D5.3! …...MARSite((GA(308417)(D5.3((5" " Top! of layer (km)! Vp(m/s)! Vs(m/s)! Density!(kg/m3)!Q*!...

MARSite (GA 308417) D5.3 1

This project has received funding from the European Union’s Seventh Programme for research, technological development and demonstration under grant agreement No [308417]”.

New Directions in Seismic Hazard Assessment through Focused Earth Observation

in the Marmara Supersite

Grant Agreement Number: 308417 co-‐funded by the European Commission within the Seventh Framework Programme

THEME [ENV.2012.6.4-‐2] [Long-‐term monitoring experiment in geologically active regions of Europe prone to natural hazards: the

Supersite concept]

D5.3 Performance assessment of finite-‐fault inversion

codes in the Marmara configuration

Project Start Date 1 November 2012 Project Duration 36 months Project Coordinator /Organization Nurcan Meral Özel / KOERI Work Package Number WP5 Deliverable Name/ Number Performance assessment of finite-‐fault

inversion codes in the Marmara configuration/D5.3

Due Date Of Deliverable 2015 April 30 Actual Submission Date Organization/Author (s) INGV, GFZ, BRGM/ Cirella A., Piatanesi A.,Diao

F., Wang R., Aochi H. Dissemination Level

PU Public

PP Restricted to other programme participants (including the Commission)

RE Restricted to a group specified by the consortium (including the Commission)

CO Confidential, only for members of the consortium (including the Commission)

MARSite (GA 308417) D5.3 2

TABLE OF CONTENTS

1. Introduction………………………………………………………………………………………………3

2. Checker board test…………………………………………………………………………………….3

2.1 Description of checker board test……………………………………………………………………………….3

2.2 GFZ’ TEAM RESULTS……………………………………………………………………………………………………5

2.2.1 Methodology…………………………………………………………………………………………………………..5

2.2.2 Geometrical setting-‐ Fault parametrization -‐ Data processing……………..…………………..6

2.2.3 Results………..…………………………………………………………………………………………………………..7

2.3 INGV’ TEAM RESULTS………………………………………………………………………………………………..12

2.3.1 Methodology………………………………………………………………………………………………………….12

2.3.2 Geometrical setting-‐ Fault parametrization -‐ Data processing……………..………………….12

2.3.3 Results………..………………………………………………………………………………………………………….13

3. Conclusions………..……………………………………………………………………………………23

MARSite (GA 308417) D5.3 3

1. Introduction

Two different finite-‐fault techniques have been proposed and tested, from BRGM, INGV and GFZ researchers’ team. In order to assess the performances of the inversion codes, in term of accuracy of the solution for the Marmara Sea tectonic setting and observational network; BRGM’ team formulated a checker-‐board test and produced two synthetic datasets, by taking into account a planned stations configuration (strong motion, cGPS, GPS, BB), in the Marmara region and by assuming a given velocity 1D profile and a 3D crustal structure. INGV and GFZ teams used this datasets to perform kinematic inversion of the slip distribution on the corresponding finite-‐fault. These tests have the principal aim to assess the performance of the inversion codes in the Marmara configuration.

2. Checker board test

2.1 Description of checker board test The area of interest is prepared for a dimension of 200 km (East-‐West) x 120 km (North-‐South) around the Sea of Marmara, including the region of Istanbul (WP5.3). The wave propagation in the elastic medium is calculated using a 3D finite difference method (Aochi et al., 2013). Any finite source scenario can be included anywhere in the model. For the verification test, we prepare a kinematic source model whose slip distribution is checker-‐board-‐like feature on a fault plane. The given parameters are in the following (also informed to the participants). In particular, the distribution of four asperities and rupture time are shown in Figure 1.

Parameter Value

Hypocenter 28.3965°E, 40.8549°N, 7 km depth

Fault Plane 96 km (length) x 24 km (width)

Mechanism Strike N86°E, Dip 90°, Rake 180°

Eastern end of fault plane 28.870°E, 40.880°N

Top of fault plane 0.5 km

Asperity size 16 km x 8 km

Slip amount 1 m on asperities/ 0 m elsewhere

Rupture velocity 2.1 km/s

Rise time (triangle function) 2.5 s

MARSite (GA 308417) D5.3 4

Figure 1. The given slip distribution (left) and rupture time (right) on a fault plane.

Figure2. Illustration of the hypocenter, the fault position (strike N86°E) and the 47 receiver positions. The eastern part of the fault (running to SE) is not used in this inversion test. The color presents a Peak Ground Velocity (PGV) from the simulation using a 3D structure.

For the purpose of the inversion, the synthetic ground motions are calculated and saved for the 47 receivers according to the real locations from the observation network of the region (Figure 2). The seismograms (velocity in m/s) are not filtered, with a time step of 0.01 seconds for a duration of 120 seconds for the three components. These data can be downloaded from http://aochi.hideo.perso.neuf.fr/marmara/ (scenario 7). The ground motions are calculated in the two structure models:

1. 1D layered structure,

2. 3D structure.

The used 1D layered structure is the following:

MARSite (GA 308417) D5.3 5

Top of layer (km)

Vp (m/s) Vs (m/s) Density (kg/m3) Q*

0 2250 1100 2150 300

-‐1 5700 3200 2700 300

-‐6 6100 3600 2750 300

-‐20 6800 3850 2800 300

-‐33 8000 4550 2850 300

* Qualitiy factor Q is included as a dumping factor in the finite difference formulation.

The 3D structure (Figure3) was prepared in WP5.3, by a combination of 3D tomography result around the Sea of Marmara (Bayrakci et al., 2013) and a regional 1D model (Karabulut, personal comm.). We also adjusted the bathymetry from the 30 seconds-‐model of GEBCO so that we introduce a layer of Sea of Marmara, letting Vp = 1500 m/s and Vs = 0 m/s (Aochi and Ulrich, ECEES, 2014; Aochi and Ulrich, BSSA, 2015).

Figure3. The used 3D structure model (Vp structure).

The two cases allow each partner to check the resolution in their inversion procedure.

2.2 GFZ’ TEAM RESULTS

2.2.1 Methodology To reduce user’s subjective influence on the inversion results, Zhang et al. (2014) proposed a new kinematic inversion scheme, called the iterative deconvolution and stacking (IDS) method. We use this method for all of our inversions in the Marsite Project. In the IDS method, synthetic Green’s function deconvolution is applied to the waveform data to obtain apparent subfault source-‐time-‐functions from different stations. In most cases, the

MARSite (GA 308417) D5.3 6

deconvolution and stacking procedure needs to be applied iteratively to resolve a complex multiple rupture process. In fact, the IDS method benefits from the complementary advantages between the traditional least-‐squares inversion and the array back-‐projection techniques. The work flow that shown in Fig. 4 draw the main steps of the IDS inversion method:

Figure4. Simple work flow of the IDS inversion method.

The IDS method aims at robust and rapid estimation of the main character of the rupture process. Because it is largely free of empirical constraints such as rupture velocity, rise time and shape of subfault source time function, the inversions by using IDS don’t give a direct estimation of rupture velocity and rise time.

2.2.2 Geometrical setting-‐ Fault parametrization -‐ Data processing Based on the 1D model provided by H. Aochi, a Green’s function database is prepared using the code QSEIS developed by Wang (1999). The fault plane (120 km × 24 km) that constructed based on the focal mechanism (strike=86°, dip=90°) was discretized to 180 subfaults of 4 km × 4 km size, each being treated as a point source. We fixed the rake angle to be 180°. We did not constraint the rupture velocity, rise time and the shape of the rise time. Firstly we filter the original synthetic data in form of velocity seismograms first with a 3rd-‐order Butterworth high-‐pass filter of 0.02 Hz to remove the influence caused by the low-‐frequency numerical drift in the data, and then with a causal low-‐pass filter corresponding to Brunes near-‐field velocity spectrum;

𝐿 𝑓 = !(!!!"/!!)

! (1)

MARSite (GA 308417) D5.3 7

where 𝑓! is called the corner frequency. Thus, our waveform data trend to the velocity seismograms only for 𝑓 ≪ 𝑓!. They become proportional to the displacement seismograms when 𝑓~𝑓! and to the integral of displacement seismograms when 𝑓 ≫ 𝑓!. The advantage of using the low-‐pass filter defined by Eq. (1) is that no sharp high-‐frequency cutoff is necessary. With higher corner frequencies used, more high-‐frequency components will be used in our inversion. So, the frequency band used in our inversion is different with many previous studies of kinematic source inversions.

2.2.3 Results The results by inverting different synthetic data are shown in Fig. 5, from which we found that the input four asperities are clearly identified. Based on the data processing approach used in our study, inversions were carried out by using the three different corner frequencies (0.05 Hz, 0.10 Hz and 0.20 Hz). The inverted results (Fig. 5) indicate that the input asperities can be well captured, although different corner frequencies were used for the inversions. Moreover, it is not surprising to find a better resolution by using higher corner frequencies.

Figure 5. Rupture model inverted based on IDS method and synthetic datasets that simulated from 1D and 3D earthquake structure. (a) – (c) show rupture models inverted based on synthetic data generated from 1D earth structure, while (d) – (f) show rupture models inverted based on synthetic data generated from 3D earth structure. (g) shows the input model of the checkerboard test.

MARSite (GA 308417) D5.3 8

As an accurate 1D earth structure was applied, the data fit is very good for rupture models inverted by using synthetic data that generated from the same 1D earth structure (Fig. 6 – Fig. 8). However, the data fit decrease when synthetic data generated from 3D earth structure was used (Fig. 9 – Fig. 11), which perhaps mainly induced by 3D wave propagation effect. The computation time used for inversion of the six rupture models varies between 75 second to 200 seconds, which increase as higher frequencies were used. All inversions were done with a PC (Intel(R) Core (TM) i7-‐3770 CPU @3.4G Hz, 32 GB RAM). We highlight the good efficiency of the IDS inversion, which may play an important role in earthquake rapid response.

Figure 6. Waveform comparison between synthetic (blue) and inverted (red) velocity seismograms, corresponding to rupture model shown in Fig. 5(a).

MARSite (GA 308417) D5.3 9

Figure 7. Waveform comparison between synthetic (blue) and inverted (red) velocity seismograms, corresponding to rupture model shown in Fig. 5(b).

Figure 8. Waveform comparison between synthetic (blue) and inverted (red) velocity seismograms, corresponding to rupture model shown in Fig. 5(c).

MARSite (GA 308417) D5.3 10

Figure 9. Waveform comparison between synthetic (blue) and inverted (red) velocity seismograms, corresponding to rupture model shown in Fig. 5(d).

Figure 10. Waveform comparison between synthetic (blue) and inverted (red) velocity seismograms, corresponding to rupture model shown in Fig. 5(e).

MARSite (GA 308417) D5.3 11

Figure 11. Waveform comparison between synthetic (blue) and inverted (red) velocity seismograms, corresponding to the rupture model shown in Fig. 5(f).

MARSite (GA 308417) D5.3 12

2.3 INGV’ TEAM RESULTS

2.3.1 Methodology The inversion methodology, developed at INGV, is a two-‐stage nonlinear technique (Piatanesi et al., 2007, Cirella et al., 2012), which involves the joint inversion of strong motion records and geodetic data. To account for rupture complexity, the model is described by four spatially variable fault parameters -‐ peak slip velocity, slip direction, rupture time and rise time. The final slip distribution is derived by the inverted parameters. The finite fault is divided into sub-‐faults with model parameters assigned at the corners, whereas the parameters within each sub-‐fault are allowed to vary through bilinear interpolation of the nodal values. Each point on the fault can slip only once (single window approach) and the source time function can be selected among different analytical forms (box-‐car; cosine; regularized Yoffe function). The forward modeling is performed with a discrete wavenumber technique (Compsyn, Spudich and Xu, 2003), whose Green's functions include the complete response of the vertically varying Earth structure. The nonlinear global inversion consists of two stages. During the first stage of the inversion, a heat-‐bath simulated-‐annealing algorithm explores the model space to generate an ensemble of models that efficiently sample the good data-‐fitting regions. In the second stage (appraisal), the algorithm performs a statistical analysis of the model ensemble providing us the best-‐fitting model, the average model and the associated standard deviation, computed by weighting all models of the ensemble by the inverse of their cost function values.

2.3.2 Geometrical setting-‐ Fault parametrization -‐ Data processing We used the 47 synthetic seismograms provided for Scenario7 (http://aochi.hideo.perso.neuf.fr/marmara/). Original ground velocity time histories are band-‐pass filtered in two different frequency ranges; between 0.01 and 0.5 Hz and between 0.01 and 0.25Hz, by using a two-‐pole and two-‐pass Butterworth filter. We invert 60 seconds of each waveform, including body and surface waves. We assumed a fault plane consistent with the hypocentre location and the focal mechanisms given for Scenario7 (strike: 86°; dip=90°; strike=180°). The fault plane is 100 km long and 24 km width, along strike and down-‐dip direction, respectively. We invert simultaneously for kinematic parameters at nodal points equally spaced (4.0 km) along strike and down-‐dip directions. During the inversion, we fix a given range of variability for each model parameter. Peak slip velocity values can range between 0 and 1.5 m/s at 0.25 m/s interval; the rise time between 1.5 and 3 sec at 0.25 sec interval. Rupture velocity is fixed to 2.1 km/s and rake angle is fixed to 180°. In this study, the adopted source time function is a modified cosine function (Ji et al. , 2002).

MARSite (GA 308417) D5.3 13

2.3.3 Results

We performed eight different inversions:

1. by using the dataset given for the 1D velocity model and by inverting for peak slip velocity. Rise time is fixed to 2.5 s; the inversion is performed in the frequency band 0.01-‐0.25Hz (results are referred as case ‘1DonlySlip_0.25Hz’); 2. by using the dataset given for the 1D velocity model and by inverting for peak slip velocity and rise time; the inversion is performed in the frequency band 0.01-‐0.25Hz (results are referred as case ‘1DSlipRise_0.25Hz’); 3. by using the dataset given for the 1D velocity model and by inverting for peak slip velocity. Rise time is fixed to 2.5 s; the inversion is performed in the frequency band 0.01-‐0.5Hz (results are referred as case ‘1DonlySlip_0.5Hz’); 4. by using the dataset given for the 1D velocity model and by inverting for peak slip velocity and rise time; the inversion is performed in the frequency band 0.01-‐0.5Hz (results are referred as case ‘1DSlipRise_0.5Hz’); 5. by using the dataset given for the 3D velocity model and by inverting for peak slip velocity. Rise time is fixed to 2.5 s; the inversion is performed in the frequency band 0.01-‐0.25Hz (results are referred as case ‘3DonlySlip_0.25Hz’); 6. by using the dataset given for the 3D velocity model and by inverting for peak slip velocity and rise time; the inversion is performed in the frequency band 0.01-‐0.25Hz (results are referred as case ‘3DSlipRise_0.25Hz’); 7. by using the dataset given for the 3D velocity model and by inverting for peak slip velocity. Rise time is fixed to 2.5 s; the inversion is performed in the frequency band 0.01-‐0.5Hz (results are referred as case ‘3DonlySlip_0.5Hz’); 8. by using the dataset given for the 3D velocity model and by inverting for peak slip velocity and rise time; the inversion is performed in the frequency band 0.01-‐0.5Hz (results are referred as case ‘3DSlipRise_0.5Hz’). For each inversion we show (see Figures 12-‐14-‐16-‐18-‐20-‐22-‐24-‐26) the retrieved rupture model, given in terms of rise time and slip distribution on the fault plane (bottom and upper panel, respectively) and the corresponding comparison between synthetic (blue lines) and inverted (red lines) velocity time histories (see Figures 13-‐15-‐17-‐19-‐21-‐23-‐25-‐27).

MARSite (GA 308417) D5.3 14

Case: 1DonlySlip_0.25Hz

Figure12. Retrieved rupture model, in terms of slip and rise time distribution (upper and bottom panel, respectively), obtained by inverting for peak slip velocity, the 1D dataset in the frequency band 0.01-‐0.25Hz.

Figure13. Misfit between synthetic ground velocities (blue lines) with those computed from the inverted rupture model displayed in Figure12 (red lines).

0.0

2.5sec

rise timeE W

-20

0

(km)

0 20 40 60 80 100(km)

0.000.250.500.751.001.251.501.752.00

mslipE W

-20

0

(km)

0 20 40 60 80 100(km)

MARSite (GA 308417) D5.3 15

Case: 1DSlipRise_0.25Hz

Figure14. Retrieved rupture model, in terms of slip and rise time distribution (upper and bottom panel, respectively), obtained by inverting for peak slip velocity and rise time, the 1D dataset in the frequency band 0.01-‐0.25Hz.


1.0

1.5

2.0

2.5

3.0sec

rise timeE W

-20

0

(km)

0 20 40 60 80 100(km)

0.000.250.500.751.001.251.501.752.00

m

slipE W

-20

0

(km)

0 20 40 60 80 100(km)

MARSite (GA 308417) D5.3 16




0.0

2.5secrise timeE W

-20

0

(km)

0 20 40 60 80 100(km)

0.000.250.500.751.001.251.501.752.00

mslipE W

-20

0(km)

0 20 40 60 80 100(km)

MARSite (GA 308417) D5.3 17




1.0

1.5

2.0

2.5

3.0secrise timeE W

-20

0

(km)

0 20 40 60 80 100(km)

0.000.250.500.751.001.251.501.752.00

mslipE W

-20

0(km)

0 20 40 60 80 100(km)

MARSite (GA 308417) D5.3 18




0.0

2.5sec

rise timeE W

-20

0

(km)

0 20 40 60 80 100(km)

0.000.250.500.751.001.251.501.752.00

mslipE W

-20

0

(km)

0 20 40 60 80 100(km)

MARSite (GA 308417) D5.3 19




1.0

1.5

2.0

2.5

3.0sec

rise timeE W

-20

0

(km)

0 20 40 60 80 100(km)

0.000.250.500.751.001.251.501.752.00

m

slipE W

-20

0

(km)

0 20 40 60 80 100(km)

MARSite (GA 308417) D5.3 20




0.0

2.5secrise timeE W

-20

0

(km)

0 20 40 60 80 100(km)

0.000.250.500.751.001.251.501.752.00

mslipE W

-20

0(km)

0 20 40 60 80 100(km)

MARSite (GA 308417) D5.3 21




1.0

1.5

2.0

2.5

3.0sec

rise timeE W

-20

0

(km)

0 20 40 60 80 100(km)

0.000.250.500.751.001.251.501.752.00

m

slipE W

-20

0(km)

0 20 40 60 80 100(km)

MARSite (GA 308417) D5.3 22

The inverted models are similar to the target one (Figure1) ; the positions of the asperities are correctly imaged and the slip values well estimated. In order to quantify and to compare the obtained results, for each performed inversion we show in Table1 the cost function values associated to the waveforms’ comparison (second and third column) and the ‘SLIP-‐FIT’ comparison (fourth and fifth column). The last ‘misfit’ is computed by the equation:

𝑆𝐿𝐼𝑃_𝐹𝐼𝑇 =1

𝑁𝑠𝑢𝑏𝑠𝑙𝑖𝑝𝑇 − 𝑠𝑙𝑖𝑝𝑅 !

!

𝑠𝑙𝑖𝑝𝑇𝑖! + 𝑠𝑙𝑖𝑝𝑅𝑖!

!"#$

!!!

where 𝑠𝑙𝑖𝑝𝑇 and 𝑠𝑙𝑖𝑝𝑅 are the target and retrieved slip values and 𝑁𝑠𝑢𝑏, is the number of sub-‐faults by which the fault plane has been parameterized. This quantity allows us to quantify the reliability of a retrieved model in respect to the target one and to indicate a preferred performed inversion, in terms of resolution capability. The cost function values’ analysis show how the waveform-‐fit improves with the lower frequency band (see also Figures 13-‐15-‐21-‐23) and it decreases when synthetic data from a 3D earth structure is inverted (Figures 25-‐27). This is probably due to the difficulty in modelling 3D wave propagation effect. From the ‘SLIP-‐FIT’ analysis emerges how the most simple inversion case (‘1DonlySlip’), in which we invert just for one parameter (peak slip velocity) and in the lower frequency band (0.01-‐0.25hz), by using the synthetic dataset generated with a 1D velocity profile, is able to yield a well resolved slip model.

Table1. Cost function values

0.01-‐0.5Hz 0.01-‐0.25Hz 0.01-‐0.5 Hz 0.01-‐0.25Hz 1DonlySlip

0.33

0.052

0.67

0.59

1DSlipRise

0.31

0.048

0.66

0.60

3DonlySlip

0.45

0.16

0.66

0.61

3DSlipRise

0.44

0.14

0.67

0.61

WAVEFORM-‐FIT SLIP-‐FIT

MARSite (GA 308417) D5.3 23

3. Conclusions The obtained results show how the proposed inversion techniques guarantee a reliable and accurate reconstruction of earthquake source rupture process on finite fault, in the Marmara tectonic and observational setting configuration. The proposed analysis represents a useful tool to assess the performance of a finite-‐fault inversion code, by taking into account the actual or future planned stations configuration (strong motion, cGPS, GPS, BB) in the Marmara Sea and earthquake scenarios.

References

Aochi, H., T. Ulrich, A. Ducellier, F. Dupros, D. Michea, Finite difference simulations of seismic wave propagation for unerstanding earthquake physics and predicting ground motions: Advances and challenges, J. Phys: Conf. Ser., 454, 012010, doi: 10.1088/1742-‐6596/454/1/012010, 2013.

Aochi, H., and T. Ulrich (2014). Dynamic rupture and ground motion simulations in the Sea of Marmara, 2nd European Conference of Earthquake Engineering and Seismology (ECEES), Istabul, Turkey, 24–29 August 2014.

Aochi, H. and T. Ulrich, A probable earthquake scenario near Istanbul determined from dynamic simulations, Bull. Seism. Soc. Am. , 105, doi:10.1785/0120140283, 2015.

Bayrakci, G., M. Laigle, A. Bécel, A. Hirn, T. Taymaz, S. Yolsal-‐Cevikbilen an SEISMARMARA team, 3-‐D sediment-‐basement tomography of the Northern Marmara trough by a dense OBS network at the nodes of a grid of controlled source profiles along the North Anatolian fault, Geophys. J. Int., 194, 1335-‐1357, 2013.

Cirella A., Piatanesi A., Tinti E. Chini M. and M. Cocco (2012), "Complexity of the rupture process during the 2009 L’Aquila, Italy, earthquake", Geophysical Journal International.190, 607-‐621, doi:10.1111/j.1365-‐246X.2012.05505.x

Ji, C., D. J. Wald, and D. V. Helmberger (2002), Source description of the 1999 Hector Mine, California, earthquake, Part I: Wavelet domain inversion theory and resolution analysis, Bull. Seismol. Soc. Am., 92 (4), 1192-‐1207.

Piatanesi A., A. Cirella, P. Spudich and M. Cocco (2007), “A global search inversion for earthquake rupture history: Application to the 2000 western Tottori, Japan earthquake”, J. Geophys. Res., 112(B7), B07314, doi:10.1029/2006JB004821

Spudich, P. and L. Xu (2003), Software for calculating earthquake ground motions from finite faults in vertically varying media, in International Handbook of Earthquake and Engineering Seismology, Academic Press.

MARSite (GA 308417) D5.3 24

Tinti, E., E. Fukuyama, A. Piatanesi and M. Cocco (2005a), A kinematic source time function compatible with earthquake dynamics, Bull. Seismol. Soc. Am., 95(4), 1211-‐1223, doi:10.1785/0120040177.

Zhang, Y., Wang, R., Zschau, J., Chen, Y. T., Parolai, S., & Dahm T., 2014. Automatic imaging of earthquake rupture processes by iterative deconvolution and stacking of high-‐rate GPS and strong motion seismograms, J. Geophys. Res., 119, 5633–5650, doi:10.1002/2013JB010469.

Wang, R., 1999, A simple orthonormalization method for stable and efficient computation of Green’s functions, Bull. Seismol. Soc. Am., 89(3), 733–7410.

MARSite (GA 308417) D5.4 1

This project has received funding from the European Union’s Seventh Programme for research, technological development and demonstration under grant agreement No [308417]”.

New Directions in Seismic Hazard Assessment through Focused Earth Observation

in the Marmara Supersite

Grant Agreement Number: 308417 co-‐funded by the European Commission within the Seventh Framework Programme

THEME [ENV.2012.6.4-‐2] [Long-‐term monitoring experiment in geologically active regions of Europe prone to natural hazards: the

Supersite concept]

D5.4 Near-‐real time estimation of most relevant

earthquake source parameters

Project Start Date 2012 November 1 Project Duration 36 months Project Coordinator /Organization Nurcan Meral Özel / KOERI Work Package Number WP5 Deliverable Name/ Number Near-‐real time estimation of most relevant

earthquake source parameters/D5.4 Due Date Of Deliverable 2015 April 30 Actual Submission Date Organization/Author (s) INGV, GFZ, BRGM/ Cirella A., Piatanesi A.,Diao

F., Wang R., Aochi H. Dissemination Level

PU Public

PP Restricted to other programme participants (including the Commission)

RE Restricted to a group specified by the consortium (including the Commission)

CO Confidential, only for members of the consortium (including the Commission)

MARSite (GA 308417) D5.4 2

TABLE OF CONTENTS

1. Introduction………………………………………………………………………………………………3

2. Blind test…………….…………………………………………………………………………………….3

2.1 Description of blind test…………………………….……………………………………………………………….3

2.2 GFZ’ TEAM RESULTS……………………………………………………………………………………………………4


2.2.2 Results………..…………………………………………………………………………………………………………..5

2.2.3 Test for near real-‐time kinematic inversions..............................................................5

2.3 INGV’ TEAM RESULTS………………………………………………………………………………………………….8


2.3.2 Results………..…………………………………………………………………………………………………………..7

3. Conclusions………..……………………………………………………………………………………16

MARSite (GA 308417) D5.4 3

1. Introduction

The main goal of this deliverable was the rapid determination of the most relevant earthquake source parameters, with special focus on their finite-‐fault characteristics, in case of large earthquakes in the Marmara region. To this goal we performed a blind test for kinematic source inversion. BRGM research group generated a synthetic dataset, by considering near-‐field strong-‐motion and high-‐rate GPS data, obtained by dynamic modelling of a single earthquake scenario; and provided these synthetics to the other teams (GFZ, INGV), to invert for the rupture process, by using the different codes, described in deliverable D5.3. This approach allowed us to assess the resolution and efficiency of the different inversion techniques, also in terms of the execution quickness, in the Marmara Sea configuration. GFZ’ team performed an additional test, for near real-‐time kinematic inversions; the related results are described in Section 2.2.3.

2. Blind test

2.1 Description of blind test In order to carry out the realistic inversion in this region, we provide a synthetic, but

more probable scenario, dynamically simulated taking into account of the fault geometry, stress and friction condition. We use one of the scenarios of Mw7.04 from Aochi and Ulrich (BSSA, 2015; LP fault model, T = 0.66, hypocenter location at Center; See Figure 1). The rupture propagates unilaterally to the east. The slip distribution remains relatively simple (briefly one big asperity, illustrated again in Figure 2), but rupture time and slip time function are not imposed. The fault geometry used here is not a single plane fault, but a geometrically irregular fault. All these dynamic factors might have made the wave radiation complex. The ground motion is calculated in the 3D structure model (See D5.3) under the same situations.

Figure 1. A set of the dynamic rupture simulations from Aochi & Ulrich (BSSA, 2015). The scenario used for the blind test corresponds to the case at the upper-‐right corner.

MARSite (GA 308417) D5.4 4

Figure 2. Slip distribution (top) and rupture time (bottom) from the dynamic rupture simulation, projected on a plane fault.

The data set of the same 47 receivers can be downloaded from http://aochi.hideo.perso.neuf.fr/marmara/ (Scenario 1). However for the purpose of blind test, the above information was not informed to the participants before their inversion. Only the following information is provided. Thus, estimating a reasonable magnitude is also an objective of the inversion here.

Informed Parameter Quantity Hypocenter 28.5274°E, 40.8717°N, 9.75 km depth Rupture It seems that the main Central Marmara part

of the North Anatolian fault takes place an earthquake. No fault geometry is given but the same fault orientation as the checker-‐board test is inferred.

2.2 GFZ’ team results

2.2.1 Geometrical setting-‐ Fault parametrization -‐ Data processing

We used a rectangular fault plane (80 km × 24 km) that constructed based on the focal mechanism (strike=86°, dip=90°). This fault geometry was discretized to 120 subfaults of 4 km × 4 km size, each being treated as a point source. We fixed the rake angle to be 180°. The data was processed using the same approach as mentioned in D5.3 (Section 2.2.1). We did not constraint the rupture velocity, rise time and the shape of the rise time function in the inversion. The corner frequency used for the blind test is 0.10 Hz.

2.2.2 Results

MARSite (GA 308417) D5.4 5

The inverted source model of the blind test is shown in Fig. 3. The rupture model suggests that the moment magnitude is 7.03 and the rupture process last for 20 second. With a rupture scale of ~ 35 km and a slip maximum of 5 m, the fault slip are mostly located at depths less than 16 km. The data fit is shown in Fig. 4, from which we found that the large misfit mainly lies at the high frequency components that induced by 3D wave propagation effect.

The computation time used for inversion of the rupture models is 120 seconds with a PC (Intel(R) Core (TM) i7-‐3770 CPU @3.4G Hz, 32 GB RAM).

Figure 3. Rupture model inverted based on synthetic datasets that simulated from 3D earth structure.

2.2.3 Test for near real-‐time kinematic inversions

For purposes of tsunami early warning and earthquake rapid response, time-‐dependent finite-‐fault source inversions are of great importance. Here we address the question how fast such source models can be obtained theoretically for the Marmara Sea region. The time delay for kinematic source inversion addressed here is not the computation time but the time for seismic wave propagation from the source to the seismic stations.

For this purpose, we adopt the IDS inversion scheme and predetermined fault geometry (Fig. 5) for a real-‐time data processing. Here the term “real-‐time” means using only the data which become available. We used a different earthquake scenario with moment magnitude of 7.19, which is the scenario 2 from Aochi H. (http://aochi.hideo.perso.neuf.fr/marmara/). We assume that large earthquakes (M 7+) can only occurred on the main Marmara Sea fault. For this reason, all major seismogenic faults in the Marmara Sea region identified by Armijo et al. (2005) were assumed to have the same potential to nuclear large-‐scale earthquakes in the inversion. The right-‐lateral strike-‐slip mechanism is assigned uniformly according to various geological and geomechanical investigations (Armijo et al., 2005; Hergert et al.,

MARSite (GA 308417) D5.4 6

2011). These faults are then divided into 4 km × 4 km subfaults. Each subfault is allowed to rupture at any time after the origin of the event.

For the real-‐time inversions, we repeat the inversion each 10 s. The results are the time-‐dependent slip distributions shown in Fig. 5 and Fig. 6. From the real-‐time magnitude curve, the final moment magnitude (Mw 7.19) of the utilized scenario earthquake seems to be available at about 30 s after the true event occurrence (Fig. 6). However, it is partly contributed by the numerical noises on neighboring non-‐causal faults. Figure 5 shows the slip distribution on the causal fault stabilizes not earlier than 50 s after the event occurrence, but the causal fault can be recognized already from 20-‐30 s. Note that the rupture duration of the earthquake is 20 s. Thus, the actual time delay for getting the final source model is at least 30 s, a large part of which is caused by the travel times of S waves to the seismic stations.

Figure 4. Waveform comparison between synthetic (blue) and inverted (red) velocity seismograms for the rupture model shown in Fig. 3.

MARSite (GA 308417) D5.4 7

Figure 5. Comparison between the input (top left) and real-‐time reconstructed slip models. The moment magnitudes outside the brackets are inferred from the seismic moment distributed on the whole fault system, while that in the brackets corresponds to seismic moment located on the main rupture fault.

Figure 6. Moment magnitude curves obtained by the retrospective (red) and real-‐time reconstruction (blue) in comparison with the input one (black).

MARSite (GA 308417) D5.4 8

2.3 INGV’ TEAM RESULTS

2.3.1 Geometrical setting-‐ Fault parametrization -‐ Data processing We used the 47 synthetic seismograms provided for Scenario1 (http://aochi.hideo.perso.neuf.fr/marmara/). Original ground velocity time histories are band-‐pass filtered in two different frequency ranges; between 0.01 and 0.5 Hz and between 0.01 and 0.25Hz, by using a two-‐pole and two-‐pass Butterworth filter. We invert 60 seconds of each waveform, including body and surface waves. We assumed a fault plane consistent with the hypocentre location and the focal mechanisms given for the Central Marmara segment (Scenario1 strike: 86°; dip=90°). The fault plane is 95 km long and 24 km width, along strike and down-‐dip direction, respectively. We invert simultaneously for all the parameters at nodal points equally spaced (4.5 km) along strike and dip directions. During the inversion, we fix a given range of variability for each model parameter. In particular, in this study we adopt the following variability intervals: peak slip velocity values can range between 0 and 7.0 m/s at 0.25 m/s interval; the rise time between 1.0 and 4 sec at 0.25 sec interval and the rupture time at each grid node is constrained by the arrival time from the hypocentre of a rupture front having a speed comprised between 2 and 4 km/s. Rake angle is fixed to 180°. In this study, the adopted source time function is a regularized Yoffe function having a constant time to peak slip velocity (Tacc) equal to 0.225 sec (Tinti et al., 2005). We adopted the inversion technique described in Section 2.3.1 (D5.3).

2.3.2 Results

We performed six different inversions:

1. by inverting for peak slip velocity, in the frequency range 0.01-‐0.25 Hz. Rise time, rake angle and rupture velocity are fixed; CPU=13minutes; (results are referred as case ‘1par0.25Hz’);

2. by inverting for peak slip velocity, in the frequency range 0.01-‐0.5 Hz. Rise time, rake angle and rupture velocity are fixed; CPU=1h; (results are referred as case ‘1par0.5Hz’);

3. by inverting for peak slip velocity and rise time, in the frequency range 0.01-‐0.25 Hz. Rake angle and rupture velocity are fixed; CPU=2h; (results are referred as case ‘2par0.25Hz’);

4. by inverting for peak slip velocity and rise time, in the frequency range 0.01-‐0.5 Hz. Rake angle and rupture velocity are fixed; CPU=6h ; (results are referred as case ‘2par0.5Hz’);

5. by inverting for peak slip velocity, rise time and rupture time, in the frequency range 0.01-‐0.25 Hz. Rake angle is fixed to 180°; CPU=3.30h ; (results are referred as case ‘Allpar0.25Hz’);

MARSite (GA 308417) D5.4 9

6. by inverting for peak slip velocity, rise time and rupture time, in the frequency range 0.01-‐0.5 Hz. Rake angle is fixed to 180°; CPU= 8.30h; (results are referred as case ‘Allpar0.5Hz’).

For each inversion we show (see Figures 7-‐9-‐11-‐13-‐15-‐17) the retrieved rupture model, given in terms of rise time, slip, peak slip velocity and rupture times distribution (depending on the set of inverted kinematic parameters) on the fault plane, and the corresponding comparison between synthetic (blue lines) and inverted (red lines) velocity time histories (see Figures 8-‐10-‐12-‐14-‐16-‐18).

MARSite (GA 308417) D5.4 10

Case: 1par0.25Hz

Figure7. Retrieved rupture model, in terms of slip and rise time distribution (upper and bottom panel, respectively), obtained by inverting for peak slip velocity, in the frequency band 0.01-‐0.25Hz.


0.00.51.01.52.02.53.03.54.0sec

rise timeE W

-18(km)

(km)

0.00.51.01.52.02.53.03.54.04.55.0m

slipE W

-18(km)

(km)28.5 66.59.5 47.5 85.50 95

28.5 66.59.5 47.5 85.50 95

MARSite (GA 308417) D5.4 11

Case: 1par0.5Hz

Figure9. Retrieved rupture model, in terms of slip and rise time distribution (upper and bottom panel, respectively), obtained by inverting for peak slip velocity, in the frequency band 0.01-‐0.5Hz.


0.00.51.01.52.02.53.03.54.0

secrise timeE W

-18(km)

(km)

0.00.51.01.52.02.53.03.54.04.55.0m

slipE W

-18(km)

(km)28.5 66.59.5 47.5 85.50 95

28.5 66.59.5 47.5 85.50 95

MARSite (GA 308417) D5.4 12

Case: 2par0.25H

Figure11. Retrieved rupture model, in terms of slip and rise time distribution (upper and bottom panel, respectively), obtained by inverting for peak slip velocity and rise time, in the frequency band 0.01-‐0.25Hz.


0.00.51.01.52.02.53.03.54.04.55.0

m

(km)

slipE W

-18

(km)

0.00.51.01.52.02.53.03.54.0

sec

rise timeE W

-18

(km)

(km)

28.5 66.59.5 85.50 9547.5

28.5 66.59.5 85.50 9547.5

MARSite (GA 308417) D5.4 13

Case: 2par0.5Hz

Figure13. Retrieved rupture model, in terms of slip and rise time distribution (upper and bottom panel, respectively), obtained by inverting for peak slip velocity and rise time, in the frequency band 0.01-‐0.5Hz.


0.00.51.01.52.02.53.03.54.0

secrise timeE W

-18

(km)

(km)

0.00.51.01.52.02.53.03.54.04.55.0

mslipE W

-18

(km)

(km)28.5 66.59.5 85.50 9547.5

28.5 66.59.5 85.50 9547.5

MARSite (GA 308417) D5.4 14

Case: Allpar0.25Hz

Figure15. Retrieved rupture model, in terms of slip, rise time, peak slip velocity and rupture time distribution (upper, middle, and bottom panel, respectively), obtained by inverting for peak slip velocity, rise time and rupture time, in the frequency band 0.01-‐0.25Hz. White contours in bottom panel show the retrieved rupture times.


0.00.51.01.52.02.53.03.54.04.55.05.56.06.57.0

m/speak slip velocityE W

2

3

-18

(km)

(km)

0.00.51.01.52.02.53.03.54.0

secrise timeE W

-18

(km)

(km)

0.00.51.01.52.02.53.03.54.04.55.0

mslipE W

-18

(km)

(km)28.5 66.59.5 85.50 9547.5

28.5 66.59.5 85.50 9547.5

28.5 66.59.5 85.50 9547.5

MARSite (GA 308417) D5.4 15

Case: Allpar0.5Hz

Figure17. Retrieved rupture model, in terms of slip, rise time, peak slip velocity and rupture time distribution (upper, middle, and bottom panel, respectively), obtained by inverting for peak slip velocity, rise time and rupture time, in the frequency band 0.01-‐0.5Hz. White contours in bottom panel show the retrieved rupture times.


0.00.51.01.52.02.53.03.54.04.55.05.56.06.57.0

m/speak slip velocityE W

1

2

3

-18

(km)

(km)

0.00.51.01.52.02.53.03.54.0

secrise timeE W

-18

(km)

(km)

0.00.51.01.52.02.53.03.54.04.55.0

mslipE W

-18

(km)

(km)28.5 66.59.5 85.50 9547.5

28.5 66.59.5 85.50 9547.5

28.5 66.59.5 85.50 9547.5

MARSite (GA 308417) D5.4 16

All the retrieved rupture models are characterized by a main patch with a maximum slip value of 4 m that extends from the right-‐upper fault plane border down to a depth of 14 km. The associated moment magnitudes are all around 7 (see fourth and seventh column in Table1). The total duration of the rupture is about 20s; and the slip asperity is characterized by rise time of about 2-‐3.5 s (Figures 15-‐17). Table1 also shows, for each performed inversion, the corresponding waveform’ cost function (second and fifth column in Table1) and the computation time (CPU), (third and sixth column in Table1). All inversions were done with a PC (Intel Xeon E5-‐2680 processor running CPU @2.7GHz, 128 GB RAM). As we can see, our code is able to retrieve a good rupture model (cost=0.35) in 13minutes. We consider this result a rapid and reliable estimation of the main rupture process’ features. Table1. CPU, waveform Cost function, Moment Magnitude values

cost CPU

Mw cost CPU

Mw

Allpar

0.23

3.30h

7.10

0.42

8.30h

7.07

2par

0.29

2.00h

7.08

0.56

6.00h

7.06

1par

0.35

13min

7.01

0.50

1.00h

7.00

3. Conclusions Information about the extended source properties are needed for performing the ground motion simulation associated to the earthquake rupture on the causative fault. The main goal of this task was the fast determination of the earthquake source, with special focus on its finite-‐fault characteristics. The obtained result, for the blind test, show that, by inverting near-‐field strong-‐motion and high-‐rate GPS data in the Marmara Sea we are able to provide a rapid (CPU between 2 and 13 minutes) and reliable reconstruction of the rupture process of large earthquakes, by retrieving the most relevant earthquake source parameters. These analyses can be done in near real time and are particularly suited for capturing near-‐source large earthquakes. The proposed approach represents a helpful tool to improve rapid ground-‐motion simulations in case of large earthquakes in the Marmara region. Moreover, according to the test results, performed by the GFZ

0.01 – 0.25 Hz 0.01 – 0.5 Hz

MARSite (GA 308417) D5.4 17

team, near real-‐time source characterization of large-‐scale earthquakes (Mw ≥ 7) under the Marmara Sea is feasible. Providing the real-‐time data acquisition for the current network and a good database of the active fault system, all key source parameters that are relevant for purpose of the rapid hazard assessment can be estimated without substantial uncertainties. The theoretical time delay between what can be resolved and what has been really occurred on the earthquake source is in the order of 10-‐15 s. The cause of this time delay is mainly physical, namely by the S wave propagation from the source to the network. In practice, a slightly larger time delay should be considered because of the time to be required additionally for the data transmission and inversion. The latter, however, can be generally reduced to a few seconds through parallelization of the IDS inversion technique.

References

Aochi, H. and T. Ulrich, A probable earthquake scenario near Istanbul determined from dynamic simulations, Bull. Seism. Soc. Am. , 105, doi:10.1785/0120140283, 2015.

Armijo, R. & 21 others, 2005. Submarine fault scarps in the Sea of Marmara pull-‐apart (North Anatolian Fault): Implications for seismic hazard in Istanbul, Geochem. Geophys. Geosyst., 6, Q06009, doi:10.1029/2004GC000896.

Hergert, T., Heidbach, O., Bécel, A. & Laigle, M., 2011. Geomechanical model of the Marmara Sea region -‐ I. 3-‐D contemporary kinematics, Geophys. J. Int., 185, 1073-‐1089, doi:10.1111/1365-‐246X.2011.04991.x.

Tinti, E., E. Fukuyama, A. Piatanesi and M. Cocco (2005a), A kinematic source time function compatible with earthquake dynamics, Bull. Seismol. Soc. Am., 95(4), 1211-‐1223, doi:10.1785/0120040177.

D5.3! …...MARSite((GA(308417)(D5.3((5" " Top! of layer (km)! Vp(m/s)! Vs(m/s)! Density!(kg/m3)!Q*!...

Documents

Transcript of D5.3! …...MARSite((GA(308417)(D5.3((5" " Top! of layer (km)! Vp(m/s)! Vs(m/s)! Density!(kg/m3)!Q*!...