1 Rapid Source Inversion using GPS and Strong-Motion Data -A case Study on the 2011 Mw 9.0 Tohoku-Oki Earthquake R. Wang, S. Parolai, M. Ge, M. Jin, T.R. Walter, J. Zschau Helmholtz Centre Potsdam GFZ German Research Centre for Geosciences ESF-COST High-Level Research Conference 27 November - 2 December 2011 Sant Feliu de Guixols, Spain

2 Introduction Near-field ground-motion information Strong motion dataStrong constraint on fault size and rupture distribution Less sensitive to earth structure than teleseismic data  Most useful for early warning and rapid response systems Strong motion data Real-time, simple data processing Fine resolution at high frequencies Baseline errors High-rate GPS data Near real-time High accuracy at low frequencies Large noise at high frequencies This study: Comparison of GPS and strong-motion derived co-seismic displacement Comparison of strong-motion and high-rate GPS records Joint use of GPS and SM data for rapid source inversion

3 for Real Time Earthquake Risk ReductionStrategies and Tools for Real Time Earthquake Risk Reduction From description of the REAKT project (2011)

4 Japan has one of the densest geodetic and seismic networks of the world:GeoNet ~ 1,200 permanent GPS stations F-Net 84 broadband stations Hi-Net 777 high-sensitivity borehole seismic stations K-Net ~ 1,000 strong-motion stations Kik-Net 777 strong-motion stations (co- installed with Hi-Net), each with both surface & borehole sensors Data used in this study (more details see:

5 The 2011 Mw 9.0 Sendai (Japan) earthquakePGA ~ 3g! From the NIED report (2011)

6 Baseline errors of digital strong-motion recordsOriginal strong-motion data with pre-event baseline offset Dominantly by shaking induced ground tilt, but also by Instrumental effect All seismic sensors with a spring-mass system do not measure the ground motion directly but gravity and inertial force acting on the sensor. Velocity seismogram (integrated from the strong-motion data after correction of the pre-event offset) Drift due to co- and post-event baseline shifts Inertia Tilt Gravity changes

7 Empirical baseline correction based on Iwan’s ideaWang et al. (BSSA, 2011)

8 An example from 2010 Mw 8.8 Maule earthquake

9 Comparison of coseismic static displacementsComparison with GPS K-Net: many stations on soft ground  large uncertainties KiK-Net (surface): coherent spatial variability except for a few outliers KiK-Net (borehole): most stable with less outliers

10 GPS and SM-derived coseismic displacement dataSource distribution inverted from GPS and SM-derived coseismic displacement data

11 SDM2008 A software code for slip inversionbased on the constrained least-squares method  Minimization of objective function by Steepest Descent Method Joint inversion of different datasets (d1, d2, …) Green’s functions (M) based on layered half-space Earth model Simultaneous determination of unknown offsets (d1o, d2o, …) included in the datasets Optional slip (s) or stress-drop () Smoothing ()  weak constraint A-priori conditions on slip amplitude and rake angle  hard constraint Convex objective function  fast convergence to global minimum

12 Geodetic slip models This study Simons et al. (2011)Land GPS + Tsunami Data Geodetic slip models This study Pollitz et al. (2011) Land + OB GPS Data

13 Forward modelling of tsunamiModel with max. 25m slip Did the fault slip exceed 50m? Most likely! Tsunami model by A. Babeyko (2011) Model with max. 50m slip The underestimate of fault rupture led to the underestimate of tsunami height: Less than 10m predicted instead of 20-30m observed!

14 from SM-derived coseismic displacement dataSlip models inverted from SM-derived coseismic displacement data Large uncertainties of the empirical baseline correction  Outliers in the derived displacements that may affect the source inversion dramatically.

15 A model-based detection of outliers in the SM datasetDefinition: An outlier is detected if it deviates from the preliminary model by larger than a given threshold (here 15o in direction) Applicability: Dense network Isolated outliers, particularly in the far-field

16 Joint inversion using the GPS and SM dataOnly 10% of the data randomly selected  To simulate the situation in other regions with much sparser monitoring networks than in Japan.

17 Comparison Joint Data Processingof strong-motion and high-rate GPS records and their for the 2011 M9 9.0 Tohoku-Oki earthquake - Joint Data Processing

18 Strong-motion versus high-rate GPSExample from a nearby GPS/SM station pair 14 nearby GPS/SM stations ( 4 km)

19 Strong-motion versus high-rate GPS for all 14 nearby stations ( 4 km)

20 Complementary information from GPS/SM20 m distance High-rate GPS Low sampling rate, high-frequency noise  Late event detection, underestimated ground shaking Strong-motion data Small baseline errors  Large displacement trend

21 Joint processing of the GPS/SM dataAssume a low-order polynomial baseline error in the SM data Determine the polynomial so that the derived displacement history after the baseline correction best fits the GPS data Make the polynomial baseline correction Integrate the corrected acceleration to velocity and displacement Estimate the noise in the GPS data “Tightly Integrated Processing of High-Rate GNSS and Seismic Sensor Data” A single-frequency GPS co-installed with A low-cost accelerometer Near real-time and complete-band ground motion information

22 Conclusions Strong-motion derived coseismic displacementLarge uncertainties for stations at soft sediment sites, but Stable results for borehole and hard-rock stations Few outliers can be detected using the model-based approach, improving constraints on the source Source inversion High resolution using both onshore and offshore GPS data Robust solution for magnitude, fault location and size can also be obtained using the SM-derived coseismic displacement data It is evident that the fault slip exceeded 50m in places Complementary information from strong-motion and high-rate GPS data Complete-band ( Hz) ground motion information (displacement, velocity and acceleration) by combined data processing Feasible using low-cost but co-installed GPS and strong-motion sensors Useful for near real-time source reconstruction, i.e. soon after the earthquake or even during the rupture process is still going on (REAKT project WP4.1)

23 Thank you!