Figure S1. Ray paths and examples of data analysis. (A) Geometry of ray paths from a deep Philippines earthquake (21724, depth = 555 km) to the USArray. Geometrical ray paths of (black) and SKKS (cyan) assuming the reference model PREM at epicentral distances of to 125 are displayed along with a less conventional path, SP d KS and SKP d S(red). Note the similarity in mantle paths with this phase and at ranges less than 115. (B) Surface projection of the ray paths at two different azimuthal ranges. The USArray stations are marked as black triangles with over 4 stations available. The sampling region at CMB of SP d KS and SKP d S are indicated as blue and green patches. Note that the sampling area in our study is away from the Central Pacific LLSVP. (C) The sampling region of SP d KS (left) and SKP d S (right) at the CMB for three events used in this study. Figure S2. The comparison between data records (black) and MPD simulations (red) for + d phases at different distance ranges. Shaded zones indicate anomalous waveforms, which suggest strong lateral variations. The data in the MPD process are filtered with a bandpass filter (5-5 sec). Figure S3. Examples of the height inversion process. The synthetics (black) library is generated with GRT method for one-side low velocity layer model with different thickness H. In the layer: (A) Vs = -6%, Vp = -2%; (B) Vs = -8%, Vp = -5%. The red traces are the stacked records within azimuth 29 o -37 o as in Figure 2C. For a particular distance, we cross-correlated the data with the synthetics across the library to find the H value with the best cross-correlation coefficient. On the top left, combined heightdistance profiles are displayed. Figure S4. Inverted elevation map of the Low Velocity Zone above the CMB for the three events. The elevation map was migrated to CMB on both source side (top row) and receiver side (bottom row). To accommodate the Fresnel zone, we apply smoothing on the original inverted results with 2.5 deg radius to get the combined map (right column). Figure S5. Validation of synthetics generated with the FD against those from (A) f- k and (B) GRT method. FD agrees with the two analytical-type methods quite well. The GRT results only include multiple reflections in the anomalous layer, which might cause the incomplete results of GRT. (C) displays the 2D and 3D synthetics along profile AA in Figure 4A. 3D synthetics are generated by DWKM method (Helmberger and Ni, 25), which approximates 3D wave propagation by summing the contribution from the Lit and the Diffraction zones. Here we only considered two Lit zones. The top diagram displays the ray paths sampling the structure with the magenta dash lines. In the bottom row, 3D synthetics are the summations of the left Lit and right Lit contributions. Figure S6. 2D synthetics for moving the low velocity structure AA in Figure 4E to source side. Figure S7. (A) Data (black) and PREM synthetics (red) for three events along the
azimuth to the dense network (TriNet) in Southern California. They waveforms appear to be PREM-like as in Figure 2A. (B) The SP d KS entering points (black crosses) at the CMB for all three events. The pink shading area indicates the sampling region with PREM-like records. This area includes the dense dataset from stations at Southern California. The blue dash lines outline the size of the Fresnel zone (FZ) with assuming the source period of 8 secs. The FZ is bigger than the lateral dimension of sampling region. If there s significant structure in the source region, we will predict some disturb waveforms for all azimuths. For the receiver side, the sampling region is much larger than the FZ as in Fig. S1B. If the same size anomaly structure is presented at receiver side, it will only affect profile along particular azimuth. Thus, it appears that the source region is PREM-like. Figure S8. The measured SKKS- differential travel time related to IASP model for the three events. The different colors relate to the azimuths for particular records. Figure S9. Stacked records along two distinct azimuths (green: 29-37 o and red: 43-6 o ) for the three events. The records are aligned on the arrivals. Figure S1. The same stacked records as in Figure S1 but aligned on the SKKS arrivals. Note that the d phase tracks the timing of SKKS very well.
Reference Helmberger, D. V., Ni, S., 25. Approximate 3D body-wave synthetics for tomographic models. Bull. Seismol. Soc. Am. 95, 212-224.
(A) 11 ay 5 Arr 12 US 15 SKKS 15 21724 1 ay Arr 25 US 1 1 11 SKKS S dk SP SKPdS (B) 6 Az = 29-37 4 Az = 43-52 2 Pacific LLSVP 14 16 18 16 14 1 8 (C) 4 21724 4 21729 21724 2917 2 Receiver side Source side 21729 2917 14 16 2 14 1 Figure S1
Distance (o) 113 115 117 119 121-5 5 1 15 2 25-5 5 1 15 2 25-5 5 1 15 2 25-5 5 1 15 2 25-5 5 1 15 2 25 Figure S2
(A) H= km H=1 km H=2 km H=3 km H=4 km Distance ( ) Height 3 6 9 km above CMB H=5 km H=6 km H=7 km H=8 km H=9 km H=1 km Distance ( ) δvs = -6% δvp = -2% 2 4 Figure S3
(B) H= km H=1 km H=2 km H=3 km H=4 km Distance ( ) Height 3 6 9 km above CMB H=5 km H=6 km H=7 km H=8 km H=9 km H=1 km Distance ( ) δvs = -8% δvp = -5% 2 4 Figure S3
21724 21729 2917 Combine three events at source side 36 Japan Source side 32 28 24 2 16 12 128 132 136 14 144 148 152 128 132 136 14 144 148 152 128 132 136 14 144 148 152 128 132 136 14 144 148 152 Receiver side 56 52 48 44 4 36 136 132 128 18 136 132 128 18 136 132 128 18 136 128 Figure S4
(A) Uniform layer: H = 6km, dvs = -1%, dvp = -5% 18 FD f-k 2 2 Figure S5
(B) Source side layer: H = 6km, dvs = -1%, dvp = -5% 18 FD 18 GRT 2 2 Figure S5
(C) 6 Left Lit 56 A Right Lit 52 A 48 44 4 36 136 132 128 18 Left Lit Right Lit 3D (left + Right lit) 2D along AA 16 18 Figure S5
Black: stacked data along AA as in Fig. 4E Red: synthetics for structure along AA at source side Distance ( ) 126 Figure S6
(A) 21724 21729 2917 Figure S7
(B) 35 Japan 3 25 Fresnel zone 2 15 PREM-like 13 135 14 145 15 Figure S7
Azimuth( ) (T SKKS T )(DATA IASP) (s) 4 4 3 4 5 2917 (T SKKS T )(DATA IASP) (s) 4 21729 4 (T SKKS T )(DATA IASP) (s) 4 21724 4 Figure S8
Event 21724 18 SKKS Azimuth 29~37º Azimuth 43~6º 2 4 6 8 1 14 Event 21729 18 SKKS 2 4 6 8 1 14 SKKS Event 2917 18 2 4 6 8 1 14 Figure S9
Event 21724 18 SKKS Azimuth 29~37º Azimuth 43~6º 1 8 6 4 2 2 4 Event 21729 Event 2917 18 SKKS 1 8 6 4 2 2 SKKS 18 4 1 8 6 4 2 2 4 Figure S1