Yucca Mountain Project

Teaming up with scientists from LBL's Earth Sciences Division, we have embarked on a project to create a visualization of the Yucca Mountain facility. This page serves to document many of the along-the-way steps we have taken, both to document our progress, but also to give a glimpse into the type of close collaborations the Visualization group enjoys with researchers.

Too Much Data!

Mark Feighner from ESD provided approximately thirty different geologic horizons. Each horizon shows the location of some surface of geologic interest. For example, topography is one horizon, the water table is another horizon, different rock types form other horizons.

Each horizon is represented using a two-dimensional mesh grid that exists in three-space. The size of each horizon weighs in at 185 by 312. Each horizon requires 57,720 quadrilaterals, or 115,440 triangles for the purposes of display. This poses a problem, since on our largest graphics machine, and using the AVS software, the best performance rates we get are on the order of 100,000 triangles per second (we can do better using the Khoros software, and have achieved rates of up to 2 million triangles a second with that package, but the initial development work for this project was done using AVS).

One of the first steps taken in this project was to create a software module which performs what is called "multiresolution" represenation of a mesh. What this means is that we can trade off size against accuracy. So, if we are willing to accept a little bit of error, we can represent the original horizin by using an approximation. For example, let's look at the topography horizon.

The left column of images shows gouroud-shaded triangles while the right column shows a wireframe representation. The top row of the table is the full resolution horizon. The second row of the table was created using a 0.02 level of error, measured in units of standard deviation. The top left picture (full res, triangles) has 91896 triangles (some clipping of the horizon was performed), while the bottom-left picture has only 34,948 triangles. By looking at the wireframe images, you can get a good feel for how regions consisting of many triangles were combined into larger regions. Such recombination accounts for the "gaps" or "cracks" in the lower-left image.

This multiresolution technique is an adaption of the one described by Hanrahan and Laur in the 1991 Siggraph paper, adapted to two-dimensional meshes such as we see here. This technique is computationally very inexpensive, although a lot of memory is consumed by the quad-tree representation of the mesh. Also, this technique doesn't work as well as the "best" multires tools used by commercial terrain modeling applications. We will be exploring the use of alternate tools for multiresolution analysis in the near future, as applied to this project. However, this tool produces minimally acceptable results (they're not wrong) in the sense that we can do interactive rotations with workstation class graphics hardware while working on this project.

Faults were added to the model. A single fault horizon was brought into a simple model consisting of a few horizons. The fault is mostly a vertical structure, whereas the geologic horizons tend to be more like pancakes. Only a few of the values in the fault horizon are meaningful. The rest are either "positive infinity" or "negative infinity" (depth in meters), which means, basically, that the fault doesn't exist there. In order to accomdate this kind of data, we wrote a custom AVS module which will take one input horizon and clip it against two other input horizons. The two images below show the raw and clipped fault horizons.

After adding some site information, such as the tunnels for the facility (in red) and the outline of the repository (shown in purple), pale green cones showing water velocity, as well as adjusting viewpoints and other parameters for asthetics, we have a couple of images requested by ESD for use on the cover of a magazine (will insert the name as soon as I know what it is). The view on the left is looking at the facility from the Southeast, the view on the right is from the Southwest.

Streamlines tracking and particle advection are tools that are used to explore vector, or flow fields. This image shows where water travels if released from the boundaries of the storage facility.

Click here to check out an MPEG movie which shows the path of particles flowing through the water velocity field.

Perched water simulations produce velocity vectors on an unstructured grid. This image shows a comparison of these velocity vectors, represented in whiite, with the velocity vectors which have been kriged to a regular grid (red arrows). The kriged grid has cells which are fifty meters on a side, and is 50 cells in X, 1000 cells in Y and 20 cells in Z. This computation took approximately 130 CPU minutes on a Sun Ultra 1.