PostDoc in physics-constrained deep learning for land surface modelling

Like much of Europe, Switzerland has seen a recent increase in summer droughts and heatwaves, prompting the development of a national drought monitoring and early warning system. To support this system, it is crucial that long-term gridded soil moisture estimates are also available everywhere over the territory, and not only at measuring stations. Your task will be to develop a land surface modelling framework that can provide such information and fulfills the needs of a very diverse user community, ranging from numerical weather forecasts to agricultural decision-making.

The 2-year project will aim to develop a physics-constrained land surface model emulator to produce gridded land surface parameters for any location in Europe. This foundation model is intended to serve as a basis for many potential applications, such as numerical weather forecast initialization, quality-checking of in situ soil moisture measurements, near-real time monitoring, or smart irrigation systems.

The project is embedded within the broader Swiss national drought monitoring program, a coordinated effort between BAFU, MeteoSwiss and swisstopo to establish a state-of-the-art drought monitoring and warning system, which will become operational in 2025. Your work will directly contribute to the improvement and further development of this system.

We are seeking a motivated scientist with strong skills in machine learning and hydro-meteorology. The main tasks of the successful candidate will be to

• translate the most recent developments in the fields of machine learning and Earth system modeling into actionable products used in a productive environment
• build an ML-framework for developing physics-constrained emulators based on gridded spatial climatological data sets (e.g. reanalyses from numerical models, satellite data) and local measurements
• produce and evaluate gridded soil moisture analyses for a climatological period and for near realtime monitoring, thereby contributing to the climate division’s operational duties
• interact with the project team and distribute knowledge on the emulator framework in MeteoSwiss and at ETH Zurich

You will also utilize your knowledge to support the division with ML knowledge. You are expected to transfer science into portable, documented, testable, and performant software according to modern coding standards.

The position is available as of July 1, 2024 (or as soon as possible thereafter). The appointment is limited to 2 years. The working place will at MeteoSwiss (Zurich airport, easily accessible by public transport). The position will remain open until filled. 

In general, postdoctoral lresearchers at ETH Zurich have a full-time employment. A part-time employment may only be considered in exceptional cases (e.g. child- or familycare, other projects or employment).

• University degree with a dissertation in environmental sciences, hydrology, atmospheric and climate sciences, physics, mathematics, or comparable course of study.

• In-depth knowledge in at least one of the following topics with demonstrated affinity for the other:

  1. Modern artificial intelligence and machine learning techniques (e.g., foundation models and ML architectures)

  2. Weather, hydrology, climate, and Earth system modelling

• Mandatory skills:
  

  • Good programming skills in Python

  • Experience with Linux

• Additional desirable skills:
  

  • Familiarity with R

  • Experience with ML packages

  • Knowledge of parallel computing

  • Advantageous knowledge: Modern software development (e.g., unit testing, code review, versioning systems, Git, mixing compiled and scripted code, automated documentation)

• You are used to work independently, thinking analytically and you are a goal-oriented, enthusiastic team player.

• You have very good oral and written skills in English, good knowledge in German or French is a plus.

This is an engaging position, which provides the opportunity to work at the interface of research and applications on a societally highly relevant topic. The position aims to bring applied research in ML-based land surface modelling to real-time operations that will benefit thousands of end-users. We offer a vibrant environment in an interdisciplinary project team with workplace at the interface between a leading national weather service and a well-known research institution. ETH Zurich is a family-friendly employer with excellent working conditions. In particular, we offer flexible working hours, opportunities for training, and peer networks.