MAELSTROM is a large-scale R&D project, aiming to fundamentally improve weather and climate prediction.
It will join the powers of high performance computing (HPC) and machine learning (ML) to cope with the extreme complexity inherent in weather and climate forecasts.
MAELSTROM will produce an entire set of solutions to fill the vacuum in this field: a novel development infrastructure for HPC-based data science, suitable hardware configurations, and the actual forecasting applications.
MAELSTROM aims to impact the world of computer and data science far beyond weather & climate.
Weather and climate have a huge impact on human safety, health, economy, and peace. This applies to short-term weather effects, causing serious hazard, and to medium and long term effects of climate change – now an omnipresent concern, likely to overthrow economic, social, and political power structures.
Effects of both weather and climate could be managed better, if better predictions were available. Yet even the most advanced weather prediction models are struggling to provide reliable predictions of extreme weather events with more than a week forecast lead time. Even the most advanced climate prediction systems fail to provide reliable predictions for local weather patterns changes due to climate change.
With novel forecasting solutions, harnessing machine learning and high-performance computing in combination, MAELSTROM aims to achieve a step change in forecasting precision and reliability.
Machine leaning (ML) applications (part of “artificial intelligence”) provide powerful opportunities to learn the behavior of complex systems from data. ML tools have shown an outstanding capability to finish difficult tasks much quicker or better than humans, e. g. in image and speech recognition, healthcare, gaming, or finance.
A big barrier to date is, however, the lack of comprehensive workflow tools that would allow to efficiently develop and benchmark scalable applications. MAELSTROM will produce ML development tools to provide data scientists a more convenient and powerful dev infrastructure - supercomputing-ready.
Developments for computer hardware are moving at a breath-taking pace. Recent machines are about to cross the threshold to exascale computing.
Weather & climate forecasts, requiring to deal with the greatest possible level of non-linear complexity (the entire earth) at very high resolution, could be the natural application area for high-performace computing (HPC). Yet, productive approaches harnessing both, ML and HPC in combination, for W&C are in their absolute infancy.
MAELSTROM will develop high-performance compute systems and hardware configurations for ML-based weather & climate prediction that perform benchmarkably better .