As artificial intelligence (AI) revolutionizes many aspects of our daily lives, accurate local predictions are also found on the radar. The US National Oceanic and Atmospheric Administration (NOAA) Global Systems Laboratory (GSL) has prepared a new regional weather forecasting system, HRRR-Cast.
This innovative AI-equipped system offers highly localized predictions over the 1-6 hour range. It provides detailed and timely predictions essential for local weather phenomena.
As the first of its kind developed by NOAA, HRRR-Cast represents the main step in integrating artificial intelligence into environmental modeling. HRRR-Cast is built as a data-driven counterpart of NOAA's well-known high-resolution Rapid Rapid Refresh (HRRR) model, aims to deliver results faster and lower computational costs while maintaining comparable prediction accuracy.
HRRR-Cast is developed under the umbrella of NOAA's Project Eagle (an experimental artificial intelligence global and limited area ensemble prediction system). By leveraging a comprehensive three-year dataset generated by physics-based HRRR models, AI systems learn to enhance prediction technology.
The project is a joint effort that includes NOAA's GSL, the Cooperative Institute for Environmental Science Research (CIRES), and the Institute for Research in the Atmospheric (CIRA).
Predict severe weather phenomena
One of the key innovations in HRRR-Cast is its ability to generate ensemble predictions using AI-driven methods such as diffusion technology.
These ensembles generate multiple prediction scenarios and help improve predictions for harsh weather events. Early tests have shown promising results by capturing the dynamics of changing weather patterns. This is important for timely warnings.
While HRRR-Cast demonstrates the power of artificial intelligence, researchers emphasize that it is not intended to replace traditional physics-based models.
Instead, the goal is a hybrid approach that combines AI's computational efficiency with the reliability of physics-driven simulations, potentially transforming the future of weather forecasting.
This breakthrough is part of NOAA's broader initiative to integrate AI technology into weather research and is part of a partnership between the NOAA Artificial Intelligence Center (NCAI) and industry leaders like Google.
Europe leads prices with global AI weather forecasts
It's not just countries that have made this progress. Several countries and organizations are moving forward in the field of AI-based weather forecasting by developing models that operate primarily at a global scale, exploring regional applications similar to the HRRR-Cast model.
These AI-driven systems utilize cutting-edge machine learning architectures to improve prediction accuracy, efficiency and handling of extreme weather phenomena.
One notable example is GraphCast, developed by Google DeepMind in the UK. Using Graph Neural Networks (GNNS), GraphCast, trained on ECMWF's ERA5 reanalysis data, generates global forecasts up to 10 days ahead at a resolution of approximately 25 km.
Surprisingly, it offers higher accuracy than traditional numerical prediction (NWP) models like ECMWF's Integrated Prediction System (IFS), completing predictions in minutes on a single TPU. NOAA's Project Eagle is built on Graphcast by tweaking with operational data and improving practical applicability.
China and other countries are catching up
China's initiative, Huawei's Pangue-Weather, employs a 3D Earth-specific transformer architecture for forecasts up to 7 days. Excellent in high resolution and efficient tropical cyclone tracks and extreme event predictions.
NOAA also uses global data systems to test Pangu-Weather, demonstrating its potential as a competitive global alternative that balances calculation speed and accuracy.
Nvidia's Fourcastnet in the US employs a vision trans model for high-resolution weather forecasts for up to seven days. Designed for GPU scalability, it supports both global and experimental regional forecasting, providing the same versatility as HRRR-Cast.
NOAA researchers have incorporated FourCastNet into AI-driven predictive research and adapted it to the global dataset.
European ECMWF contributes through an AI Prediction System (AIFS), which integrates trans-architecture-based AI models into existing prediction pipelines. AIFS complements traditional NWPs, like IFS, faster and more data-driven predictions, and drives local AI innovation through initiatives such as the AI Weather Quest.
Finally, China's Feng Shui, developed by the Shanghai Institute of Artificial Intelligence, provides high-precision global forecasts for medium-range ranges. With a transformer-based model that trains a wide range of reanalytic data, Fengwu highlights use for scalable operational purposes and tailors global trends towards efficient, AI-enhanced weather.
Where does India stand?
India is also in the advanced stages of developing its own AI-based weather forecasting models. The institutions under the Ministry of Earth Science (MOE) are actively integrating artificial intelligence (AI) and machine learning (ML) technologies into both research and operational frameworks.
These techniques have been applied to monitor and predict tropical cyclone pressure (TCHP), which is a key factor predicting cyclone strength.
The Indian Meteorological Department (IMD) uses AI and ML to correct bias in the output of the Numerical Weather Forecast (NWP) model. The government is providing dedicated financial support to accelerate the development of Indigenous forecasting systems.
Additionally, the ministry has established a virtual centre at the Tropical Meteorological Institute (IITM) in India, focusing on AI, ML and Deep Learning (DL).
The centre is leading efforts to develop AI/ML-based applications specifically designed for localized weather and climate analysis.
These systems focus on predictive accuracy, efficiency and improving extreme weather handling through advanced machine learning architectures.
– end
I'll adjust it
