Postdoctoral Researcher - Probabilistic Deep Learning for Urban Air Quality (AEON-UP)

Postdoctoral Researcher - Probabilistic Deep Learning for Urban Air Quality (AEON-UP)

Beginn der Ausschreibung 14.08.2026

Ende der Ausschreibung 03.09.2026

Project start October 01, 2026 until September 30, 2028.

Air pollution remains one of the major environmental health challenges in urban areas. At the same time, observations are often sparse and unevenly distributed. In the AEON-UP project, we develop a transferable AI system for high-resolution urban air quality prediction across European cities and combine:

- physics-based chemistry transport models (CTMs)

- with probabilistic deep learning methods, in particular neural processes

to generate spatially resolved predictions of NO2, PM2.5, and ultrafine particles (UFP), including uncertainty estimates.

• long-standing expertise in urban air quality modelling
• with growing activities in machine learning and data-driven methods

You will work closely with:

• Dr. Martin Ramacher (machine learning for environmental applications)
• Dr. Matthias Karl (urban air quality modelling and emissions) and collaborate within a Helmholtz-wide network with expertise in:
• air quality measurements and exposure science (RIFS Potsdam)

We value a collaborative, inclusive, and respectful working environment and actively encourage diverse perspectives in research.

Equal opportunity is an important part of our personnel policy. We would therefore strongly encourage qualified women to apply for the position. In principle, the full time position (39 h/week) is also shareable and limited to 2 years, starts at 1st of October 2026.

You will play a key role in the methodological development of the AEON-UP system.

Your focus will be on:

• building the data infrastructure
• designing and implementing the neural-process architecture

while contributing to the overall modelling pipeline, evaluation, and transfer to new cities.

Your focus will be on:

• you will develop scalable data pipelines for heterogeneous environmental datasets
• you will design and implement neural-process-based architectures for spatio-temporal prediction
• you will combine gridded CTM data and sparse observations in a unified probabilistic framework
• you will develop and evaluate uncertainty quantification methods
• you will train and scale models on HPC systems
• you will benchmark your approach against existing methods (e.g. XGBoost, Gaussian Processes)
• you will publish your work and contribute to open-source releases

Your profile

• PhD in Machine Learning, Computer Science, Physics, Environmental Sciences, or a related field
• experience with deep learning frameworks (e.g. PyTorch)
• strong programming skills in Python
• intrinsic motivation to work on complex, real-world problems with societal relevance

Nice to have:

• experience with probabilistic modelling / Bayesian methods
• experience with spatio-temporal or geospatial data
• experience with HPC

Our offer

• an exciting and varied job in a research center with around 1,000 employees from more than 60 nations
• International Welcome Office with guest accommodation
• 6 weeks holiday per year; company holidays between Christmas and New Year's Day
• social benefits according to the collective agreement of the public service and remuneration up to pay group 13 (according to EntgO Bund)
• very well work-life balance; options for mobile and flexible work
• a well-connected research campus (public transport) and best networking opportunities, subsidy for the Deutschlandticket if certain conditions are met (job ticket), charging facilities for electric vehicles
• individual opportunities for further training
• an excellent technical infrastructure and modern workplace equipment
• family-friendly company policy with childcare facilities, e. g. nursery close to the company
• Operational Health Management, e. g. Employee Assistance Program
• a varied offer in the canteen on campus

Severely disabled persons and those equaling severely disabled persons who are equally suitable for the position will be considered preferentially within the framework of legal requirements.