Doctoral Position in Machine Learning Avalanche Forecasting

The Swiss Data Science Center (SDSC) is seeking a PhD student for a Swiss National Science Foundation (SNSF) project, starting May 1st.

This role sits at the intersection of applied machine learning, natural hazards, and snow/avalanche physics. The project is named Towards high-resolution, intelligent, spatiotemporal avalanche forecasting (THIRST in short).

The position will be based at the SDSC Zurich office (Andreasturm), with co-supervision from researchers at the WSL Institute for Snow and Avalanche Research (SLF) in Davos and the Chair of Alpine Mass Movements at ETH Zurich.

The host institution, the SDSC, is a national research infrastructure in data science and artificial intelligence (AI) of the ETH Domain, with EPFL and ETH Zurich as founding partners. Its mission is to enable data-driven science and innovation for societal impact, and it drives its initiatives in research projects, knowledge and technology transfer, and education. With a large multidisciplinary team of professionals in Lausanne, Zurich and Villigen, the SDSC provides expertise and services to various domains, such as health and biomedical sciences, energy and sustainability, climate and environment, and large-scale scientific infrastructures.

For more information please visit our website.

The project team includes a second PhD student (hosted at SLF Davos and the University of Zurich) and a dedicated software engineer (based at SLF Davos). The PhD will be supervised by Dr. Michele Volpiand officially registered under Prof. Johan Gaume at ETH Zurich’s Department of Civil, Environmental, and Geomatic Engineering (d-BAUG).

Predicting snow avalanches—both in time and space—remains a major challenge, despite its critical role in saving lives and reducing infrastructure and mobility disruptions. This project aims to support human expert forecasters by developing machine learning models and a related infrastructure, capable of processing large, heterogeneous datasets with complex spatio-temporal dynamics and underlying physical processes.

While machine learning models now approach human-level performance on average, they still struggle with rare and critical conditions due to a lack of physical grounding. This PhD project aims to enhance traditional machine learning methods by integrating physics, empirical rules, heterogeneous observations and from different modalities, all aiming at improving the prediction quality and reliability of key avalanche formation and propagation parameters—critical for forecasters and downstream models.

The project will leverage a wealth of newly available data sources, including, but not limited to, avalanche observations from seismic sensing and real time detections, remote sensing data, physical and numerical models and simulations, empirical laws and past avalanche danger level bulletins.

The proposed system which the PhD student will help developing will incorporate a human-in-the-loop approach, ensuring expert forecasters retain full control—such as dynamically correcting errors and retraining models.

The PhD student is expected to carry out independent research and propose creative, grounded, and well thought solutions to a range of questions involving modelling, processing, forecasting and processing of key parameters representing snow, climate and avalanche processes, by relying on state-of-the-art machine learning models.

Specifically:

• Design, develop and implement baselines machine learning models.
• Extend, design, develop and implement machine learning models which include knowledge about physics (e.g., physics constrained neural networks, scientific machine learning).
• Design, develop and implement multivariate time series forecasting models.
• Collaborate on dedicated and existing codebases with other project members and collaborators, and potentially members of the open source community.
• Collaborate with experts from different backgrounds (machine learning, physics, natural hazards, software engineering).
• Effectively present and communicate scientific…