Wielding Artificial Intelligence, the National Labs Take a Stab at Disaster Resilience

Key Takeaways:

• The U.S. Department of Energy’s National Laboratories are leveraging artificial intelligence (AI) to support innovations in natural disaster and extreme weather forecasting.
• Tools like Argonne National Laboratory’s ADDA and Pacific Northwest National Laboratory’s ChatGrid and RADR are identifying flood risks to individual neighborhoods, tracking wildfire boundaries and burn severity, and calculating risks to transmission lines in a winter storm and heat wave alike. 
• State governments, city planners, utility companies, and emergency agencies alike can access and apply the data from these tools to act quickly to protect people and infrastructure in the face of extreme weather.

 

From hurricanes and wildfires to flooding and extreme heat, billion-dollar disasters are on the rise in the United States, with average annual losses from extreme weather events totaling $149.3 billion between 2020 and 2024. As disasters and their price tags increase, predictive planning and adaptation tools have become increasingly critical to resilience. The U.S. Chamber of Commerce Foundation estimates that every dollar invested in community disaster preparedness can save $13 in damages, cleanup costs, and economic impacts down the line.

At the U.S. Department of Energy’s National Laboratories, artificial intelligence (AI) is supporting innovations in advanced predictive modelling crucial to disaster forecasting, preparedness, and risk mitigation. Among the labs at the forefront of AI-driven climate resilience are the Argonne National Laboratory and the Pacific Northwest National Laboratory.

 

Community-Level Disaster Prediction at the Argonne National Laboratory

In 2023, researchers at the Argonne National Laboratory (ANL) in Lemont, Illinois, unveiled the Argonne Downscaled Data Archive (ADDA) to help fill a crucial gap in understanding extreme weather risk. ADDA uses artificial intelligence and advanced computing to predict the risk of wildfires, drought, extreme heat and cold, hurricanes, rainfall, flooding, and sea level rise at the national, regional, city, and community levels.

Aerial view of Argonne National Laboratory in Illinois. Credit: Argonne National Laboratory via Flickr

Unlike global climate models, which can only calculate risk for resolutions of 62 to 124 miles, ADDA can make risk projections for areas spanning just 2.5 miles (4 kilometers). The data behind these predictions are also particularly comprehensive—both future-looking and retrospective, based on 20 years of historic weather data and a wide range of past and future global greenhouse gas emissions scenarios. Most importantly, ADDA’s data is accessible, translating and condensing massive datasets into multiple formats of information that are usable by utilities, governments, and the public. They can also access the information through ANL’s Climate Risk and Resilience (ClimRR) portal.

The latest version of ADDA, known as ADDA v2, ups its climate resilience game in two ways. First, the tool is now able to provide updates by the hour, offering the type of real-time data needed to keep people and critical infrastructure safe during emergencies. Second, ADDA v2 accounts for previously underrepresented areas by expanding its coverage both outward and inward, providing data for broader North America, from Alaska to the Caribbean islands, and using a spatial resolution of 2.5 miles (the earlier version of ADDA—ADDA v1—worked at a resolution of 7.5 miles).

This type of information has crucial real-world applications for state governments, city planners, utility companies, and emergency agencies. The governments of California, Texas, and Portland, Maine, used ADDA v1 to identify risks to their critical infrastructure as part of their respective regional resiliency assessment programs. Commonwealth Edison, Illinois’s largest electric utility, used ADDA v1 to identify the impacts of future extreme heat and wind gusts on its grid infrastructure, while the New York Power Authority used the tool to assess the impact of extreme rainfall and flooding on its power plants and the impact of extreme heat on its transmission lines. The Department of Energy applied ADDA data to weigh how climate change will impact soil and groundwater at 118 contaminated “legacy sites” across the country. It has also been used to assess risk from hurricanes, wildfires, droughts, and extreme cold.

 

Bolstering Grid Resilience with the Pacific Northwest National Laboratory’s ChatGrid

The U.S. electric grid comprises thousands of power lines, substations, and control centers that work together to keep electricity flowing to our homes, schools, and businesses. Every day, grid operators have to make quick decisions when problems arise—from storms and equipment damage to sudden spikes in electricity use. To identify the issue and its solution, operators usually sort through massive amounts of technical data, which can be time intensive, slowing down response efforts.

Aerial view of Pacific Northwest National Laboratory in Washington State. Credit: Pacific Northwest National Laboratory via Flickr

To help accelerate this process, researchers at Pacific Northwest National Laboratory (PNNL) created a tool called ChatGrid, which was made available to the public on Github in 2024. ChatGrid answers questions about the power grid in real time, using the same kind of advanced language model used by tools like ChatGPT. This means users can ask questions in plain English and get clear and immediate answers. For example, a user might ask, “How much power is Wind Generator A producing in the northwest?” ChatGrid will then create an easy-to-follow map that identifies electricity levels, voltage, and how power is flowing through different parts of the grid.

Rather than relying on live grid data, which is often inaccessible, ChatGrid uses simulated data from a platform called the Exascale Grid Optimization model, which mimics how the nation’s power grid behaves under different conditions. This allows operators and planners to explore "what-if" scenarios—for example, how a major storm or outage might affect the system. As Chris Oehmen, one of the lead researchers on the project, puts it, “if you have a hurricane coming, you need to know where to send trucks and equipment right away. You don’t have a week.”

By making complex grid information easier to access, ChatGrid helps utility operators and planners make smart decisions more quickly. As the tool continues to improve, it has the potential to play a major role in keeping the U.S. power system reliable—especially in the face of extreme weather and growing energy demand.

 

Rapid Analytics for Disaster Response at Pacific Northwest National Laboratory

The Rapid Analytics for Disaster Response (RADR) system is a disaster assessment tool developed by PNNL in 2014 to illustrate the impact of natural disasters on infrastructure. When a wildfire, flood, hurricane, or earthquake hits, RADR pulls together satellite imagery, AI, and cloud computing to rapidly depict what is happening on the ground.

RADR takes large volumes of high-resolution, open-access satellite images and processes them using machine learning models that are trained to detect damage and risks to infrastructure. Within minutes of receiving images, RADR can identify which areas have been affected and where critical infrastructure might be in danger. This allows planners and emergency responders to move quickly, without waiting for lengthy manual assessments.

In response to the increasing frequency and reach of wildfires, PNNL created a specialized version of their system called RADR-Fire. Fully automated, cloud-based, and AI-driven, RADR-Fire maps wildfire boundaries, tracks burn severity, and identifies how close fires are to key infrastructure like power lines, roads, and utilities. The result? Better access to the information emergency managers need to respond to wildfires.

RADR is already being used by the Department of Energy, the Federal Emergency Management Agency, the National Interagency Fire Center, the U.S. Geological Survey, utilities, and the private sector. These users rely on RADR’s ability to quickly turn raw satellite data into clear, actionable visualizations. For federal agencies, RADR supports faster disaster response and recovery planning. For utilities and private industry, it helps teams make decisions to protect critical infrastructure and reduce service disruptions.

By automating what would otherwise be a slow and resource-heavy process, RADR is helping decision-makers get ahead of fast-moving disasters and improve how they respond to and manage infrastructure risks.

AI-driven disaster prediction and preparedness tools developed by national labs are revolutionizing how we anticipate and respond to extreme weather events. These technologies can save lives, reduce economic losses, and strengthen community resilience.

Authors: Raneem Iftekhar and Nicole Pouy