Real-time Analytics
Colorado State University needed a practical way to forecast severe hailstorms earlier and across multiple radar systems, but traditional nowcasting approaches were largely limited to 30- to 60-minute windows and difficult to generalize across networks.
Reported impact
CSU moved operational hailstorm nowcasting from 30- to 60-minute horizons to skillful ensemble forecasts with 2- to 3-hour lead times, delivered a regional deployment within a few months, and created a model that can be rolled out across multiple radar networks.
CSU built a specialized severe-hail version of NVIDIA Earth-2 StormScope so its radar-meteorology researchers could move from short-horizon advection models to AI-assisted operational nowcasting. The team trained the model on NEXRAD composite data for the Colorado-Wyoming region, integrated auxiliary CHIVO radar observations through the open StormScope architecture in PhysicsNeMo, and used Earth2Studio libraries plus H100 and A100 GPUs for deployment and inference. That stack let CSU generate forecasts in minutes rather than hours, preserve a high-resolution composite-radar workflow, and ship a working regional model within only a few months. Because the composite technique can ingest different radar systems, the same architecture gives CSU a deployable path to other US, European, and Asian radar networks rather than confining the work to one local installation.
Composite NEXRAD and auxiliary CHIVO radar data are used to train the specialized StormScope model through PhysicsNeMo. Forecast jobs then run through Earth2Studio on H100 and A100 GPU infrastructure, and the resulting nowcasts are served back to CSU's operational severe-weather workflows across the Colorado-Wyoming region.
CSU's composite radar technique is flexible and easily integrates different radar systems, making the solution deployable across multiple radar networks.
Now, the new model generates skillful ensemble forecasts for severe hailstorm prediction with a 2- to 3-hour lead time.
The CSU team trained and implemented a version of NVIDIA Earth-2 StormScope Nowcasting built on the StormScope architecture that targets severe hailstorm prediction.
The inference pipeline in Earth2Studio helps generate forecasts in minutes rather than hours.
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
NVIDIA Sample Feb 4, 2026 VinuesaLab at the University of Michigan VinuesaLab needed high-fidelity, city-scale urban airflow simulation for planning and resilience work, but traditional CFD workflows were too slow, too resource intensive, and largely constrained to 2D approximations. 10x Faster Training Diff-SPORT · NVIDIA A100 Tensor Core GPUs · NVIDIA PhysicsNeMo NVIDIA Sample Oct 9, 2026 California Department of Transportation (Caltrans) Caltrans needed to inspect and monitor roughly 1,200 active transportation projects more frequently and safely, but manual site visits were slow, inconsistent, and risky for staff working around traffic, bridges, and… Remote operations up to 100 miles Esri ArcGIS · NVIDIA CUDA · NVIDIA Jetson Orin NVIDIA Sample Sep 30, 2026 EPRI EPRI needed to run advanced AI for sensitive utility documents and weather-planning workflows locally, because energy-sector research data could not leave controlled environments and shared compute delays limited how… 150 tokens/sec aggregate across 10 concurrent Power Chat users DGX Spark · Dell Pro Max with GB10 · NVIDIA CBottle 