grl50369-sup-0001-auxiliary.txtplain text document, 8K | Supporting Information |

grl50369-sup-0002-figureS1.epsPS document, 4893K | Map showing the distribution of events (green circles) used to generate receiver functions. Yellow star indicates the location of the Ryukyu arc. |

grl50369-sup-0003-figureS2.epsPS document, 1340K | Receiver function backazimuthal gather for northern station TKO. Receiver functions are plotted as described in Figure . Estimated arrival curves (green) are for a *P*-to-*S*conversion from an interface at 45 km depth with a strike of 190° and a dip of 55° |

grl50369-sup-0004-figureS3.epsPS document, 1282K | Receiver function backazimuthal gather for northern station KYK. Receiver functions are plotted as described in Figure . Estimated arrival curves (green) are for a *P*-to-*S*conversion from an interface at 50 km depth with a strike of 200° and a dip of 50° |

grl50369-sup-0005-figureS4.epsPS document, 3376K | Receiver function backazimuthal gather for central station AMM (left) compared to synthetic receiver functions generated from the model shown in Figure for station ZMM (right). Receiver functions are plotted as described in Figure and estimated arrival curves (green) are for a *P*-to-*S* conversion from an interface at 35 km depth with a strike of 210° and a dip of 45° There are several shared features between the synthetic receiver functions and receiver functions from station ZMM that are not apparent in the data from station AMM. This may be due to the changing morphology of the slab, including a slight change in strike (e.g., Figure ) and a change in the depth to the slab (~35 km beneath AMM, opposed to ~45 km beneath ZMM). However, we do see some evidence of matching features, including a transverse component polarity flip at ~4 s and 140‒150° backazimuths (yellow triangle) and a positive-negative(-positive?) polarity reversal at ~6 s (orange line). |

grl50369-sup-0006-figureS5.epsPS document, 3364K | Receiver function backazimuthal gather for central station KGM (left; also in Figure ) compared to synthetic receiver functions generated from the model shown in Figure for station ZMM (right). Receiver functions are plotted as described in Figure and estimated arrival curves (green) are for a *P*-to-*S* conversion from an interface at 40 km depth with a strike of 210° and a dip of 45° We see several commonalities between receiver functions from station KGM and those generated by our model: a transverse component polarity reversal at ~4 s and 140° (yellow triangle), a polarity reversal at ~3 s and 320°‒340° (blue star), and a positive-negative-positive polarity reversal at ~6 s (orange line). |

grl50369-sup-0007-figureS6.epsPS document, 1321K | Receiver function backazimuthal gather for southern station YNG. Receiver functions are plotted as described in Figure . Estimated arrival curves (green) are for a *P*-to-*S*conversion from an interface at 60 km depth with a strike of 270° and a dip of 45° |

grl50369-sup-0008-figureS7.epsPS document, 4994K | Synthetic receiver functions generated using our preferred model and the actual station-event geometry for station ZMM (left), compared to using one representative slowness value (5.5E-5 s/m) for all backazimuths (right). Overall, there is very little difference between the two receiver function gathers, which implies that the effect of varying incidence angle or slowness does not significantly blur the results of backazimuthal gathers. For simplicity in the ensuing synthetic tests, we use this representative slowness value (5.5E-5 s/m) for all backazimuths. |

grl50369-sup-0009-figureS8a.epsPS document, 3338K | Synthetic receiver functions generated using an altered version of our preferred model, in which we vary parameters associated with the anisotropic super-slab layer (UM3). Receiver functions from the unaltered preferred model are overplotted as thick black lines. Specific pulses that are referred to in the figure caption are circled in green. (a) If UM3 is removed completely from the model (but the layer above is made 6 km thicker to ensure that all layer interfaces remain at the same depth), the polarity reversal at ~240° and ~4 s on the transverse component is lost. Additionally, the amplitude of the negative pulse on the radial component at ~5 s and ~90‒180° is greatly diminished. (b) If we change the thickness of the layer from 6 km to 8 km, there emerges a negative pulse doublet on the radial component at ~4 s and backazimuths of 240‒290° that does not exist in the actual data. The same negative pulse is also poorly defined at backazimuths <50° (c) If we weaken the anisotropy in UM3 from −10% to −6%, we see either a complete loss or loss of amplitude in the positive pulse on the transverse component at ~4 s and backazimuths less than 60° (d) Finally, if we vary the tilt of the anisotropic symmetry axis in UM3 by 10°, the emergence of the negative pulse at ~4 s and 240° on the transverse component disappears, instead this pulse appears to be continuous with a negative pulse at lesser backazimuths. Also in the altered version of the model, we see diminished amplitude of the negative pulse at ~4 s and ~120‒180° on the radial component. |

grl50369-sup-0010-figureS8b.epsPS document, 5213K | Supporting information |

grl50369-sup-0011-figureS8c.epsPS document, 3286K | Supporting information |

grl50369-sup-0012-figureS8d.epsPS document, 3403K | Supporting information |

grl50369-sup-0013-figureS9a.epsPS document, 3258K | Synthetic receiver functions generated using an altered version of our preferred model, in which we vary parameters associated with the upper mantle wedge layers UM1 and UM2. Plotting is as described in Figure S8. (a) If we combine layers UM1 and UM2 into one layer with the properties of UM1, this results in a very weak emergence of the negative pulse at ~240° and ~4 s on the transverse component. (b) Similarly, if UM1 and UM2 are merged into one layer with the properties of UM2, the transverse component loses a negative pulse at ~3 s and 10‒50°, and the positive pulse at ~3 s at ~230‒260° |

grl50369-sup-0014-figureS9b.epsPS document, 3334K | Supporting information |

grl50369-sup-0015-figureS10a.epsPS document, 4310K | Synthetic receiver functions generated using an altered version of our preferred model, in which we vary parameters associated with what we interpret to be the subducting oceanic crust. Plotting is as described in Figure S8. (a) Weakening the anisotropy in the subducting oceanic crust from −6% to −2% causes significant variations in *Ps* phase amplitudes, especially on the transverse component ~4‒7 s and backazimuths greater than ~210° (b) Similarly, strengthening the anisotropy from −6% to −10% also causes subtle variations in *Ps* phase amplitude, including an increase in transverse component amplitudes at ~6 s and 250‒320° backazimuths, that is not evident in the data. (c) Aside from some relatively small variations in amplitude, changing the tilt of the anisotropic symmetry axis by 20° results in few changes. However, we do note that the low amplitude positive pulse on the radial component at ~6 s and 90° to 180° backazimuths disappears. |

grl50369-sup-0016-figureS10b.epsPS document, 4076K | Supporting information |

grl50369-sup-0017-figureS10c.epsPS document, 3241K | Supporting information |

grl50369-sup-0018-figureS11.epsPS document, 583K | To ensure that the signal from primary *P*-to-*S* conversions was not contaminated by multiple phases from shallow interfaces, we also examined receiver functions gathered by epicentral distance. One way to distinguish between primary *P*-to-*S* conversions and multiple phases is by examining pulse moveout as a function of epicentral distance. As epicentral distance increases, the delay time of a *P*-to-*S* conversion will decrease. Conversely, as epicentral distance increases, the delay time of a multiple phase will increase. In this figure we show receiver functions gathered by epicentral distance for central Ryukyu station KMU. For phases that we interpret as real *P*-to-*S* conversions (green lines), arrival time either decreases or remains approximately the same as epicentral distance increases. For multiple phases (purple lines), delay time increases with increasing epicentral distance. |

grl50369-sup-0019-TableS1.xlsxExcel 2007 spreadsheet, 46K | Model Parameters Used to Generate the Synthetic Receiver Functions Shown in Figures S8‒S10. Table Parameters are the Same as Described in Table (Main Text). Bold Font Denotes Parameters Whose Values Have Been Deviated From the Preferred Model |

grl50369-sup-0020-TableS2.xlsxExcel 2007 spreadsheet, 48K | Model Parameters Used to Generate the Synthetic Receiver Functions Shown in Figures S8‒S10. Table Parameters are the Same as Described in Table (Main Text). Bold Font Denotes Parameters Whose Values Have Been Deviated From the Preferred Model |

grl50369-sup-0021-TableS3.xlsxExcel 2007 spreadsheet, 47K | Model Parameters Used to Generate the Synthetic Receiver Functions Shown in Figures S8‒S10. Table Parameters are the Same as Described in Table (Main Text). Bold Font Denotes Parameters Whose Values Have Been Deviated From the Preferred Model |

grl50369-sup-0022-TableS4.xlsxExcel 2007 spreadsheet, 47K | Model Parameters Used to Generate the Synthetic Receiver Functions Shown in Figures S8‒S10. Table Parameters are the Same as Described in Table (Main Text). Bold Font Denotes Parameters Whose Values Have Been Deviated From the Preferred Model |

grl50369-sup-0023-TableS5.xlsxExcel 2007 spreadsheet, 47K | Model Parameters Used to Generate the Synthetic Receiver Functions Shown in Figures S8‒S10. Table Parameters are the Same as Described in Table (Main Text). Bold Font Denotes Parameters Whose Values Have Been Deviated From the Preferred Model |

grl50369-sup-0024-TableS6.xlsxExcel 2007 spreadsheet, 47K | Model Parameters Used to Generate the Synthetic Receiver Functions Shown in Figures S8‒S10. Table Parameters are the Same as Described in Table (Main Text). Bold Font Denotes Parameters Whose Values Have Been Deviated From the Preferred Model |

grl50369-sup-0025-TableS7.xlsxExcel 2007 spreadsheet, 48K | Model Parameters Used to Generate the Synthetic Receiver Functions Shown in Figures S8‒S10. Table Parameters are the Same as Described in Table (Main Text). Bold Font Denotes Parameters Whose Values Have Been Deviated From the Preferred Model |

grl50369-sup-0026-TableS8.xlsxExcel 2007 spreadsheet, 46K | Model Parameters Used to Generate the Synthetic Receiver Functions Shown in Figures S8‒S10. Table Parameters are the Same as Described in Table (Main Text). Bold Font Denotes Parameters Whose Values Have Been Deviated From the Preferred Model |

grl50369-sup-0027-TableS9.xlsxExcel 2007 spreadsheet, 47K | Model Parameters Used to Generate the Synthetic Receiver Functions Shown in Figures S8‒S10. Table Parameters are the Same as Described in Table (Main Text). Bold Font Denotes Parameters Whose Values Have Been Deviated From the Preferred Model |

grl50369-sup-0028-supplement_revised2_clean.docxWord document, 110K | Supporting information |