IEEE Visualization 2006 Design Contest

This page describes the format and semantics of the data files for the contest. See the Data Download page to get copies of the actual data.

Data Set Description for TeraShake 2.1

TeraShake 2.1 represents a physics-based simulation of a magnitude 7.7 earthquake on the Southern San Andreas Fault. In the TeraShake 2.1 simulation, a Mw 7.7 earthquake begins South of Palm Springs and propagates Northwest on the San Andreas Fault. The rupture begins at time 0 in the simulation and continues for approximately 60 seconds. At time 60 seconds, the fault rupture stops but the earthquake waves continue to propagate throughout the region for several minutes. The simulation stops at time 250 seconds after the earthquake started by which time most strong ground motions have subsided.

This particular earthquake, a Mw 7.7 Southern San Andreas earthquake was simulated because such earthquakes are known to have occurred in the past and they are reasonably likely to occur again. The simulation results are of significant interest to scientists and the public in the Southern California region. The simulation produces high resolution ground motion records for the entire simulation area including the many large cities in the region. The distribution of strong ground motions and the levels of ground motion in areas of significant populations have important societal implications.

TeraShake 2.1 was performed by propagating earthquake waves through an anelastic volume using finite difference codes on San Diego Supercomputer Center (SDSC) DataStar computer. The material characteristics of the volume are specified prior to running the simulation using the SCEC Community Velocity Model 3.0, which is a realistic 3D velocity structure for the region that was developed by SCEC. The edges and the bottom of the simulation volume are designed to act as absorbing boundaries to reduce or eliminate reflections of waves as they reach the edges of the simulation volume. The surface of the ground is treated as a free surface, not an absorbing boundary. Ground motions at the surface are of particular interest because that is where people live.

This type of simulation has been validated and shown to produce accurate results for earthquake waves at 0.5Hz and longer periods. Validation has been done by running similar simulations for historic earthquakes and comparing the resulting synthetic seismograms produced by the simulation to the seismograms there were recorded by instruments in the region during the actual earthquake.

Region

TeraShake 2.1 simulates earthquake propagating through a volume that represents a region in Southern California. The region used in the TeraShake simulation is a 600x300x80km3 area of Southern California. The surface area of this region is shown in to the right (click on the image for a larger rendition). The geographical origin of this region is Latitude: 34.5, Longitude: -121.0 at depth of 80km. The four coordinates of the box corners are (121W, 34.5N), (118.9511292W, 36.621696N), (116.032285W, 31.082920N), 113.943965W, 33.122341N).

The result is a rectangular region 600 km by 300 km in extent, with long axis oriented N50W and short axis N40E. We extend the domain to 80 km depth, well into the upper mantle. The X coordinates increase moving to the Southeast. The Y coordinates increase moving to the Northeast. The Z coordinates increase moving up towards the surface. This is a right-handed coordinate system.

The image to the right shows three regions of interest. The largest (red) region shows the extent of the simulation volume. The smaller blue region includes the fault zone as well as most of the basins and the large majority of the populated areas in the simulation region. The smallest box (black) indicates the Los Angeles area including the Los Angeles basin. The two smaller regions are of greatest interest to the scientists for this simulation and visualizations may want to focus on these areas.

Data Format

Coordinate System

In this simulation, the region is treated as a flat Cartesian box. The curvature of the earth and the topography are ignored. This volume is divided into regular squares with a 200 meter spacing in all directions in the pre-decimated mesh, creating a 3000x1500x400 point regular mesh. This results in a mesh consisting of 1.8 billion regular cubes. Material properties are specified for each cube at the start of the simulation. The output data is saved as ground velocity for each cube.

The region is specified using geographical coordinates to begin. The geographical coordinates are projected onto a flat surface and then the geographically-based material properties are determined.
Given the geo-referenced material properties, the simulations are then conducted in the internal mesh coordinate system with point 0,0,0 as the origin at the Southwest corner at the bottom of the volume.

Time: The original pre-decimated simulation stored the state of the entire volume every 0.011 seconds. This was calculated for a total of 250.03 seconds for a total of 22730 time steps.

Data Layout

TeraShake 2.1 output data is time-varying velocity vector data. All TeraShake 2.1 simulation data is big-endian, 4 byte, IEEE floating point binary format data. The data is saved in binary files with no headers. To enable proper interpretation of the data, we specify the x,y,z dimensions of the mesh, the geo-referenced coordinates for the mesh, and data types that the floating points represent, the units of the data, and the order in which the indices change.

The TeraShake 2.1 data values are velocity in meters/sec. The velocities created by the simulation are broken into 3 components (x,y,z) and are divided into files by component and time step. Thus, there are 3 files (x,y,z) for each time step.

There are two types of TeraShake 2.1 data presented here: Volumetric+time output wave-propagation data and Volumetric input-mesh data.

Volumetric Wave-Propagation Data Sets

The 4D volumetric velocity data represents the vector velocity state at each point in the volume throughout the simulation. This type of data provides insights into the subsurface paths along which the earthquake waves propagate. Because of the large storage requirements, this data has been decimated in both space and time to produce the data sets for the visualization contest. The 4D volumetric data is decimated by 100 in time (1 volume every 1.1 seconds) and is decimated by a factor of 4 in each spatial dimension. The resulting volume meshes are 750x375x100 in space for a total of 28,125,000 cubes per time step and a total of 227 time steps.

There are 227 files for each component (x,y,z) for a total of 681 files. Each file represents a single velocity component for a volume of earth at a particular time in the simulation. These reduced volume files are named TS21z_Component_R2_Timestep.bin (for example, TS21z_X_R2_008000.bin corresponds to time-step 80 seconds for the X velocity data).

The data in these are 4-byte floats ordered such that the x index varies fastest, then y, then z. Therefore, floats in the file will be ordered like this:

X Mesh Coordinate, Y Mesh Coordinate, Z Mesh Coordinate
0,0,0 <--- 1st 4 bytes of data are for this coordinate
1,0,0 <--- 2nd 4 bytes of data are for its neighbor in x
2,0,0
...
749,0,0
0,1,0
1,1,0
2,1,0
...
749,374,0
0,0,1
1,0,1
2,0,1
...
749,374,99

Other Information:
250.00 TMAX = propagation time (.011 x 22730)
800.00 DH = spatial step for x, y, z (meters)
1.1 DT = time step (seconds)
750 NX = x model dimension in nodes
375 NY = y model dimension in nodes
100 NZ = z model dimension in nodes

Volumetric Input-Mesh Data Set

The material characteristic (stiffness) of the volume through which the waves propogate is specified prior to running the simulation using a 3D model of the Southern California region, derived from the SCEC Community Velocity Model 3.0. The file TS21VelocityMesh_VS_R2.bin contains a scalar field representing material stiffness properties from this simulation, and it is used to determine basins and other features of interest against which the wave-propagation data will be compared. This data set determines the type of material at each voxel.

The data set that we have represents the "S-wave velocity magnitudes" that the scientists believe describe the geology in Southern California. This is a scalar field that is related to the stiffness of the material.

Basins are regions in the data with a VS value of <2500; when we do an isosurface at that level, it shows several sloped indentations in the data set that look like basins, as seen in the image to the right. These are the basis for the various domain-science questions that ask about the behavior of waves in and around basins. The particular basins of interest are labeled. Click on the image for a larger rendition. A map of the basin locations with respect to land features can be found on the Tasks and Judging page.

The format of this volume data set is the same as that of the Volumetric Wave_Propagation Data Sets listed above.

Deriving your own data sets

Further processing of data into displacement, or calculation of data types such as gradient, divergence, and curl are possibly of use in displaying answers to the scientists' questions. The frequency contents of the original and decimated data should be considered if such processing is considered.

Sample Readers and Converters

As part of the data validation process, Tcl scripts were created to load the data into the open-source Visualization ToolKit (VTK). The first script (available here) converts one of the X data sets into a VTK-format volumetric scalar-field data set. The second script (available here) will load one of the X data sets into VTK and display it using an image viewer. Very small thumbnail images of slices 0 and 99 can be seen to the right. Click on them to get a full-sized image. The displayer reads from the bin file rather than the VTK file so that you can see how to put together a full chain of VTK commands to read directly from the raw files.

These scripts are included here to provide concrete examples of how to load or convert this data. Note that the script needs to be run from the same directory as the data file named "TS21z_X_R2_008000.bin" or else the name of the file to be loaded must be changed in each script. Note that the data files from SDSC do not have the ".bin" extension on them. They are compressed with Gzip and must be uncompressed before being loaded by these scripts.

Note that the image data is inverted around the Y axis when it loaded into VTK to make it match the simulation data space (although the VTK viewer is right-handed like the data itself, the image reader assumes a left-handed image file format whose origin is at the upper-left corner of the image).

Another script has been created to load the Velocity Progagation Input Mesh data (available here). It brings the data into the same space as the time-dependendent data above.

What the data should look like when properly loaded

The best place to look to get an understanding of the structure of the data is TeraShake 2.1 Visualization web page. This page contains still shots and videos of the earthquake in action. These videos are for reference only: their authors are are not eligible to submit to the contest. Note that some videos on that page show derived data (summed displacement rather than instantaneous velocity). The visualization of this data set to date have been almost exclusively surface-based. They went to a lot of trouble to store all of the volumetric data, and they have questions about volumetric effects, so this is our chance to show them new things they have not yet ever seen!

The image below shows a view from VTK of two isosurfaces, one at -0.02 and one at 0.02 in the TS21z_X_R2_008000.bin file (click on the image for a larger view). This represents the x component of the instantaneous velocity for the TeraShake 2.1 sample dataset at time step 80. The viewpoint here looks up from below the data set, with X increasing to the right, Y towards the back, and Z up. The positive isosurface is colored red and the negative is blue. This is not necessarily a recommended technique, but is provided to let you see if the orientation of your conversion matches the actual data's orientation.

Sample images of a top down view of data as visualized by Marcus Thiebaux of the TeraShake 2.1 sample dataset for time step 80 are available. There is an X data set, a Y data set, and a Z data set view of the top layer (ground level) seen in the isosurface image. These two images also include an overlay of the California border and the California highways data sets described in the metadata section below. The X image is drawn below.

The input volumetric data set is displayed in a spinning volume rendering with different surfaces shown in a movie created by Amit Chourasia. If that link doesn't work, look for the image to the right and view the movie associated with it on the following web page: http://visservices.sdsc.edu/projects/scec/terashake/conceptual. Click on the image for a larger rendition.

Metadata

Two additional data sets are available in case they are useful. They are a map of the California border (shown in blue in the three sample images above) and a map of California highways (shown in red in the sample images above). These are ASCII-formatted files that contain one or more segments with X and Y coordinates in lattitude and longitude. Lattitude is listed first on each line in both files. The data from these files, when plotted, should look like the image to the right.

Registration: Remember that the data is rotated 40 degrees, as described under the Region section of this document; this must be taken into account when aligning the maps with the data sets. Also, the simulation grid is a rectangular approximation to the curvilinear grid in which lat/long are specified, so registration is slightly off. Below is the fit used by Marcus Thiebaux to align them. 40 degrees is the geographic angle, but 39.5 is part of a linear approximation to do a best-fit of the data..

rotate euler 0.0 0.0 39.5
scale 0.1539 0.183 1.0
translate 121.000000 -34.500000 0.133333