VN 26/24 Research Fellowship at ETH Zürich/MeteoSwiss

Recent advances in AI-based weather prediction have demonstrated remarkable skill and computational efficiency. However, most current machine-learning weather prediction (MLWP) systems rely primarily on NWP analyses for initialization and only partially exploit the wealth of available satellite observations. With the advent of the Meteosat Third Generation (MTG), new high-frequency and high-resolution measurements of clouds, moisture, temperature, and lightning activity provide unprecedented opportunities for substantially improving regional forecasts.

Within the framework of a Research Fellowship funded by EUMETSAT, we are advancing the integration of geostationary satellite observations into a next-generation regional MLWP system. The project builds on an existing graph-based, stretched-grid regional forecasting model developed at MeteoSwiss and is embedded in the Anemoi framework initiated by ECMWF. The objective is to develop and evaluate novel multi-encoder-decoder architectures capable of ingesting various satellite data streams (e.g. radiances, cloud products, lightning observations, hyperspectral soundings) and integrating them into high-frequency forecasting cycles with lead times from short-range to up to 10 days ahead. The work will be carried out in close collaboration with national and international partners and contributes directly to the future operational exploitation of MTG data.

We are looking for a Scientific Programmer / Software Developer to join our motivated and interdisciplinary team.

In this role, you will:

• Develop and implement machine-learning model architectures enabling the direct ingestion of next generation satellite data (e.g. MTG FCI, LI, IRS) into state-of-the-art regional forecasting models in Anemoi
• Contribute to the evolution of a multi-encoder-decoder MLWP framework within the Anemoi ecosystem

• Train, fine-tune, and evaluate models using large-scale meteorological and satellite datasets
• Quantify the impact of satellite data on forecast skill across variables and lead times
• Collaborate closely with scientists, ML researchers, and operational forecasting teams to ensure that forecast outputs meet the needs of diverse users
• Disseminate results through scientific publications, conference presentations, and exchanges with EUMETSAT and partner institutions

We welcome applications from candidates with diverse backgrounds who meet most (not necessarily all) of the following criteria:

• PhD in computer science, data science, natural sciences (e.g. physics, meteorology) or a related field. Candidates with an MSc and proven professional experience may also be considered
• Experience working with satellite data (e.g. geostationary observations, radiances, retrieval products)
• Strong programming skills in Python
• Experience in machine learning, ideally including deep learning architectures such as graph neural networks, transformers, or spatio-temporal models
• Experience with high-performance or distributed computing environments
• Good understanding of meteorological processes and numerical weather prediction
• Interest in DevOps practices and sustainable software engineering
• Ability to work independently on research questions while contributing to a collaborative team environment
• Motivation to work in a diverse, interdisciplinary, and international environment
• Good communication skills (oral and written) in English and one of the Swiss national languages

We look forward to receiving your online application with the following documents:

• Motivation letter
• CV (including a link to your latest thesis and your publication record)