Presentation
Exascale Systems Present a Vision for Weather and Climate Prediction - Can we Meet the Challenges?
Speaker
Event Type
Focus Session
Extreme-Scale Computing
TimeTuesday, June 18th2:52pm - 3:15pm
LocationPanorama 2
DescriptionA qualitative change of our models towards much enhanced predictive skill requires running simulations at much finer resolutions than today, with more realistic Earth-system models and with much larger ensembles. Our data assimilation methods need to follow this trend to provide accurate initial conditions at such scales exploiting information from unprecedented volumes of observational data. Meeting these requirements translates into at least 1000 times bigger high-performance computing and data management resources than available today – towards what’s generally called ‘exascale’. However, our current codes only reach about 5% efficiency on supercomputers so that we are actually far away from true exascale computing.
Achieving the next step in forecasting therefore needs a significant investment in code efficiency, which entails one of the most radical changes in weather and climate prediction since the first implementation of numerical techniques on computers in the 1950s. This step encompasses a fundamental redesign of mathematical algorithms and numerical methods, the adaptation to new programming models, the implementation of dynamic and resilient workflow management, and the efficient post-processing and handling of big data. While artificial intelligence methods cannot overcome the main bottlenecks of efficient computing they can help alleviate algorithmic cost and support information extraction from both observational and simulated data.
The weather and climate community is increasingly becoming connected to rethink their approach to efficient forecasting - as this challenge is too big to be solved by individual organisations or communities.
Achieving the next step in forecasting therefore needs a significant investment in code efficiency, which entails one of the most radical changes in weather and climate prediction since the first implementation of numerical techniques on computers in the 1950s. This step encompasses a fundamental redesign of mathematical algorithms and numerical methods, the adaptation to new programming models, the implementation of dynamic and resilient workflow management, and the efficient post-processing and handling of big data. While artificial intelligence methods cannot overcome the main bottlenecks of efficient computing they can help alleviate algorithmic cost and support information extraction from both observational and simulated data.
The weather and climate community is increasingly becoming connected to rethink their approach to efficient forecasting - as this challenge is too big to be solved by individual organisations or communities.
Speaker
