Physics-informed data-driven weather forecasts using neural networks

Applications are invited for an Industrial PhD fellowship at Graduate School of Natural Sciences, Aarhus University, Denmark, within the Physics and Astronomy programme. The successful candidate will be part of an application for an industrial PhD from the Innovation Fund Denmark.  The position is available from 1 November 2026 or later. 

Title:
Physics-informed data-driven weather forecasts using neural networks 

Research area and project description:
Data driven models has been shown to produce reliable nowcasts, i.e. weather forecasts valid for the next 12 hours. In this regime such models, usually implemented through neural networks running on GPUs, outperform physical models with respect to both precision and computation requirements. However, in contrast to physcial models, the model predictions are usually not required to satisfy the laws of physics. Thus, we have two main questions:

1. To what extent are physical constraints violated in the regime of long extrapolation time (≫ 1 hour) and unusual input data which is not well-sampled in the training data?
2. What is the impact on the fidelity of model predictions when physical constraints are violated?

The goal of this project is to first quantify violations of physical constraints and their impact on the models predictive power, and then to mitigate this error-mode by implementing a physics-informed model.

More concretely, we will focus on models that are allowed to learn based on empirical data, while being regularised by physical constraints. This could be through a more direct wiring of physical parameters either in the model architecture or in the loss function. Besides a more robust and precise model prediction, this could also lead to faster training times, more stable output results, and a model that generalises better to sparsely sampled regions of the input parameter space.

This is an industrial PhD project in collaboration with Cordulus. Cordulus is a Danish weather company located in Aarhus, with their office in Agro Food Park in Aarhus N, and they currently employ more than 25 people, Cordulus builds their own weather stations, which are utilized primarily by farmers, but also in other industries like construction, road maintenance and racing. This gives Cordulus access to a unique dataset from their proprietary network of ~5000 weather stations. Cordulus has from the start utilized deep learning as the primary foundation of their modelling framework.

For more information about doing an industrial PhD research project, see both the website of Aarhus University and of Innovation Fund Denmark.

For technical reasons, you must upload a project description. When - as here - you apply for a specific project, please simply copy the project description above, and upload it as a PDF in the application. If you wish to, you can indicate an URL where further information can be found.

Qualifications and specific competences:
Applicants must either hold an MSc degree in Physics or expect to finish the MSc within 6 months. 

Note that Industrial PhD candidates must have a relevant Master’s degree.

Place of employment and place of work:
The place of employment is Aarhus University, and the place of work is Department of Physics and Astronomy, Ny Munkegade 120, 8000 Aarhus C and Cordulus, Agro Food Park 13, 8200 Aarhus, Denmark. 

Since it is an Industrial PhD project, the Industrial PhD candidate will be employed in the company and enrolled at Graduate School of Natural Sciences. The Industrial PhD candidate is expected to split the time equally between the company and the university, i.e. work equally in both places – the actual scheduling will be decided within the framework of the project.   

Please note:

• The programme committee may request further information or invite the applicant to attend an interview.
• Admission is contingent upon full external funding from Innovation Fund Denmark, thus, if approved by GSNS, it will be a conditional approval until full external funding is secured. 

All interested candidates are encouraged to apply, regardless of their personal background.

Contacts:
Applicants seeking further information for this project are invited to contact the main supervisor: Assoc. Professor, Thomas Tram, thomas.tram@phys.au.dk. 

The co-supervisors are: Professor, Steen Hannestad, steen@phys.au.dk and CTO, Morten Birk, mb@cordulus.com 

How to apply:

For information about application requirements and mandatory attachments, please see the Application guide. Please read the Application guide thoroughly before applying.

When ready to apply, go to https://phd.nat.au.dk/for-applicants/apply-here/ (Note, the online application system opens 1 June 2026)

1. Choose August 2026 Call with deadline 1 August 2026 at 23:59 CEST.
2. You will be directed to the call and must choose the programme “Physics and Astronomy”.
3. In the boxed named “Study”: In the dropdown menu, please choose: “Physics-informed data-driven weather forecasts using neural networks (Pdwfun)” 

Please note:

• The programme committee may request further information or invite the applicant to attend an interview.

At the Faculty of Natural Science at Aarhus University, we strive to support our scientific staff in their career development. We focus on competency development and career clarification and want to make your opportunities transparent. On our website, you can find information on all types of scientific positions, as well as the entry criteria we use when assessing candidates. You can also read more about how we can assist you in your career planning and development.

Aarhus University’s ambition is to be an attractive and inspiring workplace for all and to foster a culture in which each individual has opportunities to thrive, achieve and develop. We view equality and diversity as assets, and we welcome all applicants. All interested candidates are encouraged to apply, regardless of their personal background. Salary and terms of employment are in accordance with applicable collective agreement.