ESCAPE-2 develops world-class, extreme-scale computing capabilities for European operational numerical weather and climate prediction, and provide the key components for representative benchmarks to be deployed on extreme-scale demonstrators and beyond.
→ to design scientifically flexible and sustainable weather and climate prediction systems.
Achieved by:
- implementing data structures and tools supporting parallel computation of dynamics and physics on multiple scales and multiple levels;
- combining highly-scalable spatial discretization with proven large time-stepping techniques to optimize time-to-solution;
- applying machine learning for accelerating complex sub-components;
- combining multi-grid tools, iterative solvers, and overlapping computations with flexible-order spatial discretization to strengthen algorithm resilience against soft or hard failure.
→ to maximize flexibility, programmability and performance portability to heterogeneous hardware solutions across different weather and climate models.
Achieved by:
- defining a weather and climate DSL concept for a comprehensive set of models;
- developing and demonstrating an open source toolchain for code adaptation and performance portability to different hardware architectures;
- sustaining community-wide code usability and maintainability beyond the lifetime of the project.
to enable deployment on energy efficient and heterogeneous HPC architectures, in particular Extreme-scale Demonstrators (EsD).
Achieved by:
- setting up a benchmark hierarchy of representative models for atmosphere and ocean together with machine-wide simulated, domain specific workflows;
- incorporating seamlessly novel and disruptive numerical algorithms and mathematics;
- ensuring portability through a weather and climate domain-specific language;
- representing (world-) leading European weather and climate prediction models for both atmosphere and ocean.
→ to establish exascale-ready verification and uncertainty quantification tools for weather and climate prediction and beyond.
Achieved by:
- implementing highly non-linear and multi-dimensional weather & climate dwarfs in a cross-disciplinary VVUQ framework;
- estimating mathematical, numerical and data parameter related uncertainties on simulation performance;
- demonstrating VVUQ capability across a hierarchy of high-dimensional modelling systems.
→ to accelerate mathematical algorithm development, foster continued leadership of European weather and climate prediction models and sustain commercialisation of weather-dependent innovative products and services in Europe.
Achieved by:
- providing an open-source DSL toolchain software and support beyond the project lifetime to sustain & accelerate novel algorithm development and ensuring performance portability to emerging HPC hardware;
- reforming fundamentally the way weather and climate modelling is performed and verified;
- supporting formally incremental upgrades from novel mathematical and algorithmic concepts into large-scale weather and climate model legacy codes;
- disseminating and providing training on novel code development concepts through direct public engagement, stakeholder communication and through the ESiWACE Centre of Excellence;
- providing sustainability for Copernicus services that deliver vital information for European society.