1 Seismic Velocity Reduction and Accelerated Recovery Due to Earthquakes on the Longmenshan Fault

1 Seismic velocity reduction and accelerated recovery due to earthquakes on the Longmenshan fault

2 Shunping Pei1,2, Fenglin Niu3,4, Yehuda Ben-Zion5, Quan Sun2, Yanbing Liu2, Xiaotian Xue2, Jinrong Su6,

3 Zhigang Shao7

4 1CAS Center for Excellence in Tibetan Plateau Earth Sciences, Chinese Academy of Sciences (CAS), 5 Beijing 100101, China

6 2 CAS Key Laboratory of Continental Collision and Plateau Uplift, Institute of Tibetan Plateau Research, 7 Chinese Academy of Sciences(CAS), Beijing 100101, China

8 3State Key Laboratory of petroleum Resource and Prospecting, and Unconventional Gas Institute, China 9 University of Petroleum at Beijing, Beijing 102249, China

10 4Department of Earth, Environmental and Planetary Sciences, Rice University, 6100 Main Street, 11 Houston, TX 77005, USA

12 5Department of Earth Sciences, University of Southern California, Los Angeles, CA 90089, USA

13 6Earthquake Administration of Sichuan Province, Chengdu 610041, China

14 7Institute of Earthquake Science, China Earthquake Administration, Beijing 100029, China

15 Correspondence and requests for materials should be addressed to F.N. ([email protected]).

17 Various studies report on temporal changes of seismic velocities in the crust and attempt to relate

18 the observations to changes of stress and material properties around faults. While there is growing

19 number of observations on co-seismic velocity reductions, detailed observations of healing phases

20 are generally lacking. Here we report on pronounced co-seismic reduction of velocities around two

21 locked sections (asperities) of the Longmenshan fault with large slip during the 2008 Mw 7.9

22 Wenchuan earthquake, and subsequent healing of the velocities. The healing phase is accelerated

23 significantly at the southern asperity right after the nearby 2013 Mw 6.6 Lushan earthquake. The

24 results are obtained by joint inversions of travel time data at four different periods across the

25 Wenchuan and Lushan earthquakes. The rapid acceleration of healing in response to the Lushan

1 26 earthquake provides unique evidence of the high sensitivity of seismic velocities to stress changes.

27 We suggest that stress redistribution plays an important role in rebuilding fault strength.

29 Earthquakes are caused by the rapid conversion of stresses to inelastic strain (rock damage) along

30 faults (1-10). Recent studies show that fault failure can manifest as a small or large earthquake (11), as

31 aseismic slip (12), or as non-volcanic tremor (13-15). It is also found that fault interactions and other

32 processes can significantly affect the long-term stress build up by plate tectonics (16,17). Stress transfer

33 can be calculated in principle with elastic or viscoelastic modelling (16-21); however, estimating stress

34 changes from data is notoriously difficult, particularly at seismogenic depths. One promising approach is

35 to accurately monitor changes of subsurface seismic velocities (4,22), which are shown in laboratory

36 studies to be sensitive to the stress field (23-25) due to stress-induced changes in properties of cracks.

37 Indeed, there are an increasing number of observations on temporal changes of seismic velocities

38 associated with the occurrence of tectonic events, such as earthquakes (3-8, 26-28) and volcanic eruptions

39 (29,30).

40 The Longmenshan fault zone is located at a pronounced topographic boundary between the eastern

41 margin of the Tibetan plateau and the western Sichuan basin (Figure 1), where elevation changes from Fig. 1

42 ~5000 m to ~500 m within a distance of ~50 km. Geologically, the fault zone manifests itself as the thrust

43 front of the Himalayan orogen and consists of a series of long-angle transpressional faults that extend

44 from southwest to northeast for approximately 300 kilometers. Fault motion is dominated by thrust at the

45 southwestern section and gradually transitions to strike slip at the northeastern end. Over the last decade,

46 two major earthquakes, the 2008 Mw7.9 Wenchuan earthquake and the 2013 Mw6.6 Lushan earthquake,

47 ruptured the northeastern part and the southern end of the fault zone, respectively. The section between

48 with a length of ~60 km remained intact and is associated with seismic risk yet to be determined. The

49 area has been well instrumented before and after the earthquakes, providing unique opportunities to study

50 temporal variations of seismic properties and interaction among different segments of the fault zone.

2 51 The Longmenshan fault zone is in a seismically active region, which is closely monitored by the

52 regional seismic network operated by the Earthquake Administration of Sichuan Province (EASP). The

53 seismicity before the 2008 Mw7.9 earthquake was rather diffuse and spread widely across the entire

54 margin (black crosses in Figure 1b). It was replaced by a much more condensed aftershock seismicity

55 along the Longmenshan fault after the main shock (circles in Figure 1b). These small events are well

56 recorded and located by the EASP seismic network due to the good station coverage in both azimuth and

57 distance.

58 Coseismic velocity reduction during the 2008 Wenchuan Earthquake

59 We selected the first P-wave arrivals in the distance range between 0.1° and 2.0°, which are known as

60 the Pg waves traveling through the upper crust, recorded in the period of 2000-2014. The Pg travel times

61 exhibit a linear relationship with epicentral distance (Supplementary Fig. S1), and the slope of the linear

62 trend corresponds to the average velocity of the upper crust sampled by the source-receiver raypaths. We

63 noticed a small yet systematic change in the slope of the traveltime curve. We organized the traveltime

64 data in a chronological order, and divided the15-year period into time intervals with roughly the same

65 amount of earthquakes. In particular, we used one-year and one-month intervals before and after the

66 Wenchuan earthquake, respectively, due to the large number of aftershocks. Using linear regressions of

67 the Pg traveltimes compiled in each period, we compute the corresponding average P-wave velocities and

68 observe significant variation across the 15-year period (Figure 2a). The estimated average P-wave Fig. 2

69 velocity remains more or less the same at ~6 km/s before the Wenchuan earthquake, and drops abruptly

70 nearly 4% to ~5.75 km/s right after the mainshock. It then rises gradually to ~5.85 km/s before the 2013

71 Mw6.6 Lushan earthquake, where a small coseismic drop is observed. The influence from the Lushan

72 earthquake, however, appears to have very short duration and the recovery of P-wave velocity seems to be

73 present one month after the Lushan earthquake and continues to grow nearly to the level prior to the

74 Wenchuan earthquake.

3 75 To further locate the lateral distribution of the observed average velocity changes, we developed a

76 tomography technique that jointly invert the travel times observed at different time periods for 2-D

77 subsurface velocity changes. We analyzed four time periods that sample right before and after the two

78 earthquakes, which are marked by black solid horizontal lines in Figure 2a. The length of each period is

79 chosen such that the four time periods have roughly the same amount of Pg traveltime data. Because of

80 the large numbers of aftershocks right after the Wenchuan earthquake, a very short time period P2 is

81 sufficient to accumulate the required amount of data, producing a ~3-year gap between P2 and P3. We

82 treated the Pg travel times in each period as independent observations and employed a 2-D traveltime

83 tomography method (31, 32) to jointly invert the data from two consecutive periods for the background

84 velocities of each lateral block, as well as their changes between the two periods (Methods).

85 The co-seismic velocity changes of the Wenchuan earthquake (Figure 3a) were obtained from the

86 joint inversion of the Pg data of the first two time periods, P1 and P2. Large velocity drops are clustered Fig. 3

87 at the Yingxiu town of the Wenchuan County and Beichuan County, which are hereafter referred to as the

88 Wenchuan asperity and Beichuan asperity, respectively. We note that they are spatially coincident with

89 the two areas of large coseismic slip (33-35) and surface deformation (36,37). Results from the joint

90 inversion of Pg data of the two time periods, P2 and P3, reveal postseismic velocity changes that occurred

91 in the four-year period following the mainshock and right before the 2013 Lushan earthquake (Figure

92 3b). During this period, most of the velocity recoveries are observed around the Beichuan asperity,

93 suggesting that fault healing during this period took place at this part of the fault. The Wenchuan

94 asperity, on the other hand, showed very little to no velocity change during this period.

95 Accelerated healing following the 2013 Lushan earthquake

96 The co-seismic velocity changes of the Lushan earthquake inverted from P3 and P4 (Figure 3c)

97 show a large velocity increase in the southern portion of the Wenchuan rupture zone, particularly in the

98 area near the Wenchuan asperity. Velocity changes in other regions including the Beichuan asperity area

99 are insignificant. We further computed the total amount of velocity recovery along the Longmenshan fault

4 100 over the six-year period after the Wenchuan earthquake by jointly inverting the P2 and P4 data (Figure

101 3d). Most of the co-seismic velocity drops of the Wenchuan earthquake that centered at the two asperity

102 zones appear to be nearly recovered during this period.

103 To examine how seismic velocity has evolved in the two asperity regions, we computed the average

104 velocities and their standard deviations using cells in the two black boxes marked in Figure 3. The results

105 are shown in Figure 2b and can be summarized as follows. The Wenchuan earthquake caused a

106 coseismic velocity drop of ~0.2 km/s in both asperity zones. Then the two asperities had very different

107 paths in their recovery process. The Beichuan asperity has been recovering continuously, and does not

108 seem to be affected by the occurrence of the Lushan earthquake. The Wenchuan asperity, on the other

109 hand, had a slow recovery at the beginning, followed by a rapid healing likely triggered by the Lushan

110 earthquake.

111 Checkerboard resolution test indicates that spatial and temporal variations in the velocity structure

112 along the Longmenshan fault zone can be well resolved by the Pg data (Supplementary Fig. S2-S6).

113 Bootstrap analysis (38) also suggests that the observed temporal velocity changes are not produced by

114 difference in earthquake location, source depth and station distribution (Supplementary Fig. S7-S8). The

115 observed coseismic velocity changes along the Longmenshan fault are consistent with previous estimates

116 derived from ambient noise data (8-10). This supports the interpretation that the temporal velocity

117 changes revealed by the Pg data reflect evolution of the Longmenshan fault zone during the major

118 earthquake failures and the initial phase of the following healing and stress buildup.

119 Coseismic velocity changes associated with large earthquakes have been widely observed, especially

120 after the emergence of ambient noise imaging techniques. However, the depth affecting primarily the

121 observed seismic velocity changes is less clear. In general, velocity changes can reflect rock damage

122 caused by strong ground motion, which is expected to occur primarily in the top few hundreds of meters,

123 structural changes within the rupture zone at depth, opening of cracks induced by changes of the stress

5 124 field that can occur at various depth sections, and related changes in the fluid content at different depth

125 sections. Seismic velocity drops associated with shallow rock damage are observed with amplitudes of a

126 few percent to tens of percent (1-7), while stress induced velocity changes are expected to be around a

127 fraction of a percent. Several seismic studies found temporal velocity changes of ~0.4% after the

128 Wenchuan earthquake using ambient noise data (8,9). As these analyses involved 10-25 s Rayleigh

129 waves, the results were interpreted to reflect deformation that extends to seismogenic depth. A recent

130 numerical study (39), however, shows that large changes in the shallow crust can affect Rayleigh wave

131 phase velocities up to 20 s. Therefore it is unclear whether the ~0.4% coseismic velocity drop observed

132 by ambient noise studies truly reflects stress-induced velocity changes at seismogenic depth.

133 Shaking from nearby or even distant great earthquakes can cause shallow fault damage when the fault

134 has been weakened by a large local earthquake. Fault healing was observed to be occurring continuously

135 after the 1992 Landers earthquake, but was temporally interrupted when the nearby Mw 7.1 Hector Mine

136 occurred in 1999 (40). Strength recovery of the 2004 Parkfield earthquake fault also seems to be

137 weakened by the distant Mw9.1 Sumatra earthquake that occurred three months later (41). Our

138 observations of changes along the Longmenshan fault zone, however, reveal a completely opposite

139 scenario. Fault healing can be accelerated by a nearby earthquake, which is difficult to explain by

140 shaking-induced damage on weakened faults.

141 Since most of the earthquakes used in this study occurred between 5 and 20 km, we expect that the

142 observed temporal velocity changes at the asperity zones are not entirely caused by rock damage within

143 the top a few hundreds of meters. To confirm this, we conducted inversions using only data of stations

144 located ~50 km away from the two asperities, and found that significant amount of temporal changes are

145 still imaged at the two asperities (Supplementary Fig. S8). We also conducted a 3D tomographic

146 inversion with the same data and found consistently significant changes in the depth range 2-20 km

147 (Supplementary Fig. S9). The weighted average velocity changes in the 3D tomography are very similar

6 148 to those derived by the 2D inversion. This implies that a 2D inversion, which has less number of

149 parameter, is suitable to image velocity changes with the types of data and geometry of the study area.

150 Stress induced fault strengthening

151 The coseismic velocity drop observed at the Wenchuan and Beichuan asperities could be caused by

152 both strong shaking and stress changes. In particular, stress is expected to play a major role for the

153 observed accelerated healing occurred at the Wenchuan asperity after the Lushan earthquake (Figure 2b).

154 We computed Coulomb stress transfer at the Wenchuan earthquake rupture zone caused by the Lushan

155 earthquake, and found that changes in both the normal and shear stress are nearly negligible (less than 1

156 Pa). Therefore, the observed velocity increase at the Wenchuan asperity is unlikely caused by the

157 coseismic elastic stress transfer due to the slip on the Lushan earthquake fault. It is also implausible that

158 the velocity increase was induced by dynamic stress changes associated with passing seismic waves of the

159 Lushan earthquake, which can be larger than the static Coulomb stress transfer (42), but are expected to

160 have transient rather than residual effects on the medium.

161 We speculate that the required larger stress change may be produced by the adjustment of stress

162 loading between the plateau and the basin along their boundary, the Longmenshan fault, due to the

163 occurrence of the Lushan earthquake. In general, earthquake faults are highly heterogeneous and tend to

164 be supported primarily by a few major locked asperities surrounded by partially creeping regions. The

165 failure of the Lushan asperity likely produced additional aseismic deformation that increased the load

166 carried by the nearby Wenchuan asperity. In other words, the observed sudden velocity increase at the

167 Wenchuan asperity is likely caused by the combined seismic and aseismic afterslip following the Lushan

168 rupture zone. Such changes of deformation should be observable if near-field GPS data across the

169 Longmenshan fault near the Lushan and Wenchuan asperities were available. Interestingly, in-situ stress

170 measurements at a borehole (solid red circle in Figure 3c) located ~35 km north of the Lushan earthquake

171 epicenter suggest that the minimum horizontal stress at ~160 m depth increases from ~5 MPa to ~12 MPa,

172 while the maximum horizontal stress grows from ~5 MPa to ~22 MPa (43). This gives an increase of

7 173 stress by ~7-17 MPa. If we assume that the stress change at the Wenchuan asperity is of the same order,

174 and use observed P-wave velocity change at the Wenchuan asperity of about 1.7%, the calculated

175 velocity-stress sensitivity (dlnV/dP) is around 1.01-2.45x10-9 Pa-1. Laboratory measurements (23-25) and

176 field observations (2,22) indicate that the stress sensitivity of the P-wave velocity at low confining

177 pressures equivalent to shallow depths is in the range 10-10-10-6 Pa-1depending on the crack density in the

178 rocks. It is therefore plausible that the observed velocity increase at the Wenchuan asperity after the

179 Lushan earthquake was produced by stress change likely involving aseismic afterslip around the Lushan

180 asperity.

181 The observed seismic velocity changes between different periods reveal the evolving coupling

182 between the Tibetan plateau and the Sichuan basin along the Longmenshan fault zone during the last 10

183 years (Figure 4). Before the Wenchuan earthquake and during period P1, the two blocks are locked and Fig. 4

184 stress is likely supported by the Beichuan, Wenchuan and Lushan asperities (Figure 4a). In the next

185 period P2 when the Wenchuan earthquake occurred, a large portion of the Longmenshan fault in the north

186 failed, leading to coseismic velocity decrease across all the ruptured section, with two peaks collocated

187 with two areas having large coseismic slips, the Wenchuan and Beichuan asperities (Figure 4b). The

188 seismic velocity drop probably resulted from a combination of increasing rock damage in the fault zone

189 and reduction of normal stress. The Wenchuan earthquake also caused stress increase in the Lushan

190 asperity, which shows a small velocity increase in Figures 2a (blue lines) and 3a. In the post-Wenchuan-

191 earthquake period P3, velocity increase around the Beichuan asperity took place much faster than at the

192 Wenchuan asperity and almost reached the pre-earthquake level before the occurrence of the Lushan

193 earthquake (Figure 4c). In the period P4, shortly after the Lushan earthquake, the velocity recovery

194 around Wenchuan asperity increased rapidly and reached nearly ~75% the pre-earthquake level

195 approximately one year after the Lushan earthquake. At the same time, a coseismic velocity drop

196 occurred around the Lushan asperity (blue line in Figure 2b), suggesting that the southwestern end of the

8 197 Longmenshan fault is now unlocked, while the Wenchuan earthquake section is nearly locked (Figure

198 4d).

199 Our results also imply that the aftershock gap between the Wenchuan and Lushan earthquakes (red

200 ellipse in Figure 1) is unlikely to be a seismic asperity, since this section appears to play little role in the

201 stress redistribution between the Wenchuan and the Lushan segments. We expect the stress redistribution

202 would have been largely shielded if a strong asperity were present between the two seismic segments. A

203 weak gap section is also consistent with recent seismic tomography results (32), which indicate that the

204 gap region is under-laid by a lower seismic velocity compared to the surrounding regions.

205

206 1. Niu, F, Silver, P. G. Nadeau, R. M. & McEvilly. T. V. Migration of seismic scatterers associated 207 with the 1993 Parkfield aseismic transient event, Nature, 426, 544-548, doi:10.1038/nature02151 208 (2003).

209 2. Schaff, D.P. & Beroza, G.C. Coseismic and postseismic velocity changes measured by repeating 210 earthquakes, J. Geophys. Res., 109, B10302, doi:10.1029/2004JB003011 (2004).

211 3. Peng, Z. & Ben-Zion, Y. Temporal changes of shallow seismic velocity around the Karadere-Duzce 212 branch of the north Anatolian fault and strong ground motion, Pure Appl. Geophys. 163, 567-599 213 (2006).

214 4. Niu, F., Silver, P. G., Daley, T. M., Cheng, X. & Majer, E. L. Preseismic velocity changes observed 215 from active source monitoring at the Parkfield SAFOD drill site. Nature 454, 204-208 (2008).

216 5. Brenguier, F. et al. Postseismic relaxation along the San Andreas fault at Parkfield from continuous 217 seismological observations. Science 321, 1478-1481 (2008).

218 6. Wu, C. Q., Peng, Z. G. & Ben-Zion, Y. Non-linearity and temporal changes of fault zone site 219 response associated with strong ground motion. Geophys. J. Int. 176, 265-278 (2009).

220 7. Bonilla, L. F., Guéguen, P. & Ben-Zion, Y. Monitoring co-seismic temporal changes of shallow 221 material during strong ground motion with interferometry and autocorrelation, Bull. Seism. Soc. Am., 222 109, 187-198, doi: 10.1785/0120180092 (2019).

9 223 8. Cheng, X., Niu, F. & Wang, B. S. Coseismic Velocity Change in the Rupture Zone of the 2008 Mw 224 7.9 Wenchuan Earthquake Observed from Ambient Seismic Noise. B. Seismol. Soc. Am. 100, 2539- 225 2550 (2010).

226 9. Froment, B.M., Campillo, M., Chen, J.H. & Liu Q.Y. Deformation at depth associated with the 12 227 May 2008 MW 7.9 Wenchuan earthquake from seismic ambient noise monitoring, Geophys. Res. 228 Lett. 40, 78–82 (2013).

229 10. Obermann, A., et al. Noise correlations to image structural and mechanical changes associated with 230 the Mw 7.9 2008 Wenchuan earthquake. J. Geophys. Res. 119, 3155–3168 (2014).

231 11. Lay, T. & Kanamori. H. Insights from the great 2011 Japan earthquake, Physics Today 64, 33-39 232 (2011).

233 12. Murray, J. R. & Segall, P. Spatiotemporal evolution of a transient slip event on the San Andreas fault 234 near Parkfield, California. J. Geophys. Res. 110, B09407, doi:10:1029/2005JB003651 (2005).

235 13. Obara, K. Nonvolcanic deep tremor associated with subduction in southwest Japan. Science 296, 236 1679–1681 (2002)

237 14. Rogers, G. & Dragert, H. Episodic tremor and slip on the Cascadia subduction zone: The chatter of 238 silent slip. Science 300, 1942–1943 (2003)

239 15. Nadeau, R. M. & Dolenc, D. Nonvolcanic tremors deep beneath the San Andreas fault. Science 307 240 389 (2005).

241 16. Xiong, X., et al., Coulomb stress transfer and accumulation on the Sagaing Fault, Myanmar, over the 242 past 110 years and its implications for seismic hazard, Geophys. Res. Lett. 44, 4781–4789 (2017)

243 17. Freed, A. M. & Lin, J. Delayed triggering of the 1999 Hector Mine earthquake by viscoelastic stress 244 transfer. Nature 411, 180-183 (2001).

245 18. Toda, S., Stein, R. S., Reasonberg, P. A. & Dieterich, J. H., Stress transferred by the Mw=6.5 Kobe, 246 Japan, shock: effect on aftershocks and future earthquake probabilities. J. Geophys. Res. 103, 24543– 247 24565 (1998).

248 19. Stein, R. S. The role of stress transfer in earthquake occurrence. Nature 402, 605-609 (1999).

249 20. Pollitz, F., Vergnolle, M. & Calais E., Fault interaction and stress triggering of twentieth century 250 earthquakes in Mongolia. J. Geophys. Res. 108, 2503, doi:10.1029/2002JB002375 (2003).

251 21. Parsons, T., Ji, C. & Kirby, E. Stress changes from the 2008 Wenchuan earthquake and increased 252 hazard in the Sichuan basin. Nature 454, 509-510 (2008).

10 253 22. Yamamura, K. et al. Long-term observation of in situ seismic velocity and attenuation. J. Geophys. 254 Res. 108, doi:10.1029/2002JB002005 (2003)

255 23. Birch, F. The velocity of compressional waves in rocks to 10 kilobars, part 1. J. Geophys. Res. 65, 256 1083–1102 (1960).

257 24. Birch, F. The velocity of compressional waves in rocks to 10 kilobars, part 2. J. Geophys. Res. 66, 258 2199–2224 (1961).

259 25. Nur, A. & Simmons, G. The effect of saturation on velocity in low porosity rocks. Earth Planet. Sci. 260 Lett. 7, 183–193 (1969).

261 26. Yamada, M., Mori, J. & Ohmi, S. Temporal changes of subsurface velocities during strong shaking 262 as seen from seismic interferometry, J. Geophys. Res. 115, B03302 (2010).

263 27. Sawazaki, K., Sato, H., Nakahara, H. & Nishimura, T. Time-lapse changes of seismic velocity in the 264 shallow ground caused by strong ground motion shock of the 2000 Western-Tottori earthquake, 265 Japan, as revealed from coda deconvolution analysis, Bull. Seismol. Soc. Am. 99, 352-366 (2009).

266 28. Nakata, N. & Snieder, R. Near-surface weakening in Japan after the 2011 Tohoku-Oki earthquake, 267 Geophys. Res. Letters. 38, L17302 (2011).

268 29. Sens-Schönfelder, C. & Wegler, U. Passive image interferometry and seasonal variations of seismic 269 velocities at Merapi volcano, Indonesia, Geophys. Res. Lett. 33, L21302, 2006.

270 30. Brenguier, F., et al. Towards forecasting volcanic eruptions using seismic noise, Nat. Geosci. 1, 126- 271 130 (2008).

272 31. Pei, S. P. & Chen, Y. J. Link between Seismic Velocity Structure and the 2010 Ms 7.1 Yushu 273 Earthquake, Qinghai, China: Evidence from Aftershock Tomography. B. Seismol. Soc. Am. 102, 445- 274 450 (2012).

275 32. Pei, S. P., Zhang, H. J., Su, J. R. & Cui, Z. X. Ductile Gap between the Wenchuan and Lushan 276 Earthquakes Revealed from the Two-dimensional Pg Seismic Tomography. Sci. Rep. 4, 6489 (2014).

277 33. Ji, C. & Hayes, G. Preliminary result of the May 12, 2008 Mw 7.9 eastern Sichuan, China 278 earthquake, http://earthquake.usgs.gov/eqcenter/eqinthenews/2008/us2008ryan/finite_fault.php 279 (2008).

280 34. Feng, G. C., Hetland, E. A., Ding, X. L., Li, Z. W. & Zhang, L. Coseismic fault slip of the 2008 Mw 281 7.9 Wenchuan earthquake estimated from InSAR and GPS measurements. Geophys. Res. Lett. 37, 282 L01302 (2010).

11 283 35. Diao, F., Xiong, X., Wang, R., Zheng, Y. & Hsu, H. Slip model of the 2008 M w 7.9Wenchuan 284 (China) earthquake derived from co-seismic GPS data, Earth Planets Space, 62, 869–874, 2010

285 36. Fu, B. et al. Surface deformation related to the 2008 Wenchuan earthquake, and mountain building 286 of the Longmen Shan, eastern Tibetan Plateau. J. Asian Earth Sci. 40, 805-824 (2011).

287 37. Xu, X. W. et al. Coseismic reverse- and oblique-slip surface faulting generated by the 2008 Mw 7.9 288 Wenchuan earthquake, China. Geology 37, 515-518 (2009).

289 38. Efron, B. & Tibshirani, R. Bootstrap methods for standard errors, confidence intervals, and other 290 measures of statistical accuracy. Statistical Science, 1, 54-77 (1986).

291 39. Yang, C., Li, G., Niu, F. & Ben-Zion, Y. Significant influence of changes in shallow sedimentary 292 structure on phase velocities of 5-25s Rayleigh waves, Geophys. J. Int., 176, doi: 10.1007/s00024- 293 018-2075-7 (2019).

294 40. Vidale, J. E. & Li, Y. G. Damage to the shallow Landers fault from the nearby Hector Mine 295 earthquake. Nature 421, 524-526 (2003).

296 41. Taira, T., Silver, P. G., Niu, F. & Nadeau, R.M. Remote Triggering of Fault-Strength Changes on the 297 San Andreas Fault at Parkfield, Nature 461, 636-639 doi:10.1038/nature08395 (2009).

298 42. Freed, A.M. Earthquake Triggering by Static, Dynamic and Postseismic Stress Transfer, Annu. Rev. 299 Earth Planet. Sci. 33, 335-367 (2005).

300 43. Qiu, J., et al. Comparative analysis of in-situ stress state in the southwestern segment of the 301 Longmenshan fault zone before and after the Lushan earthquake, Acta Geologica Sinica 91, 969-978 302 (2017).

303

304 Correspondence and requests for materials should be addressed to F.N. ([email protected]).

305 ACKNOWLEDGEMENTS

306 We thank the Earthquake Administration of Sichuan Province and the Center of the China Earthquake

307 Networks for providing the seismic data of this study. This work was supported by the Strategic Priority

308 Research Program of Chinese Academy of Sciences (XDA20070302), the National Key R&D Program of

309 China (2016YFC0600303) and the National Natural Science Foundation of China (41674090).

12 310 Author contributions

311 Q.S., Y.L., X.X., J.S. and Z.S. were responsible for collecting the travel time data; S.P. developed the

312 new 4D tomography method and conducted the inversions; S.P., F.N. and Y.B.-Z. contributed to the

313 interpretations and writing; F.N. took the lead on writing the manuscript.

314

315 Competing financial interests

316 The authors declare no competing financial interests.

317

318 Figure 1. Topographic map showing tectonic blocks around the 2008 Mw7.9 Wenchuan

319 Earthquake (WCEQ) and 2013 Ms7.0 Lushan Earthquake (LSEQ). (a) Color thick lines and white

320 thin lines indicate block boundaries and active faults, respectively. (b) Solid blue triangles, crosses, and

321 circles represent, seismic stations, earthquakes occurring before and after the WCEQ. Blue lines indicate

322 active fault, and red lines show the surface traces of the WCEQ. The dashed blue and black boxes in (a)

323 and (b) indicate the two areas used to compute the average velocity changes shown in Figure 2b. The

324 white solid box outlines the Longmenshan fault zone.

325 Figure 2. Temporal variations of crustal velocity structure beneath the Longmenshan fault zone. (a)

326 The averaged velocities calculated in the Longmenshan and Lushan areas are shown in black and blue

327 dots, respectively. Error bars represent the corresponding standard deviations. Four black line segments

328 show four time periods, in which the Pg travel times are used to image the temporal-spatial velocity

329 changes. (b) Observed velocity changes of the Wenchuan (cyan) and Beichuan (blue) asperities are shown

330 together with healing models. The velocities are computed by taking the averages of two square areas

331 shown in Figure 3.

332 Figure 3. Spatiotemporal evolution of velocity structure along the Longmenshan fault zone. (a) Co-

333 seismic velocity changes of the Wenchuan Earthquake are shown in color; red and blue represents areas

13 334 with a velocity decrease and increase, respectively. The dashed line indicates the surface displacements

335 along the faults (thick grey lines). (b) Spatial variations of the post-seismic velocity recovery before the

336 Lushan earthquake. (c) Co-seismic velocity changes of the Lushan Earthquake. The solid red circle

337 indicates the Qiaoqi borehole stress-meter (42). (d) Spatial variations of the post-seismic velocity

338 recovery after the Lushan earthquake. BCA and WCA refer to the Beichuan and Wenchuan asperities,

339 respectively.

340 Figure 4. Evolution diagram of the Longmenshan fault zone in four different stages. (a) The

341 Beichuan, Wenchuan, and Lushan asperities (BCA, WCA, and LSA) were locked before the two

342 earthquakes. (b) The BCA and WCA were broken by the WCEQ and showed significant coseismic

343 velocity drops. The earthquake also caused a slight velocity increase inside the LSA region. (c) In ~5

344 years before the LSEQ, the BCA has mostly healed but the WCA is still under healing. (d) The LSEQ

345 unlocked the LSA, causing a coseismic velocity drop in the asperity area, and a rapid acceleration in the

346 healing at the WCA.

347

348 Methods

349 As all the earthquakes occurred in the upper to middle crust and were recorded at distance less than

350 2°, we first assume that the Pg waves propagate through a straight line connecting the source and the

351 receiver (31, 32). We further project the hypocenters to the epicenters and approximate the raypaths with

352 horizontal lines from epicenters to the receivers. The difference in the true raypath and horizontal raypath

353 is expected to be small when the source depth (h) is much smaller than the epicentral distance (Δ), which

354 is generally true for most of the raypaths used in our inversion. We further make a correction to the

355 observed travel time due to this difference:

1/2 356 t c = t − {}()h2 +Δ2 −Δ ⋅s (1) o o 0

14 c 357 Here to and t0 are the original and corrected Pg travel time, and s0 is the average slowness computed from

358 the linear regression described above. We further introduce a station (ar) and a source term (be) to the

359 corrected observed Pg travel time

c = + + 360 to ttt ar be , (2)

361 where ttt is the true travel times along the horizontal raypath. The station correction term is expected to

362 absorb site effects including instrument uncertainties and anomalies related to the very shallow velocity

363 structure around the station (44). The event correction term mainly takes care of the errors in focal depth

364 and origin time of an earthquake.

365 We further discretize the upper crust of the study area into small 2-D cells, and each cell has a

366 slowness perturbation δsk with respect to the average slowness computed from slope of the Pg travel

367 times. There the traveltime residual for the Pg raypath between event j and station i can be written as:

368 δ = c − = + + ⋅δ , (3) tij to tc ai bj dijk sk k

369 where dijk is the segment length of the Pg ray within the k-th cell, and tc is the computed traveltime. Based

370 on the Pg traveltime data, we set up two sets of linear equations (3) for two different periods. For the

371 second period, we write the slowness perturbation in the second period as the summation of the

372 background perturbation (i.e., the slowness perturbation in the first period, δsk) and a temporal change

373 term, Δsk. More specifically we can express the two sets of equations that share some common unknowns:

δs δs 374 Δs (4) Aa⋅=⋅= Axδδ t B ⋅=⋅= Bx t 111 22  a2 b1  b2

375 Note here the station terms of the two periods (a1 and a2) are solved jointly if the stations used in the two

376 periods are exactly the same. The two sets of the equations in (4) are solved jointly with a spatial

377 smoothing and temporal damping:

λ ⋅L ⋅δs = 0 378 1 (5) λ ⋅Δ = 2 s 0

15 379 Here, L is the Laplacian operator, and λ1 and λ2 are the smoothing and damping parameters, respectively.

380 We then solve the equations (4) and (5) using a LSQR algorithm (45) with a spatial preconditioning to

381 find the least-squares solution that minimizes the following function:

2 2 2 2 382 ⋅ − δ + ⋅ − δ + λ 2 ⋅δ + λ 2 Δ . (6) A x1 t1 B x2 t2 1 L s 2 s

383 As shown in Supplementary Fig. S5-S6, the checkerboard tests indicate that the lateral resolution of

384 our Pg dataset is ~0.25º×0.25º. In order to keep the 3-D blocks having roughly the same size in the depth

385 and lateral directions, we decided to treat the ~20-km upper crust as a single layer and employed a 2-D

386 tomographic inversion in mapping subsurface velocity changes.

387

388 Code availability

389 The codes used to generate individual results are available through the contact information from

390 the original publications. Requests for further materials should be directed to S.P. ([email protected]).

391

392 Data availability

393 The travel time data are provided by the Sichuan Earthquake Administration and the China

394 Earthquake Data Centre, and are available at http://data.earthquake.cn/.

395

396 References

397 44. Pujol, J. Comments on the joint determination of hypocenters and station corrections. B. Seismol. Soc. 398 Am. 78, 1179-1189 (1988).

399 45. Paige, C. C. & Saunders, M. A. LSQR: An algorithm for sparse linear equations and sparse least 400 squares. ACM Trans. Math. Software 8, 43-71 (1982).

16 (a) (b) 33˚ Elevation(m) 4000 40˚ 3000 2000 Alashan Block 1000 Tibetan Plateau 0 Ordos Block 32˚ Qaidam Basin

35˚ Kunlun Eq. M8.1

Yushu Eq. M7.1 31˚ WCEQ Tibetan Plateau Sichuan aftershock Sichuan Basin gap 30˚ Basin M=8 LSEQ M=7 SCTQ 30˚ M=6

M=5 25˚ 90˚ 95˚ 100˚ 105˚ 110˚ 103˚ 104˚ 105˚ 106˚ 2000 2002 2004 2006 2008 2010 2012 2014 (a) 6.1 WCEQ LSEQ

6.0

5.9

5.8

P−wave Velocity (km/s) P1: 2000.01 - 2008.04 P3: 2011.07 - 2012.12 5.7 P2: 2008.05 - 2008.07 P4: 2013.05 - 2014.09 5.6 (b) Beichuan 0.0 asperity

-0.1

Wenchuan -0.2 asperity Vp changes (km/s) WCEQ LSEQ -0.3 2000 2002 2004 2006 2008 2010 2012 2014 Year (a) P2-P1 (b) P3-P2 33˚ 33˚

32˚ BCA BCA 32˚

(m) 6 5 4 3 2 1 0

31˚ WCA WCA 31˚

-0.2 -0.1 0.0 0.1 0.2 -0.2 -0.1 0.0 0.1 0.2 Vp Changes (km/s) Vp Changes (km/s) 30˚ 30˚ (c) P4-P3 (d) P4-P2 33˚ 33˚

32˚ BCA BCA 32˚

WCA 31˚ WCA 31˚ WCEQ Qiaoqi borehole LSEQ -0.2 -0.1 0.0 0.1 0.2 -0.2 -0.1 0.0 0.1 0.2 Vp Changes (km/s) Vp Changes (km/s) 30˚ 30˚ 103˚ 104˚ 105˚ 103˚ 104˚ 105˚ P1: Before Wenchuan Earthquake (a) WCEQ BCA NE Locked LSEQ WCA Locked LSA Locked

P2: After Wenchuan Earthquake (b) WCEQ

Unlocked LSEQ

Unlocked

Locked

Healing Locked

(d) WCEQ P4: After Lushan Earthquake Nearly healed LSEQ

Nearly healed

Unlocked 1 Seismic velocity reduction and accelerated recovery due to earthquakes on the Longmenshan fault

2 Supplementary Information

4 Shunping Pei1, Fenglin Niu2, Yehuda Ben-Zion3, Quan Sun1, Yanbing Liu1, Xiaotian Xue1, Jinrong Su4, Zhigang

5 Shao5

6 1Key Laboratory of Continental Collision and Plateau Uplift, Institute of Tibetan Plateau Research, 7 Chinese Academy of Sciences, Beijing 100101, China

8 2Department of Earth Science, Rice University, 6100 Main Street, Houston, TX 77005, USA

9 3Department of Earth Sciences, University of Southern California, Los Angeles, CA, USA

10 4Earthquake Administration of Sichuan Province, Chengdu 610041, China

11 5Institute of Earthquake Science, China Earthquake Administration, Beijing 100029, China

12 S1. Data Selection and Inversions

13 We selected the Pg-wave traveltime data in the distance of 0.1°-2.0° recorded by the regional seismic 14 network operated by the Earthquake Administration of Sichuan Province in the period between 2000 and 15 2014. Since Pg is the first arrival with relatively high frequency (5-20 Hz), it can be easily picked with an 16 uncertainty less than 0.1 s. To invert for the coseismic and postseismic seismic velocity changes of the 17 Wenchuan earthquake (WCEQ) and Lushan earthquake (LSEQ) along the Longmenshan fault, we 18 selected four periods, which have roughly the similar amount of raypaths (Table S1). As shown in Figure 19 S1, the Pg travel times exhibit a well-defined linear trend in the selected distance range. The average P- 20 wave velocity can be obtained from the slope of the linear trend. To investigate the temporal changes of 21 the average P-wave velocity, we used a one-year and a one-month or two-month interval before and after 22 the Wenchuan earthquake, respectively. The numbers of seismic stations, earthquakes, and raypaths as 23 well as the calculated the average P-wave velocity and one-sigma standard deviation of each period are 24 listed in Table S2.

25 For the 2-D tomographic inversion, the travel time data were divided into four periods (P1, P2, P3, 26 and P4). In all the four periods, the Longmenshan fault zone was well sampled by the selected Pg rays

1 27 (Figure S2). We discretized the study area with 0.05º×0.05º cells, and most of the cells along the 28 Longmenshan fault are well sampled by a few hundreds to a thousand Pg rays (Figure S3).

29 Table S1. Data of the four periods No. No. No. Distance§ Depth¶ Period Time Window Stations Events Raypaths (km) (km) P1 (before WCEQ) 2000.01 - 2008.04 27 1498 11814 108.2±52.8 6.9±4.5 P2 (after WCEQ) 2008.05 - 2008.07 34 2095 14846 92.8±47.9 9.3±7.6 P3 (before LSEQ) 2011.07 - 2012.12 31 1864 17746 85.7±46.5 11.7±4.5 P4 (after LSEQ) 2013.05 - 2014.09 50 1739 19003 82.5±48.4 13.5±4.7 30 §: Average epicentre distance and one-sigma standard deviation; ¶: average focal depth and one-sigma 31 standard deviation.

32 Table S2. The average velocities with error in Figure 2 for WCEQ and LSEQ area.

Year Month No. Stations No. Events No. Raypaths Vp (km/s) Error (km/s)

2001 21 164 1201 6.013 0.020 2002 21 205 1538 5.995 0.017 2003 20 219 1565 6.043 0.018 2004 20 210 1608 5.996 0.016 2005 20 143 1288 6.012 0.017 2006 23 173 1565 5.957 0.014 2007 23 201 1728 6.034 0.014 2008 5 22 1033 6032 5.772 0.004 2008 6 22 778 5646 5.762 0.004 2008 7 34 735 6522 5.755 0.004 2008 8 37 1560 14786 5.746 0.003 2008 9 45 1071 13800 5.749 0.003 2008 10 45 869 10990 5.757 0.003 2008 11 48 747 9793 5.755 0.003 2008 12 39 626 7621 5.741 0.004 2009 1 39 572 6364 5.754 0.004 2009 2 39 470 5555 5.751 0.004 2009 3 39 500 5832 5.745 0.004 2009 4 42 436 5108 5.785 0.005 2009 5 41 398 4475 5.774 0.005 2009 6 38 383 4372 5.786 0.005 2009 7 36 333 3736 5.776 0.005 2009 8 37 285 3363 5.759 0.006 2009 9 36 274 3229 5.784 0.005 2009 10 36 248 2891 5.753 0.006 2009 11 34 243 2701 5.735 0.006 2009 12 36 239 2632 5.759 0.006 2010 1 31 200 2100 5.746 0.007

2 2010 2 34 175 2133 5.806 0.007 2010 3 35 236 2709 5.779 0.006 2010 4 34 181 2161 5.769 0.007 2010 5 34 220 2582 5.808 0.007 2010 6 34 190 2158 5.799 0.007 2010 7 35 226 2552 5.803 0.007 2010 8 31 181 1973 5.834 0.008 2010 9 30 163 1706 5.805 0.008 2010 10 33 170 1697 5.793 0.008 2010 11 28 158 1474 5.817 0.008 2010 12 25 146 1177 5.846 0.010 2011 1 26 162 1501 5.822 0.008 2011 2 26 142 1460 5.816 0.008 2011 3 27 148 1339 5.795 0.009 2011 4 27 149 1399 5.807 0.009 2011 5 27 110 1002 5.783 0.010 2011 6 27 123 1108 5.785 0.010 2011 7 24 113 987 5.777 0.010 2011 8 27 105 876 5.814 0.011 2011 9 26 106 886 5.832 0.011 2011 10 27 100 806 5.843 0.012 2011 11 28 113 885 5.860 0.011 2011 12 28 96 772 5.852 0.012 2012 1 22 146 1281 5.907 0.010 2012 3 27 200 1815 5.814 0.008 2012 5 27 230 2251 5.844 0.007 2012 7 28 211 1953 5.787 0.008 2012 9 28 219 2109 5.831 0.008 2012 11 28 175 1694 5.806 0.008 2013 1 34 180 1977 5.832 0.008 2013 3 50 916 8317 5.725 0.004 2013 5 47 354 4588 5.895 0.004 2013 7 27 227 1971 5.899 0.007 2013 9 27 187 1568 5.886 0.008 2013 11 26 156 1322 5.890 0.008 2014 1 26 128 1325 5.917 0.008 2014 3 26 154 1394 5.897 0.008 2014 5 27 137 1165 5.893 0.008 2014 7 27 130 1137 5.979 0.010 2014 9 28 156 1553 5.907 0.008 2000.01-2008.04* 8 410 1705 6.016 0.048 2008.05-2013.03* 10 285 1491 6.046 0.044 2013.04-2014.12* 22 1126 8057 5.763 0.009 33 *Here for travel time data in LSEQ area (black dashed square in Figure 1b) and the others for WCEQ area 34 (blue dashed square in Figure 1a).

3 36

37 Figure S1. Pg traveltimes of the four periods picked from the local seismic network. The four periods are: P1, 2000.01- 38 2008.04; P2, 2008.05-2008.07; P3, 2011.07~2012.12; P4, 2013.05~2014.09. Traveltime data with residuals <2.0 s are used in 39 inverting spatial and temporal variations of the upper crustal velocity structure around thhe Longmenshan fault zone.

4 40

41 Figure S2. Raypath coverage of each period with crosses and triangles representing the earthquakes and stations, respectively. 42 The pink squares show the Wenchuan asperity (WCA) and Beichuan asperity (BCA), respectively.

5 43

44 Figure S3. Maps showing the density of raypath coverage of the four periodds. The color represents the hit counts of ray path at 45 each cell with size of 3 by 3 minutes. The number of color scale indicates logarithm of the hit counts. The densest ray coverage 46 occurred in Longmenshan faf ult zone at all four periods.

47 S2. Smoothing and damping parameters

48 We employed a smoothing parameter λ1 and a damping parameter λ2 to regularize the spatial (δs) and 49 temporal (Δs) variations of the slowness field, respectively. As we expect that temporal velocity changes 50 outside the Longmenshan fault zone is likely insignificant, therefore we set the damping parameter away

6 51 from the Longmenshan fault zone (cells out of the white box in Figure 1b) four times larger than that in

52 the box (λ2). We employed four different λ1, 200, 300, 500, and 1000, as well as four different λ2, 10, 50, 53 100, and 200, so we conducted a total of 16 joint inversions for the seleccted two periods. The resulting

54 temporal variations are shown in Figure S4. After a series of tests, we found that (λ1, λ2) =(500,50) seems 55 to lead to results with good resolution and stability.

57 Figure S4. Inverted coseismic velocity changes of the WCEQ (difference between P2 and P1) are shown as a function of

58 smoothing (λ( 1, vertical axis) and damping (λ2, horizontal axis) parameters.

7 59 S3. Resolution tests 60 We conducted checkerboard tests to evaluate the spatial resolution of the selected Pg raypath 61 coverage. We created a test checkerboard velocity model by adding sinusoidal velocity perturbations with 62 amplitude of 0.4 km/s to a uniform velocity model with a velocity of 6 km/s. We then computed the travel 63 times using the checkerboard model for all the Pg raypaths used in the real data inversion. We also added 64 Gaussian noise with standard deviation 0.1 s, based on the uncertainty in picking the Pg arrival time to 65 synthetic travel time dataset. Finally we inverted the synthetic travel time dataset for velocity variations in 66 a similar way that we inverted the real data. The inverted velocity perrturbations from a 0.25º×0.25º 67 checkerboard model are shown in Figure S5. We found that our Pg data have a lateral resolution of 68 0.25º×0.25º within the Longmenshan fault zone.

70 Figure S5. Maps showing the results of 0.25º×0.25º checkerboard tests of thhe four periods, which indicate that the checker-board 71 anomalies can be nearly recovered around the Longmenshan fault zone.

8 72 We also conducted checkerboard test using a depth dependent 3-D velocity model and 3-D ray tracing 73 to evaluate the lateral resolution of the Pg dataset. More specifically, we chose a 1-D background model 74 consisting of three layers. The top sediment layer is 2 km thick with a P-wave velocity of 4.8 km/s, and 75 the bottom crystalline layer is 15 km thick with a P-wave velocity of 6.0 km/s. There is a 3-km thick 76 transition layer in the middle with a velocity increasing linearly from 4.8 km/s to 6.0 km/s. This 1-D 77 reference model was constructed based on the tomographic studies of Zhao et al. (1997) and Pei et al. 78 (2010). We then added a depth-independent checkerboard pattern of velocity perturbations to the above 1- 79 D background velocity model to create a 3-D velocity model. The grid size and the amplitude of velocity 80 perturbations are 0.25º×0.25º and ±6%, respectively. We employed the pseudo-bending technique (Zhang 81 and Thurber, 2003; Pei et al., 2010) to calculate the 3-D synthetic traveltimes, which are treated as the 82 travel time data of P2. The 1-D traveltimes of the background velocity model are considered as the data of 83 P1. We then inverted the two synthetic datasets similarly to the real data inversion to obtain the changes 84 between P1 and P2. The result is shown in Figure S6a, which clearly shows that velocity anomalies in the 85 WCA and BCA areas can be well resolved. The inverted checkerboard anomalies from the 3-D synthetic 86 traveltime data are very similar to those inverted from 2-D synthetic traveltimes computed with 87 hypothetic straight-line raypaths (Figure S6b). This 3-D test not only suggests that our data have a lateral 88 resolution of 0.25º×0.25º within the Longmenshan fault zone, but also implies that our 2-D method under 89 the assumption of straight-line Pg raypaths is accurate and valid.

9 90

91 Figure S6. Maps showing the comparison of 0.25º×0.25º checkerboard tests of P2-P1 between the 3-D and 2-D methods. (a) 92 Resolution from a 3-D velocity model using 3-D ray tracing. (b) Resolution from the 2-D method. (c) The histogram of 93 traveltime differences between the 3-D and 2-D method. In general, the travel time differences are insignificant as compared to 94 their residuals, and have a mean of -0.01s and a standard deviation 0.29s.

95 S4. Robustness tests 96 We performed a number of tests to evaluate the stability of the inversion results. We first applied a 97 bootstrap technique to estimate the uncertainties in the inverted spatial and temporal velocity variations. 98 We created a new dataset from the original Pg data by using sampling with replacement, and then 99 inverted the data similarly as the real inversion. We repeaatted this procedure for 100 times, and then 100 computed the standard deviation of all the solutions, whicch is shown in Figure S7. The maximum 101 variation is less than 0.03 km/s, which is far less than the observed temporal velocity changes observed at 102 the Wenchuan and Beichuan asperities (~0.2 km/s).

10 103

104 Figure S7. Maps showing the results of the bootstrap variations of the four periods, which indicate the maximum uncertainty to 105 be less than 0.03 km/s. The error is far less than the observed velocity changes at the Wenchuan and Beichuan asperities, which 106 has amplitude around 0.2 km/s.

107 We further investigated how our results could be affected by data of stations in the two asperity areas. 108 This is done by sequentially removing the traveltime data from stations closest to one of the two asperities 109 before the inversion. The resulting coseismic velocity changes are shown in Figure S8a, and Figure S8b, 110 respectively. The inverted velocity drops at the two asperity areas are still distinct, although their 111 amplitudes are significantly reduced. We also removed data from the newly installed stations after the 112 Wenchuan earthquake and redid the inversion with the same stations across the first and second periods. 113 The inverted velocity changes remain more or less the same (Figure S8c). Figure S8d shows the 114 coseismic velocity changes if we treat all stations in period P1 and P2 as different stations, i.e. using 115 different station terms in these two periods. The fact that the observed velocity changes at the WCA and

11 116 BCA are still robustly shown (Figure S8d) even with introducing time-dependent site corrections suggests 117 that the observed data cannot be explained with such time-dependent site corrections

118

119 Figure S8. Maps showing the coseismic velocity changes using data from different station sets. (a) removing the data recorded 120 by three stations (white triangles) close to the WCA, (b) removing the data recorded by three stations (white triangles) closse to 121 the BCA, (c) Using only same stations (black triangles) in period P1 and P2, (d) Using different station term in period P1 and P2.

122 We also conducted 3D tomography to check our 2D resuult. The complete same data set used in 2D 123 tomography was used to image 3D structure changes beforee and after Wenchuan earthquake. The same 124 strategy was adopted in 3D tomography that station terms and event terms were added in the travel time 125 equations to represent very local structure near stations and earthquake depth or origin time errors, and 126 two regularization parameters were used in LSQR for smoothing lateral and vertiical variations and 127 damping velocity changes between period P1 and P2. The 1D initial model was set to that in 3D 128 checkerboard test and consisted of three layers with the thickkness of 2km, 3km and 15km, respectively. 129 The velocity of these layers was set to constant 4.8km/s, gradient from 4.8 to 6.0km/s and constant

12 130 6.0km/s, respectively. Finally, the velocity changes in all three layers were obtained by a 3D tomographic 131 inversion. The coseismic velocity changes of the three layers are shown in Figure S9a, S9b and S9c, 132 respectively. The Figure S9d show the weighted average velocity changes of three layers byy their 133 thickness. The velocity drops at the two asperity areas, WCA and BCA, are clearly shown in the second 134 and third layers, which have good ray coverage as compareed to the first layer. The weighted average 135 velocity changes, which are primarily inherited from those of the third layer due to its large weighting 136 (75%), are almost the same as those derived from the 2D inversion shown in Figure 3a, suggesting that a 137 2D inversion is suitable to image velocity changes with the tyype of data and geometry of the study area.

138

139 Figure S9. Maps showing the coseismic velocity changes between period P1 and P2 from 3D tomography at different depths: (a) 140 velocity changes at depth of 0~2km, (b) velocity changes at depth of 2~5km, (c) velocity changes at depth of 5~20km. The 141 average velocity changes of the three layers computed by using layer thickness as the weighting factor are shown in (d).

142 Data availability

13 143 Seismological data used in this study are available from http://data.earthquake.cn/.

144 References 145 Zhao, Z., Fang, J., Zhen, S, Hasegawa, A., Horiuchi, S. Crustal structure and accurate hypocenter 146 determination along the Longmenshan fault zone. Acta Seismol. Sin. 6, 761–768 (1997) (in Chinese).

147 Pei, S. et al. Three-dimensional seismic velocity structure across the 2008 Wenchuan Ms8.0 earthquake, 148 Sichuan, China. Tectonophysics 491, 211–217 (2010).

149 Zhang, H. & Thurber, C.H. Double-Difference Tomography: The Method and Its Application to the 150 Hayward Fault, California. B. Seismol. Soc. Am. 93, 1875–1889 (2003)

151