the mantle in the same regions (Fig. 1c, Supplementary Fig. S1b). These low-seismic-velocity anomalies are found at depths up to 1,000 km and have absolute amplitudes that are at least as large as the deeper high-velocity anomalies (Fig. 1d). A global analysis of the correlation between different tomog- raphy models and the geoid (Supplementary Fig. S1e) reveals that slow seismic velocities in the upper half of the mantle and fast seismic velocities in the lower half of the mantle are both correlated with geoid lows, whereas geoid highs have a similar strength of correlation with fast seismic velocities in the upper approximately 800 km of mantle (actively subducting slabs) and slow seismic velocities in the lower mantle (inferred to be super- plumes and hotspots). The relatively high values of correlation between velocity anomalies and geoid lows, combined with our tectonic interpretations of cross-sections through the tomographic models (Fig. 1e, Supplementary Fig. S1c–d), lead us to suggest that buoyant upwellings above slab graveyards play a significant role in producing the global pattern of geoid lows. Similar suggestions have been made previously to explain regional features of tomography models and the geoid. It was suggested that increased water content in the upper mantle beneath the US east coast 10 was supplied by the subducted Farallon slab and caused an extensive zone of low seismic velocities in the upper mantle. The geoid low between Antarctica and New Zealand is best explained by middle and upper mantle upwelling that followed the cessation of subduction beneath Gondwanaland 11 . To further investigate this hypothesis, we developed instantaneous models of global mantle flow based on density structures scaled from seismic tomography. Previous global geodynamic studies have used the geoid and gravity as constraints on seismic velocity–density scaling, and on mantle viscosity as a function of radius 3,12,13 , or more complex functions with radial and lateral parameterizations based on tectonic regionalizations 14,15 . Regional models of the Tonga–Kermadec and Aleutian subduction zones 8,16 show that significant lateral viscosity variations are required between upper mantle slabs and adjacent mantles wedges to match the geoid, whereas some global studies indicate that weak plate margins in the lithosphere are needed to match positive geoid anomalies over subduction zones 14 . Viscosity variation is defined differently for two sets of models: (1) viscosity variation as a function of radius and temperature; and (2) lateral viscosity variation in the upper mantle based on tectonic regionalization in addition to radial and temperature viscosity NATURE GEOSCIENCE | VOL 3 | JUNE 2010 | www.nature.com/naturegeoscience 435 © 2010 Macmillan Publishers Limited. All rights reserved.