The Institute of Solar Research develops innovative technologies for the utilisation of solar energy. The focus is on electricity generation and the provision of heat and fuels. The primary goal is to use solar energy to contribute to the heat transition and a reduction in fossil fuels.
What to expect
This master thesis explores the challenge of creating physically interpretable representations of atmospheric states using sparse, multimodal observations. It explores learning-based approaches for continuous three-dimensional representation and reconstruction, such as neural field representations and probabilistic generative models, to infer continuous atmospheric variables from heterogeneous data sources, including all-sky imager observations, satellite imagery, and weather radar measurements. The emphasis is on self-supervised learning using geometric, temporal, and physical consistency constraints to enable 3D reconstruction without direct ground-truth volumes. This study aims to advance uncertainty-aware representation learning for underdetermined atmospheric imaging problems and contribute to physics-informed generative modelling in geophysical systems.
You will be part of a diverse and motivated team working on energy-transition topics and contributing to climate protection. Close collaboration with supervisors and colleagues will support you in exchanging ideas and solving challenges, so you will gain hands-on experience in machine learning, software development, automated testing, version control and modern image-processing technologies. A particular highlight of the project is the opportunity to work in Almería, Spain, one of the sunniest locations in Europe.
Your tasks
- get familiar with existing methods for 3D reconstruction and machine learning
- analyze and preprocess multi-modal atmospheric observation data
- implement and test a 3D atmospheric reconstruction model using modern deep learning tools
- improve model performance by incorporating physical constraints and consistency between different data sources
- analyze and visualize the reconstructed 3D atmospheric fields and their uncertainty
- document methodology and results in a well-structured Master’s thesis
Your profile
- You have a strong academic record in a master's program in computer science, physics, mathematics engineering or a related field.
- experience in Python and basic knowledge about machine learning
- ability to work independently and collaborate in an international team
- prior experience in data analysis, computer vision and git versioning systems
- confident in speaking and writing English
We look forward to getting to know you!
If this sounds like an exciting opportunity for you, please apply by sending us a cover letter and your CV!
If you have any questions about this position (Vacancy-ID 4086) please contact:
Stefan Wilbert
Tel.: +49 2203 601 4619
