Oliver Fuhrer

Operational Numerical Weather Prediction (NWP) workflows are increasingly challenged by rapidly growing data volumes, expanding product diversity, and the need for timely and scalable access to model data. At the same time, modern Earth system services are evolving toward open data policies that require not only standardized access to model output for internal and external users, but also flexible mechanisms to extract and process relevant information in a FAIR (Findable, Accessible, Interoperable, and Reusable) manner. In this context, MeteoSwiss, in collaboration with the European Centre for Medium-Range Weather Forecasts (ECMWF), is developing a modernized data workflow to improve access to NWP model data for internal and external downstream users. The redesigned workflow shifts from a product-centric dissemination model toward a scalable data-as-a-service approach. Rather than relying on the generation and distribution of numerous predefined products, recent ICON forecast output is organized in the Field Database (FDB) and exposed through Polytope, which provides semantic data access and feature extraction capabilities. The workflow automates the ingestion, indexing, access control, and on-demand extraction of forecast fields, and integrates these steps into existing HPC-based production workflows and downstream processing pipelines. By replacing file-based product generation with database-backed access, the workflow enables deterministic data extraction, explicit provenance tracking, and consistent versioning of datasets, so that identical data requests can be reproduced reliably across time and environments. We present recent developments in Earthkit and Polytope that, for the first time, enable such automated workflows on the icosahedral grids used by ICON. Standardized interfaces and modern processing tools from the Earthkit Python ecosystem enable downstream users and applications to retrieve and process tailored subsets of NWP data on demand. Our use of open-source, community-developed software (FDB, Polytope, Earthkit) as core workflow components illustrates how ECMWF technologies can be integrated into national weather service environments. Operational experience gained in this context contributes to improving the maturity and usability of these tools and supports their broader adoption by other ECMWF Member States, facilitating the transfer of FAIR, workflow-based data access concepts across the weather and climate community.

Kilometre-scale weather and climate datasets are invaluable for quantifying, forecasting and projecting hazards in areas of complex topography, such as the Alps. However, producing such datasets using traditional numerical weather prediction (NWP) models is becoming prohibitively expensive, particularly for climate-timescale simulations and large ensembles. Probabilistic generative downscaling offers a potential alternative, as it learns the conditional mapping from coarse global drivers to kilometre-scale regional fields.Here, we evaluate a modified conditional generative correction–diffusion model (CorrDiff) for downscaling the ERA5 and IFS-ENS datasets over the Greater Alpine Region. The modified CorrDiff model was trained using a 20-year, 1-km resolution dataset produced with the ICON numerical model, with precipitation constrained to Swiss radar observations using a latent-heat nudging scheme. This setup allows us to make a direct comparison with MeteoSwiss' operational NWP ensemble.Verification against observations and gridded products reveals that CorrDiff achieves competitive performance following substantial targeted adaptations to the model. Although not explicitly encoded in the loss function, the adapted model reproduces emergent climatological indices, including the diurnal cycle of land precipitation and exceedance probabilities for heavy precipitation. It also captures the spatial patterns of consecutive dry and wet days, as well as prevailing wind direction and directional variability.

Numerical weather prediction and climate models require continuous adaptation to take advantage of advances in high-performance computing hardware. This paper presents the port of the ICON model to GPUs using OpenACC compiler directives for numerical weather prediction applications. In the context of an end-to-end operational forecast application, we adopted a full-port strategy: the entire workflow, from physical parameterizations to data assimilation, was analyzed and ported to GPUs as needed. Performance tuning and mixed-precision optimization yield a 5.5× speed-up compared to the CPU baseline in a socket-to-socket comparison. The ported ICON model meets strict requirements for time-to-solution and meteorological quality, in order for MeteoSwiss to be the first national weather service to run ICON operationally on GPUs with its ICON-CH1-EPS and ICON-CH2-EPS ensemble forecasting systems. We discuss key performance strategies, operational challenges, and the broader implications of transitioning community models to GPU-based platforms.

A refactored atmospheric dynamical core of the ICON model implemented in GT4Py, a Python-based domain-specific language designed for performance portability across heterogeneous CPU-GPU architectures, is presented. Integrated within the existing Fortran infrastructure, the new GT4Py dynamical core is shown to exceed ICON OpenACC performance. A multi-tiered testing strategy has been implemented to ensure numerical correctness and scientific reliability of the model code. Validation has been performed through global aquaplanet and prescribed sea-surface temperature simulations to demonstrate model's capability to simulate mesoscale and its interaction with the larger-scale at km-scale grid spacing. This work establishes a foundation for architecture-agnostic ICON global climate and weather model, and highlights poor strong scaling as a potential bottleneck in scaling toward exascale performance.

Numerical weather prediction and climate models require continuous adaptation to take advantage of advances in high-performance computing hardware. This paper presents the port of the ICON model to GPUs using OpenACC compiler directives for numerical weather prediction applications. In the context of an end-to-end operational forecast application, we adopted a full-port strategy: the entire workflow, from physical parameterizations to data assimilation, was analyzed and ported to GPUs as needed. Performance tuning and mixed-precision optimization yield a 5.6x speed-up compared to the CPU baseline in a socket-to-socket comparison. The ported ICON model meets strict requirements for time-to-solution and meteorological quality, in order for MeteoSwiss to be the first national weather service to run ICON operationally on GPUs with its ICON-CH1-EPS and ICON-CH2-EPS ensemble forecasting systems. We discuss key performance strategies, operational challenges, and the broader implications of transitioning community models to GPU-based platforms.

Numerical weather prediction and climate models require continuous adaptation to take advantage of advances in high-performance computing hardware. This paper presents the port of the ICON model to GPUs using OpenACC compiler directives for numerical weather prediction applications. In the context of an end-to-end operational forecast application, we adopted a full-port strategy: the entire workflow, from physical parameterizations to data assimilation, was analyzed and ported to GPUs as needed. Performance tuning and mixed-precision optimization yield a 5.6x speed-up compared to the CPU baseline in a socket-to-socket comparison. The ported ICON model meets strict requirements for time-to-solution and meteorological quality, in order for MeteoSwiss to be the first national weather service to run ICON operationally on GPUs with its ICON-CH1-EPS and ICON-CH2-EPS ensemble forecasting systems. We discuss key performance strategies, operational challenges, and the broader implications of transitioning community models to GPU-based platforms.

Numerical weather prediction and climate models require continuous adaptation to take advantage of advances in high-performance computing hardware. This paper presents the port of the ICON model to GPUs using OpenACC compiler directives for numerical weather prediction applications. In the context of an end-to-end operational forecast application, we adopted a full-port strategy: the entire workflow, from physical parameterizations to data assimilation, was analyzed and ported to GPUs as needed. Performance tuning and mixed-precision optimization yield a 5.6x speed-up compared to the CPU baseline in a socket-to-socket comparison. The ported ICON model meets strict requirements for time-to-solution and meteorological quality, in order for MeteoSwiss to be the first national weather service to run ICON operationally on GPUs with its ICON-CH1-EPS and ICON-CH2-EPS ensemble forecasting systems. We discuss key performance strategies, operational challenges, and the broader implications of transitioning community models to GPU-based platforms.

The Swiss government introduced an Open Government Data (OGD) policy for all its federal offices starting on January 1st, 2024, with a deadline for implementation at the end of 2026. For MeteoSwiss, the OGD policy took effect with the revised Ordinance for Meteorology and Climatology, entering into force on April 1st, 2025. Building on the experience of several European countries, MeteoSwiss started the project for implementation already in 2022, before the governing law was finalized. The objective was to have API interfaces in place at the start of the revised ordinance, in order to prevent overloading of the current push delivery system for meteorological and climatological data.The project was divided into the sub-projects ‘Legal’, ‘Data Policy’, ‘Technical Implementation, and ‘Organization and Customer Orientation’. It involved the management, customer support and data production units from the beginning in order to draw from the full know-how of the organization as well as to create awareness of the broad effects of OGD on the service. MeteoSwiss teamed up with the Federal Office of Topography swisstopo and the Swiss National Supercomputing Centre CSCS for the hosting and data distribution. Furthermore, the experience of many European met services and the work of the RODEO project were an invaluable help.This presentation will focus on the course of action regarding the interface design for different user groups of data literacy, the layout of the documentation, the strategy for sequencing the implementation steps as well as the design of the support organization. It will present the lessons learned during the project and the first months of operation, and the challenges of setting up a mixed infrastructure including a public cloud, having only vague assumptions of usage and a limited budget at hand. Finally, it will highlight the benefits of cooperating with partners and the community to design and speed up the introduction of open data.

To manage Earth in the Anthropocene, new tools, new institutions, and new forms of international cooperation will be required. Earth Virtualization Engines is proposed as an international federation of centers of excellence to empower all people to respond to the immense and urgent challenges posed by climate change.

The north and south of the Alps, as well as the eastern shores of the Adriatic Sea, are hot spots of severe convective storms, including hail and lightning associated with deep convection. With advancements in computing power, it has become feasible to simulate deep convection explicitly in climate models by decreasing the horizontal grid spacing to less than 4 km. These kilometer-scale models improve the representation of orography and reduce uncertainties associated with the use of deep convection parameterizations. In this study, we perform km-scale simulations for eight observed cases of severe convective storms (seven with and one without observed hail) over the Alpine-Adriatic region. The simulations are performed with the climate version of the regional model Consortium for Small-scale Modeling (COSMO) that runs on graphics processing units (GPUs) at a horizontal grid spacing of 2.2 km. To analyze hail and lightning we have explored the hail growth model (HAILCAST) and lightning potential index (LPI) diagnostics integrated with the COSMO-crCLIM model. Comparison with available high-resolution observations reveals good performance of the model in simulating total precipitation, hail, and lightning. By performing a detailed analysis of three of the case studies, we identified the importance of significant meteorological factors for heavy thunderstorms that were reproduced by the model. Among these are the moist unstable boundary layer and dry mid-level air, the topographic barrier, as well as an approaching upper-level trough and cold front. Although COSMO HAILCAST tends to underestimate the hail size on the ground, the results indicate that both HAILCAST and LPI are promising candidates for future climate research.