Scientist sharpens weather forecasts with AI | Tech Rasta

[ad_1]

Editor’s Note: This is the first in a series of blogs from researchers who are advancing the science in the expanding universe of high performance computing.

Random raindrops fall inside Dale Duran’s three-foot metal ring outside his front door (shown above). It is a symbol of his passion for finding order in the chaos of the planet’s atmosphere.

A part-time sculptor and full-time professor of atmospheric science at the University of Washington, Duran has co-authored dozens of papers describing patterns in Earth’s ever-changing sky. It is a field for those who seek the chaotic challenge of trying to express mathematically the endless dance of wind and water.

Meteorologist Dale Duran
Dale Duran

In 2019, Duran acquired a new tool called AI. He teamed up with a grad student and a Microsoft researcher to create the first model to demonstrate deep learning’s ability to predict the weather.

Although crude, the model overcomes the complex equations used for the first computer-based instruction. Descendants of those equations now run on the world’s largest supercomputers. In contrast, AI reduces the traditional load of required calculations and can perform faster on very small systems.

“It was a dramatic revelation, and we better jump in with both feet,” Duran recalls.

A sunny outlook for AI

Last year, the team took their work to the next level. Their latest neural network can process 320 six-week samples in less than a minute on four NVIDIA A100 Tensor Core GPUs on an NVIDIA DGX station. That’s 6 times more than the 51 forecasts today’s supercomputers are synthesizing to make climate forecasts.

In a demonstration of how rapidly the technology is advancing, the model was able to predict the path of Hurricane Irma in the Caribbean in 2017, almost as well as conventional methods. A tenth of a second on a single NVIDIA V100 Tensor Core GPU.

AI predicts Hurricane Irma's path
Duran’s latest work uses AI to predict Hurricane Irma’s path in Florida more efficiently and almost as accurately as traditional methods.

Duran envisions AI crunching thousands of predictions simultaneously to provide a clearer statistical picture with radically fewer resources than traditional equations. Some suggest that performance advances should be measured by five orders of magnitude and use a fraction of the energy.

AI acquires satellite data

The next big step could radically expand the lens for weather observers.

The complex equations used by today’s forecasters cannot easily handle the growing wealth of satellite data on details such as cloud patterns, soil moisture and drought stress in plants. Duran believes AI models can.

One of his graduate students hopes to present an AI model this winter that incorporates direct satellite data on global cloud cover. If successful, it could pave the way for AI to improve predictions using the flood of data types now being collected from space.

In a separate effort, researchers at the University of Washington are using deep learning to apply the grid used by astronomers to track stars to understand the atmosphere. The novel mesh will help map out a new style of weather forecasting, Duran said.

Good season’s harvest

In nearly 40 years as an educator, Duran mentored dozens of students and wrote two highly rated textbooks on fluid dynamics, the mathematics used to understand weather and climate.

One of his students, Gretchen Mullendore, now leads a lab at the US National Center for Atmospheric Research, working with top researchers to improve climate prediction models.

“I was fortunate to work with Dale in the late 1990s and early 2000s on adapting numerical climate predictions to the latest hardware at the time,” says Mullendore. “I’m so grateful to have a mentor who showed me that it’s cool to be excited by science and computers.”

Continuing the legacy

Duran is scheduled to receive the American Meteorological Society’s most prestigious honor, the Jules G. Charney Medal, in January. Named after the scientist who worked with John von Neumann in the 1950s, these algorithms are still used by climate forecasters today.

Charney authored one of the first scientific papers on global warming in 1979. Following in his footsteps, Duran wrote two editorials last year The Washington Post To help a wider audience understand the effects of climate change and rising CO2 emissions.

The editorials reflect the passion he discovered in his first job in 1976, creating computer models of air pollution trends. “I decided to work on the front end of that problem,” he says of his career change to meteorology.

It’s a field notorious for effects as subtle as a butterfly’s wings, which inspires his passion for advancing science.

[ad_2]

Source link