Physics of the Earth and Planetary Interiors 146 (2004) 113–124 Constraining large-scale mantle heterogeneity using mantle and inner-core sensitive normal modes Miaki Ishii a,∗,1 , Jeroen Tromp b a Department of Earth & Planetary Sciences, Harvard University, Cambridge, MA 02138 USA b Seismological Laboratory, California Institute of Technology, Pasadena, CA 91125 USA Received 2 September 2002; received in revised form 5 February 2003; accepted 18 June 2003 Abstract Attempts to resolve density heterogeneity within the mantle using normal-mode data revealed an unexpected feature near the core-mantle boundary: regions of strong sheer velocity reduction, known as superplumes, are characterized by heavier than average material [Science 285 (1999) 1231]. Thus far, free-oscillation studies of mantle structure have relied upon modes which sample only the mantle (e.g., [J. Geophys. Res. 96 (1991) 551; Science 285 (1999) 1231; J. Geophys. Res. 104 (1999a) 993; Geophys. J. Int. 143 (2000b) 478; Geophys. J. Int. 150 (2002) 162]). In order to better constrain mantle heterogeneity, we add inner-core sensitive modes to our data set, and invert for mantle structure and inner-core anisotropy simultaneously. The additional modes do not alter the pattern of the density anomalies relative to model SPRD6 [Geophys. J. Int. 145 (2001) 77]. However, they help constrain the amplitude of lateral variations in density. In a previous study, obtaining a density model with reasonable amplitudes required supplemental gravity data, which are not included in the current study. Nonetheless, the new density model is generally weaker, especially at the bottom of the mantle. The root-mean square (RMS) amplitude of the model exhibits two maxima around 600 and 2300 km depth. Experiments indicate that these are robust components of the model and neither artifacts of the choice of radial basis functions nor the damping scheme. Comparisons of the density model with shear- and compressional-velocity models show negative or nearly zero correlation in the transition zone and near the base of the mantle, where the root-mean square density amplitude is high. Finally, the anomalous features near the core-mantle boundary are confirmed: strong anti-correlation between shear and bulk-sound velocities, and increased density at the locations of superplumes. © 2004 Elsevier B.V. All rights reserved. Keywords: Seismic tomography; Normal modes; Seismic velocities; Density 1. Introduction Body-wave travel times, the most extensively used data in seismic tomography, have revealed large ∗ Corresponding author. Tel.: +1-858-534-4643. E-mail address: [email protected] (M. Ishii). 1 Present address: Institute of Geophysics and Planetary Physics, Scripps Institution of Oceanography, University of California, La Jolla, CA 92093, USA. low-velocity structures near the core-mantle boundary (Dziewoński, 1984). Although images of these velocity anomalies have been refined using larger data sets and improved techniques (e.g., Inoue et al., 1990; Li and Romanowicz, 1996; Dziewoński et al., 1997; van der Hilst et al., 1997; Masters et al., 2000c), mantle density anomalies have remained elusive. This is because high-frequency body-wave data are insensitive to density variations. In contrast, the gravitational restoring force is important for low-frequency normal 0031-9201/$ – see front matter © 2004 Elsevier B.V. All rights reserved. doi:10.1016/j.pepi.2003.06.012 114 M. Ishii, J. Tromp / Physics of the Earth and Planetary Interiors 146 (2004) 113–124 modes, and hence they possess sensitivity to lateral variations in density. Knowledge of the density distribution within the mantle would constitute a substantial step toward understanding its dynamics and provides important constraints on separating effects due to temperature and chemistry (Forte and Mitrovica, 2001). Although controversial (Resovsky and Ritzwoller, 1999b; Masters et al., 2000b; Romanowicz, 2001), a previous inversion of normal-mode data resulted in density variations with an unexpected pattern near the core-mantle boundary (Ishii and Tromp, 1999). Underneath the Pacific and Africa, where seismic velocities are anomalously low (features identified as superplumes), the density is found to be higher than average. This observation is consistent with convection simulations where chemically distinct, and heavier, material accumulates under regions of active upwellings with higher temperatures and lower shear velocities (Christensen, 1984; Davies and Gurnis, 1986; Hansen and Yuen, 1988; Tackley, 1998). The disadvantage of normal-mode tomography is that only large-scale even-degree structure can be resolved. Nonetheless, the mantle is generally dominated by large-scale features, and there are structures with a dominant even-degree component, such as the superplumes. Therefore, a comparison between velocity and density heterogeneity should give us insight into the petrology and dynamics of these structures. Thus far, modeling of density variations within the mantle has relied upon mantle sensitive modes (Ishii and Tromp, 1999; Masters et al., 2000b; Kuo and Romanowicz, 2002). Despite their sensitivity to mantle structure, inner-core sensitive modes have been ignored in studies of the mantle, because data from these modes exhibit a strong inner-core signature. To overcome this dilemma, we invert mantle and inner-core sensitive modes simultaneously for mantle heterogeneity and inner-core anisotropy. The results for the inner core are discussed elsewhere (Ishii et al., 2002), and we focus on models of the mantle in this paper. 2. Theory In a seismic spectrum, peaks associated with the normal modes of the Earth can be easily observed. Some of them split visibly due to three factors: rotation, excess ellipticity, and non-spherical material properties (e.g., lateral heterogeneity and anisotropy). The first two effects can be determined accurately using the precisely known rotation rate and ellipticity of the Earth (Woodhouse and Dahlen, 1978), facilitating the isolation of splitting due to internal structure. This process provides “splitting-function coefficients” for every mode which combine to produce the “splitting function” σ(r̂) = s cst Yst (r̂), s=0 t=−s where r̂ denotes a point on the unit sphere, cst is the splitting-function coefficient at spherical degree s and order t, and Yst represents fully-normalized spherical harmonics (Edmonds, 1960). This splitting function represents a radial average of the structure beneath the point r̂ as uniquely sampled by a given mode. It is closely related to surface-wave phase-velocity maps, with the important difference that the splitting function of an isolated mode is limited to even degrees. This is because waves traveling in opposite directions destructively interfere to cancel out the odd-degree signal. Because a splitting function is a radial average of three-dimensional heterogeneity, its coefficients are related to internal properties δm and topography δd by a cst = δmKm δdst Ksd , (1) s dr + 0 m d where Ks denotes the degree-dependent sensitivity kernel of a given mode. The first term on the right-hand side of Eq. (1) is the volumetric contribution with an integration over radius from the center to the surface (a) of the Earth. Both δm and Ksm are dependent on radius and the summation over m represents a sum over material properties, such as lateral variations in velocity or anisotropy. The second term on the right-hand side of Eq. (1) is a contribution from topography δd on various boundaries represented by a sum over d, e.g., the core-mantle boundary and Moho. The sensitivity kernel for such undulations, Ksd , tends to be small, hence this term is often neglected. Because the trade-off between density and topography is not significant (Ishii and Tromp, 2001), we also choose to ignore this term in this study. Previous inferences of density heterogeneity within the mantle are based upon mantle sensitive modes (Ishii and Tromp, 1999; Masters et al., 2000b; Kuo and M. Ishii, J. Tromp / Physics of the Earth and Planetary Interiors 146 (2004) 113–124 Romanowicz, 2002), i.e., modes for which the sensitivity kernels Ksm are zero in the inner core (Fig. 1). In this case, the volumetric term in Eq. (1) is integrated over the mantle, with density δρ, shear and compressional velocities δβ and δα, respectively, as material properties δm in Eq. (1): a β ρ cst = (δβst Ks + δαst Ksα + δρst Ks ) dr, b where b is the radius at the core-mantle boundary. In contrast, inner-core sensitive modes have non-zero sensitivity in the inner core (Fig. 1), and they introduce an additional volumetric term associated with inner-core anisotropy, a β ρ cst = (δβst Ks + δαst Ksα + δρst Ks ) dr b c γ + δγKst dr, (2) 0 γ where c is the radius of the inner-core boundary, and δγ represents parameters which describe inner-core anisotropy (Tromp, 1995). Eq. (2) defines the linear problem dm = Gm m, (3) 115 where dm is the data vector containing splitting-function coefficients from various modes, Gm a matrix containing sensitivity kernels, and m the model vector with mantle and inner-core components, i.e., m = (δβstk δαkst δρstk δγ)T for different values of s and t (T denotes the transpose). The additional superscript k is the index of the radial basis functions within the mantle. In most of this paper, we use Chebyshev polynomials (Su, 1992) up to order 13 (kmax = 13) as the radial basis function. Using both mantle and inner-core modes, we can simultaneously invert for mantle and inner-core structure. In order to better constrain inner-core anisotropy, we include PKP travel times in our data set. The relationship between the anisotropic parameters δγ and travel times is also linear, i.e., dt = Gt m, (4) where the data vector dt contains travel times and Gt describes their sensitivity to anisotropy in the inner core (see Ishii et al., 2002 for a detailed discussion). Combining Eqs. (3) and (4), we have d = Gm, where d = (dTm dTt )T , and G = (GTm GTt )T . Since this is generally an under-determined inverse problem, Fig. 1. Comparison of sensitivity kernels. Sensitivity kernels (K2 ) for isotropic variations in shear and compressional velocities and density for mantle and inner-core sensitive modes. The kernels are for modes 1 S4 , a mantle sensitive mode (left), and 6 S3 , an inner-core sensitive mode (right). Mode 1 S4 is insensitive to structure within the inner core, but is strongly sensitive to lateral variations in shear velocity (dashed curve) in the lower mantle, compressional velocity (dotted curve) near the surface, and density (solid curve) in the upper mantle. On the other hand, 6 S3 has a non-zero kernel within the inner core as well as significant sensitivity to heterogeneity within the mantle, especially to density. 116 M. Ishii, J. Tromp / Physics of the Earth and Planetary Interiors 146 (2004) 113–124 weighting and damping are introduced in the inversions, a detailed description of which can be found in Ishii and Tromp (2001) and Ishii et al. (2002). Inner-core sensitive modes are strongly sensitive to the compressional-velocity structure of the mantle, and we are forced to damp model parameters for compressional velocity twice as hard as shear velocity or density. Otherwise the compressional-velocity model in the mantle acquires a strong zonal pattern which we know to be erroneous based upon other tomographic models. 3. Data In the past several years, various groups have measured mode splitting using a multitude of methods (e.g., Tromp and Zanzerkia, 1995; Resovsky and Ritzwoller, 1998; Masters et al., 2000a). The currently available data set includes isolated spheroidal and toroidal modes, and efforts have been made to determine the splitting of coupled modes (Resovsky and Ritzwoller, 1995), which provide valuable information about the odd-degree structure of the Earth’s interior. The maximum angular degree, s, of the splitting function has been extended to degree 12 (Ritzwoller and Resovsky, 1995), a three-fold improvement compared to earlier studies (e.g., Giardini et al., 1988). From this wealth of data we select splitting coefficients of isolated modes at degrees 2, 4, and 6. We considered higher-degree and coupled-mode measurements to be insufficient at this time for deriving reliable mantle models. These selection criteria reduce the number of modes in our data set to 123 spheroidal and 42 toroidal modes, of which 55 are inner-core sensitive. For some modes, measurements of splitting functions are available from independent research groups. Rather than excluding data, we keep multiple measurements of splitting functions for the same mode by different groups in our data set, much the same way as data for the same ray path are used in body-wave tomography. The data set for this study differs from that of our previous study (Ishii and Tromp, 2001) in that it incorporates measurements from the study of Masters et al. (2000b), which includes surface-wave equivalent as well as other mantle sensitive modes, and inner-core sensitive modes from He and Tromp (1996), Resovsky and Ritzwoller (1998), and Durek and Romanowicz (1999). Before inversion, these data are corrected for crustal structure using model Crust5.1 (Mooney et al., 1998). Because the data set includes splitting coefficients of isolated modes up to and including spherical harmonic degree 6, corresponding mantle models will consist only of the even degrees 2, 4, and 6. The travel-time data included in this study only constrain inner-core anisotropy, hence we omit a discussion of this data set. A detailed description of the data processing applied to the body-wave data can be found in Ishii et al. (2002). 4. Results Statistical results of inversions with different parameterizations within the mantle are presented in Table 1 Mode and travel-time inversions Model χ2 /(N − M) χ2 /(N − M7 ) VR (DF) VR (BC) VR (AB) VR (modes) VR (IC) S SP SPR 3.3 3.0 2.7 3.4 3.0 2.7 84.4 83.2 84.4 71.3 72.2 70.5 74.7 75.0 74.2 89.6 91.1 92.0 76.5 79.9 82.6 Table summarizing the fit for different model parameterizations within the mantle. Inner-core anisotropy is assumed to be uniform. χ2 -tests for the overall fit are given by χ2 /(N − M) and χ2 /(N − M7 ), where N denotes the number of data, and M and M7 are the number of model parameters with kmax = 13 and 7, respectively. VR (DF) is the variance reduction for PKPDF data, VR (BC) the variance reduction for PKPBC − PKPDF data, and VR (AB) the variance reduction for PKPAB − PKPDF data. VR (modes) is the fit to the entire normal-mode data set; VR (IC) the fit to inner-core sensitive modes. The “S” model involves an inversion for mantle and inner-core structure assuming that compressional velocity and density are related to shear velocity through scaling. The scaling values are δ ln ρ = 0.2 δ ln β and δ ln α = 0.55 δ ln β, where ρ is the density, β the shear velocity, and α the compressional velocity. “SP” indicates that data are inverted for shear- and compressional-velocity structure, but density is a scaled shear-velocity model. Finally, the “SPR” model has shear-velocity, compressional-velocity, and density variations. M. Ishii, J. Tromp / Physics of the Earth and Planetary Interiors 146 (2004) 113–124 117 Fig. 2. Models of shear velocity, compressional velocity, density, and bulk-sound velocity within the mantle. Three-dimensional models of the mantle plotted in map view at six discrete depths. Blue indicates regions where the relative perturbation is higher than average and red indicates values that are lower than average. The scale is fixed at a saturation level of ±1% for all maps. These maps are plotted using even spherical harmonic degrees 2, 4, and 6, hence some of the common features seen in tomographic maps such as the contrast between oceans and continents at shallow depths is not obvious. (Shear Velocity) Comparison of the shear-velocity model derived from mantle and inner-core sensitive modes (left) and SKS12WM13 (right). (Compressional Velocity) Comparison of the compressional-velocity model derived from mantle and inner-core sensitive modes (left) and P16B30 (right). (Density) Comparison of density models based upon mantle and inner-core sensitive modes (left) and SPRD6 (Ishii and Tromp, 2001) which is constrained only by mantle sensitive modes (right). (Bulk Sound Velocity) Comparison of bulk-sound velocity models derived using shear- and compressional-velocity models from this study (left) and from SPRD6 (right). Table 1. An increase in the number of parameters from a shear-velocity only (S) to a shear-velocity, compressional-velocity, and density (SPR) inversion is supported by a systematic decrease in χ2 /(N − M), where N is the number of data and M is the number of model parameters. In a previous study, this decrease in χ2 /(N − M) is obtained when the maximum radial parameterization (kmax ) is 7 (Ishii and Tromp, 2001), but the improved modal database supports a decrease in χ2 /(N − M) even when kmax = 13. The fit to the body-wave data does not change more than a couple of percent, suggesting that additional parameters in the mantle do not trade-off significantly with inner-core anisotropy. On the other hand, the fit to inner-core sensitive modes improves with an increased number of mantle parameters, confirming that mantle structure makes a substantial contribution to the splitting coefficients of these modes. In Fig. 2, velocity and density models of the mantle are shown at discrete depths. These models are obtained using the following set of starting models: shear-velocity model SKS12WM13 (Dziewoński et al., 1997), compressional-velocity model P16B30 (Bolton, 1996), and a zero density model. Corresponding statistical results appear in Table 1 under the model name SPR. Fig. 2 also includes plots of bulk-sound velocity obtained using the shear- and compressional-velocity models from the inversion and reference model PREM (Dziewoński and Anderson, 1981). Unlike compressional velocity, bulk-sound velocity has no dependence on shear velocity, which tends to mask anomalies associated with incompressibility in compressional-velocity maps. The models obtained from our inversions are not readily comparable to existing mantle models because they consist only of the even degrees 2, 4, and 6. For example, the ocean-continent distribution observed in all tomographic models near the surface is difficult to see with only even degrees. We therefore compare our models with degrees 2, 4, and 6 of SKS12WM13, 118 M. Ishii, J. Tromp / Physics of the Earth and Planetary Interiors 146 (2004) 113–124 P16B30, the density model of SPRD6 (Ishii and Tromp, 2001), and a bulk-sound velocity model based upon the shear- and compressional-velocity models of SPRD6. In general, these models are similar to SPRD6, with average correlation coefficients between SPRD6 and the new set of models of 0.9, 0.8, and 0.7 for shear velocity, compressional velocity, and density, respectively. Even though the pattern of the density model obtained in this study is similar to that of SPRD6, there are two regions within the middle mantle, at around 1000 and 1800 km depth, where the correlation is relatively low (0.5 and 0.6, respectively). These are noticable in Fig. 2 as slight differences in the density distribution pattern in the maps at depths of 1300 and 1800 km. Furthermore, strong anti-correlation of bulk-sound velocity with shear velocity at the base of the mantle is observed, as well as the regional anti-correlation of shear-velocity and density underneath the central Pacific and Africa. In a previous study based upon mantle-sensitive modes, the amplitude of density was highly uncertain and the addition of gravity data was required to Fig. 4. Correlation coefficients between models of shear and compressional velocities (solid), shear velocity and density (dotted), and compressional velocity and density (dashed) as a function of depth. For the number of free parameters in these models, the correlation is statistically significant at the 90% confidence level if the correlation coefficient is greater than 0.25. Fig. 3. Comparison of the RMS amplitude of the density models from SPRD6 (Ishii and Tromp, 2001; dashed curve), and this study (solid curve). constrain it to a reasonable value (Ishii and Tromp, 2001). Here, we do not use gravity data, yet the amplitude of the density model is comparable to or smaller than that of the compressional-velocity model. In particular, unlike the density model of SPRD6, the new model has smaller root-mean square (RMS) amplitude near the core-mantle boundary. As a consequence, the RMS profile as a function of depth contains two peaks: one within the transition zone and another at around 2300 km depth (Fig. 3). As we demonstrate in the next section, these characteristics are independent of the radial basis functions used in the inversion. As shown in Fig. 4, the correlation between velocities and density is rather poor. Furthermore, some trends observed previously in SPRD6 are strengthened, such as the decorrelation of density and velocities in the transition zone. In fact, the correlation coefficients in these regions suggest that the models may be significantly anti-correlated. In the bulk lower M. Ishii, J. Tromp / Physics of the Earth and Planetary Interiors 146 (2004) 113–124 119 Fig. 5. Map views of laterally varying ratios of density-to-shear-velocity (left; R-to-S), density-to-compressional-velocity (center; R-to-P), and Poisson’s ratio (right) at six discrete depths. Blue indicates regions where the relative perturbation is higher than average and red indicates values that are lower than average. Perturbations from the reference ratio as in PREM (Dziewoński and Anderson, 1981) are shown at 0.05, 0.02, and 0.05% levels for the three ratios, respectively. mantle, the correlations between density and seismic velocities initially decrease monotonically but flatten or increase at the core-mantle boundary. Simple radially-dependent scaling relations between density and velocity have been calculated both theoretically and experimentally (e.g., Anderson et al., 1968; Anderson, 1987; Karato, 1993) and have been used extensively in geodynamic modeling (e.g., Hager and Clayton, 1989; Forte et al., 1994), but their meaning is ambiguous when the models are not highly correlated (Ishii and Tromp, 2001). Therefore, we calculate laterally varying ratios of density-to-shear-velocity, density-to-compressional-velocity, and Poisson’s ratio (Fig. 5). Note that Poisson’s ratio is another representation of the ratio between shear and compressional velocities. The ratios are usually dominated by the pattern of models with stronger lateral variations, although effects due to density variations can be discerned in the density-to-compressional-velocity ratio. Anomalies associated with superplumes clearly stand out in these plots. 5. Discussion The resolution of these mantle models can be addressed most conveniently by looking at the resolution matrix of the inversion (see Ishii and Tromp, 2001, for the definition of the resolution matrix and for various resolution tests). If the model parameters are perfectly resolved, i.e., if no damping is used in the inversion, the resolution matrix becomes the identity matrix, therefore deviations of the resolution matrix from the identity matrix reflect the effects of damping on 120 M. Ishii, J. Tromp / Physics of the Earth and Planetary Interiors 146 (2004) 113–124 Fig. 6. Resolution matrix of the SPR inversion when Chebyshev polynomials are used as the radial basis functions. The elements are arranged by model, i.e., shear (S) velocity, compressional (P) velocity, or density, and in increasing Chebyshev polynomial order (left to right). Parts of the matrix corresponding to model parameters at degree 2 order 1 (top), degree 4 order 1 (middle), and degree 6 order 1 (bottom). M. Ishii, J. Tromp / Physics of the Earth and Planetary Interiors 146 (2004) 113–124 model parameters. In Fig. 6 components of the resolution matrix are shown. Because neither the sensitivity kernels nor the applied damping depend upon angular order t, the resolution for different values of t within the same angular degree is nearly identical. The effect of stronger damping for higher-order Chebyshev polynomials is evident, because higher-order model coefficients are poorly resolved (diagonal elements with high values of k are close to zero). In addition, stronger damping on higher spherical harmonic degrees and the compressional-velocity models manifest themselves as poorer resolution of the related parameters. Fig. 6 also shows that there does not appear to be substantial leakage from model to model (indicated by smaller off-diagonal elements in comparison to diagonal components). To address the effects of off-diagonal terms associated with density model parameters and to understand the full effects of damping and the starting model, we perform a resolution test. In this test, splitting-function coefficients of all modes used in the inversion are calculated using shear- and compressional-wave models SKS12WM13 and P16B30, respectively, a zero model for density, and an inner-core model described in Ishii et al. (2002). These synthetic data (without noise) are inverted with the same damping, weighting, and starting models as in inversions based upon the real data. This test focuses on the leakage of power from velocity heterogeneity to density, because any density variations observed in the retrieved model are due to the inversion process. The RMS amplitudes and correlation between the two density models obtained from the real and synthetic data are compared in Fig. 7. The influence of the velocity models on the density is small, and the density distribution obtained from this resolution test is considerably different from that presented in Fig. 2. Both the RMS amplitude and correlation indicate that the depths affected most strongly by velocity structure are between 2000 and 2500 km. The weak trade-off between the velocity models and density is in contrast with a similar analysis by Kuo and Romanowicz (2002), who used less than 20% of the modes considered in our study. We assume in our analysis that the mantle exhibits only isotropic variations. In the bulk mantle, this assumption is reasonable given that most modes have kernels with broad sensitivity in radius. However, this is not a valid assumption in the upper mantle, where 121 Fig. 7. Resolution test for density model. Comparison of density models obtained from the inversion of data and from a resolution test where shear and compressional velocity models are used as the input models. The RMS amplitudes of the density model (solid curve) and that obtained from the resolution test (dashed curve) are shown in the left panel, and the correlation between the two models is shown in the right panel. there is evidence of significant anisotropic variations (e.g., Ekström and Dziewoński, 1998). To investigate the effects of anisotropy near the Earth’s surface, we perform an inversion without the splitting-function coefficients of surface-wave equivalent modes. Two alterations to the models are observed when we use this subset of the data. The first is a change in pattern. When correlation coefficients are calculated between models with and without the Masters et al. (2000b) data, they are smaller near the surface, as expected. The minimum correlation coefficient is observed for the density models and occurs around 150 km depth. However, it is still large at 0.83, so the effect of pattern modification due to the omission of the surface waves is not very significant, especially in the lower mantle. The second effect is related to the amplitude of the models. Because we have excluded a significant number of data, the constraints on amplitude are weaker and hence the resulting mantle models have amplitudes that are closer to those of Ishii and Tromp (2001). The choice of Chebyshev polynomials as the radial basis function is somewhat arbitrary, although their 122 M. Ishii, J. Tromp / Physics of the Earth and Planetary Interiors 146 (2004) 113–124 Fig. 8. Comparison of density models based upon various radial basis functions. Density models using layers (left), Chebyshev polynomials (center), and cubic B-splines (right) at six discrete depths. Blue indicates regions where the relative perturbation is higher than average and red indicates values that are lower than average. The scale is fixed at a saturation level of ±0.5% for all maps. gradual variation is consistent with the smoothly varying sensitivity kernels. The disadvantage of global basis functions, such as Chebyshev polynomials, is that the termination of polynomials at some order (in our case, kmax = 13) can lead to structure due to “ringing” near the end points (i.e., near the surface and the core-mantle boundary). In what follows, we present results for density models using two local basis functions: layers and cubic B-splines (de Boor, 1978). In these inversions, the number of unknowns in the radial direction is kept constant (i.e., kmax = 13) and both layers and B-spline knots are spaced evenly throughout the mantle. It should be remembered that damping in the radial direction has an entirely different meaning for local and global basis functions. Therefore, damping parameters in the radial direction are chosen so that the traces of the resolution matrices (i.e., the number of resolved model parameters) are similar, while keeping damping in the lateral directions the same. The fits to data with local basis functions are then similar to that obtained with Chebyshev polynomials. In general, the observed patterns in the density distribution are compatible between models with different radial bases, including features near the core-mantle boundary (Fig. 8). The models are least correlated near the surface and the core-mantle boundary, as expected, but the correlation coefficients remain well above the 95% significance level. Furthermore, the amplitudes of these models are relatively consistent with one another (Fig. 9). The two peaks in RMS amplitude in the transition zone and around 2300 km depth are present regardless of the choice of basis function or damping scheme, implying that they are robust features and not the results of “ringing”. M. Ishii, J. Tromp / Physics of the Earth and Planetary Interiors 146 (2004) 113–124 123 resolution matrix indicates that the model parameters are well-resolved, with limited leakage between models. Unlike our previous inversions, where the amplitude of the density model needed to be constrained by gravity data, the amplitude of the new density model is stable using only the modal data set. The density RMS amplitude is generally smaller than that of the velocity models, and exhibits two peaks around 600 and 2300 km depth. These depths coincide with poor or negative correlations between density and velocities. Experiments with local radial basis functions demonstrate that the two maxima in the density RMS amplitude are robust, and that the pattern of the density distribution does not depend on the choice of the basis or damping scheme. The presence of heavier material at the locations of superplumes suggests that these large-scale anomalies differ from the ambient mantle not only in temperature but also in chemistry. Acknowledgements Fig. 9. Comparison of the RMS amplitude of density based upon various radial basis functions. Plot of the RMS amplitudes of density models based upon Chebyshev polynomials (grey curve), cubic B-splines (solid black curve), and layers (dashed black curve). M.I. was supported by a Julie Payette Research Scholarship from the Natural Sciences and Engineering Research Council of Canada. Contribution Number 8887, Caltech Division of Geological and Planetary Sciences. 6. Conclusions We present three-dimensional models of shear velocity, compressional velocity, and density within the mantle based upon a mode data set that includes inner-core sensitive free oscillations. The low-frequency inner-core modes possess substantial sensitivity to the mantle, and we simultaneously invert for mantle heterogeneity and inner-core anisotropy. The mantle models obtained from this inversion do not trade-off significantly with inner-core anisotropy, and the improved database supports inversions for shear-velocity, compressional-velocity, and density structure with a maximum radial expansion of order 13. The models obtained from our inversions exhibit a similar distribution of positive and negative anomalies compared to our previous study, including the regional anti-correlation of velocities and density near the core-mantle boundary. Analysis of the References Anderson, D.L., 1987. A seismic equation of state II. Shear properties and thermodynamics of the lower mantle. Phys. Earth Planet. Inter. 45, 307–323. Anderson, O.L., Schreiber, E., Liebermann, R.C., Soga, N., 1968. Some elastic constant data on minerals relevant to geophysics. Rev. Geophys. 6, 491–524. Bolton, H., 1996. Long Period Travel Times and the Structure of the Mantle. Ph.D. Thesis, U.C. San Diego. Christensen, U., 1984. Instability of a hot boundary layer and initiation of thermo-chemical plumes. Ann. Geophys. 2, 311– 320. Davies, G.F., Gurnis, M., 1986. Interaction of mantle dregs with convection: lateral heterogeneity at the core-mantle boundary. Geophys. Res. Lett. 13, 1517–1520. de Boor, C., 1978. A Practical Guide to Splines. Springer Verlag, New York. Durek, J.J., Romanowicz, B., 1999. Inner core anisotropy inferred by direct inversion of normal mode spectra. Geophys. J. Int. 139, 599–622. 124 M. Ishii, J. Tromp / Physics of the Earth and Planetary Interiors 146 (2004) 113–124 Dziewoński, A.M., 1984. Mapping the lower mantle: determination of lateral heterogeneity in P velocity up to degree and order 6. J. Geophys. Res. 89, 5929–5952. Dziewoński, A.M., Anderson, D.L., 1981. Preliminary reference Earth model. Phys. Earth Planet. Inter. 25, 297–356. Dziewoński, A.M., Liu, X.-F., Su, W.-J., 1997. Lateral heterogeneity in the lowermost mantle. In: Crossley D.J. (Ed.), Earth’s Deep Interior. Gordon and Breach, pp. 11– 50. Edmonds, A.R., 1960. Angular Momentum in Quantum Mechanics. Princeton University Press, Princeton, NJ. Ekström, G., Dziewoński, A.M., 1998. The unique anisotropy of the Pacific upper mantle. Nature 394, 168–172. Forte, A.M., Mitrovica, J.X., 2001. Deep-mantle high-viscosity flow and thermochemical structure inferred from seismic and geodynamic data. Nature 410, 1049–1056. Forte, A.M., Woodward, R.L., Dziewoński, A.M., 1994. Joint inversions of seismic and geodynamic data for models of three-dimensional mantle heterogeneity. J. Geophys. Res. 99, 21857–21877. Giardini, D., Li, X.-D., Woodhouse, J.H., 1988. Splitting functions of long-period normal modes of the Earth. J. Geophys. Res. 93, 13716–13742. Hager, B.H., Clayton, R.W., 1989. Constraints on the structure of mantle convection using seismic observations, flow models, and the geoid. In: Peltier, W.R. (Ed.), Mantle Convection: Plate Tectonics and Global Dynamics. Gordon and Breach, Newark, NJ, pp. 657–763. Hansen, U., Yuen, D.A., 1988. Numerical simulations of thermal-chemical instabilities at the core-mantle boundary. Nature 334, 237–240. He, X., Tromp, J., 1996. Normal-mode constraints on the structure of the Earth. J. Geophys. Res. 101, 20053–20082. Inoue, H., Fukao, Y., Tanabe, K., Ogata, Y., 1990. Whole mantle P-wave travel time tomography. Phys. Earth Planet. Inter. 59, 294–328. Ishii, M., Tromp, J., 1999. Normal-mode and free-air gravity constraints on lateral variations in velocity and density of the Earth’s mantle. Science 285, 1231–1236. Ishii, M., Tromp, J., 2001. Even-degree lateral variations in the mantle constrained by free oscillations and the free-air gravity anomaly. Geophys. J. Int. 145, 77–96. Ishii, M., Tromp, J., Dziewoński, A.M., Ekström, G., 2002. Joint inversion of normal mode and body wave data for inner core anisotropy: 1. Laterally homogeneous anisotropy, J. Geophys. Res. 107, 10.1029/2001JB000712. Karato, S., 1993. Importance of anelasticity in the interpretation of seismic tomography. Geophys. Res. Lett. 20, 1623– 1626. Kuo, C., Romanowicz, B., 2002. On the resolution of density anomalies in the Earth’s mantle using spectral fitting of normal mode data. Geophys. J. Int. 150, 162–179. Li, X.-D., Romanowicz, B., 1996. Global mantle shear-velocity model developed using nonlinear asymptotic coupling theory. J. Geophys. Res. 101, 22245–22272. Li, X.-D., Giardini, D., Woodhouse, J.H., 1991. Large-scale threedimensional even-degree structure of the Earth from splitting of long-period normal modes. J. Geophys. Res. 96, 551–577. Masters, G., Laske, G., Gilbert, F., 2000a. Autoregressive estimation of the splitting matrix of free-oscillation multiplets. Geophys. J. Int. 141, 25–42. Masters, G., Laske, G., Gilbert, F., 2000b. Matrix autoregressive analysis of free-oscillation coupling and splitting. Geophys. J. Int. 143, 478–489. Masters, G., Laske, G., Bolton, H., Dziewoński, A.M., 2000c. The relative behavior of shear velocity, bulk sound speed, and compressional velocity in the mantle: Implications for chemical and thermal structure. In: Karato, S., Forte, A.M., Libermann, R.C., Masters, G., Stixrude, L. (Eds.), Earth’s Deep Interior: Mineral Physics and Tomography From the Atomic to the Global Scale, vol. 117. American Geophysical Union, Washington DC, pp. 63–87. Mooney, W.D., Laske, G., Masters, T.G., 1998. Crust 5.1: a global crustal model at 5◦ × 5◦ . J. Geophys. Res. 103, 727–747. Resovsky, J.S., Ritzwoller, M.H., 1995. Constraining odd-degree Earth structure with coupled free-oscillations. Geophys. Res. Lett. 22, 2301–2304. Resovsky, J.S., Ritzwoller, M.H., 1998. New constraints on deep Earth structure from generalized spectral fitting: application to free oscillations below 3 mHz. J. Geophys. Res. 103, 783–810. Resovsky, J.S., Ritzwoller, M.H., 1999a. A degree 8 mantle shear velocity model from normal mode observations below 3 mHz. J. Geophys. Res. 104, 993–1014. Resovsky, J.S., Ritzwoller, M.H., 1999b. Regularization uncertainty in density models estimated from normal mode data. Geophys. Res. Lett. 26, 2319–2322. Ritzwoller, M.H., Resovsky, J.S., 1995. The feasibility of normal mode constraints on higher degree structures. Geophys. Res. Lett. 22, 2305–2308. Romanowicz, B., 2001. Can we resolve 3D density heterogeneity in the lower mantle? Geophys. Res. Lett. 28, 1107–1110. Su, W.-J., 1992. The Three-Dimensional Shear-Wave Velocity Structure of the Earth’s Mantle. Ph.D. Thesis, Harvard University Tackley, P.J., 1998. Three-dimensional simulations of mantle convection with a thermo-chemical basal boundary layer: D”? In: Gurnis, M. et al. (Eds.), The Core-Mantle Boundary Region, Geodynamics Series, vol. 28. American Geophysical Union, Washington DC, pp. 231–253. Tromp, J., 1995. Normal-mode splitting due to inner-core anisotropy. Geophys. J. Int. 121, 963–968. Tromp, J., Zanzerkia, E., 1995. Toroidal splitting observations from the great 1994 Bolivia and Kuril Islands earthquakes. Geophys. Res. Lett. 22, 2297–2300. van der Hilst, R.D., Widiyantoro, S., Engdahl, E.R., 1997. Evidence for deep mantle circulation from global tomography. Nature 386, 578–584. Woodhouse, J.H., Dahlen, F.A., 1978. The effect of a general aspherical perturbation on the free oscillations of the Earth. Geophys. J. R. Astron. Soc. 53, 335–354.
© Copyright 2026 Paperzz