Categories: FeaturedLos Alamos

Enlightening predictions

In comparing the simulations and the real hurricanes, one thing already is clear: It’s going to take a lot more initial detail to accurately predict hurricane intensification.

“At the heart of the issue is that you need a lot of data,” Reisner says. “A hurricane is a big animal. It’s not like a super-cell thunderstorm that produces a tornado; it’s much bigger than that. It’s composed of many super cells or thunderstorms. We’re finding that you need approximately 200,000 data points to really constrain a hurricane model in terms of properly initializing it and constraining key parameters that maybe can be used to predict a hurricane.”

To run such an enormous simulation, Reisner’s team used Jaguar, a Cray XT at DOE’s Oak Ridge Leadership Computing Facility.

“Jaguar enabled us to do stuff that no other computer at the time could do,” Reisner says. In total, Reisner’s group ran an ensemble of 120 simulations for Hurricane Guillermo that used approximately 118,000 Jaguar processors.

The simulations used a “matrix-free Ensemble Kalman Filter,” a technique that enables the integration of large amounts of real-world data. And, Reisner says, it appears that not just more but also different types of initial real-world input, such as lightning data, will be required to forecast changes in hurricane intensity.

“The thing that worries me now, based on my research, is that we may have to characterize somewhat the ocean surface,” Reisner says. “This means the complex surface wave field, sea spray and water temperature that is responsible for producing the water vapor flux that gets fed up into the hurricane.”

Prediction frontier

Although Reisner’s results show a link between a hurricane’s building strength and increasing rates of intra-cloud lightning, it’s too early to use this as a predictor.

“There’s a lot of pressure on the people who run these predictive models (when there’s a real hurricane), and my research is setting the tone for what goes into them 10 years down the road,” he says.

Reisner points to two improvements in data and modeling that will facilitate this.

The growth of new lightning networks, such as the World Wide Lightning Location Network, will provide an unprecedented level of detail on tropical-storm lightning activity, including hints at the microphysical conditions driving the lightning.

More accurately integrating this microphysical data into models is the next step. 
The HIGRAD simulations involved a range of scales from millionth-of-a-meter-sized water droplets to the entire span of the Gulf of Mexico. Integration across these scales will be aided by improved high-performance computing software and hardware that will be able to handle increasingly data-intensive models.

The jury’s still out on whether changes in a hurricane’s lightning bursts will be the cue to evacuate a coastal city, but the storms Reisner has swirling in supercomputers today will definitely help hurricane forecasters see more clearly.

Page: 1 2

Share
Published by

Recent Posts

Planet-friendly plastics

A Los Alamos team applies machine learning to find environmentally benign plastics. Read More

June, 2023

A split nanosecond

Sandia supercomputer simulations of atomic behavior under extreme conditions advances materials modeling. Read More

May, 2023

A colorful career

Argonne’s Joe Insley combines art and computer science to build intricate images and animations from… Read More

April, 2023

Mapping the metastable

An Argonne National Laboratory group uses supercomputers to model known and mysterious atomic arrangements, revealing… Read More

March, 2023

The human factor

A PNNL team works to improve AI and machine learning tools so that grid operators… Read More

January, 2023

Simulation looped in

CogSim, a machine-learning approach modeled on the brain, coordinates simulations, experiments and data-analysis to yield… Read More

December, 2022