Introduction

PKP precursors were first observed in the 1930's [Gutenberg and Richter, 1934], but it has taken more than sixty years to establish their origin. Array analyses of arrival times, slownesses, and spectra [Cleary and Haddon, 1972] have suggested that these precursors are scattered waves from the lower mantle rather than diffracted, reflected, or refracted waves from the core. Global simulations under the single and multiple scattering hypotheses have determined that small-scale, weak ($<\sim 1\%$) heterogeneities distributed throughout the mantle likely contribute to the PKP precursor wave-trains, with perhaps a concentration in the lowermost mantle [Hedlin et al., 1997].

Small-scale heterogeneities have important geodynamic significance in mantle convection. In particular, subducted slabs can survive for billions of years in the lower mantle due to incomplete mixing, and so regional distributions of small-scale heterogeneity in subduction or upwelling zones might help us sketch out local depth ranges of the mantle flow field and understand better the distribution and nature of heterogeneity. Given the fact that current resolution provided by seismic tomography is not high enough to image structures at scales of $\sim 1-10km$, locating and estimating the size and strength of individual scatterers responsible for PKP precursors provides a potential complementary approach.

Recent studies have derived general properties of the PKP precursor field from the analysis of high quality data from the global seismic network or from large aperture seismic arrays. The large aperture of the arrays considered prevented the use of standard array processing techniques such as the construction of vespagrams. Even when considering stacks across small-aperture arrays such as Norsar, these studies have primarily modelled stacks of the envelopes of the precursor train, and only in a statistical sense. In most cases, these authors have invoked the presence of partial melting associated with Ultra Low Velocity zones to interpret the large velocity contrasts ($\sim 10$% ) necessary to explain the observed precursor amplitudes.

However, very few studies have attempted to locate individual scatterers in the mantle, because PKP precursors are usually weak and their arrivals overlap. Doornbos [1988] tried to locate the scattering regions using the NORSAR seismic array, but he pointed out that the uncertainty in the precursor slowness measurements was unknown. The arrival time of the onset of the precursor train has also been used to try and locate the region of observed strong scattering. An added complication comes from the fact that there is ambiguity between source and receiver side scattering. In general, this is resolved indirectly, by comparing paths in different azimuths from the source or receiver side, and proposing an interpretation most compatible with all observations. Hedlin et al. showed that the ambiguity can be resolved in many regions of the lowermost mantle by inverting a global dataset of precursor average power estimates, in the framework of Rayleigh-Born scattering theory. Finally, even if the slowness and back-azimuth of a precursor can be precisely estimated using a small-aperture seismic array, it is also necessary to know if the precursor was scattered from PKPbc or PKPab on the receiver side or on the source side, in order to uniquely estimate the latitude, longitude, and depth of the corresponding scatterer. Since the amplitude of PKPbc is generally much larger than that of PKPab, it is often assumed that most of the scattering originates on the bc branch. However, until now, it was not possible to demonstrate that explicitely.

Doublet events, for which hypocenters, moment tensors, and source time history are basically identical, provide a powerful means to estimate repeatability of measurements of precursor slowness and back-azimuth. Fortunately, a very high quality earthquake doublet was reported recently [Zhang et al., 2006]. Highly similar waveforms were recorded at 102 stations with a broad coverage of epicentral distances and azimuths, and the hypocenter separation of the two events was estimated to be less than $1.0 km$. Further evidence of the unique quality of this doublet was obtained from the analysis of PP phases, which have identical waveforms in a time interval of at least 70 sec, and well into the PP coda. In this paper, we use this doublet to conduct array analyses of PKP precursors. Taking advantage of an effective stacking technique, we obtain clear and isolated doublet PKP precursors (Figure 1), which, we will argue, originate from individual scatterers in the mantle. The stability of the estimated slowness and back-azimuth enable us to obtain reliable locations of several of these scatterers in the lower mantle.

Figure 2.62: Stacked waveforms of 1993 (blue) and 2003 (red) doublet events filtered between 1 and 2 Hz. (a) Linear stack ing. Waveforms before the dashed line are amplified 10 times. PKP precursor train is apparent, but all precursors are mixed together. (b) Phase Weighted Stacking (PWS). Waveforms before the dashed line are amplified 15 times. Individual precursors A, B, and C stand out. Precursors D and E are likely a mixture of scattered energy arriving sequentially from multiple scatterers. The PKIKP phase arrives after the dashed line.

Figure 2.63: Distribution of seismic scatterers in the lower mantle. (a) Map view indicating the location of two vertical profiles in the region of our study. (b) and (c) Vertical cross-sections of the SAW24B16 shear wave tomographic model [Megnin and Romanowicz, 2000]. Profile XX' is along the great circle containing source and receivers, and profile YY' is perpendicular to it. Red dots show the projected locations of our constrained mantle scatterers into the cross-sections. (d) Map view of the distribution of seismic scatterers in the lower mantle. Black square denotes the Yellowknife seismic arry (YK). Red dots indicate our located scatterers. Solid white lines are the orizontal projections of the ray paths of precursors. The dashed white line is part of the great circle from the sources in South Sandwich Islands (SSI) to YK. The background tomographic model shows the distribution of shear-wave velocity at a depth of $2800 km$.