Roadmap | The review process | Home
   Transition zone
Constraints on the mantle transition zone structure from P-to-Sv converted waves

Benoît Tauzin, Eric Debayle & Gérard Wittlinger

Institut de Physique du Globe de Strasbourg, Ecole et Observatoire des Sciences de la Terre, Centre National de la Recherche Scientifique and Université Louis Pasteur, 67084 Strasbourg, Cedex, France ; ;

1. Introduction

Pressure-induced phase transitions of olivine give a framework for interpreting topography perturbations of mantle transition zone (MTZ) seismic discontinuities in terms of thermal anomalies in the upper mantle. Two techniques are commonly used to observe seismically at a global scale the 410- and 660-km discontinuities: underside reflections of SS precursors (Deuss, 2007; Gu & An, 2007; Ed: See also other Transition Zone pages) and P-to-Sv converted waves (Chevrot & Vinnik, 1999; Lawrence & Shearer, 2006). These two approaches have provided different images of the MTZ, because both types of data sample the MTZ differently and with a different resolution (e.g. Lawrence & Shearer, 2006). SS precursors enable good global coverage of the MTZ, but with lateral resolution limited to horizontal wavelengths longer than 2000 km, which remains insufficient for detecting narrow mantle plumes (Li, 2003). These studies generally suggest that the MTZ is thinner beneath oceanic areas than beneath continents (Gu & Dziewonski, 2002; Li et al., 2000). P-to-S converted waves provide better lateral resolution (few hundred of kilometres) beneath the stations. P-to-S converted waves are, however, less suited to global studies owing to the small-scale sensitivity of Pds and the limited geographic distribution of seismometers, especially in oceanic areas. Recent studies using both types of data suggest that the MTZ is thicker beneath subduction zones, in agreement with the Clapeyron slope of the olivine phase transformations at 410- and 660-km depths. However, evidence for a thin MTZ beneath hotspots is more elusive.

This webpage summarises work presented by Tauzin et al. (2006).

2. Method

We present a new global study of MTZ seismic discontinuities from PdS converted waves at the 410- and 660-km discontinuities. We collected several thousand three-component broadband seismograms from eight global and regional seismic networks. The data set is significantly extended compared to previous studies, with 162 high-quality stations, compared with the 82 used by Chevrot et al. (1999) and the 118 used by Lawrence & Shearer (2006). The data were low pass-filtered at 10 s. We built receiver functions by iteratively deconvolving the Sv trace using the P waveform (Ligorria & Ammon, 1999). Weak 410- and 660-km phases are enhanced by stacking in the time-slowness domain the records having the highest signal-to-noise ratio for each station. In doing this stacking, we assume the mantle has no lateral seismic velocity variations beneath the station. We estimate the data quality from the noise level, the consistency between  the different receiver functions recorded at each station and the focusing of P410s and P660s energy in the τ,p domain.

We define three major tectonic provinces within the MTZ: the "plume", "subduction" and "normal MTZ" regions (Figure 1a).  The plume province is defined from a compilation of 65 hotspot locations by Anderson & Schramm (2005). We consider 21 of our stations to be underlain by the “plume” province, which means that we hope that Pds waves recorded at these stations would detect the influence of a thermal plume on the MTZ thickness. The subduction province is defined from the RUM model of Gudmundsson & Sambridge (1998). Pds waves recorded at 23 stations sample the subduction province. The “normal MTZ” region corresponds to the part of the MTZ which does not belong to the other two provinces. The 119 remaining stations are therefore assumed to belong to the "normal MTZ" group of stations.

Figure 1. (a) Hotspots (black triangles) and major subductions zones (red contours). Depth contours are from Gundmundsson & Sambridge (1998). Hotspots are from Anderson & Schramm (2005). (b) Apparent depth of the "410" seismic discontinuity. Solid circles show topography variations of the 410-km discontinuity. Deviations are shown with respect to IASP91 (410 km). Red (blue) circles indicate a deeper (shallower) "410". The background color represents the radial average of Sv-wave velocity variations in the top 400 km of the DKP2005 model (Debayle et al., 2005). (c) Apparent MTZ thickness. The radius of each circle is proportional to the perturbation from the reference thickness in IASP91 (250 km).

We take into account the error measurements with a bootstrap resampling technique (Efron & Tibshirani, 1991). We investigate in detail whether there are significant differences between stations belonging to the different provinces, consistent with olivine phase transformations.

3. Results

We first convert our measured travel-times into depth using the IASP91 1D velocity model (Kennett & Engdahl, 1991). We observe a strong correlation between the "410" apparent topography and velocity anomalies above the MTZ (Figure 1b). The 410-km discontinuity is deflected downward beneath hotspots and back-arc basins associated with subductions in the northwest Pacific (Figures 1a and 1b). Archean  and Proterozoic regions are generally associated with a shallow apparent 410-km discontinuity. The distribution of apparent "410" depths has a maximum at the expected depth of 410 km (Figure 2a).

Figure 2. (a) Histogram of apparent "410" depths. tP410s-tP travel-times have been converted to depth in IASP91, assuming there is no lateral velocity variations in the mantle. (b) Same histogram except that tP410s-tP travel-times have been corrected for shallow heterogeneities above 410 km using the global shear-wave tomographic model DKP2005 (Debayle et al., 2005). (c) Correlation coefficients between velocity anomalies and uncorrected (dashed) or corrected (solid) 410-km discontinuity topography at each depth in DKP2005 (Debayle et al., 2005).

The apparent "410" depths shown on Figure 1b are not corrected for strong lateral heterogeneities in the top of the upper mantle. However, there is a clear negative correlation between the apparent 410-km-discontinuity topography and lateral variations in seismic velocity in the top 250 km of the mantle, with a maximum of anti-correlation between 100 and 200 km depth (dashed line in Figure 2c). Corrections for the shallow 3D velocity structure are expected to affect strongly the apparent 410- and 660-km depths. In return, the MTZ thickness is weakly sensitive to the strong 3D heterogeneities above 410 km and provide a more robust constraint.   

Figure 1c shows that lateral variations of MTZ thickness are distributed in ± 40 km around 250 km in the IASP91 model. These perturbations are similar to those obtained by Lawrence & Shearer (2006) but significantly greater than the ± 10 km obtained by Chevrot et al. (1999). Other important features of Figure 1c are:

  1. the observation of a thick MTZ beneath subduction zones (northwest Pacific, India, south America and the Farallon plate);
  2. the absence of clear correlation between a thin MTZ and hotspots;
  3. the existence of short-wavelength topography variations in some regions where a dense network of stations are available, e.g., in Europe;
  4. the existence of longer-wavelength topography variations, as evidenced by the gradual westward thinning of the MTZ from the northwest Pacific subduction zones to central Asia.

To provide a more complete view of heterogeneities in MTZ structure, we analyse jointly the apparent "410" depth and MTZ thickness beneath our 162 stations (Figure 3). Measurements at 34 stations (20% of the data set) are within ±10 km of the IASP91 reference values (shown with the vertical and horizontal dashed lines) and are outlined by the grey square. Beneath most of the “hotspot” stations, the 410-km discontinuity is deflected downward, but the apparent MTZ thickness stays within ± 15 km of the reference IASP91 value (250 km). The MTZ is generally thickened beneath "subduction” stations. We also show in Figure 3 the topography variations expected for a range of thermal anomalies having the same amplitude at the 410- and 660-km discontinuities (black line). Topography variations with respect to the IASP91 model are computed using Clapeyron slopes of 3 and -2.5 MPa/K for the 410 km and 660 km olivine phase transformations and assuming that there is no seismic heterogeneity within and above the MTZ. The tendency of our measurements is to show a negative correlation between "410" topography and the MTZ thickness, as predicted from mineral physics. The large scatter of our observations confirms previous studies (Chevrot et al., 1999,  Li et al., 2003), and suggests that Pds absolute and differential travel-times cannot be explained only by the topography of the seismic discontinuities. Other effects, such as 3D velocity variations, need to be considered.

Figure 3. MTZ thickness versus apparent depth of the "410" discontinuity. Travel-times are converted into depth in the 1D model IASP91. Red squares show measurements at the "hotspot", blue triangles at the "subduction" and black dots at the "normal mantle" groups of stations. Grey dashed lines show the "410" depth and MTZ thickness for the IASP91 velocity model. They also delimit four rectangular areas. The upper left area shows the domain where the MTZ is thick and the 410-km discontinuity is shallow, while the lower right area delimits the domain where the MTZ is thin and the 410-km discontinuity deep. The black line shows the topography variations expected for a range of thermal anomalies having the same amplitude at the 410- and 660-km discontinuities. Topography variations are computed for olivine phase transformations using the Clapeyron slopes given in the text.

Velocity heterogeneities above the MTZ affect the absolute travel times of converted waves at both discontinuities and are likely to explain at least one part of the dispersion observed in the apparent depth of the 410-km topography (Figures 2c and 3). We try to reduce this dispersion by taking into account the 3D velocity structure above the MTZ. For this purpose, we use the recent global shear-wave tomographic model DKP2005 by Debayle et al. (2005). This model provides the 3D distribution of shear-wave velocities in the top 400 km of the mantle with a horizontal resolution better than 1000 km beneath continents and a vertical resolution on the order of 50 km. Raw measurements of P410s and P660s travel-times are thus corrected for velocity perturbations relative to IASP91 and converted into depths (Figures 2b and 4).

Figure 4. Same as Figure 3, except that the absolute P410s and P660s travel times have been corrected for velocity heterogeneities above the MTZ using the global shear-wave tomographic model DKP2005 (Debayle et al., 2005), before conversion into depth in IASP91. This correction does not affect the apparent MTZ thickness but changes significantly the apparent "410" depth.

After corrections, the correlation between the "410" apparent topography and the main seismic heterogeneities between 100 and 200 km has almost disappeared (Figure 2c, solid line). The dispersion of the "410" apparent depths is significantly reduced (Figures 2b and 4) and a larger number of measurements are within ± 10 km of the IASP91 value. 54 stations, or 33% of the measurements are within the grey square of Figure 4. The whole range of observations has been shifted to shallower 410 apparent depths. This effect is particularly important for "hotspot" stations, because these hotspots are associated with slow velocities in the top 400 km of the DKP2005 model. Observations at "subduction" and "hotspot" stations are still scattered and do not align on the predicted trend from mineral physics (black line on Figure 4).

4. Discussion

We discuss the apparent depth of the 410-km discontinuity after corrections for 3D shallow heterogeneities, in addition to the MTZ thickness. After 3D shallow corrections, ~30% of our observations are within ± 10 km of the apparent 410-km depth and MTZ thickness in the IASP91 model. This proportion increases to 40%, if we restrict our observations to the normal MTZ subset of stations. The fact that at least 60% of our "normal MTZ" measurements cannot be explained by IASP91 suggests that what we call the “normal MTZ” region is in fact an heterogeneous region. Our coarse a priori regionalization assumes that the mantle is unperturbed away from active slabs and hotspots, without considering the effect of other heterogeneities such as fossil subduction zones, stagnant slabs, or plumes that may not have reached the surface. It is for example possible that the fossil Farallon subduction contributes to the NW-SE pattern of thick MTZ observed beneath the “normal MTZ” stations of north and central America. The average MTZ thickness of our measurements is 253 km while the average apparent depth of the 410-km discontinuity after 3D corrections using DKP2005 is 412 km. It is likely that these average measurements are biased by the uneven sampling of our dataset. Results obtained using SS-precursor data, which provide better sampling of oceanic regions, generally show that the MTZ is thinner beneath oceanic regions (e.g. Gu & Dziewonski, 2002). If correct, because our dataset preferentially samples continents, our average values are likely to represent an upper bound for the apparent 410-km depth and MTZ thickness. For this reason, and because most previous studies have found an average MTZ thickness within ± 10 km of IASP91, we assume that the apparent 410-km depth and MTZ thickness in IASP91 are representative of an unperturbed MTZ.

Two-thirds of the 21 “subduction” stations are within the upper left part of Figure 4, which means that for these stations, the MTZ is thicker and the apparent 410-km depth is shallower than in IASP91, consistent with the Clapeyron slope of the olivine phase transition. The dispersion in our measurements can be explained by insufficient 3D velocity corrections above the MTZ, the presence of significant 3D velocity variations within the MTZ or the fact that slabs produces different thermal anomalies in the vicinity of the piercing points at 410- and 660-km depths.

The 410-km discontinuity is deflected downward for most of the "hotspot" subset of stations, in agreement with the Clapeyron slope of the olivine phase transition at that depth (Figure 4). The deepening is compatible with hot thermal anomalies ranging from 100 to 300 K. For this range of thermal perturbations, pressure-temperature dependence of olivine phase transformations should produce a 15 - 40 km thinning of the MTZ. However, a deep 410-km discontinuity is not always correlated with a thin MTZ in our observations. Mantle plumes are thought to originate from a thermal boundary layer, and it is difficult to imagine that such plumes can originate somewhere between the 410- and 660-km discontinuities, especially as some observations of deep apparent 410-km discontinuities and a thick MTZ require strong cold anomalies at 660 km.

An alternative explanation is to consider that structural transformations of other mantle minerals might contribute to the seismic signal observed for the 660-km discontinuity. Hirose (2002) has shown that the majorite/garnet phase transformation occurs at a depth close to 660 km with a positive Clapeyron slope, in contrast to the negative slope of the postspinel phase transition. This phase transition could dominate the topography of the 660-km discontinuity in high temperature regions (>1800°C) such as plumes, resulting in a greater depth for the 660-km discontinuity. A similar explanation has been proposed by Deuss (2007) to explain SS-precursor observations. If correct, the MTZ thickness may not be a suitable discriminant parameter to detect deep mantle plumes. Because the 410-km discontinuity is not affected by the majorite/garnet phase transformation, the less robust apparent depth of the 410-km discontinuity may provide more important information to detect mantle plumes. Our ability to obtain an accurate absolute topography for the 410-km discontinuity relies on our ability to perform accurate 3D shallow velocity corrections, which is an important issue to address in future.


This work was supported by the program DyeTI of the Institut National des Sciences de l'Univers (INSU) at CNRS and by the young researcher ANR TOMOGLOB. We thank the Iris and Geoscope data centers for providing seismological data.


last updated 27th June, 2007