1. The GEM Computational System and Recent Scientific Results Andrea Donnellan Third International ACES Meeting May 10, 2002 GEM

2. Data Volumes from Observations GRACE: 50 MB/day onboard, 8GB/day derived product ECHO: 100 GB/day onboard SRTM: 12 TB raw data, ICESat: 1 GB/day onboard, 2 GB/day derived SCIGN: 250MB daily - 7.5 GB/day for real time Airborne observations: LIDAR VCL: 2 GB/day onboard, 4 GB/day derived Hyperspectral imagery: 100GB/day raw Imaging LIDAR: >20 GB/day, >40 GB/day

3. Volumes from Models Geodynamo model: –1GB of storage for one model run –2010: 5 TB/run –Minimal need of 10 runs General earthquake/lithospheric models: –1TB/run –2010: 10 PB/run (multiple scales combined, many regions) Gravity –100 GB/run –2010: 2 TB/run Mantle convection models –1 TB/run –2010: 10PB/run Geomagnetic field models –32 GB/run –2010: 300 GB/run

4. Where We Will Be in 2010 Multiple solid earth missions flying PetaBytes of data per year gathered in a distributed fashion Data analyzed by widely distributed scientists using widely distributed computational resources Growing need for integration of information from multiple sources on multiple scales into a integrated analysis

5. Goal World-wide computational systems supporting gathering of 3 PetaBytes of data per year, integrating analysis, visualization, simulation, and interpretation.

6. Requirements Onboard adaptive processing High space to ground bandwidth of TeraBytes per day per mission Data transmission and handling Reusable capabilities (framework) Data processing (100 Petaflops per mission per year)

7. Requirements (continued) Product storage (National Virtual Solid Earth Science Observatory) using cooperative federated databases Distributed computational environment for analysis (interoperable framework, portal) Software tools Hardware

8. Hardware (Hierarchical) Large central Petaflop computers with TeraBytes of memory Single sign-on seamless access Distributed computers for decomposable problems Cluster computers (e.g. Beowulf for cost performance) Heterogeneous computational capabilities (e.g. for storage, visualization, computing)

9. Software Problem Solving Environment –Visualization tools –Analysis algorithms –Data mining Framework –Supports software integration into multidisciplinary analysis –Interoperability between data,software, and computer systems

10. GEM/SERVO Components Visualization Model and algorithm development IT: GRID technologies Computational Environments/PSEs Data handling/archiving Assimilation Datamining/pattern recognition Data fusion High speed networks High end computers Clusters Laptops Cycles needed and other infrastructure Scalable system

11. Solid Earth Research Virtual Observatory (SERVO) Tier2 Center Archive SERVO … Goddard LangleyAmes Institute Fully functional problem solving environment 100 - 1000 Mbits/sec Plug and play composing of parallel programs from algorithmic modules On-demand downloads of 100 GB in 5 minutes 10 6 volume elements rendering in real-time Program-to-program communication in milliseconds Approximately 100 model codes Data cache ~TBytes/day Tier2 Center Tier 0 +1 Tier 1 Tier 3 Tier 4 Tier2 Center 1 PB per year data rate in 2010 Observations Archive Downlink Archive Downlink … … … … … … … 100 TeraFLOPs sustained Tier 2 Workstations, other portals

12. Virtual Observatory Project 20032004200520062007200820092010 Timeline Capability Architecture & technology approach Decomposition into services with requirements Prototype cooperative federated data base service integrating 5 datasets of 10 TB each Prototype data analysis service Prototype modeling service capable of integrating 5 modules Prototype 1920x1080 pixels at 120 frames per second visualization service Scaled to 100 sites Solid earth research virtual observatory (SERVO) On-demand downloads of 100 GB files from 40 TB datasets within 5 minutes. Uniform access to 1000 archive sites with volumes from 1 TB to 1 PB

13. Problem Solving Environment Project 20032004200520062007200820092010 Timeline Capability Isolated platform dependent code fragments Prototype PSE front end (portal) integrating 10 local and remote services Extend PSE to Include 20 users collaboratory with shared windows Seamless access to high-performance computers linking remote processes over Gb data channels. Integrated visualization service with volumetric rendering Fully functional PSE used to develop models for building blocks for simulations. Program-to-program communication in milliseconds using staging, streaming, and advanced cache replication Integrated with SERVO Plug and play composing of parallel programs from algorithmic modules Plug and play composing of sequential programs from algorithmic modules

14. Computational Environment 20032004200520062007200820092010 Timeline Capability 100’s GigaFLOPs 40 GB RAM 1 Gb/s network bandwidth ~100 model codes with parallel scaled efficiency of 50% ~10 4 PetaFLOPs throughput per subfield per year ~100 TeraFLOPs sustained capability per model ~10 6 volume elements rendering in real time Access to mixture of platforms low cost clusters (20-100) to supercomputers with massive memory and thousands of processors

15. The Ventura Basin is Actively Deforming

16. Northridge Example Northridge class simulation: 100,000 unknowns, 4000 time steps –> 8 hours on high end workstation. Southern California system: 0.5 km resolution –> 100,000 processor hours or 400 hours (17 days) on a dedicated 256 processor machine.

17. Steep Gradient Largely Attributable to Low Rigidity Basin Fill

18. Coseismic Removed from the Interferogram Postseismic Interferogram

19. Results from Data Inversion Show Fault Afterslip as Primary Mechanism

20. Comparison of InSAR and Seismic Anomalies Similar anomaly shows up in both the postseismic deformation indicated by GPS and InSAR (Donnellan et al) and seismic anomalies identified using Principal Component Analysis (Rundle and Tiampo). Mojave desert shows a similar correlation near Barstow and the Blackwater Fault (Rundle and Tiampo; Peltzer).

21. Recent GPS Results Similar to pre-seismic velocity field, particularly near the source.

22. Residuals

23. Anomalous Motion at JPL was Observed Related to the Northridge Earthquake Residual Geodetic Longitude (cm) JPL is several fault dimensions away from the Northridge rupture. The earthquake probably triggered slip on the Sierra Madre Fault in the upper 0.5 km. Based on additional observations collected near JPL. Later extent of anomaly is unknown due to lack of stations. Sierra Madre Fault 1 m

24. Faults are shown as light lines, the earthquakes at model year 4526 are shown as dark lines Simulations indicate that major events are clustered in time like the real events. Simulations using a realistic heterogeneous earth structure are computationally intensive. California 3D Fault Simulations

25. Modeling Faults as Interacting Systems Southern California Seismicity Courtesy John Rundle Space-time Stress Diagram Transients likely occur as a result of stress redistribution. Are observed on different faults, sometimes a few fault dimensions away.

26. Conclusions 90% of Northridge postseismic motion was aseismic. Afterslip on the mainshock rupture plane responsible for most of the deformation. No evidence for lower crustal relaxation playing a major role in postseismic motions. Recent deformation is consistent with that observed before the earthquake.

27. More Conclusions High velocity gradient largely attributable to a low rigidity basin. Lower crust is a minor player in interseismic and postseismic motion in this region – consistent with a cold lower crust. The earthquake probably triggered slip on the Sierra Madre fault.