Periodic Reporting for period 2 - ESiWACE3 (Center of excellence for weather and climate phase 3)
Periodo di rendicontazione: 2024-01-01 al 2025-06-30
Reliable weather and climate predictions are essential for society to adapt to climate change and mitigate the impacts of extreme events. Achieving this requires a new generation of efficient, optimised models capable of running on the world’s most powerful supercomputers. In this context, the Centre of Excellence in Simulation of Weather and Climate in Europe (ESiWACE) was created to strengthen the European weather and climate modelling community by providing advanced technologies, tools, training, and support to improve performance and sustainability. But the transition to the exascale era brings enormous challenges in portability, scalability, and data management that cannot be addressed by individual institutes alone. Now, the third phase of the project, ESiWACE3, plays a central role in linking and enhancing Europe’s expertise. It aims to enable more detailed simulations on upcoming exascale supercomputers and to provide the technological foundations for improved local-scale climate risk assessments. ESiWACE3 focuses on three strategic goals: 1) The project facilitates the transfer and consolidation of knowledge and technology for efficient and scalable simulations across the European Earth system modelling community; 2) It aims to close technology gaps by developing joint solutions for high-resolution modelling challenges that no single group could tackle alone; 3) It acts as a sustainable community hub for training, communication, and dissemination, supporting the education of the next generation of scientists and ensuring inclusiveness. Through these activities, ESiWACE3 promotes synergies, provides targeted HPC services, and strengthens the resilience and innovation capacity of the European climate modelling ecosystem.
Over its first two and a half years, ESiWACE3 has achieved major advances towards preparing weather and climate models for exascale computing. A flagship result is the High-Performance Climate and Weather (HPCW) benchmark, a suite of applications reflecting the complexity of climate codes. It has become a key co-design tool, offering HPC vendors performance tests and guidance for new architectures. Selected core applications, or “dwarfs”, have been deployed on European pre-exascale systems such as LUMI and LEONARDO. They have also improved the LLVM compiler for RISC-V (via EUPILOT) and assessed High Bandwidth Memory in the EPI and EUPEX projects. The project has also advanced the use of Domain Specific Languages (DSLs) to port models to heterogeneous architectures. Key results include implementing FVM\_LAM in Python with GT4Py and porting the ocean model NEMO with Psyclone, now running two configurations on GPUs. These achievements mark important steps towards portability and performance. In parallel, ESiWACE3 has addressed the data challenge of massive simulation outputs. Activities include generating uncompressed IFS datasets, creating the Online Field Compression Laboratory for interactive Jupyter analysis, and studying compression methods combined with diagnostic tests to ensure data quality. Another major strand is the creation of HPC services for the community. Fourteen proposals have been granted under the “Service on model optimisation, acceleration and HPC exploitation”, six already completed. Within the “Service on domain-specific tools and technologies”, two projects have been awarded. Training and capacity building are also a cornerstone. Two hackathons and a summer school have been held, focusing on early-career scientists and equipping them with essential skills. An online training calendar covering Earth system modelling, HPC, and related fields is continuously updated, while several online courses and teaching materials have been made available to share knowledge more broadly.
ESiWACE3 has delivered results that surpass the state of the art in European weather and climate modelling. The HPCW benchmark is a central achievement, encapsulating model complexity in a vendor-usable way that enables co-design and robust evaluation of novel architectures, from GPU-based systems to RISC-V. A governance plan ensures its sustained use on EuroHPC machines and synergies with initiatives like HANAMI. Work on DSLs has advanced portability and performance strategies, accelerating adaptation of complex codes to heterogeneous systems and ensuring sustainability of major models on future exascale platforms. On the data front, ESiWACE3 has pioneered new approaches to storage, cataloguing, and compression. FAIR Digital Objects and compression benchmarks offer transformative solutions to the massive volumes expected from exascale simulations, ensuring datasets remain accessible, interoperable, and reproducible. The project has also shown that complex models can be efficiently ported to CPUs and GPUs, supported by workflows for reduced-precision computing, automated profiling, and portability. Collectively, these innovations prepare the community for exascale computing. Beyond technical advances, ESiWACE3 has strengthened capacity through training, open-source releases, and inclusive strategies. Results are widely adopted, particularly by early-career scientists benefiting from mentoring. Looking ahead, sustained investment in software sustainability, FAIR standards, and international collaboration will be key. By delivering these results, ESiWACE3 accelerates discovery, fosters innovation, and consolidates Europe’s global leadership in high-resolution climate and weather modelling.