EPOS use case: High-resolution seismic tomography of Italy

Slides:



Advertisements
Similar presentations
INGV ShakeMaps for the lAquila earthquake Alberto Michelini, Licia Faenza & Valentino Lauciani INGV, Centro Nazionale Terremoti.
Advertisements

Real-time data for seismic monitoring at INGV National Earthquake Center S. Mazza et al. Centro Nazionale Terremoti Istituto Nazionale di Geofisica e Vulcanologia.
A Partition Modelling Approach to Tomographic Problems Thomas Bodin & Malcolm Sambridge Research School of Earth Sciences, Australian National University.
Numerical methods in the Earth Sciences: seismic wave propagation Heiner Igel, LMU Munich III The latest developments, outlook Grenoble Valley Benchmark.
10/09/2007CIG/SPICE/IRIS/USAF1 Broadband Ground Motion Simulations for a Mw 7.8 Southern San Andreas Earthquake: ShakeOut Robert W. Graves (URS Corporation)
Prague, March 18, 2005Antonio Emolo1 Seismic Hazard Assessment for a Characteristic Earthquake Scenario: Integrating Probabilistic and Deterministic Approaches.
1 High Performance Computing at SCEC Scott Callaghan Southern California Earthquake Center University of Southern California.
Astrophysics, Biology, Climate, Combustion, Fusion, Nanoscience Working Group on Simulation-Driven Applications 10 CS, 10 Sim, 1 VR.
The RSNI Seismic Network Laboratory of seismology, DISTAV - University of Genoa (Italy) RSNI STAFF RSNI STAFF : Spallarossa D.,
Jan. 14, 2008Southern Great Basin & Las Vegas1 3D Models of the Southern Great Basin and Ground Motion in Las Vegas Arthur Rodgers Seismology Group Atmospheric,
Jeroen Tromp Computational Seismology. Governing Equations Equation of motion: Boundary condition: Initial conditions: Earthquake source: Constitutive.
Surface wave tomography: part3: waveform inversion, adjoint tomography
MARsite kickoff meeting December 19-20, 2012, Istanbul WP5 - TASK 2 Near real-time determination of the earthquake finite-fault source parameters and models,
CYBERINFRASTRUCTURE FOR THE GEOSCIENCES High Performance Computing applications in GEON: From Design to Production Dogan Seber.
A Metadata Based Approach For Supporting Subsetting Queries Over Parallel HDF5 Datasets Vignesh Santhanagopalan Graduate Student Department Of CSE.
NSF Geoinformatics Project (Sept 2012 – August 2014) Geoinformatics: Community Computational Platforms for Developing Three-Dimensional Models of Earth.
GSI 1.0 by A. Elmekati M. Zeghal Geotechnical System Identification Software Framework 8/20/07 Introducing.
35 1 Introduction Since 2004, the EU funded Marie Curie Training Network "Seismic wave propagation and imaging in complex media: a European network" joins.
A Framework for Elastic Execution of Existing MPI Programs Aarthi Raveendran Tekin Bicer Gagan Agrawal 1.
SPICE Research and Training Workshop III, July 22-28, Kinsale, Ireland presentation Seismic wave Propagation and Imaging in Complex media:
An array analysis of seismic surface waves
Seismic Waveform Tomography Jeroen Tromp. Classical Tomography Theoretical limitations due to use of 1D background models Data coverage Pervasive (ab)use.
Presented by On the Path to Petascale: Top Challenges to Scientific Discovery Scott A. Klasky NCCS Scientific Computing End-to-End Task Lead.
1 SCEC Broadband Platform Development Using USC HPCC Philip Maechling 12 Nov 2012.
ICPP 2012 Indexing and Parallel Query Processing Support for Visualizing Climate Datasets Yu Su*, Gagan Agrawal*, Jonathan Woodring † *The Ohio State University.
Computational Seismology at LLNL: A National Lab Perspective Arthur Rodgers Atmospheric, Earth and Environmental Sciences Department Lawrence Livermore.
Complex earthquake directivity during the 2009 L’ Aquila mainshock Tinti E., Scognamiglio L., Cirella A., Cocco M., and A. Piatanesi Istituto Nazionale.
Fusion-SDM (1) Problem description –Each run in future: ¼ Trillion particles, 10 variables, 8 bytes –Each time step, generated every 60 sec is (250x10^^9)x8x10.
How DO they actually locate earthquakes? Alternatives to S-P Triangulation Hubenthal, M. Taber, M.
Disputable non-DC components of several strong earthquakes Petra Adamová Jan Šílený.
Lg Q Across the Continental US Dan McNamara and Rob Wesson with Dirk Erickson, Arthur Frankel and Harley Benz.
3D Seismic Imaging based on Spectral-element Simulations and Adjoint Methods Qinya Liu Department of Physics University of Toronto 1 st QUEST Workshop,
GEON2 and OpenEarth Framework (OEF) Bradley Wallet School of Geology and Geophysics, University of Oklahoma
Validation of physics-based ground motion earthquake simulations using a velocity model improved by tomographic inversion results 1 Ricardo Taborda, 1.
High Resolution Finite Fault Inversions for M>4.8 Earthquakes in the 2012 Brawley Swarm Shengji Wei Acknowledgement Don Helmberger (Caltech) Rob Graves.
CISN: Draft Plans for Funding Sources: OES/FEMA/ANSS/Others CISN-PMG Sacramento 10/19/2004.
MARsite kickoff meeting December 19-20, 2012, Istanbul INGV contribution to WP5 Tasks 2 & 4 Alessio Piatanesi.
Antonella Cirella, Alessio Piatanesi, Elisa Tinti, Massimo Cocco Ground Motion and Source Process of the 6 th April 2009 L’Aquila, central Italy, Earthquake.
Ground motion simulations in the Pollino region (Southern Italy) for Mw 6.4 scenario events.
Regional On-Demand Simulations Near real-time simulations of all M>3.5 earthquakes Analysis of past events –Scientific database: 3D synthetic seismograms.
06/22/041 Data-Gathering Systems IRIS Stanford/ USGS UNAVCO JPL/UCSD Data Management Organizations PI’s, Groups, Centers, etc. Publications, Presentations,
Visualizing TERASHAKE Amit Chourasia Visualization Scientist Visualization Services San Diego Supercomputer center Geon Visualization Workshop March 1-2,
Phase 1: Comparison of Results at 4Hz Phase 1 Goal: Compare 4Hz ground motion results from different codes to establish whether the codes produce equivalent.
Surface-wave Derived Focal Mechanisms in Mid-America R. B. Herrmann 1, C. J. Ammon 2 and H. M. Benz 3 1 Saint Louis University, 2 Pennsylvania State University,
Southern California Earthquake Center CyberShake Progress Update 3 November 2014 through 4 May 2015 UGMS May 2015 Meeting Philip Maechling SCEC IT Architect.
Unified Structural Representation (USR) The primary mission of the USR Focus Area has been the development of a unified, object-oriented 3-D representation.
UCERF3 Uniform California Earthquake Rupture Forecast (UCERF3) 14 Full-3D tomographic model CVM-S4.26 of S. California 2 CyberShake 14.2 seismic hazard.
Focal mechanisms and moment tensors of micro-earthquakes in the Malé Karpaty (Little Carpathians) Mts., Slovakia Lucia Fojtíková 1, Václav Vavryčuk 2,
A Blind Test of Traveltime and Waveform Inversion Colin A. Zelt 1, R. Gerhard Pratt 2, Andrew Brenders 2, Sara Hanson-Hedgecock 1 and John A. Hole 3 1.
VERCE Cloud presentation. EGI CF VERCE & Cloud Steffen Claus Lion Krischer Andre Gemünd Virtual Earthquake and seismology.
Southern California Earthquake Center CyberShake Progress Update November 3, 2014 – 4 May 2015 UGMS May 2015 Meeting Philip Maechling SCEC IT Architect.
A brief introduction to NERA – NA3 work package ‘Networking accelerometric networks and strong- motion data users’ Sinan Akkar On behalf of entire NERA-NA3.
Fast Least Squares Migration with a Deblurring Filter 30 October 2008 Naoshi Aoki 1.
SCEC Capability Simulations on TeraGrid
What is SeisFlows? -Provides a complete, customizable waveform inversion workflow -So far, has been used for 3D production runs with up to 10 9 model parameters.
Alessandro Spinuso, Andreas Rietbrock, Andrè Gemuend,
Susan L. Beck George Zandt Kevin M. Ward Jonathan R. Delph.
Kinematic Modeling of the Denali Earthquake
In-situ Visualization using VisIt
Scott Callaghan Southern California Earthquake Center
The SCEC Broadband Platform: Computational Infrastructure For Transparent And Reproducible Ground Motion Simulation Philip J. Maechling [1], Fabio Silva.
High-F Project Southern California Earthquake Center
Philip J. Maechling (SCEC) September 13, 2015
2University of Southern California, 3San Diego Supercomputer Center
17-Nov-18 Parallel 2D and 3D Acoustic Modeling Application for hybrid computing platform of PARAM Yuva II Abhishek Srivastava, Ashutosh Londhe*, Richa.
Skeletonized Wave-Equation Surface Wave Dispersion (WD) Inversion
High-Performance Computing (HPC) IS Transforming Seismology
Growing importance of metadata for synthetics: Calculating and Sharing Synthetic Seismic Data Dogan Seber University of California, San Diego San Diego.
Full-waveform Inversion of GPR Data and its Frequency
CyberShake Study 18.8 Technical Readiness Review
Presentation transcript:

EPOS use case: High-resolution seismic tomography of Italy Giuseppe Fiameni with the collaboration of Alberto Michelini, Federica Magnoni, Emanuele Casarotti

Outline Objective Motivation EPOS’s use case Requirements Actual status Proposal

Objective High resolution tomographic velocity model of Italy (i.e., invert the recorded seismic waveform data of earthquakes in Italy and adjacent regions for the 3D, laterally heterogeneous velocity structure of Italy.)

Motivation Big scientific challenge: waveform tomography of te 3D seismic velocity structure of the Italian region up to frequencies of 0.5 Hz (i.e., maximum spatial resolution ~2 Km) has never been feasible before Very relevant implications for modeling at high resolution strong ground shaking (large earthquakes) scenarios with evident seismic engineering applications The use case workflow conjugates data storage retrieval, staging, HPC modeling and stage-out of large volumes of simulated data, detailed 3D velocity model and adjoint kernels

EPOS Workflow Observed Data Forward Modeling Inverse Visualization Input Parameters

Forward Modeling velocity model Synthetic 3D wavefield source model 2009 L’Aquila earthquake simulation for 3D velocity model and finite fault La struttura del progetto è articolata in due fasi velocity model source model topography/geology RUN Synthetic 3D wavefield simulations

Iteratively improve the velocity model Inverse Modeling velocity model source model topography/geology TOMOGRAPHY Misfit between synthetics and data Iteratively improve the velocity model observed data

Remarks on Computational Resources All the computing times and storage presented refers to: Central Italy adjoint tomography (Magnoni 2012) using SPECFEM3D INGV HPC Resources: ELIOS 64 compute nodes, each with 2 quad-core AMD Opteron 2374 processors at 2.4 GHz and with 16 GB RAM (512 total cores, 2 GB RAM/core) SELENE 32 compute nodes, each with 2 oct-core AMD Opteron 6136 processors at 2.4 GHz and with 32 GB RAM (512 total cores, 2 GB RAM/core) then extrapolated for Italy

DATA & Input Parameters Forward Modeling Inverse Visualization Observed Data Input Parameters Observed waveforms Source solutions Geology i dati li prendiamo in FULL SEED format tutti e tre gli elementi sono ottenuti ‘manualmente’ geology: include known information depending on frequency x x Quality check: no gaps no saturation signal/noise station metadata Selection: spatial distribution magnitude MESH

MESH Geological Discontinuities (Moho, basins...) Topography Forward Modeling Inverse Visualization Observed Data Input Parameters Geological Discontinuities (Moho, basins...) Topography Velocity model MESHER (CUBIT/GEOCUBIT)

MESH Geological Discontinuities (moho, basin...) Topography Velocity model MESHER (CUBIT/GEOCUBIT)

Db of saved state variables for the last timestep Forward Modeling Observed Data Forward Modeling Inverse Visualization Input Parameters Station location Source solutions Mesh Partitioning SCOTCH Synthetic seismogram db (ASCII - SAC) Generate Database Db of saved state variables for the last timestep (binary) SPECFEM3D Solver SPECFEM3D

Processing (user routines) Inverse Modeling Synthetic seismogram db Observed Processing (user routines) Source timing Instrument metadata Synchronization Instrument response Filtering Data Forward Modeling Inverse Visualization Input Parameters station metadata --> strumento

Inverse Modeling Syncronization Instrument response Filtering Synthetic seismogram db Observed Processing Source timing Instrument metadata Syncronization Instrument response Filtering Data Forward Modeling Inverse Visualization Input Parameters Text file with selected windows 55DATA/MN.AQU..HHR.D.SAC.6.20.realSYN/AQU.MN.FXR.semd.sac.6.20.synt 2 -1.8396 29.4404 29.4404 42.8104DATA/MN.AQU..HHT.D.SAC.6.20.realSYN/AQU.MN.FXT.semd.sac.6.20.synt 1 -1.5896 25.7604 ... Time window selection Serial fortran code FLEXWIN (Maggi et al 2009) Adjoint source db (ASCII) Misfit measurement and adjoint source calculation Serial fortran code measure_adj (Tape et al. 2009) Text file with measurement estimates + Adjoint source db (ASCII)

(Forward simulation only if attenuation is included) Inverse Modeling Observed Data Forward Modeling Inverse Visualization Input Parameters Adjoint source db (ASCII) Station location Source solutions Mesh Saved state variables for the last timestep (Forward simulation only if attenuation is included) Adjoint event kernel db (binary) Adjoint Simulations SPECFEM3D

Inverse Modeling Adjoint event kernel db (binary) Adjoint misfit Observed Data Forward Modeling Inverse Visualization Input Parameters Adjoint event kernel db (binary) Adjoint misfit kernel db (binary) Regularized misfit kernel db (binary) SUM SUM Sum SPECFEM3D fortran utilities Preconditioning - Smoothing SPECFEM3D fortran utilities

Inverse Modeling Regularized misfit kernel Starting velocity model Observed Data Forward Modeling Inverse Visualization Input Parameters Regularized misfit kernel db (binary) Starting velocity model (binary) SPECFEM3D fortran utilities computationally inexpensive 1.7 external_mesh.bin 3 vp vs rho in vtk UPDATED velocity models (binary & VTK ASCII) Forward Simulations for steplength test SPECFEM3D

Seismic source inversion Input Parameters Observed Data Inverse Modeling Forward Simulations Forward Modeling Forward Modeling Inverse Modeling Visualization 1 2 3 Updated velocity model db 4 5 Iterative improvement of the tomographic model 6 Seismic source inversion FUTURE TASK

KERNEL SUMMATION and REGULARIZATION Inverse Modeling Wall Time on 256 CPUs Cluster C. Italy 256 cores 63 events Italy 2300 cores 150 events Forward Simulations Forward Modeling Forward Modeling Inverse Modeling Visualization FORWARD SIMULATIONS 42 h + 0.5/31 h 84 h 3 h 169/231 h 4 days + 0.5/75 h 8 days 3 h 16/22 days WINDOW SELECTION MEASUREMENTS ADJOINT SOURCES Updated velocity model db (binary & vtk ASCII) ADJOINT SIMULATIONS -il numero prima dello slash è se faccio selez win e meas (compreso quello nello steplength test) in parallelo cioè per tutti gli ev contemp -steplength test ha sia le sim forw che il calcolo delle meas Per l’Italia è fino a 0.5 Hz KERNEL SUMMATION and REGULARIZATION Iterative improvement of the tomographic model STEP LENGTH TEST ~ 13/16 days ~ 28/37 days TOTAL TIME 1 ITERATION

KERNEL SUMMATION and REGULARIZATION Inverse Modeling Storage C. Italy 256 cores 63 events Italy 2300 cores 150 events Forward Modeling Forward Simulations Forward Modeling Inverse Modeling Visualization FORWARD SIMULATIONS 60 GB + 15 GB 70 GB 7 GB 240 GB + 600 GB WINDOW SELECTION MEASUREMENTS ADJOINT SOURCES 144 GB Updated velocity model db (binary & vtk ASCII) ADJOINT SIMULATIONS -lo storage per la win select è pratic nullo, è tutto nel calcolo meas (cioè stima adj src) -lo storage per adj sim comprende sia ev kernel che sismogrammi -nella voce kernel c’è anche lo storage x model update di 1 iteraz (slide 14) -steplength test = forw*4 825 GB KERNEL SUMMATION and REGULARIZATION 30 GB Iterative improvement of the tomographic model STEP LENGTH TEST 2400 GB 392 GB 4 TB TOT STORAGE 1 ITERATION

Visualization Seismograms (SAC, GMT, ObsPy, Google Earth) Observed Data Forward Modeling Inverse Visualization Input Parameters Seismograms (SAC, GMT, ObsPy, Google Earth) Mesh - Velocity Models (Paraview) parallelo spt per ultimi 3 punti Movies (Paraview) Kernels (Paraview) Presently using ‘manual’ scripts Parallel environment