1. Purpose

The goal of this preprint is to provide researchers and research programmers with examples of the use of advanced technology in visualizing atmospheric data sets. The first involves high resolution animation for use in the OMNIMAX theater environment and for subsequent display on the high resolution NII Wall. The second utilizes the virtual reality room known as the CAVE where the viewer can interact with the display around them.

2. Storm and Tornado Visualization for Stormchasers

2.1 Overview

In the past decade, the OMNIMAX film and theater industry reached the point where the number of space theaters was growing while the number of newly produced, high-quality films was decreasing. To guarantee first-rate films for exhibitors and museum-goers alike, ten distinguished science museums came together to form the Museum Film Network. This new organization invited NOVA, the Public Broadcasting Service's acclaimed science series, to act as executive producer. MacGillivray Freeman Films, a company with 20 years' experience creating IMAX films, was asked to serve as the producer. The partnership dedicated itself to creating entertaining and effective science films.

"To The Limit" was the Network's first film and celebrates the extraordinary workings of the human body. A second film entitled "Stormchasers" debuted in September of 1995. It is expected that some 15 million people will see this film, at least 40 percent of them are children who will leave the OMNIMAX theater with a new and exciting impression of science.

Four weather events are highlighted in the film "Stormchasers", one for each season of the year: monsoons, hurricanes, winter blizzards, and tornadoes. Current and affiliated members of the Storm Group at the University of Illinois along with several professional animators (Arrott, Bajuk, and McNeill) contributed a 90 s segment in which a numerically modeled storm is followed from its inception through the generation of a tornado. Some of the techniques used in and software developed for the award winning visualization entitled "Study of a Numerically Modeled Severe Storm" by Wilhelmson et al. (1990) were used in this new animation. In addition, a new particle trajectory code was written for advecting thousands of particles through time in order to illuminate the air flow within one of the tornadoes that developed after the storm entered its mature stage.

The storm simulation was carried out using the COMMAS nested grid nonhydrostatic numerical model by Wicker and Wilhelmson (1995). Horizontal resolution in the finest grid encompassing the tornado was 200 m while the coarsest resolution in the outermost grid was 1,800 meters. The storm evolves into a supercell and shortly thereafter several tornadic events occurred. The animation focuses on the most spectacular tornado that lasted approximately 10 minutes, had ground relative wind speeds of over 60 m/s (134 mph) near the ground, and had a 40 millibar pressure drop.

Isosurfaces of the growing thunderstorm reveal common shapes associated with supercell storms (dome, vault, and anvil). The viewer sees the early development of the precipitation region of the storm (the region typically seen by radar which is somewhat different than seen by an observer watching a real storm develop) from above the ground and to the east of the storm. An abstract prairie landscape with a blue horizon is used to anchor the audience with familiar real world cues. The main source of lighting came from the west in tone evocative of afternoon sun light. This environment was designed to create the time and place in which a tornado can occur.

As the storm develops, the viewer begins to descend and move around the southern part of the storm in order to gain a different perspective and prepare for moving closer to the tornado as it develops. The tornado is revealed through the use of thousands of particles after the lower part of the precipitation isosurface is removed. This is necessary in order to "see" the tornado which is embedded in the precipitation region but near its edge. Examples of images from the complete animation can be found on the World Wide Web at http://redrock.ncsa.uiuc.edu/AOS/imax.html / .

2.2 Image Technology

The images in the animation were computed from the approximately 40 Gbytes of simulation data saved from the numerical simulation. This is about four times more than the amount saved for the 1989-1990 storm video mentioned above.

The storm isosurfaces were rendered using the Wavefront Technology's Advanced Visualizer. There are actually two isosurfaces used in every image defining surfaces very close to one another. This was done to soften the surface and give a perception of depth. Most of the data representation software was developed by the Visualization Group at NCSA. The particle representation and shading techniques were developed especially for this project.

A weightless particle moving with the flow field was represented as a view aligned, texture mapped polygon. The polygon started at the location of the particle for a given time and then stretched back for some specified duration to previous locations of the particle. The shading of a particle was based on a simple reflection / transmission model for a sphere. Each particle was shadowed by all the other particles between it and the light source. The goal of the shading technique was to make the volume of the tornado more prevalent than the individual particles.

A flow of forty thousand particles was built up and maintained to represent the evolution of the tornado. These particles were progressively revealed in the animation of the tornado as a horizontal plane descending from cloud base in time.

The final frames were composed of up to four separate images: the prairie landscape, the shadow, the cloud, and the particles. Each image was produced at 2048x1536 pixel resolution. These final frames were produced after substantial image experimentation in the presentation of the model data as well as after many massive particle release strategies were investigated.

2.3 The Particle Advection Code

Special code was developed for creating the particles from the velocity data stored every 5 seconds of model time. In this code the release of particles was based on location, vertical vorticity magnitude and other factors. The removal of a particle from the advection was based on age of the particle and its location. This technique for releasing and removing particles focused the visible flow in the main tornado region.

The massive non-interactive semi-time-dependent particle advection package has been called Partadv. The input data comes from HDF or netCDF files of time dependent velocity data. Sequences of files representing multiple times and multiple nested grids can be handled automatically. A sophisticated parameter file interpreter provides 105 different parameters controlling advection, particle release and other activities. It allows the adjustment of parameters as a function of time based on the use of conditional statements, equations and control points (key frames).

Adjustable selection criteria determines where particles are released and when they should be removed from the system. The code is designed to allow a continuous flow of particles into the system while maintaining an adjustable cap (limited only by memory and time constraint) on the total number of particles allowed in the system at a given time.

Each time step may be checkpointed to allow restart from any point. Multiple output formats include lists of particle positions at any given time, or lists of adjustable length trajectories (a sequence of particle positions over time). Optional SGI Inventor file output allows quick visualization of trajectories and particles on SGI workstations.

Advection in the code is accomplished using a 4th order Runge-Kutta method and full linear spatial interpolation over nested grids. For the OMNIMAX sequence, 100,000 particles were advected with a 0.5 s time step. About 24 hours are needed for a 3600 time step calculation on a 150Mhz R4400 SGI Indigo^2. Less than half of this time goes to advection, the rest is spent processing the release and selection criteria as well as reading the velocity data. Release criteria were varied during the tornado's life. During the mature stage particles were released into the surface region having a vorticity value greater than 0.1. Particles were purged if they went out of the domain of interest or if they fell below certain speed or vorticity thresholds.

2.4 Use of the NII Wall at Supercomputing '95 and at the IIPS SGI Exhibit

The NII/Wall is a large-screen, high-resolution, stereo or mono (2560 x 2048 in mono) display meant for fairly large audiences. It will be demonstrated at the Supercomputing '95 conference and a version will be available at NCSA early next year. It uses four Reality Engines spread across two Power Onyxes to achieve a high-resolution, high-intensity display. Ultimately, the NII/Wall will be driven by POWER CHALLENGE Arrays from dozens of shared-memory processors over ATM networks. The OMNIMAX animation will be reformatted for display using a version of the Wall in the SGI booth at the '96 IIPS meeting.

3. Interactive CAVE Visualization

3.1 Storm Applications in the CAVE

The CAVE (Cave Automatic Virtual Environment) is a multi-person, room-sized (10 feet cubed), high-resolution, 3D video and audio environment, in which graphics are rear-projected in stereo onto two or three walls and front-projected onto the floor. As a viewer wearing a location sensor moves within its display boundaries, the correct perspective and stereo projections of the environment are updated, and the image moves with and surrounds the viewer. The other viewers in the CAVE, wearing only stereo glasses, are like passengers in a bus, along for the ride. Also within the CAVE is a computer-controlled audio providing sonification capability to multiple speakers.

The CAVE was introduced by Tom Defanti and others in the Electronic Visualization Laboratory at the University of Illinois at Chicago. More information can be obtained on the WWW at http://www.ncsa.uiuc.edu/EVL/docs/html/CAVE.references.html . The CAVE at NCSA is powered by a 12 processor Onyx with three Reality Engines. Control can be achieved through a three-button wand or by voice command.

Isosurface and particle trajectory data can be interactively viewed in the CAVE under the control of the viewer with the headset. The data in our project will come from the OMNIMAX simulation and from a non-supercell tornado simulation made with a high resolution, constant grid, convective cloud model run on NCSA's CM-5 (Lee and Wilhelmson, 1996). Instabilities that form near the ground along thunderstorm outflow boundaries have been shown to lead to tornado development beneath growing cumulonimbus which do not possess the strong mid-level rotation of supercell thunderstorms. The project involves current and affiliated members of the Storm Group at the University of Illinois along with NCSA staff members Terstriep and Jaswal.

The viewer will be immersed within the storm environment, first by letting them get an overall view of the storm/thunderstorm boundary and its different features and then by zooming in on specific features, particularly the tornado. The viewer will be able to move around the storm, either from a ground or airplane position and release large numbers of tracer particles to look at how air moves in different parts of the storm. When tracers move inside the storm, the surface will become transparent so the tracers can still be seen.

The particle motion will be computed using SGI Challenge and Power Challenge machines. These machines will communicate with the CAVE software, supplying new coordinates as fast as they can be computed. In San Diego at Supercomputing '95, these computers will be located halfway across the country in Champaign, Illinois. In addition, software will allow multiple CAVEs across the country to be connected creating a virtual collaboratory in which researchers at different CAVE sites may interact both with the data and also with each other.

3.2 CAVE Application Components

Our application consists of four modules, CAVEvis (the CAVE display program), Iso (an isosurface generator), Tracer (a particle convector) and the Accumulation Server. Each of the modules runs independently and can be run on the CAVE machine or on separate machines. In addition to these modules, it is possible to add an active storm simulation generating data in real-time.

The Accumulation Server is the data repository. It controls the flow of data to other modules and helps keep the other modules synchronized. The Accumulation Server should be thought of as a large data source to which the user is able to send 'queries.' In addition to identifying which data is of interest, the queries allow the user to specify transformation to be performed on the data before it is sent to the visualization modules. The Accumulation Server currently stores data on a HIPPI attached RAID disk. The data can be built from precomputed data sets or collected in real-time from a simulation.

The visualization modules perform the mapping between the scientific data being sent from the Accumulation Server and polygonal geometries needed by the CAVEvis program. The visualization modules are free running and generate new geometries as necessary. This is generally required when new data arrives from the Accumulation Server or when user requested visualization parameters change (e.g., an isosurface threshold level).

The CAVEvis program accepts incoming geometries and displays them within the CAVE. Through the use of buttons on the wand and pop-up menus, the CAVE program also acts as the user interface to all the other modules in the system. This allows users to change the isosurface threshold, release new tracer particles, or stop at a particular time step.

The CAVEvis program has been enhanced to allow collaboration between caves. Geometries being sent from the visualization modules can easily be replicated for any number of CAVEvis programs. This allows researchers to share a common display within their respective caves. In addition, head and wand information is communicated between the caves. Finally, a CAVE user will be able to see a representation of other users in the remote CAVE, placed appropriately within the scientific domain.

3.3 ImmersaDesk

A smaller, software compatible, drafting-table form of the CAVE, called the ImmersaDesk, has been designed to enable head-tracked users to see stereo imagery both forward and down, a technique that greatly enhances the feeling of immersion. The ImmersaDesk is powered by a deskside Power Onyx and rear-projected onto a large (approximately 4 x 5 feet) diagonal screen in stereo. It will fit through a door, and deploy in an office. With appropriate video/ATM gear, it can be used for remote ImmersaDesk-to-ImmersaDesk collaboration.

4.0 Acknowledgments

This work was supported in part by NSF through Grant ATM 92-14098, through NCSA, and through support of the OMNIMAX film. The computer simulations were carried out on NCSA's CM-5 and Pittsburgh Supercomputing Center's C90. Visualization for the OMNIMAX film was accomplished using up to 12 SGI processors at NCSA.

References

Lee, B. D., and R. B. Wilhelmson, 1996: The numerical simulation of non-supercell tornadogenesis. Preprints, 18th Conf. on Severe Local Storms, San Francisco, California, AMS .

Wilhelmson, R.B., B. Jewett, C. Shaw, L. Wicker, M. Arrott, C. Bushell, M. Bajuk, J. Thingvold, J. Yost, 1990: A Study of the Evolution of a Numerically Modeled Severe Storm. International Journal of Supercomputing Applications. Summer Issue.

Wicker, L. J., and R. B. Wilhelmson, 1995: Simulation and analysis of tornado development and decay within a three-dimensional supercell thunderstorm. J. Atmos. Sci., 52, 2675-2703.