Funded Masters Internships at UNIL’s Expertise Center for Climate Extremes

Scroll Down

Applications should be sent to tom.beucler@unil.ch and will be accepted on a rolling basis after December 1st, 2025 (see below for guidelines on how to apply). The preferred start date is anytime in 2025, with a start in June 2025 at the latest to ensure the internship concludes by the end of the academic year.

Internship 1: Towards Physics-Informed Neural Operators for Climate Modeling

Collaborators: Julien Le Sommer (UGA)

Neural networks are increasingly used to improve climate simulations due to their ability to approximate complex nonlinear operators and their low inference costs (Eyring et al., 2024). However, these models often struggle to generalize in scenarios far beyond their training conditions (Beucler et al., 2024). Neural operators, which can learn mesh-free mappings between spatial fields, offer improved generalization across different spatial resolutions (Li et al., 2020). Nonetheless, it remains uncertain whether they address persistent challenges in machine learning for climate modeling, such as ensuring the long-term temporal stability of dynamical systems and generalizing across a wide range of environmental conditions, including domain shifts induced by climate change.

To tackle these challenges, we propose accelerating the development of Physics-Informed Neural Operators (PINO, Li et al., 2024). Akin to physics-informed neural networks (Daw et al., 2022), PINOs integrate domain knowledge into the loss function to enhance physical consistency. Unlike previous efforts focused on general-purpose partial differential equations with soft constraints (Goswami et al., 2023), our approach will specifically incorporate knowledge of climate processes to improve the robustness of neural operators in atmospheric and oceanic applications. 

To achieve this goal, this internship will focus on atmospheric column physics, for which prototype neural networks have already been developed (Hu et al., 2024), along with software that enables their testing in realistic climate models once trained (Yu et al., 2024). Given its synergies with work at the University of Grenoble Alpes, this internship may also include participation in Grenoble’s Artificial Intelligence for Physical Sciences summer workshop, depending on timing.

Some references:

• Beucler, T., et al. “Climate-invariant machine learning.” Science Advances 10.6 (2024): eadj7250.

• Daw, A., et al. “Physics-guided neural networks (pgnn): An application in lake temperature modeling.” Knowledge Guided Machine Learning. Chapman and Hall/CRC (2022): 353-372.

• Eyring, V., et al. “Pushing the frontiers in climate modelling and analysis with machine learning.” Nature Climate Change (2024): 1-13.

• Hu, Z., et al. “Stable Machine-Learning Parameterization of Subgrid Processes with Real Geography and Full-physics Emulation.” arXiv preprint arXiv:2407.00124 (2024).

• Goswami, S., et al. “Physics-informed deep neural operator networks.” Machine Learning in Modeling and Simulation: Methods and Applications. Springer International Publishing (2023): 219-254.

• Li, Z., et al. “Fourier neural operator for parametric partial differential equations.” arXiv preprint arXiv:2010.08895 (2020).

• Li, Z., et al. “Physics-informed neural operator for learning partial differential equations.” ACM/JMS Journal of Data Science 1.3 (2024): 1-27.

• Yu, S., et al. “ClimSim: A large multi-scale dataset for hybrid physics-ML climate emulation.” Advances in Neural Information Processing Systems 36 (2024).

Internship 2: Video Super-Resolution for Precipitation Downscaling

Collaborators: Mathieu Vrac (LSCE), Stephan Mandt (UCI), Daniele Nerini (Meteo Swiss), Filippo Quarenghi (UNIL)

Empirical downscaling increases the resolution of physical variables through data-driven approaches, enabling decision-making at local scales that are crucial for early-warning systems and climate adaptation (Rampal et al., 2024). Machine learning (ML)-based image super-resolution, which generates high-resolution images from low-resolution counterparts, has gained popularity due to its accuracy and low computational cost after training (Yang et al., 2019). Super-resolution algorithms improve the spatial resolution of precipitation forecasts and climate projections through various strategies, ranging from simple ML approaches (Vandal et al., 2019) to deep learning-based generative models, including diffusion probabilistic models (Mardani et al., 2023) and unpaired methods where models are trained solely on high-resolution target data and conditioned during inference (Hess et al., 2024; Bischoff et al., 2024). However, spatial super-resolution alone may be insufficient when temporal resolution is limited to six-hourly or daily intervals. Such limitations, often due to data storage constraints or model design choices for long-term projection, can distort the representation of precipitation extremes.

This internship hence proposes deploying video super-resolution algorithms (Liu et al., 2022) to improve both the spatial and temporal resolutions of precipitation fields. We will leverage high-fidelity products that combine radar data and rain gauge observations over the Swiss and French domains (Sideris et al., 2014; Champeaux et al., 2009; Laurantin et al., 2012). We will evaluate current video super-resolution frameworks previously applied to downscale precipitation in both space and time, using meteorological reanalyses (Leinonen et al., 2024), radar data (Glawion et al., 2023; Scher et al., 2021) and storm-resolving simulations (Srivastava et al., 2023). Key topics of interest include adapting object-based evaluation metrics to account for temporal consistency, assessing the usability of video outputs in generative contexts where information overload is a risk, transfer learning, and exploring the added value of spatiotemporal super-resolution versus spatial super-resolution alone for capturing precipitation extremes relevant to climate adaptation.

Some references:

Your Responsibilities

  • Conduct independent, innovative, and responsible research
  • Disseminate results through a publicly-available report at the end of the project, which may be your Masters thesis
  • Ensure clear communication and productive interactions with project collaborators through regular in-person or Zoom meetings
  • Openly discuss research obstacles and seek advice/help in the lab 
  • Make your research/code open and reproducible (e.g. through Github/Gitlab/arXiv)
  • Informally share research progress at the weekly ∂3AWN lab meeting
  • Actively participate in creating a collaborative and inclusive culture in the ∂3AWN lab and ECCE
  • Actively participate in the various ECCE events and activities (e.g., research seminars, informal seminars for early-career scientists, social events)
  • (If within the internship’s scope) Actively develop the modeling capabilities or the new products of the Expertise Center for Climate Extremes at UNIL

Your Qualifications

  • Bachelor degree in atmospheric/climate science, computer science or electrical engineering, physics, fluid mechanics or mechanical engineering, statistics, applied mathematics, or a closely related discipline
  • Strong background in physics and applied mathematics (preferentially including calculus, differential equations, statistics, mechanics, and thermodynamics) 
  • Experience reading, writing, and manipulating scientific datasets
  • Excellent skills in written and spoken English
  • Enthusiasm for atmospheric/climate science, scientific machine learning, and statistics
  • (Preferred but not required) Background in environmental science
  • (Preferred but not required) Experience in high-performance computing

Note that proficiency in French is not required and free French courses are available at UNIL during the academic semester.

Position perks

  • Monthly salary of CHF 1’774.50 for the duration of the internship (4-6 months)
  • Fully-funded personal research equipment (desktop computer, monitor, etc.)
  • Fully-funded open-access publication costs
  • Access to UNIL’s high-computing facilities (Millions of CPU/GPU hours, 1PB of storage, etc.), additional CPU hours available for computationally-intensive atmospheric simulations through collaborations
  • International collaboration network
  • A friendly and cohesive culture at the Institute of Earth Surface Dynamics including 1-day Summer and Winter retreats, usually in the Alps
  • A very interdisciplinary environment provided by the Expertise Center for Climate Extremes, including social events
  • Access to UNIL Campus facilities (Sports center with 100+ recreational options, campus-grown food, etc.)
  • (If applicable) Fully-funded research-related travel (conferences, collaborations, etc.)

Application Documents

  • Curriculum Vitae
  • Copy of all university degree certificates and transcripts
  • Name, affiliation, and email address of 1 reference (e.g. Bachelor thesis advisor(s), previous teachers, previous employers)
  • Copy of one first-author research report (can be a manuscript, a thesis, or a class paper) 
  • Personal Statement (no more than 1 page excluding bibliography), including:

– A ~200-word description of which project(s) you are interested in and your strategies to tackle it/them within a 4-6 months timeframe

– Your professional goals, and how this position might further them

– Your strategies to collaborate with colleagues of diverse backgrounds and experience

Applications should be sent to tom.beucler@unil.ch and are accepted on a rolling basis. The preferred start date is anytime in 2025, with a start in June 2025 at the latest to ensure the internship concludes by the end of the academic year.

What is ECCE?

The recently established Expertise Center for Climate Extremes (ECCE) fosters collaboration between the Faculty of Business and Economics (HEC) and the Faculty of Geosciences and Environment (FGSE). Its focus is on researching the prediction of climate extremes and their impacts, along with assessing risks associated with these extremes in current and future climates. “Climate extremes” encompass weather extremes, climate anomalies, water extremes, and weather-related natural hazards. ECCE, grounded in interdisciplinarity, aims to leverage the faculties’ expertise in meteorology, climatology, natural processes, statistics (especially extreme-value theory), machine learning, insurance risk, and finance to create synergies. For more details, visit https://www.unil.ch/ecce/home.html. These internships will be hosted at the Data-Driven Atmospheric & Water Dynamics (∂3AWN) lab (https://wp.unil.ch/dawn/) of the FGSE, one of the research groups affiliated with ECCE.

Additional Information

  • The University of Lausanne is committed to equal opportunity and stands firm against all forms of discrimination.
  • The Faculty of Geosciences and Environment of the University of Lausanne adheres to the DORA agreement and follows its guidelines in the evaluation of applications (in short, quality over quantity)