Professors Collaborate with NVIDIA for Improved Weather Forecasting System
Two professors in the Department of Civil and Environmental Engineering (CEE) are helping NVIDIA integrate atmospheric composition into its Earth-2 AI foundation model ecosystem to establish an AI framework for a combined weather-air quality prediction.
Current forecasting systems have advanced rapidly, but many still have limited representation of aerosols and other atmospheric composition processes like dust and smoke that critically influence weather extremes, climate feedback, and air quality.
Chancellor’s Professor, John D. Tickle Professor, and James G. Gibson Professor Joshua S. Fu and Research Associate Professor Jia Xing are using their own atmospheric chemistry weather model to try and help improve NVIDIA’s current model.
Earth-2 is an open AI-platform for accelerated global weather and climate simulation that generates high-resolution, real-time forecasts. The platform integrates AI models with tools like Earth2Studio and PhysicsNeMeo to predict extreme weather and climate risk.
NVIDIA is supporting Fu and Xing’s research by providing free resources, including eight graphics processing unit (GPU) nodes and cloud computing for the AI application, for a 16-month period.
“They are really looking forward to our results, so hopefully our model can help improve their model and help local communities get better performance,” Fu said. “Hopefully, NVIDIA might be interested in further investigating this area and we can establish long-term collaborations.”
Benefits of Atmospheric Accuracy
The atmospheric chemistry model created by Fu and Xing provides more detailed predictions that are hyper local. The increased accuracy can better protect communities from extreme weather while improving environmental decision-making and public health prediction at a global scale.
“Current systems have a very limited precipitation forecast performance. Implementing the information that we include can definitely improve that,” Xing said. “Also, having straightforward air quality forecast can definitely benefit for the air quality forecasting.”
To create their system, Fu and Xing pretrained AI weather foundation models with UFS-DeepAQM, a machine learning-based atmospheric composition surrogate under development within National Oceanic and Atmospheric Administration (NOAA). The integration enables two-way interactions between meteorology and atmospheric composition.
“Jia has some preliminary results that are pretty promising,” Fu said. “I’m very confident that we are the leading university at this point working in this space with the results we have.”
More accurate weather forecasting has widespread benefits that hit every sector of society, including helping farmers predict season forecasts and assisting various emergency departments with their preparations for hot summer months.
Fu and Xing hope the collaboration with NVIDIA can lead to improvements in weather forecasting that benefit everyone involved. “This is a very good opportunity,” Fu said. “If they think that we are helpful—and this is a purpose they want to use their computer power and resources for—it may lead to an expanded partnership in the future, which would be exciting.”
Contact
Rhiannon Potkey ([email protected])