WeatherGenerator Service Call 2026

This is an open call for service requests to promote experiments with the WeatherGenerator, to include it in existing workflows, and to design novel use cases. The execution of projects will take place through 2027.

The WeatherGenerator is a state-of-the-art foundation model designed to improve weather and climate predictions, enhance renewable energy planning, and provide better tools for mitigating the effect of extreme events like floods and heatwaves. It will combine the latest machine learning with high performance computing, creating a robust tool for scientific discovery and real-world applications.

The aims of the WeatherGenerator Service Call are to stimulate the uptake of WeatherGenerator in the wider community and promote knowledge transfer from the consortium to the user community. This open call supports researchers that want to benefit from the time and expertise of dedicated Research Software Engineers to:

• Gain early access to the WeatherGenerator deployed/integrated in their own existing workflows
• Experiment with predictions from the model trained on their own (potentially private) data
• Acquire the knowledge and skills to continue development and maintenance after the project end

We expect to award 3-4 projects for this round (to run through the year 2027). Projects that are not initially selected may be placed on a “waiting list” with the possibility of being contacted if engineering hours become available.

Guidelines for applicants

Eligibility: who can apply?

European research groups and industry partners (outside of the WeatherGenerator consortium and affiliated projects).

What kind of applications would be a good fit for this call? The WeatherGenerator currently has 22 test applications that can be explored here: Applications . These give an indication of the range and diversity of applications the project aims to support, but many are very challenging. Simple, well-motivated applications are welcome for this call due to the relatively early phase of development. Note that currently the Global Forecasting application is most mature. Fine-tuning of decoders is at an experimental phase so will have a more research and exploration aspect. Tail networks are not yet available.

Project scope and budget: what can be applied for?

For this round, we are looking to provide one of two types of project: WeatherGenerator deployment support

• Focuses on simple deployment (no fine-tuning) of the WeatherGenerator in an existing workflow
• The applicant provides infrastructure
• Aim: operational usage

WeatherGenerator research application

• Focuses on fine-tuning Weather Generator for new AI application
• Applicant provides fine-tuning data
• Aim: research and exploration

Awarded projects will receive up to 3 Person-Months of time from Netherlands eScience Engineers to support the proposed application, with the possibility to extend up to 6 months for very promising projects at our discretion. Early termination is possible once the objective is met, or if unforeseen circumstances prevent timely completion.

Timeline: when to apply?

7 September 2026 Call published 10 September 2026 Information presentation at EMS, Utrecht 16 October 2026, 14:00 CET Deadline submissions December 2026 Notification of results Early Q1 2027 Projects start

Submitting your proposal: how to apply?

Fill out the google form here: Apply

Conditions for projects

Projects awarded under this call are subject to the conditions set out below. By submitting a proposal, applicants declare that they agree to comply with these conditions if the project is awarded.

Data availability

Any required training data must be available (in a ready-to-go state) by the agreed start date of the proposed work. This means e.g. any data to be used should have been transferred to and be available on the intended (private or public) computing infrastructure by the project start date.

Digital infrastructure

Any required access to (non-public) infrastructure must be arranged and active before the planned start date to allow a timely start. GPUs (or equivalent accelerators) will be required for any fine-tuning work.

Time commitment

The eScience Center engineers will provide help and expertise with setting up the WeatherGenerator for the requested application and infrastructure. This includes knowledge transfer to the applicants such that they may continue development after the end of the project time. Some time commitment is therefore required on the side of the applicant to facilitate the work and knowledge transfer.

Assessment criteria

Applications are reviewed based upon

1. Scientific or societal impact (max 100 words)
2. Technical feasibility

• Proposed work (max 250 words)
  • Describe the intended use case/research/workflow where the WeatherGenerator model’s predictions could be used
  • What inputs would be provided to the model and what physical output (predictions) are required? How would the output be used? What prediction window/timescale is expected for this application?
• Infrastructure
  • Outline the computational infrastructure where the intended work will take place
  • How will eScience engineers be granted access in order to carry out their work?
• Datasets
  • Describe any datasets that will be necessary, in particular size-on-disk, format, resolution, intended use (as appropriate)
  • State any intellectual property restrictions on datasets that will be provided and confirm their availability for the project

Technical feasibility and scientific merit will be reviewed by a panel consisting of experts from within the consortium as well as external.

Contact details

Specific questions about this call For questions about this call for proposals and the assessment procedure, please contact: WeatherGenerator Service Call Coordination Email: [email protected]