Very early identification of a bimodal frictional behavior during the
post-seismic phase of the 2015 M<sub><i>w</i></sub>8.3 Illapel, Chile, earthquake

Twardzik, Cedric; Vergnolle, Mathilde; Sladen, Anthony; Tsang, Louisa L. H.

doi:https://doi.org/10.5194/se-2021-6

Preprints

https://doi.org/10.5194/se-2021-6

Preprints

05 Feb 2021

Review status: a revised version of this preprint is currently under review for the journal SE.

Very early identification of a bimodal frictional behavior during the post-seismic phase of the 2015 M_w8.3 Illapel, Chile, earthquake

Cedric Twardzik^1,a, Mathilde Vergnolle¹, Anthony Sladen¹, and Louisa L. H. Tsang^1,b Cedric Twardzik et al.

Show author details

¹Université Côte d’Azur, CNRS, Observatoire de la Côte d’Azur, IRD, Geoazur, UMR 7329, Valbonne, France
^anow at: Institut de Physique du Globe de Strasbourg, UMR 7516, Université de Strasbourg, EOST, CNRS, Strasbourg, France
^bnow at: University of Surrey, International Study Centre, Guildford, United Kingdom

Received: 25 Jan 2021 – Accepted for review: 02 Feb 2021 – Discussion started: 05 Feb 2021

Abstract. It is well-established that the post-seismic slip results from the combined contribution of seismic slip and aseismic slip. However, the partitioning between these two modes of slip remains unclear due to the difficulty to infer detailed and robust descriptions of how both evolve in space and time. This is particularly true just after a mainshock when both processes are expected to be the strongest. Using state-of-the-art sub-daily processing of GNSS data, along with dense catalogs of aftershocks obtained from template-matching techniques, we unravel the spatiotemporal evolution of post-seismic slip and aftershocks over the first 12 hours following the 2015 M_w8.3 Illapel, Chile, earthquake. We show that the very early post-seismic activity occurs over two regions with distinct behaviors. To the north, post-seismic slip appears to be purely aseismic and precedes the occurrence of late aftershocks. To the south, aftershocks are the primary cause of the post-seismic slip. We suggest that this difference in behavior could be inferred only few hours after the mainshock, and thus could contribute to a more data-driven forecasts of long-term aftershocks.

Download & links

Preprint (PDF, 11666 KB)

Supplement (24336 KB)

Cedric Twardzik et al.

Status: final response (author comments only)

CC1:
'Comment on se-2021-6', Sylvain Barbot, 11 Feb 2021

Dear Dr. Twardzik,
Your study "Very early identification of a bimodal frictional behavior during the post-seismic phase of the 2015 Mw8.3 Illapel, Chile, earthquake" unravels the evolution of early afterslip during the first 12 hours that follow the 2015 Illapel earthquake. You find two separate regions of afterslip, north and south of the mainshock rupture, that accumulate slip with different mechanisms. The southern region is almost entirely seismic, as shown by the tight correlation with the moment of aftershocks. Indeed, an Mw 7.1 aftershock occurred there in the first 20 minutes following the mainshock and it was followed by an Mw 6.8 aftershock some 5 hours later. When the coseismic offsets of these aftershocks are removed, the geodetic inversion does not resolve significant slip at that location. In the northern patch, there is spatial overlap between afterslip and aftershocks, but no correlation between the evolution of afterslip and the evolution of aftershocks (either considering the number or moment). Hence, these observations provide clear evidence for the spatial heterogeneity of the constitutive behavior. The study provides observational evidence for afterslip to be virtually entirely seismic at some locations. These results are important, providing clear evidence against the common assumption that stable or strengthening frictional regimes are dominant during the postseismic period. In hindsight, this is not surprising because Bath Law indicates that Mw 7+ earthquakes are likely after an Mw 8+ earthquake, but this study sheds nice new light on the phenomenon.
The study is clear and well-targeted. I just have a few minor comments.
Abstract lines 1-3: a lot of "slip" in just a few sentences.
Line 34: unclear what "potentially reducing the propagation of errors" means.
Line 46-48: detection of early aftershocks after the Gorkha earthquake was discussed in the study
Wang, X., Wei, S. and Wu, W., 2017. Double-ramp on the Main Himalayan Thrust revealed by broadband waveform modeling of the 2015 Gorkha earthquake sequence. Earth and Planetary Science Letters, 473, pp.83-93.
Line 64: 15 GNSS stations within 350 km does not sound like much. Discussion of resolution and sensitivity is in order.
Lines 104-113: Not sure why a Monte Carlo sampling method is used here as the problem is entirely linear and can be solved by least squares with Laplacian regularization. It would be useful to document the resolution of the inverse problem or to characterize it with a checkerboard test.
Line 137: I can't recall an example of the opposite. Do we have examples of afterslip distributions that are firmly not time/space separable?
Line 143-144: Note the work of
Salman, R., Hill, E.M., Feng, L., Lindsey, E.O., Mele Veedu, D., Barbot, S., Banerjee, P., Hermawan, I. and Natawidjaja, D.H., 2017. Piecemeal rupture of the Mentawai Patch, Sumatra: the 2008 mw 7.2 North Pagai earthquake sequence. Journal of Geophysical Research: Solid Earth, 122(11), pp.9404-9419.
that indicates afterslip in the area that ruptures coseismically. The study provides a numerical model based on rate-and-state friction that explains the phenomenon by the fact that the coseismic rupture extended significantly into a velocity-strengthening region.
Lines 156-158: Is it possible that this deep slip patch may in fact represent strain on crustal faults above the megathrust?
Line 181: Wasn't the geodetic moment 8E19 Nm after the deep patch is removed? So the seismic moment is actually greater than the geodetic moment? Also, shouldn't the comparison be with the geodetic moment at the time of the Mw 7.1 earthquake instead of at the end of the 12 hours?
Figure 2: It would be useful to show the "time since mainshock" as a second x-axis. Please also indicate the moment magnitude of the two large aftershocks next to their dashed blue lines.
Figures 3 & 5: the repetitive degrees around every subplots are redundant. Consider showing only the left and bottom ones. Consider better showing the trench with the usual chevrons. Indicate the meaning of the blue area in the legend. Add the moment magnitude of the aftershock next to the respective star.
Figure 4: Remove the title "postseismic 12 hours" as it shows afterslip distribution for longer periods.
Figure 7: This should be replaced by a composite with Figures S7.1 and S7.2. The corresponding discussion of the number of aftershocks and the logarithm of the same in the main text is not particularly useful. Instead, focus on the obvious difference between Figures S7.1 and S7.2.
Finally, please consider commenting the phenomenology shown in Figure S7.1. Why is the cumulative moment of aftershocks increasing so much in the northern segment around 6-7 hours? How does that translate in terms of fault slip? It does not seem clear from the various figures. It is hard to tell if the moment is significant because the plots use "normalized" time dependence. Since the geodetic and afterslip moment are so similar to the south, why not using moment (Nm) as the y-axis?
Best wishes,
Sylvain Barbot

Citation: https://doi.org/10.5194/se-2021-6-CC1
- AC1:
  'Reply on CC1', Cédric Twardzik, 16 Jul 2021
  Dear Sylvain Barbot,
  We are very grateful for the community comments that you have made regarding our manuscript se-2021-6 submitted to EGU Solid Earth. You will find below our responses to the various points that you have raised.
  Abstract lines 1-3: a lot of "slip" in just a few sentences.
  
  We have rewritten the beginning of the abstract so that we avoid such repetitions of the word slip.
  Line 34: unclear what "potentially reducing the propagation of errors" means.
  
  What we meant to say was that the approach of forecasting aftershocks using Coulomb stress changes relies on the modeling of the slip distribution as well as the modeling of the Coulomb stress changes. Thus, for the modeling part alone, there are potentially two sources of errors. On the contrary, forecasting aftershocks solely based on afterslip implies only one source of error for the modeling part. We have rephrased that sentence to make that point clearer (line 33-35 of the new manuscript).
  Line 46-48: detection of early aftershocks after the Gorkha earthquake was discussed in the study
  
  Thank you for drawing our attention to that study. We have included it in the article to answer that comment, which has also been raised by reviewer Bernd Schurr (line 51-53 of the new manuscript).
  Line 64: 15 GNSS stations within 350 km does not sound like much. Discussion of resolution and sensitivity is in order. & Lines 104-113: Not sure why a Monte Carlo sampling method is used here as the problem is entirely linear and can be solved by least squares with Laplacian regularization. It would be useful to document the resolution of the inverse problem or to characterize it with a checkerboard test.
  
  We have added a resolution analysis to the Supplementary Material S5. More specifically, we show a map of the resolution in Figure S5.1 (i.e., diagonal elements of the resolution matrix -- Tarantola and Valette 1982). We also show on Figure S5.1 a map of the restitution for the two major patches discussed in the main text to show the potential smearing effect when slip is imaged in these areas.
  Line 137: I can't recall an example of the opposite. Do we have examples of afterslip distributions that are firmly not time/space separable?
  
  After doing some literature review, we have to agree with the reviewer that it is difficult to find examples showing afterslip migration. So far, we could only find one study suggesting hints for afterslip migration : Jiang et al. (2021). Even in that study, the migration is not so much on the causative fault plane, but more migration of afterslip to adjacent faults. Therefore, it is true that the most common observation is that afterslip is stable over time. We have modified the text to reflect that (line 152-155 of the new manuscript).
  Line 143-144: Note the work of Salman, R., Hill, E.M., Feng, L., Lindsey, E.O., Mele Veedu, D., Barbot, S., Banerjee, P., Hermawan, I. and Natawidjaja, D.H., 2017. Piecemeal rupture of the Mentawai Patch, Sumatra: the 2008 mw 7.2 North Pagai earthquake sequence. Journal of Geophysical Research: Solid Earth, 122(11), pp.9404-9419.
  
  We have added that reference in the main text as a potential mechanism for explaining an overlap between co-seismic slip and afterslip (line 162-166 of the new manuscript). However, we want to emphasize that, as stated in the text, our conclusion is that the resolution of our afterslip model along with the variability of coseismic models does not provide enough evidences to conclude with certainty if this is the case or not.
  Lines 156-158: Is it possible that this deep slip patch may in fact represent strain on crustal faults above the megathrust ?
  
  This is an interesting point that is raised here. However, we do not think that this is the case. We have looked at the contribution of this patch alone on the surface displacements (Supplementary Materials S5). When looking at that, we find that this patch do not generate a significant signal at the surface, i.e., displacements with amplitudes larger than the noise level of the time series. In addition, we also provide in Supplementary Material S5 two qualitative arguments that in our opinion reflect the fact that this patch is unreliable : (1) its spatio-temporal evolution is not very stable compared to the main areas of afterslip (see Figure 3 in the main text) and (2) it almost completely disappears when we use the position time series corrected from the two large aftershocks (see Figure 7 in the main text). This is why we have reached the conclusion that this patch is rather uncertain and more likely an artefact from the inversion.
  Line 181: Wasn't the geodetic moment 8E19 Nm after the deep patch is removed? So the seismic moment is actually greater than the geodetic moment? Also, shouldn't the comparison be with the geodetic moment at the time of the Mw 7.1 earthquake instead of at the end of the 12 hours?
  
  ndeed, the estimated geodetic moment over the whole fault is 8.3E19 Nm while the estimated seismic moment over the whole fault is 9.5E19 Nm. We have changed the text to provide a better explanation for that discrepancy (line 203-218 of the new manuscript). In particular, we find that most of the seismic moment is released in the southern patch (~9.0E19 Nm). In that same region the geodetic moment is even lower (~4.5E19 Nm). We provide several explanation for that difference. First, seismic moment of the earthquakes in the GCMT catalog are obtained using PREM for the Earth structure, a model that differs from ours and that can over-estimate the rigidity especially at shallow depths (Beck and Lay , 1999). Then, the Kalman filter used to process the GNSS has been tuned to properly recover slow processes such as afterslip. Therefore, it might not be suited to recover static offsets from large earthquakes, which can distort the recovery of the real ground motion in that case (Choi, 2007). Finally, as pointed out by Konca et al. (2007), there is also a moment-dip trade-off when using near-field geodetic data. Thus, the fact that our fault plane only approximates the real geometry of the Slab1.0 model could also explain the discrepancy.
  Figure 2: It would be useful to show the "time since mainshock" as a second x-axis. Please also indicate the moment magnitude of the two large aftershocks next to their dashed blue lines.
  
  We do not think that it is relevant to show a second x-axis with « time since the mainshock » as the orange dots are by construction n-hours after the mainshock. However, we have added the moment magnitude of the two large aftershocks next to the dashed lines on the plots.
  Figures 3 & 5: the repetitive degrees around every subplots are redundant. Consider showing only the left and bottom ones. Consider better showing the trench with the usual chevrons. Indicate the meaning of the blue area in the legend. Add the moment magnitude of the aftershock next to the respective star.
  
  Done
  Figure 4: Remove the title "postseismic 12 hours" as it shows afterslip distribution for longer periods.
  
  Done
  Figure 7: This should be replaced by a composite with Figures S7.1 and S7.2. The corresponding discussion of the number of aftershocks and the logarithm of the same in the main text is not particularly useful. Instead, focus on the obvious difference between Figures S7.1 and S7.2.
  
  We have taken that into account especially since this was also pointed out by all of the other reviewers. Figure 6 now shows the evolution of the geodetic moment of afterslip along with the seismic moment released by aftershocks. The curves are on the same sub-figures by are displayed using distincts y-axis. That way, they look like normalized so that we can compare the temporal evolution, but the distinct y-scales allow the reader to get the actual value of the moment. We have kept a comparison between the geodetic moment of afterslip and the time evolution of the number of aftershocks (Figure 8) as we still discuss that relationship in the main text.
  Finally, please consider commenting the phenomenology shown in Figure S7.1. Why is the cumulative moment of aftershocks increasing so much in the northern segment around 6-7 hours? How does that translate in terms of fault slip? It does not seem clear from the various figures. It is hard to tell if the moment is significant because the plots use "normalized" time dependence. Since the geodetic and afterslip moment are so similar to the south, why not using moment (Nm) as the y-axis?
  
  We think that this comment comes from the fact that the curves were normalized giving the false impression of a very large increase of seismic moment. We believe that with Figure 6 in the new manuscript, the fact that this increase is in fact very small should be more obvious.
  
  Citation: https://doi.org/10.5194/se-2021-6-AC1
- AC5:
  'Reply on CC1', Cédric Twardzik, 18 Jul 2021
  Dear Sylvain Barbot,
  We are very grateful for the community comments that you have made regarding our manuscript se-2021-6 submitted to EGU Solid Earth. You will find below our responses to the various points that you have raised. We have attached as a .zip file an annotated pdf of the manuscript so that changes can be tracked as well as an updated version of the Supplementary Materials.
  Abstract lines 1-3: a lot of "slip" in just a few sentences.
  
  We have rewritten the beginning of the abstract so that we avoid such repetitions of the word slip.
  Line 34: unclear what "potentially reducing the propagation of errors" means.
  
  What we meant to say was that the approach of forecasting aftershocks using Coulomb stress changes relies on the modeling of the slip distribution as well as the modeling of the Coulomb stress changes. Thus, for the modeling part alone, there are potentially two sources of errors. On the contrary, forecasting aftershocks solely based on afterslip implies only one source of error for the modeling part. We have rephrased that sentence to make that point clearer (line 34-36 of the new manuscript).
  Line 46-48: detection of early aftershocks after the Gorkha earthquake was discussed in the study
  
  Thank you for drawing our attention to that study. We have included it in the article to answer that comment, which has also been raised by reviewer Bernd Schurr (line 52-54 of the revised manuscript).
  Line 64: 15 GNSS stations within 350 km does not sound like much. Discussion of resolution and sensitivity is in order. & Lines 104-113: Not sure why a Monte Carlo sampling method is used here as the problem is entirely linear and can be solved by least squares with Laplacian regularization. It would be useful to document the resolution of the inverse problem or to characterize it with a checkerboard test.
  
  We have added a resolution analysis to the Supplementary Material S5. More specifically, we show a map of the resolution in Figure S5.1 (i.e., diagonal elements of the resolution matrix -- Tarantola and Valette 1982). We also show on Figure S5.1 a map of the restitution for the two major patches discussed in the main text to show the potential smearing effect when slip is imaged in these areas.
  Line 137: I can't recall an example of the opposite. Do we have examples of afterslip distributions that are firmly not time/space separable?
  
  After doing some literature review, we have to agree with the reviewer that it is difficult to find examples showing afterslip migration. So far, we could only find one study suggesting hints for afterslip migration : Jiang et al. (2021). Even in that study, the migration is not so much on the causative fault plane, but more migration of afterslip to adjacent faults. Therefore, it is true that the most common observation is that afterslip is stable over time. We have modified the text to reflect that (line 157-161 of the revised manuscript).
  Line 143-144: Note the work of Salman, R., Hill, E.M., Feng, L., Lindsey, E.O., Mele Veedu, D., Barbot, S., Banerjee, P., Hermawan, I. and Natawidjaja, D.H., 2017. Piecemeal rupture of the Mentawai Patch, Sumatra: the 2008 mw 7.2 North Pagai earthquake sequence. Journal of Geophysical Research: Solid Earth, 122(11), pp.9404-9419.
  
  We have added that reference in the main text as a potential mechanism for explaining an overlap between co-seismic slip and afterslip (line 172 of the new manuscript). However, we want to emphasize that, as stated in the text, our conclusion is that the resolution of our afterslip model along with the variability of coseismic models does not provide enough evidences to conclude with certainty if this is the case or not.
  Lines 156-158: Is it possible that this deep slip patch may in fact represent strain on crustal faults above the megathrust ?
  
  This is an interesting point that is raised here. However, we do not think that this is the case. We have looked at the contribution of this patch alone on the surface displacements (Supplementary Materials S5). When looking at that, we find that this patch do not generate a significant signal at the surface. In addition, we also provide in Supplementary Material S5 two qualitative arguments that in our opinion reflect the fact that this patch is unreliable : (1) its spatio-temporal evolution is not very stable compared to the main areas of afterslip (see Figure 3 in the main text) and (2) it almost completely disappears when we use the position time series corrected from the two large aftershocks (see Figure 7 in the main text). This is why we have reached the conclusion that this patch is rather uncertain and more likely an artefact from the inversion.
  Line 181: Wasn't the geodetic moment 8E19 Nm after the deep patch is removed? So the seismic moment is actually greater than the geodetic moment? Also, shouldn't the comparison be with the geodetic moment at the time of the Mw 7.1 earthquake instead of at the end of the 12 hours?
  
  Indeed, the estimated geodetic moment over the whole fault is 8.3E19 Nm while the estimated seismic moment over the whole fault is 9.5E19 Nm. We have changed the text to provide a better explanation for that discrepancy (line 203-218 of the new manuscript). In particular, we find that most of the seismic moment is released in the southern patch (~9.0E19 Nm). In that same region the geodetic moment is even lower (~4.5E19 Nm). We provide several explanation for that difference. First, seismic moment of the earthquakes in the GCMT catalog are obtained using PREM for the Earth structure, a model that differs from ours and that can over-estimate the rigidity especially at shallow depths (Bilek and Lay , 1999). Then, the Kalman filter used to process the GNSS has been tuned to properly recover slow processes such as afterslip. Therefore, it might not be suited to recover static offsets from large earthquakes, which can distort the recovery of the real ground motion in that case (Choi, 2007). Finally, as pointed out by Konca et al. (2007), there is also a moment-dip trade-off when using near-field geodetic data. Thus, the fact that our fault plane only approximates the real geometry of the Slab1.0 model could also explain the discrepancy. We have added this discussion in the main text (lein 210-231 in the revised manuscript)
  Figure 2: It would be useful to show the "time since mainshock" as a second x-axis. Please also indicate the moment magnitude of the two large aftershocks next to their dashed blue lines.
  
  We do not think that it is relevant to show a second x-axis with « time since the mainshock » as the orange dots are by construction n-hours after the mainshock. However, we have added the moment magnitude of the two large aftershocks next to the dashed lines on the plots.
  Figures 3 & 5: the repetitive degrees around every subplots are redundant. Consider showing only the left and bottom ones. Consider better showing the trench with the usual chevrons. Indicate the meaning of the blue area in the legend. Add the moment magnitude of the aftershock next to the respective star.
  
  Done
  Figure 4: Remove the title "postseismic 12 hours" as it shows afterslip distribution for longer periods.
  
  Done
  Figure 7: This should be replaced by a composite with Figures S7.1 and S7.2. The corresponding discussion of the number of aftershocks and the logarithm of the same in the main text is not particularly useful. Instead, focus on the obvious difference between Figures S7.1 and S7.2.
  
  We have taken that into account especially since this was also pointed out by all of the other reviewers. Figure 6 in the revised manuscript shows the evolution of the geodetic moment of afterslip along with the seismic moment released by aftershocks. The curves are on the same sub-figures by are displayed using distincts y-axis. That way, they look like normalized so that we can compare the temporal evolution, but the distinct y-scales allow the reader to get the actual value of the moment. We have kept a comparison between the geodetic moment of afterslip and the time evolution of the number of aftershocks (Figure 8) as we still discuss that relationship in the main text.
  Finally, please consider commenting the phenomenology shown in Figure S7.1. Why is the cumulative moment of aftershocks increasing so much in the northern segment around 6-7 hours? How does that translate in terms of fault slip? It does not seem clear from the various figures. It is hard to tell if the moment is significant because the plots use "normalized" time dependence. Since the geodetic and afterslip moment are so similar to the south, why not using moment (Nm) as the y-axis?
  
  We think that this comment comes from the fact that the curves were normalized giving the false impression of a very large increase of seismic moment. We believe that with Figure 6 in the new manuscript, the fact that this increase is in fact very small should be more obvious.
  
  Citation: https://doi.org/10.5194/se-2021-6-AC5
RC1:
'Comment on se-2021-6', Bernd Schurr, 04 Mar 2021

Twardzik and co-authors investigate the early, i.e. first 12 h of co-seismic deformation following the 2015 M8.3 Illapel Chile earthquake. For this purpose they use 2 already published data sets: 1) 30 sec sampled position time series from 15 cont. GNSS sites earlier published by the authors (Twardzik et al. 2019) and 2 earthquake catalogs based on template matching to obtain high completeness even during the early phase of the aftershock sequence, published by others (Huang et al. and Frank et al. both 2017). The authors derive hourly displacements from the position time series and invert them for hourly afterslip maps. These are compared to aftershock catalogs to distinguish between seismic and aseismic deformation during the early post-seismic phase. They find that the cumulative afterslip during the first 12 h skirts around the lower end of co-seismic slip, a pattern that persists also during the later and longer post-seismic period. There are two main lobes north and south of the co-seismic slip region that behave differently. The northern patch apparently slips mostly aseismically, whereas for the southern patch moment release is dominated by aftershocks (incl. an M7.1 and M6.8 aftershock).

The manuscript is well written and well illustrated. I have only few minor points I comment in the following.

l.58 “seismic noise” – I think this is not really seismic noise but overwhelming signal, i.e. numerous often simultaneous aftershocks, that is causing problems to most detectors.

l.28: change “activity” to “deformation”. I think there is a “itself” missing after “express”.

l.49: change “highly” to “more”

l.89ff: “The cumulative surface displacements are calculated at every hour since the mainshock origin time by computing the average positions over a 1-hour time window centered on the time of interest.” Does cumulative refer here for cumulative during the one hour processed or cumulative since the mainsock. I assume the earlier, but please clarify.

l.126: “yellow circles” should be “purple stars”?

l.129: There is a word missing after “second”. Maybe “patch”.

l.140ff: “When we look more closely, we see that some of the post-seismic slip might have penetrated inside the co-seismic rupture area (Figure 3).”

The general fuzziness of both co-seismic and post-seismic slip models makes this assertion difficult to maintain (and the authors actually relativize it later in the paragraph). In particular, different fault model used (simple plain slab like the authors or varying slab dip based on e.g. slab2.0) in the modeling will shift location of slip. To start interpreting this, at least the modeling set-up of co- and post-seismic slip should be the same.

l.186ff: “Our models show that post-seismic slip is now only observed north of the co-seismic rupture area and that the patch to the south has completely vanished.” And also Figure 3 and Figure 6.

First of all, slip in Fig. 3 and 6 I assume is the slip during the respective hour, not the cumulative slip added up also from the previous hours (must be based on the amplitudes and the fact that some patches vanish). I don’t understand why, if only the offsets of the 2 largest aftershocks are corrected (hour 1 and hour 5), all slip vanishes in the southern patch also during the other hours. Please explain.

Fig. S7.1: The second M6.8 aftershock occurs during hour 5 and clearly shows up as a step in the graphed moment. However a step in slip seems to occur mainly in hour 6? Is this an averaging effect?

l.197ff: Please elaborate in one or two sentences what this predicted accelation phase signifies.

l.201ff: Please mention where the cited studies were based (Ecuador and Japan).

l.209: Change “rate-and-state law” to “rate-and-state friction law” here and everywhere else.

l.257: I wonder, are aftershocks anywhere actually operationally forecasted based on some models (maybe a citation would be good)? If so, I would assume that models have to be simple and robust. Here e.g. CFS would naturally predict aftershocks around the co-seismic rupture area, where they do occur, for the Illapel eqk and also for many other subduction zone earthquakes. I wonder, how realistic and it is to actually do the hindsight analysis outlined here in near real time and if it really adds value. Of course, this could be tested.

l.271ff: “Our additional finding is that the slip patterns that we observe after 12 hours persists over the first 2 months. When that is the case, information about very-early post-seismic slip can help to characterize longer-lasting post-seismic slip, which can prove to be useful to include for the forecast of aftershocks locations.”

But can this really be generalized?

Citation: https://doi.org/10.5194/se-2021-6-RC1
- AC2:
  'Reply on RC1', Cédric Twardzik, 17 Jul 2021
  Dear Bernd Schurr,
  
  We would like to thank you for reviewing our manuscript se-2021-6 submitted to EGU Solid Earth. Below are our responses to the different points that you have made. We have attached as a .zip file an annotated pdf of the manuscript so that changes can be tracked as well as an updated version of the Supplementary Materials.
  
  l.58 “seismic noise” – I think this is not really seismic noise but overwhelming signal, i.e. numerous often simultaneous aftershocks, that is causing problems to most detectors.
  
  We have changed that part to better highlight the issues of detecting aftershocks right after a large mainshock (line 52-54 of the revised manuscript)
  
  l.28: change “activity” to “deformation”. I think there is a “itself” missing after “express”.
  
  Done
  
  l.49: change “highly” to “more”
  
  Done
  
  l.89ff: “The cumulative surface displacements are calculated at every hour since the mainshock origin time by computing the average positions over a 1-hour time window centered on the time of interest.” Does cumulative refer here for cumulative during the one hour processed or cumulative since the mainsock. I assume the earlier, but please clarify.
  
  We agree that the terminology that we use here is rather confusing. A given position in time on the position time series represents how much surface displacement has occurred since the mainshock. This is what we meant by « cumulative ». Therefore, the average position that we compute and use during the inversions represents how much surface displacement has occurred after N hour(s). Consequently, each snapshot on the former Figure 3 and 6 represent the total amount of afterslip that has occur after N hour(s). We have attempted to clarify that point in the text (line 100-104 in the revised manuscript).
  
  l.126: “yellow circles” should be “purple stars”?
  
  Done
  
  l.129: There is a word missing after “second”. Maybe “patch”.
  
  Done
  
  The general fuzziness of both co-seismic and post-seismic slip models makes this assertion difficult to maintain (and the authors actually relativize it later in the paragraph). In particular, different fault model used (simple plain slab like the authors or varying slab dip based on e.g. slab2.0) in the modeling will shift location of slip. To start interpreting this, at least the modeling set-up of co- and post-seismic slip should be the same.
  
  We do not think that it would be relevant to the study to add our own co-seismic slip model using our own dataset. There are already plenty of co-seismic models from other groups, that we show in Supplementary Materials S6, and which already illustrates the variability of the co-seismic slip area. To the first order, our model will very likely match the ones displayed and thus it will not change the discussion. Also, as we mention in the text, we are not making any assertion regarding the penetration of afterslip inside the co-seismic area by concluding that this is an observation for which we cannot reliably address the veracity (line 180 in the revised manuscript).
  
  First of all, slip in Fig. 3 and 6 I assume is the slip during the respective hour, not the cumulative slip added up also from the previous hours (must be based on the amplitudes and the fact that some patches vanish).
  
  We think that our answer to comment l.89ff should clarify that point.
  
  I don’t understand why, if only the offsets of the 2 largest aftershocks are corrected (hour 1 and hour 5), all slip vanishes in the southern patch also during the other hours. Please explain.
  
  This relates to one of the point you have raised previously (l.89ff). The surface observations record the total amount of afterslip after n-hours. As we explain in the text, the observed afterslip to the south is for the most part due to the two largest aftershocks. So, by removing the signal from these earthquakes, there is consequently no slip in this region anymore.
  
  Fig. S7.1: The second M6.8 aftershock occurs during hour 5 and clearly shows up as a step in the graphed moment. However a step in slip seems to occur mainly in hour 6? Is this an averaging effect?
  
  This is indeed an averaging effect. The second aftershock occurs at the end of the time window used for averaging the position at hour 5. Therefore, within that window, the majority of the data used for the averaging are not affected by the offset from that aftershock. Instead, the windows for hour 6 and after are fully offset by the aftershock. This explains the shift between the seismic moment and the geodetic moment. Also, instead of showing the continuous evolution of seismic moment, we were showing the seismic moment summed over time windows of 1 hour, to match the time window covered by the afterslip. This accentuated the effect even more. Thus, we have changed that figure (see Figure 6 in the revised manuscript).
  
  l.197ff: Please elaborate in one or two sentences what this predicted accelation phase signifies.
  
  We have added a sentence to clarify what is this acceleration phase (line 256-257 in the revised manuscript).
  
  l.201ff: Please mention where the cited studies were based (Ecuador and Japan).
  
  Done
  
  l.209: Change “rate-and-state law” to “rate-and-state friction law” here and everywhere else.
  
  Done
  
  l.257: I wonder, are aftershocks anywhere actually operationally forecasted based on some models (maybe a citation would be good)? If so, I would assume that models have to be simple and robust. Here e.g. CFS would naturally predict aftershocks around the co-seismic rupture area, where they do occur, for the Illapel eqk and also for many other subduction zone earthquakes. I wonder, how realistic and it is to actually do the hindsight analysis outlined here in near real time and if it really adds value. Of course, this could be tested.
  
  It is indeed, very difficult to predict the added value of including information about very early afterslip for forecasting aftershocks, and the point that we want to make is that we have now the capability to investigate such question. However, it is true that we might have been too ambitious by having the word « operational ». In fact, we are not aware if this is done anywhere. We have rephrased that last sentence to reflect better our point (line 319-321 of the revised manuscript).
  
  l.271ff: “Our additional finding is that the slip patterns that we observe after 12 hours persists over the first 2 months. When that is the case, information about very-early post-seismic slip can help to characterize longer-lasting post-seismic slip, which can prove to be useful to include for the forecast of aftershocks locations.” But can this really be generalized?
  
  A community reviewer pointed out to us that the fact that afterslip patterns seem rather stationary in time, and that is the case for many examples. Following that question, we have investigated the literature on the question. We have only found one study that suggest a hint of afterslip migration (see line 157-161 of the revised manuscript). Therefore, although we cannot conclude with certainty that this can be generalized, it seems reasonable to make the assumption that afterslip is commonly stationary. We have added a sentence to reflect that (line 340-342 of the revised manuscript).
  
  Citation: https://doi.org/10.5194/se-2021-6-AC2
CC2:
'Comment on se-2021-6', Dietrich Lange, 22 Mar 2021

Twardzik and co-authors investigate the first 12 h of post-seismic deformation following the 2015 M8.3 Illapel Chile earthquake. For the geodetic data, they use GNSS positions of 15 stations hourly displacements are inverted for hourly afterslip on the plate interface. For the seismicity, they and earthquake catalogs based on template matching based on previous studies. Then the seismicity is compared with the aftershocks. The northern patch is essentially aseismic slip, while in the southern patch, slip is essentially seismic slip related to aftershocks. Furthermore, the authors show that the ratio of seismic to aseismic slip might change over time at some locations.
The manuscript is well written and well-illustrated. I have only a few points related to the ratio of seismic to aseismic slip. Since for subduction zones, there are just limited observations on the relation of aftershocks to afterslip (in particular for the very first hours after the co-seismic), the topic is sound and relevant, and the paper contributed new and interesting aspects on this topic. The comments and suggestions below should all be feasible to be incorporated.
In the manuscript, you sometimes refer to “south/north of the rupture area,” but it seems that the two patches you classify as south and north of the rupture are still located in the co-seismic region (e.g., between 30-32°S). I suggest clarifying the two patches exact position and how they are related to the co-seismic slip.
In particular, previous authors observed a widening of the aftershock zone of the Illapel earthquake (e.g., Lange et al., 2016, GJI, doi: 10.1093/gji/ggw218). It would be interesting to state if this is observed with the catalog based on template matching and how this relates to the inverted afterslip. I assume that the possible expansion outside the co-seismic region is too small to be robustly inverted by the hourly evolving afterslip model. I suggest adding some words to this.
In particular, the authors find that the partitioning in seismic and aseismic slip changes in time (As mentioned in Line 207). I could not feasible see this in Figure 7 and S7.1, and suggest to simply plot the displacement versus the number of events for the patches to show the relationship between both processes. For example, Lange et al. (2014, GJI, doi: 10.1093/gji/ggu292) mapped for the late postseismic (e.g. >1d) of the Maule 2010 earthquake the partitioning of seismic to aseismic slip, which was relatively stable in time.
There is a similar observation for strike-slip faulting for early observation of afterslip and aftershocks, and their relation (Savage, 2007, GRL, doi: 10.1029/2010GL042872) shows that aftershock seismicity rate is not proportional to the stress relaxation rate for the San-Andreas fault.
Formal issues:
I suggest adding a caption to Fig. S7.1 and S7.2.
Figure S7, S7.1, and S7.2 (right panels) might need additional labeling for the number of aftershocks. Currently, only the slip is labeled.
Table 1 and 2 do not contribute and might belong to the supplementary.
I suggest showing Figure S7.1 (temporal development southern patch) in the primary material since parts of the findings are difficult to understand without this figure.
Line 9: …. and thus could contribute to a more data-drivenforecasts of long-term aftershocks.
I cannot necessarily follow the argument in Line 9. Since the partitioning of seismic slip in some places changes in time prediction might be very difficult if this process remains enigmatic. Does the suggested forecast suggested here use Omori-laws and constant b-values, such as modelled by (Jonsdottir et al., 2006, Tectonophysics 424, https://doi.org/10.1016/j.tecto.2006.03.036.)?

Citation: https://doi.org/10.5194/se-2021-6-CC2
- AC4:
  'Reply on CC2', Cédric Twardzik, 17 Jul 2021
  Dear Dietrich Lange,
  First of all, we would like to thank you for the community comments that you have made on our study that we have submitted to EGU Solid Earth (manuscript se-2021-6). Below are our responses to the comments that you have made. We have attached as a .zip file an annotated pdf of the manuscript so that changes can be tracked as well as an updated version of the Supplementary Materials.
  In the manuscript, you sometimes refer to “south/north of the rupture area,” but it seems that the two patches you classify as south and north of the rupture are still located in the co-seismic region (e.g., between 30-32°S). I suggest clarifying the two patches exact position and how they are related to the co-seismic slip.
  
  We have highlighted the regions that we are referring to on Figure 3 so that the reader can clearly see the location of what we refer to as the southern patch and the northern patch. We have also added that added information in the text (line 143 and line 149 in the revised manuscript).
  In particular, previous authors observed a widening of the aftershock zone of the Illapel earthquake (e.g., Lange et al., 2016, GJI, doi: 10.1093/gji/ggw218). It would be interesting to state if this is observed with the catalog based on template matching and how this relates to the inverted afterslip. I assume that the possible expansion outside the co-seismic region is too small to be robustly inverted by the hourly evolving afterslip model. I suggest adding some words to this.
  
  This is an interesting point that you raise regarding the migration of aftershocks. Huang et al. (2017) as well as Frank et al. (2017), from which the catalogs of aftershocks used in our studies are taken, seem to make a similar observation, even when looking at the very early stage. However, regarding the afterslip, we do not see such migration. Instead, as mentioned in the text, we only observe a growth of the afterslip patches (see line 142 in the revised manuscript). It is likely that our afterslip maps don’t allow to investigate such fine details. Because we already mention in the text that we do not observe any afterslip migration, we have decided not to discuss that point in the main text.
  In particular, the authors find that the partitioning in seismic and aseismic slip changes in time (As mentioned in Line 207). I could not feasible see this in Figure 7 and S7.1, and suggest to simply plot the displacement versus the number of events for the patches to show the relationship between both processes. For example, Lange et al. (2014, GJI, doi: 10.1093/gji/ggu292) mapped for the late postseismic (e.g. >1d) of the Maule 2010 earthquake the partitioning of seismic to aseismic slip, which was relatively stable in time.
  
  We would like to clarify that we do not argue that the seismic/aseismic slip partitioning changes over time. Line 207 in the former manuscript only relates to the mechanical link that has been proposed between afterslip and aftershocks. One usual argument for the fact that afterslip is mechanically driving aftershocks is based on the fact that the shape of the time evolution of afterslip closely match the shape of the time evolution of the cumulative number of aftershocks. But, this is not what we observe. Figure 7 (now Figure 8 in the revised manuscript) shows that : the afterslip (blue line) evolves clearly differently than the number of aftershocks (orange dashed line).
  There is a similar observation for strike-slip faulting for early observation of afterslip and aftershocks, and their relation (Savage, 2007, GRL, doi: 10.1029/2010GL042872) shows that aftershock seismicity rate is not proportional to the stress relaxation rate for the San-Andreas fault.
  
  Thank you for pointing that study to our attention. At the time of writing the first version, we have attempted to find other cases exhibiting such behavior, without success. Thus, we have added that to the main text (line 268-270 in the revised manuscript).
  I suggest adding a caption to Fig. S7.1 and S7.2. Figure S7, S7.1, and S7.2 (right panels) might need additional labeling for the number of aftershocks. Currently, only the slip is labeled.
  
  This is done on Figures 6, 8 and S7.2 in the revised manuscript and Supplementary Materials
  Table 1 and 2 do not contribute and might belong to the supplementary.
  
  Done (see Supplementary Material S1)
  I suggest showing Figure S7.1 (temporal development southern patch) in the primary material since parts of the findings are difficult to understand without this figure.
  
  Done (see Figure 6 in the revised manuscript).
  Line 9: …. and thus could contribute to a more data-drivenforecasts of long-term aftershocks. I cannot necessarily follow the argument in Line 9. Since the partitioning of seismic slip in some places changes in time prediction might be very difficult if this process remains enigmatic. Does the suggested forecast suggested here use Omori-laws and constant b-values, such as modelled by (Jonsdottir et al., 2006, Tectonophysics 424, https://doi.org/10.1016/j.tecto.2006.03.036.)?
  
  We would like to clarify again that we do not argue that the seismic/aseismic slip partitioning changes over time. However, we do agree that this last sentence of the abstract might be too ambitious of an opening with respect to our finding. We have rephrased that last sentence (line 9-11 of the new manuscript).
  
  Citation: https://doi.org/10.5194/se-2021-6-AC4
RC2:
'Comment on se-2021-6', Mathilde Radiguet, 11 May 2021

Twardzik and co-author investigate the early postseismic signal of the Mw 8/3 Illapel earthquake in Chile. The author use subdaily GNSS processing to extract the postseismic signal of the first 12 hours after the earthquake. The GNSS displacements are inverted to provide an image of the slip distribution on the plate interface every hour. Two main slip patches are identified, the one located South of the rupture being mostly seismic (the cumulative moment is mostly due to aftershocks), whereas the Northern patch is mostly aseismic. They observe that the patches identified in the first 12 hours are consistent over longuer time period.The author then discuss the implications of their results in terms of frictionnel behavior and driving mechanism for aftershock occurence.

The paper is well written in general, and the figures are clear. I have a few suggestions, mostly minor, which are detailed below.

section 3:

The discussion on the relative location of the co-seismic and postseismic slip is difficult to follow since the co-seismic slip distribution is not shown. It would be useful to add on Figure 3 the coseismic slip distribution estimated from previous studies (for example the one inferred by Melgar et al. 2016, as the authors refer to it), or inverted by the authors. Of course uncertainties exist on this slip distribution based on the data, inversion scheme and fault geometry used by previous authors, but the authors could show several models if needed.

The authors could also perform their own inversion of the co-seismic slip (using only GNSS data). Even if this inversion would be constrained only by geodetic data, it would be interesting because fully consistent with the post-seismic study in terms of fault geometry, Green’s function and data with the post-seismic study. it could be added to the supplementary material.

l. 104: "We search for the spatial distribution of slip amplitude and rake angle independently for each time step": the slip amplitudes obtained are shown but not the rake angles. Do they vary from one time step to another ? The optimal rake for each time step should be given in the supplementary materials.

Section 4:

The discussion on the relationship between early aftershocks and afterslip could be clarified by a better description of the areas involved when the seismic or geodetic moments are calculated: l.181"a seismic moment of 9.5x 10^19 Nm." what is the region considered for the calculation ? Is it the same as the one shown in Fig. S6.2 ?

l. 189-190: same question, what is the area considered ?

The times series from figure S7 could be a nice illustration of the seismic/aseismic ratio on both regions, if the time series showed the seismic moment in both regions (instead of an arbitrary normalised unit). I understand the interest of the normalised unit when comparing the number of aftershock with the cumulative slip, but this is already shown in Fig 7 and S6 for both regions. In Fig. S7, the cumulative slip could be converted to seismic moment so that one can see where the values given in the text for the seismic/aseismic rations come from.

The time series from Fig.7 could also be included in the main text, as they are really relevant for the discussion of the paper.

On FigureS7.2: what is happening between 6h and 7h (strong increase in cumulative moment): is there a large aftershock at this time ?

Formal issues:

Figure 1 (legend):

- "yellow circles" should be "pink stars"

Several problems with figure/table numbers (they appear with "??"): Line 85, 136, 165...

Citation: https://doi.org/10.5194/se-2021-6-RC2
- AC3:
  'Reply on RC2', Cédric Twardzik, 17 Jul 2021
  Dear Mathilde Radiguet,
  Thank you agreeing to review our manuscript se-2021-6 submitted to EGU Solid Earth. We are providing answers to your questions/comments below. We have attached as a .zip file an annotated pdf of the manuscript so that changes can be tracked as well as an updated version of the Supplementary Materials.
  It would be useful to add on Figure 3 the coseismic slip distribution estimated from previous studies (for example the one inferred by Melgar et al. 2016.
  
  Done as the shaded blue region in Figure 3
  Of course uncertainties exist on this slip distribution based on the data, inversion scheme and fault geometry used by previous authors, but the authors could show several models if needed.
  
  Done in the Supplementary Material S6
  The authors could also perform their own inversion of the co-seismic slip (using only GNSS data). Even if this inversion would be constrained only by geodetic data, it would be interesting because fully consistent with the post-seismic study in terms of fault geometry, Green’s function and data with the post-seismic study. it could be added to the supplementary material.
  
  We do not think that it would be relevant to the study to add our own co-seismic slip model based on the data that we have. There are already plenty of co-seismic models from other groups, that we show in Supplementary Material S6, and which already illustrates the variability of the region of co-seismic slip. To the first order, our model will very likely match the ones displayed and thus it will not change anything to the discussion.
  l. 104: "We search for the spatial distribution of slip amplitude and rake angle independently for each time step": the slip amplitudes obtained are shown but not the rake angles. Do they vary from one time step to another ? The optimal rake for each time step should be given in the supplementary materials.
  
  We have included that information in Supplementary Material S4
  You do not want to add that you invert the rake in a +/- 15° of the convergence direction in the main text or the legend of the figure S4.17
  
  We have included that information in the main text (line 119-121 in the revised manuscript).
  l.181"a seismic moment of 9.5x 10^19 Nm." what is the region considered for the calculation ? Is it the same as the one shown in Fig. S6.2 ?
  
  We have done a better description of the area considered to obtain this estimate (line 208-214 in the revised manuscript)/
  l. 189-190: same question, what is the area considered ?
  
  We have highlighted the regions that we are referring to on Figure 3 in the revised manuscript so that the reader can clearly see the location of what we refer as the southern patch and the northern patch. We have also added that into the text (see line 143 and line 149 in the revised manuscript).
  In Fig. S7, the cumulative slip could be converted to seismic moment so that one can see where the values given in the text for the seismic/aseismic rations come from. The time series from Fig.7 could also be included in the main text, as they are really relevant for the discussion of the paper.
  
  We have made that change by including a new figure in the revised manuscript (Figure 6).
  On FigureS7.2: what is happening between 6h and 7h (strong increase in cumulative moment): is there a large aftershock at this time ?
  
  We think that this comment comes from the fact that the curves were normalized giving the false impression of a very large increase of seismic moment. With the new figure 6, we believe that it clarifies that the increase is in fact very small.
  "yellow circles" should be "pink stars »
  
  Done
  Several problems with figure/table numbers (they appear with "??"): Line 85, 136, 165…
  
  Fixed
  
  Citation: https://doi.org/10.5194/se-2021-6-AC3

Cedric Twardzik et al.

Supplement

https://doi.org/10.5194/se-2021-6-supplement

Data sets

30s Position Time Series for the 2015 Illapel earthquake Cedric Twardzik https://doi.org/10.5281/zenodo.4498200

Model code and software

Python Code for sidereal filtering Cedric Twardzik https://doi.org/10.5281/zenodo.4498160

Cedric Twardzik et al.

Viewed

Total article views: 600 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	Supplement	BibTeX	EndNote
455	124	21	600	45	3	3

HTML: 455
PDF: 124
XML: 21
Total: 600
Supplement: 45
BibTeX: 3
EndNote: 3

Views and downloads (calculated since 05 Feb 2021)

Month	HTML	PDF	XML	Total
Feb 2021	263	51	5	319
Mar 2021	62	13	7	82
Apr 2021	32	11	0	43
May 2021	31	11	2	44
Jun 2021	19	10	0	29
Jul 2021	48	28	7	83

Cumulative views and downloads (calculated since 05 Feb 2021)

Month	HTML	PDF	XML	Total
Feb 2021	263	51	5	319
Mar 2021	62	13	7	82
Apr 2021	32	11	0	43
May 2021	31	11	2	44
Jun 2021	19	10	0	29
Jul 2021	48	28	7	83

Viewed (geographical distribution)

Total article views: 494 (including HTML, PDF, and XML) Thereof 494 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 30 Jul 2021

Short summary

After an earthquake, the fault continues to slip for days to months. Yet, little is know about the very early part of this phase (i.e., minutes to hours). So, we have looked at what happens just after an earthquake in Chile from 2015. We find that the fault responds in two ways: south of the rupture zone it slips seismically in the form of aftershocks, while north of the rupture zone it slips slowly. Early inference of such bimodal behavior could prove to be useful for forecasting aftershocks.


Total:	0
HTML:	0
PDF:	0
XML:	0

Very early identification of a bimodal frictional behavior during the post-seismic phase of the 2015 Mw8.3 Illapel, Chile, earthquake

Supplement

Data sets

Model code and software

Viewed

Viewed (geographical distribution)

Very early identification of a bimodal frictional behavior during the post-seismic phase of the 2015 M_w8.3 Illapel, Chile, earthquake