Higher Education / Research
Colorado State University Develops Real-Time Hailstorm Forecasting Model Using StormScope
Objective
Colorado State University (CSU) is a leading research institution in radar meteorology and atmospheric science. An ongoing cross-departmental collaboration between the Department of Electrical and Computer Engineering, Atmospheric Science, and the Cooperative Institute for Research in the Atmosphere (CIRA) aims to provide critical weather observation and nowcasting capabilities for severe weather prediction—particularly for high-impact events, such as hailstorms. Recently, this CSU research team trained a specialized version of the NVIDIA Earth-2 StormScope Nowcasting model, specifically for operational severe hailstorm prediction.
Customer
Colorado State University (CSU)
Topic
Simulation / Modeling / Design
Products
Key Takeaways
Forecasting real-time weather events
- Across the Colorado-Wyoming region in a short time frame of a few months
Generating skillful ensemble forecasts
- For severe hailstorm prediction with a 2–3-hour lead time
Model easily integrates different radar systems
- Making the solution deployable across multiple radar networks
The Earth-2 Nowcasting Solution
The CSU team trained and implemented a version of NVIDIA Earth-2 StormScope Nowcasting built on the StormScope architecture that targets severe hailstorm prediction. A high-resolution version of this StormScope model was deployed in the Northern Colorado/Southern Wyoming region for a project supported by the Colorado Wyoming Engines program.
The researchers chose to utilize the StormScope model architecture from PhysicsNeMo™ for its:
- Open Architecture: The StormScope model architecture and PhysicsNeMo allowed them to train specialized versions on NEXRAD composite data and integrate their auxiliary CHIVO weather radar observations.
- Speed: The inference pipeline in Earth2Studio helps generate forecasts in minutes rather than hours, providing more up-to-date predictions for rapidly evolving severe weather.
- Generative AI Capability: The trained StormScope model captures storm evolution, movement, and intensity that is not resolved by traditional models.
- Scalability: CSU’s composite radar technique is flexible and easily integrates different radar systems, making the solution deployable across multiple radar networks.
In addition to StormScope, the NVIDIA hardware and software utilized for this project include H100 and A100 GPUs to run the model inference on, PhysicsNeMo training libraries, and Earth2Studio inference libraries for model deployment.
This specialized hailstorm version of StormScope was trained using NEXRAD radar data covering the Colorado-Wyoming region. Combined with the CSU team’s flexible composite radar technique—which uses multiple radar systems at high spatial and temporal resolution—the model was seamlessly developed, deployed, and used for forecasting real-time weather events across the Colorado-Wyoming region in a short time frame of a few months.
“StormScope is a very convenient and efficient tool to adopt for high-resolution radar based convective storm Nowcasting,” said Chandra Chandrasekar. “Our team has traditionally done Nowcasting using storm advection models, but it was easy to adopt the data-driven AI system.”
Predicting Storms With Speed at Scale
Real-time hailstorm forecasting in Northern Colorado/Southern Wyoming has been rapidly accelerated and scaled due to the implementation of StormScope. Prior nowcasting methods were effective only on a 30- to 60-minute time scale. Now, the new model generates skillful ensemble forecasts for severe hailstorm prediction with a 2- to 3-hour lead time.
With seamless deployment capabilities, CSU’s StormScope-based model is set up to integrate across multiple radar networks throughout the US, Europe, and Asia.
The CSU research team is already looking to deploy the model at other sites.
“NVIDIA and CSU are partnering to solve societal problems with AI, such as improved hail-storm forecasts. We hope to grow this partnership to tackle more problems in the future,” said Allen Robinson, Dean of Walter Scott, Jr. College of Engineering.
“NVIDIA and CSU are partnering to solve societal problems with AI, such as improved hail-storm forecasts. We hope to grow this partnership to tackle more problems in the future.”
Allen Robinson
Dean of Walter Scott, Jr. College of Engineering
Explore NVIDIA Earth-2 open, production-ready AI models and tools for global forecasting and planetary resilience.