Cloud

Federal Agencies Model Storms with Cloud Computing Power

NOAA and the Army Corps of Engineers increase forecast accuracy with public cloud resources.

Twitter

Wylie Wong is a freelance journalist who specializes in business, technology and sports. He is a regular contributor to the CDW family of technology magazines.

Two days before Hurricane Helene slammed into Florida in late September, the National Hurricane Center began issuing dire warnings that torrential rainfall and gusty winds from the Category 4 storm could cause catastrophic, life-threatening flooding and widespread power outages in Florida, Georgia and the Carolinas.

Forecasters used hurricane models, primarily relying on the next-generation Hurricane Analysis and Forecast System (HAFS), to accurately predict the hurricane’s path, intensity and storm surge. After landfall, Helene, with its 140 mph winds, caused a trail of destruction, leaving millions in the Southeast without power, more than 250 dead, hundreds missing and more than $47 billion in damage.

To forecast tropical cyclones, the National Oceanic and Atmospheric Administration runs two versions of HAFS in an on-premises supercomputer system. But for major hurricanes like Helene, NOAA runs 31 additional HAFS variants in the Microsoft Azure cloud, providing forecasters with supplementary information.

“Our forecasters love it,” says Vijay Tallapragada, senior scientist at NOAA’s Environmental Modeling Center (EMC). “They are getting much better information on the possibilities, so they don’t leave anybody out from their warnings but also they don’t warn people where it’s not needed.”

Click the banner below to learn about advancing cloud deployments.

Storm modeling not only helps the federal government predict the route, timing and strength of major storms but also allows it to save lives and reduce property damage. NOAA and others rely on on-premises high-performance computing systems, augmented by public cloud resources, to scale and meet increased workload demands.

Climate modeling — and running models and simulations in general — is a common HPC use case. Using HPC in the cloud is becoming more popular, particularly for the ability to burst in the cloud, spin up instances and take them down as needed, says Ashish Nadkarni, group vice president and general manager of IDC’s worldwide infrastructure research.

“HPC in the cloud is a very good use case because the cloud providers have a lot of compute power that can be utilized on demand,” he says. “A lot of them have gone to great lengths to have these HPC-ready instances.”

Enabling More Accurate Forecasts for Multiple Hurricanes

When NOAA’s National Hurricane Center, a division of the National Weather Service, identifies an area with a potential tropical cyclone developing, it alerts the EMC, whose scientists run two versions of the HAFS model on twin NOAA supercomputers, powered by Hewlett Packard Enterprise Cray systems.

Each version of HAFS uses slightly different physics and different ocean models, one created by NOAA and the other by the U.S. Navy, Tallapragada says.

“You will get slightly different answers because of the small differences, but they can be significant,” he says. “We try to capture the possibilities with these two versions.”

During the 2024 hurricane season, NOAA still used two older hurricane models for guidance, but HAFS enables more accurate forecasts because of its advanced technology, such as a high-resolution moving nest, which allows forecasters to track multiple hurricanes simultaneously across the Atlantic and Pacific basins.

HAFS, launched in 2023, also allows forecasters to zoom in on specific areas of the hurricane. With a resolution of 1.8 kilometers, this allows a better prediction of the hurricane’s track, structure, precipitation amounts and wind speeds, Tallapragada says.

10%-20%

The percentage improvement attributed to NOAA’s new Hurricane Analysis and Forecast System in predicting the track and intensity of storms

Source: National Oceanic and Atmospheric Administration

A numerical weather prediction model such as HAFS needs accurate initial weather conditions, so when a storm is brewing, EMC coordinates with NOAA’s Office of Marine and Aviation Operations, which sends reconnaissance planes into the developing storm to collect wind and rainfall data and capture a 3D view of the storm, he says.

EMC also uses satellite observations and gets real-time ocean data from autonomous sailing drones, underwater gliders and prepositioned buoys that provide ocean conditions.

“All of the data collected is fed into the model in real-time, and we improve our forecast based on that information,” Tallapragada says.

NOAA’s supercomputers take about an hour and 40 minutes to produce a five-day forecast. NOAA runs both versions of HAFS four times a day for as many as seven tropical cyclones at any given time.

Because the supercomputers are busy, NOAA turns to Azure to run 31 experimental variations of HAFS — called a Hurricane Ensemble in Real-time on the Cloud (HERC) — for high-priority storms like Hurricane Helene. These variations use slightly different physics, assumptions and initial weather condition data, Tallapragada says.

A Cloud-Based Work Environment for Simulating Storms

The U.S. Army Corps of Engineers also uses models to simulate storms, relying on its Coastal Storm Modeling System (CSTORM-MS) to accurately assess risk and build levees, floodwalls and beach dunes to manage flooding.

USACE runs the model in a supercomputer operated by the Army Engineer Research and Development Center. But within the next year, the agency plans to adopt a hybrid model and take advantage of Microsoft Azure to enhance its modeling operations, says Chris Massey, senior research mathematician for the USACE Research and Development Center Coastal and Hydraulics Laboratory.

“It’s a very accurate model, but it requires a tremendous amount of computing power,” he says.

DISCOVER: GAO’s CIO is no stranger to cloud computing.

Accuracy is important because it allows USACE to build a levee or floodwall at the right height to reduce a community’s flood risk, while allowing the agency to be as cost-efficient as possible, he says.

“It allows for engineering margins of safety, but if you don’t need to build it above a certain height, then you can save on construction costs,” Massey says.

A large simulation that runs CSTORM-MS on 3,000 CPUs typically takes six hours to complete. However, the on-premises supercomputer is a shared resource; if the supercomputer is operating at capacity, USACE can run its modeling system in Azure instead, Massey says.

“This allows me to quickly surge capacity in the cloud for a short time to meet a deadline,” he says.

LEARN MORE: Agencies have options when deploying digital twins.

USACE is currently building a cloud application on Azure that will serve as a next-generation work environment to run the model.

Instead of using software installed on a powerful desktop computer, as they have in the past, USACE staff will log in to the new cloud-based work environment and choose between the on-premises supercomputer or Azure to set up the model, run simulations and analyze the results, he says.

USACE plans to go live with the new work environment within six months to a year. In the meantime, the agency has tested CSTORM-MS in the cloud, and it’s ready to go.

“We evaluated the model to see how well it works in the cloud, and it turns out that it works quite well,” Massey says.

Regularly Upgrading Hurricane Models with Machine Learning

The National Oceanic and Atmospheric Administration relies on supercomputers for real-time modeling during hurricane season. But in the off-season — between December and May — the agency’s modeling team uses its high-performance computing resources to develop, test and evaluate next-generation models.

That includes upgrading its current model — the Hurricane Analysis and Forecast System — every year. The agency is also working on a next-generation weather prediction model that could be enhanced by machine learning, says Tallapragada.

“Once the season winds down, we work on developing the next version,” he says. NOAA collaborates with other federal agencies (including the Energy and Defense departments and NASA) and the academic community as it develops models for scientific research and National Weather Service operations.

Photography by Jonathan Thorpe