 |
Severe storm simulations have changed a lot in 11 years, in large part due to the work of Robert Wilhelmson's storm research groupInterview with Robert Wilhelmson |
Severe storm simulations have changed a lot in 11
years, in large part due to the work of Robert
Wilhelmson's storm research group.
On a chill February day in San Francisco, Robert Wilhelmson was feeling the warm glow of success. A postdoctoral research associate from his group, Bruce Lee, was showing the group's latest storm animation to a crowd of more than 200 experts on severe storms at the 1996 biennial severe storms conference.
Usually that crowd is difficult to impress, but Lee had captured its attention with the first high- resolution rendering of landspout tornadoes. These lesser-known cousins of supercell tornadoes are of interest to meteorologists now that housing subdivisions are dotting the landspout-prone Florida peninsula and the plains of northeastern Colorado. Landspout tornadoes are not as violent as supercell tornadoes, but their cumulative toll can be as great. People have witnessed as many as six funnels emerging simultaneously from a rapidly growing line of thunderstorms before weaving destructive paths several hundred meters wide and over 10 kilometers in length.
What impressed the experts during the five-minute animation wasn't just the striking imagery but also the resolution. In capturing the dynamics of these storms, Lee and Wilhelmson had developed a computer code for NCSA's CM-5 system that resolved the entire storm dynamics on a 60-meter horizontal grid (one data point every 60 meters over 10 kilometers). That resolution is almost twice as fine as the grids most often used for modeling tornadic circulations in severe storms. Severe storms --- thunderstorms, tornadoes, squalls -- are usually modeled on 3D grids on which partial differential equations for winds, temperature, pressure, moisture, and water are solved. In the landspout simulation, equations are updated every 0.4 seconds for just under an hour, resulting in billions of numbers.
Equally impressive were the massive numbers of particles modeled. Wilhelmson and Lee identified the air flow patterns within the tornadoes and storms by tracing the trajectories of some 6,500 particles. Other animations had used 50 particles, sometimes 100. Here were 65 times as many.
When Lee's talk ended, hands shot up around the room. Wilhelmson and his research group had another hit on their hands.
Saving lives and property is one of the reasons why Wilhelmson models storms, but his innovations have led to a string of firsts in storm visualization. In 1989 he and his NCSA colleagues produced a severe storm animation that won 14 awards and is regarded as a classic in scientific visualization. In 1995 his research team's visualization of a tornado evolving from a thunderstorm appeared in the OMNIMAX film Stormchasers, seen by an estimated 15 million people. In 1996 they simulated landspouts. Wilhelmson was already well-known in scientific circles for his computational achievements -- such as the landmark 1978 storm convection model he wrote with Joseph Klemp at the National Center for Atmospheric Research -- but he has won wider acclaim with his visualizations.
When Wilhelmson helped found NCSA in 1985, his goal was to model the evolutionary nature of storm. Animation and real-time interactions with storm data -- none of which was then possibleÑfollowed. In the last 11 years, he and his storm research group have pushed storm visualizations from static 2D and 3D images, to 2D animations, to 3D renderings of simulations in real time at greater resolution and longer time spans. Today they are exploring immersive simulations in virtual reality. They have pushed the limits of visualization, both to improve its graphical and communicative qualities and to build new tools for scientific explorations.
Here are highlights from those eventful 11 years.
This seminal animation traced a small cloud's growth into a thunderhead. The evolution was computed using time-dependent equations solved over the storm region at grid points spaced 500 to 1,000 meters apart. By today's standards that resolution is coarse -- certainly too large to capture the elusive changes in air flow that transform a storm into a tornado. In 1987 that was not an issue. It would be several years before computing speed and memory increased sufficiently to contemplate modeling a tornado together with its parent storm.
The RIVERS group produced a system for rendering 3D images at rates of up to 30 frames per second at low resolution. The equivalent of videotape, this rate is considered real time. Group members also wrote prototype software for interactive visualization, such as for calculating a storm flow trajectory in real time. Their most significant contribution, however, was a schema for computing across platforms and on architectures ranging from workstations to supercomputers. This precursor to distributed computing was demonstrated by Haber and Wilhelmson at SIGGRAPH '89 when the team hooked up a satellite dish behind NCSA to broadcast a live interview, data, and images to the convention in Boston. Data and images were displayed on a graphics workstation and projected onto a large screen.
The video, which later won 14 awards and was nominated for an Academy Award for animation, was of a storm that pummeled Oklahoma and Texas on April 3, 1964. The storm split near Wichita Falls, TX, later forming a tornado that injured 111 people and caused $15 million in damage. The animation took a year to produce and employed many sophisticated visualization techniques such as 3D surface rendering combined with 2D slices of the evolving storm. Red and blue ribbons and white spheres wound through the storm, tracing the movement of air currents. Twisting ribbons represented storm rotation; growing ribbons represented updrafts and downdrafts; the spheres showed the movement of weightless tracer particles.
The video's breakthrough was graphic, not scientific. Most of the scientific relationships shown in the animation were well known, though it did confirm several characteristics of air flow. Wilhelmson credits it with helping awaken the scientific world to the potential of computer animation. "It demonstrated a marriage between science and visualization capabilities and techniques that had been developed for other purposes." Or, says Larry Smarr, director of NCSA, "It coupled the scientific community with Hollywood."
PATHFINDER was a prototype of today's visualization system for scientists. It placed tools on their workstations for quickly producing 3D images and animations like those in the numerical thunderstorm video. "It took scientists from the place where 3D animation was something so complicated they could only do it once a year to something they could do in a matter of days," says David Wojtowicz Jr., a systems manager in UIUC's Department of Atmospheric Sciences who then was a research programmer for NCSA.
At the core of PATHFINDER was SGI's IRIS Explorer, a distributed software system for viewing 3D data. It consists of a string of modules (software building blocks) for data reading, filtering, geometry, rendering, and display. The PATHFINDER team worked closely with SGI to augment this system for Earth scientists by adding modules for contouring and more sophisticated capabilities for reading NCSA-developed Hierarchical Data Format files. A feature of PATHFINDER -- with Wilhelmson's fingerprints on it -- was a separate package for particle advection that was coupled with the rendering of an evolving storm. PATHFINDER software was demonstrated at SIGGRAPH '92, where the output from a storm was displayed on a workstation screen in the Showcase event in Chicago as quickly as the CRAY-2 supercomputer at NCSA solved the equations.
The storm simulation used COMMAS -- a code for nonhydrostatic nested grids -- written by one of Wilhelmson's former postdoctoral research assistants, Lou Wicker, who now is an assistant professor of meteorology at Texas A&M University. Both the storm and the tornado were resolved on grids with 1,800-, 600-, and 200-meter horizontal resolution to reduce computer time and data storage. As it was, the animation required 40 gigabytes of data -- about four times more than in the 1989 project. "This couldn't have been done 10 years ago," says Wilhelmson. "It has only been in the last five or six years that the computing power and modeling technology have been good enough for us to simulate both the storm and the tornado it produces."
"I think this is what the future holds," says Wilhelmson. "Within 20 to 25 years, the National Weather Service will cover the entire globe with a fixed high-resolution grid."
Last fall, immersed in a tropical squall simulation while experimenting with a new particle trajectory tool in NCSA's CAVE, Wilhelmson discovered a structure he had never before seen in the animation. As he brought the squall line away from the wall and in front of his face, he noticed a sheet of particles behaving like a rear inflow jet. Other phenomena appeared more chaotic than expected. Why the differences? One reason was the immersive nature of VR -- data surrounds you. Another, says Wilhelmson, was real-time interactivity. "The tools developed for animation let you move forward and backward. As we use this interactive, 3D software environment, we can change the parameters and produce a different output."
Real-time storm VR made its public debut in December at Supercomputing '95. WilhelmsonÕs team simplified the OMNIMAX trajectory calculations so that they could run in real time by reducing the number of particle trajectories to between 1,000 and 2,000 and by assuming that the velocity data did not vary in time. Despite the compromises, it was another first for severe storm simulations.
What Wilhelmson finds most exciting about the Web is that many of the tools being developed for this medium have dual purposes -- they inform the public and enrich science. Eleven years ago these two goals of visualization were pursued separately; now they are melding. One example is the new Java-aware Weather Visualizer recently developed at UIUC's Department of Atmospheric Sciences and NCSA. Says Wilhelmson, "It shows how important visualization technology is becoming to both the scientific world and the public."
Return to the Table of Contents.
NCSA: The National Center for Supercomputing
Applications
access / Summer 1996 issue
Email comments to NCSA Publications Group:
pubs@ncsa.uiuc.edu