Andreas Plesch, John H. Shaw, Egill Hauksson, Toshiro Tanimoto, and members of the USR Working Group

Dept. of Earth & Planetary Sciences, Harvard University20 Oxford St., Cambridge, MA 02138andreas_plesch@harvard.edu, shaw@eps.harvard.edu

Abstract

We present a new version of the SCEC Community Velocity Model (CVM-H version 5.5) that incorporates several enhancements to better facilitate its use in strong ground motion prediction and seismic hazards assessment. These improvements include updates to the background models, basin structures, geotechnical layer, and the code that delivers the CVM-H. The extent of the model was expanded to an area between longitudes 120d52'20"W and 113d19'16"W and latitudes 30d56'49"N and 36d37'17"N to accommodate larger numerical experiments.

Improvements to the basin structures include new Vp, Vs, and density parameterizations within the Santa Maria and Ventura basins that are based on direct velocity measurements from petroleum well logs and seismic reflection data. These new basin structures were used as input for the development of new P and S wave tomographic velocity models, and of a new upper mantle teleseismic and surface wave model. The CVM-H thus consists of the revised basin representations embedded in self-consistent regional crust and upper mantle models. In addition, we also enhanced the geotechnical layer (GTL) representation. The GTL in the major sedimentary basins remains unchanged, following the approach of the SCEC CVM 4.0 (Magistrale et al., 2000). In bedrock sites, however, we implemented a new GTL based on the depth-velocity relations of Boore & Joyner (1997). In this implementation, we used the empirical velocity gradient from Boore & Joyner (1997) to scale upwards from the base of the GTL (top of basement). This bedrock GTL implementation results in gradual velocity gradients and variable velocities at the surface.

The CVM-H 5.5 is delivered as code with accompanying data files (voxets and tsurfs) available through the SCEC/Harvard web sites (http://structure.harvard.edu). We have made a number of improvements to the code that were requested by the USR working group to assist with its use in developing and parameterizing meshes and grids. The code now provides the location of the nearest neighbor cell used to specify the velocity values at user defined coordinates, and outputs the topographic, top basement, and Moho elevations for each user-defined x and y location. To aid in meshing models, additional code is made available which computes the closest vectorial distance to a velocity interface, such as the top basement surface, from user defined coordinates.

In addition, we provide a series of enhancements to the C-code that delivers the CVM-H. This code specifies Vp, Vs, and density values at arbitrary points (x,y,z) defined by the user by locating the nearest neighbor grid point in the appropriate CVM-H voxet. The new model version consists of high (250m) and medium (1000m) resolutions voxets, or regular grids, defining both Vp and Vs structure (density is derived using a scaling relationship from Vp). To support the needs of SCEC scientists who employ the code to help parameterize their computational grids, we

enhanced the code to deliver several additional functions. First, the code now provides the location, in addition to the value, of the nearest neighbor grid point. This allows users to identify the locations where the values were initially parameterized in the CVM-H, ensuring data integrity and supporting the use of a variety of interpolation schemes that can be tailored to the users application. A sample implementation of an interpolation routine which handles material boundaries correctly is provided. Second, the code now provides the depths (distances) from the arbitrary points to the surfaces used to construct the CVM-H, namely the surface topography/bathymetry, the top of crystalline basement, and the Moho. This information is of particular value when using the CVM-H to guide the construction of computational meshes.

SCEC Community Velocity Model (CVM-H 5.5)

Overview of CVM-H 5.5enlarged dimensions (fits Terashake box):• 30°56'N to 36°37'N latitudes• 113°19'W to 120°53'W longitudes• topographic elevation to 200 km depth

Components:• upper crustal low resolution voxet: 1,000 m horizontal resolution, 100 m vertical resolution• upper crustal high resolution voxet for LA basin: 250 m horizontal resolution, 100 m vertical resolution• lower crustal and mantle voxet: 10,000m horizontal resolution, 1,000m vertical resolution• elevation (topography, basement, Moho) voxet: 250m horizontal resolution• elevation (topography, basement, Moho) tsurfs• code to query data• code to calculate distances to tsurfs

Projection and datum:• model was contructed in UTM zone 11, datum NAD27 • code accepts geographic coordinates• code expects input elevations relative to sealevel, positive upwards

Expanded areaResponding to requests from the modeling community to have access to a unified velocity model which covers the area of Terashake or PetaSHA like efforts completely, we expand lower resolution crustal CVM-H grid and the lower-crustal/mantle CVM-H grid. The grid is populated with E. Hauksson tomographic model and T. Tanimotos mantle model based on teleseismic surface waves where those datasets are available. Outside the area of those datasets we use the Socal 1d velocity function.

In order to provide a smooth transition without discrete velocity boundaries, we define buffer zones between domains of velocity data origins. Then we employ discrete-smooth interpolation in these buffer zones to populate the CVM-H grid. E.g. to transition between the tomographic model and the 1d background model we define a 50km wide buffer surrounding the tomographic model and fill in this zone by interpolation. Similarly, the area of the previous CVM-H version is surrounded by a 20km wide buffer.

Figure 1:

200m depth slice showing various components of the expanded CVM-H: 1) 1d background; 2) buffer between background and tomography; 3) tomography; 4) buffer between enlarged area and core area; 5) core area; 6) Terashake box

The velocity model in the added area differs from the core area in that there is no attempt to incorporate basin sediments or near surface geo-technical layer. This is significant in the San Joaquin Valley which is part of the enlarged area.

Independant Vs model in basement and basin GTL in all voxetsWe update the model to provide shear wave velocities for queries in basement material which is based on the underlying mantle and tomographic models, and not on a Vp-Vs relation. Similarly for queries into the basin geotechnical layer we provide now Vs as it is defined in the CVM 4.0 (Magistrale, 2000). Vs for basin sediments is calculated from Brocher's (2005) regression fit to Vp, and the "mudline" for shallow depths.

This update requires additional Vs voxets for each of the CVM-H components: the high resolution model, the lower resolution model, and the lower crust/mantle model. These voxets are now provided with the query code. The code itself is simplified to process Vs in parallel to the processing of Vp.

Figure 2:

perspective view of CVM-H 5.5. Cross-sections show Vs which is derived from mantle and tomographic data in basement areas, and based on Brocher (2005) in the basins. Inset shows Vs vs. Vp plot for all data points in the CVM-H 5.5.

Updated GTL in basement areasTo examine the implementation, we first produced a map showing the velocity contrast between the base of the GTL and the top of the underlying rocks. This map shows that the contrasts are generally low in the sedimentary basins, and are stronger in areas underlain by basement rocks. Moreover, the gradients between the GTL and basement rocks in the CVM-H show undesirable artifacts that are perhaps a product of their different heritages.

We are considering a strategy for completely replacing the GTL in basement areas using the generic Boore & Joyner (1997) velocity profile, and providing a smooth transition to the underlying velocities.

The first and most straightforward approach is to implement the surface velocity specified in the generic curve at the top of the GTL (generally around 50m below ground) and then use the remainder of the thickness of the GTL (generally about 200m - 300m) to linearly interpolate up to the velocity at the top of the basement rocks. This approach results in rather steep velocity gradients with depth and an almost constant velocity close to the surface.

The second approach accounts for the differences in basement velocities as sampled in the tomographic model. We start with the velocity at the base of the GTL (or top of basement) and use the shape (gradient) of the generic curve to specify velocities within the GTL. Note that this approach does not use the absolute values of the generic curve. Instead, in this implementation we scale the generic velocity profile in the GTL according to the ratio of the velocity of the profile and the velocity of the basement, at the depth at the top of the basement. This approach results in more gradual velocity gradients and variable velocities at the surface.

Either of these basic approaches could be calibrated based on simulation studies or revamped with new GTL velocity profiles. Moreover, since the Boore & Joyner profile is for Vs, we use Brocher's (2005) regression fit to calculate Vp. Vs is then calculated from Vp using the same relation.

Figure 3:

Map of Vp contrast below GTL in basins showing relatively small contrasts. In the basement areas outside the basins this contrast was high and showed interpolation artefacts. This is the area which is revised.

Figure 4:

Maps and cross-sections of revised GTL in basement area. Maps shows the top of the GTL, cross-sections are 10x vertically exaggerated. Left panel has a direct implementation of the Boore & Joyner (1997) velocity profile at the highest levels combined with an interpolation down to the top of basement velocities at about 300m depth below the surface. Right panel has a scaled implementation of the Boore & Joyner (1997) profile. The profile is scaled according to the ratio of the profile velocities and basement velocities at the top of the basement level.

Sample velocity interpolation routineThe query code returns nearest neighbour velocities from the CVM-H voxets, e.g. is does not interpolate between grid points. The query code package is updated with a sample interpolation routine which takes advantage of the fact that the query code also returns the location of the grid cells from which the velocity is provided. The sample routine uses inverse distance weighting of velocities sampled from neighbouring cells and is implemented in awk. The routine correctly deals with material interfaces by only taking into account velocities of the same material for the interpolation.

Here is the algorithm:1) run vx with the input points to be interpolated2) find the eight cells which are closest to the query point: The default cell and based on the offset from the default cell seven more. Run vx on the neighbour cells.3) Filter out neighbour cells of undesired rock types and calculates distance from query point to each returned cell center.4) Use inverse distance weighting with a power of two to calculate average velocity at query point

Associating the M5.4 Chino Hill sequence with CFM faults

By distanceAs a first means of association of an earthquake with fault representions, it is possible to measure the smallest distances from the hypocenter to the faults. After using a cut-off distance of perhaps 1km this step results in a small list of candidate faults ranked by distance.

By orientationGiven the nodal planes of a focal mechanism, one can then compare the orientations of the nodal planes with orientations of nearby fault patches. In the case of the Chino Hill sequence, the Chino fault has the orientation fitting the least with nodal planes among close faults. It would be given a lower probability. The Whittier fault and the Peralta Hills faults have orientation which match more closely.

By size and aftershocksUsing an accepted magnitude-fault-area scaling relationship it may be possible to further constrain an earthquake-fault association. This is especially true for large events. Additionally, the distribution of aftershocks may be aligned along a fault. It would be possible to calculate distances of all aftershocks to all close faults, and sum those per fault as an indication how well aftershocks fit a fault shape. In the Chino Hill sequence the aftershocks align in a plane almost perpendicular to Yuerba Linda zone of seismicity which would impact an assigned probability.

12

34

5 6