ZAMG is the fourth weather centre to select Cray to run advanced machine learning and AI. This project focuses on improving meteorological products including local nowcasts for weather, wind power yield assessments and air quality analysis. Nowcasting means forecasting the weather over short time windows, typically up to two hours. Using deep learning methods, ZAMG is leveraging its Cray CS-Storm supercomputer to optimize the orientation of wind-powered generators for maximum efficiency and to train neural networks with current and historical weather data.
"ZAMG is adopting new methodologies to advance its suite of meteorological products. By leveraging machine learning, we are able to improve the quality of our nowcasting system", stated Dipl. Ing. Mag. Günther Tschabuschnig, CIO at ZAMG. "With the Cray supercomputer, we cannot make the weather better, but we can predict it much better with a whole new level of service."
As the state meteorological and geophysical service of Austria, ZAMG provides critical information and advises on efficient use of conventional and alternative resources for the energy sector. The Institution is tasked with providing accurate weather forecasts on a variety of time scales. This is a challenge ZAMG researchers are solving by training neural networks with massive amounts of weather data from measurement stations and weather models from data collected over the years. With the Cray CS-Storm, ZAMG can improve local nowcasts for wind power, air quality and more.
"Many of the world's leading weather organizations rely on Cray advanced supercomputing technology for their forecasting", stated Ilene Carpenter, earth sciences segment director at Cray. "Our customers' workloads are becoming increasingly converged as they adopt machine learning and deep learning to augment modelling and simulation. ZAMG's new Cray CS-Storm system enables scientists to extract value from the organisation's rich, historical weather data and more efficiently train neural networks and deep learning models to improve their nowcasting capabilities."