Subject

Extreme precipitation is one of the most important natural hazards, especially in an urbanized country such as Belgium, where extensive soil sealing increases the risk of flash floods. One of the challenges is to generate reliable short-term forecasts for such extreme events, so that appropriate measures can be taken by emergency services and local authorities, and e.g. evacuations can be organized. Short-term data-driven weather forecasts or nowcasts aim to address this need.

In the past few years, there have been some very successful applications of deep learning (DL) in the field of precipitation modelling. The use of techniques such as Generative Adversarial Networks (GANs) was proven to be highly effective in this field (Leinonen et al., 2021 Ravuri et al., 2021).

Kind of work

You will implement a deep learning-based nowcasting method such as the architecture by Ravuri et al. in the pySTEPS nowcasting library, and compare it to the operational Belgian nowcasts. You will analyse extreme rainfall events to evaluate how DL-based models perform for events far outside the training data set. You investigate how to improve this architecture to make it more robust for such events.

With your research, you can contribute to the open-source pySTEPS nowcasting software, which has a friendly and international community of developers and researchers. The work will happen in collaboration with the Royal Meteorological Institute, which is responsible for the Belgian operational weather forecasts.

Framework of the Thesis

• Ravuri, S., Lenc, K., Willson, M. et al. Skilful precipitation nowcasting using deep generative models of radar. Nature 597, 672–677 (2021). https://doi.org/10.1038/s41586-021-03854-z
  
• J. Leinonen, D. Nerini and A. Berne. Stochastic Super-Resolution for Downscaling Time-Evolving Atmospheric Fields With a Generative Adversarial Network. IEEE Transactions on Geoscience and Remote Sensing, 59, 7211–7223, 2021. doi:10.1109/TGRS.2020.3032790, preprint available at ArXiv ·
• https://pysteps.readthedocs.io/
  
• Pulkkinen, S., D. Nerini, A. Perez Hortal, C. Velasco-Forero, U. Germann, A. Seed, and L. Foresti, 2019: Pysteps: an open-source Python library for probabilistic precipitation nowcasting (v1.0). Geosci. Model Dev., 12 (10), 4185–4219, doi:10.5194/gmd-12-4185-2019.
  

Expected Student Profile

Programming background (Python)

Background in machine learning

Experience with machine learning frameworks (e.g., TensorFlow or PyTorch)

Strong interest in weather, climate or hydrology