In an age where computers are commonplace, two University of Wisconsin professors are harnessing their power to run simulations of real world scenarios.
Professor Leigh Orf has been using computer models to better understand thunderstorms for over 30 years. He works to understand how thunderstorms behave, how they work and how to better predict them. Orf said there are many things scientists still do not understand about what goes on within a thunderstorm, but using supercomputers to model them can add new clarity to their inner workings — even at very small scales.
“I use an atmospheric model that simulates a cloud at really really high fidelity, high quality, high resolution. So all the small scale complex things that go on inside the cloud are revealed in the simulation,” Orf said.
Orf said prior events are often inspirations for simulations in his work. Orf also emphasized field meteorology is very important, since atmospheric and physical information from a storm cannot be collected any other way. From one particular storm, he obtained the atmospheric data collected right before the storm formed and ran it as a simulation, which then successfully produced the thunderstorm and subsequent tornado that occurred in the actual event.
Visualization of the 2011 El Reno tornado simulated on a supercomputer
Courtesy of Leigh Orf
“I was able to study the simulation independent of the real event. But I can also compare it to the real event because its based on a real event,” Orf said.
Orf’s work was recently published on the cover of Science magazine. He, along with colleagues from Stanford University, found answers to an unknown atmospheric phenomena by running a simulation of the event. Orf said these computer models can sometimes provide information about real-world scenarios and allow scientists to discover things on the computer before they are discovered in the atmosphere.
He said one growing field that will likely become more important in the future of computer modeling is machine learning. As more data is collected from simulations, processing that data will be a vital task for artificial intelligence to undertake.
“We’re going to have robots at some point, and self-driving cars, and that stuff is going to be AI,” Orf said. “So that’s one area I’d say is really interesting and important.”
Self-driving cars are precisely what professor Dan Negrut uses computer models to study. As the technical lead of the Simulation Based Engineering Lab, Negrut is currently working on two projects which involve running computer simulations of a self-driving car and a moon rover to determine how they would function in a unique terrain situation.
Negrut said computer models offer a safer and more cost-effective way to analyze the reactions of autonomous vehicles. The simulations allow researchers to create a virtual environment in which they can test the vehicle millions of times with no real-world consequences of running the tests.
“It’s really hard to generate these sort of doctored scenarios in the real world. How would you take a Tesla and make it really hit a patch of black ice? What if you want to do it ten times? What if you want to do it ten times? What if you want to change the ‘brain’ of the vehicle and hit it again?” Negrut said.
For the moon rover experiment, this advantage is even more relevant. Negrut said it is key to test the rover’s capabilities before the very expensive process of sending it to space. In this case, running computer simulations means scientists can test the rover in different terrains and obstacles in an environment — including a difference in gravity — which accurately resembles that of the moon without having to send it there and back multiple times.
As part of a larger project aimed at getting humans on the moon, the VIPER rover will be tasked with determining where water is on the moon and how difficult it will be to extract, Negrut said.
Negrut said running a computer simulation is, in essence, solving a large set of equations representing the physical system you are working with. For something like the rover project with so many components, the number of equations needed to represent the environment can be in the billions, he added.
Orf said a key aspect to computer modeling is being able to write code exploiting the full power of the computer. Programming the computer to do what you want is not an easy process, especially when supercomputers are involved, Orf said.
“Supercomputers are really just a ton of computers all networked together. Imagine a warehouse full of really powerful desktop computers,” Orf said. “How do I write a program that runs on all of them and runs all together in lock step? It’s not easy. And to even have access to these computers you have to prove that you have code that will run efficiently on them.”
Orf said using computer modeling for meteorology research requires a very specific skill set. Somebody with just computer knowledge may not know the physics enough to determine whether the model is reasonable. But an observational meteorologist may not have the knowledge to run the necessary computer programs to even create a model.
Both Orf and Negrut are using computer modeling to help improve people’s safety and lives. Whether it’s ensuring autonomous cars behave in predictable ways or helping people be better prepared for dangerous storms, computers are contributing to our society on levels much deeper than we know.