Subject

Extreme climatic events, such as the devastating floods of 2021 or the heat waves in the past years, cause enormous societal and economical damage. Numerical weather prediction (NWP) models, based on the physical equations of the atmosphere, allow us to forecast such events. However, these forecasts inevitably have errors, due to inaccurate initial conditions as well as model errors such as parametrization, representativity, and numerical truncation.

Post-processing techniques such as Model Output Statistics (MOS) aim to correct systematic forecast errors. These statistical techniques are gradually being outperformed by more advanced machine learning techniques, such as deep learning (Rasp et al., 2018) and recently, Generative adversarial networks. However, the training data for extreme events such as the one from 2021 is scarce, which makes it challenging to use ML approaches.

Kind of work

In this Master thesis, you will investigate machine learning methods and deep learning methods, in particular to better anticipate extreme events. You will perform a literature study of ML-based post-processing approaches and implement such an approach for an operational weather model used at the Royal Meteorological Institute of Belgium and the European Centre for Medium-range Weather Forecasts. You will compare the performance to state-of-the-art and operational approaches with particular attention for extreme events. You will analyse the effects of the small sample size inherent to extreme events and discuss potential mitigating strategies (e.g. few-shot learning, generating synthetic data, augmenting data sets, semi-supervised learning). With this highly societally relevant topic, you may contribute to an improved forecasting chain and help protect society against climatic extremes.

Framework of the Thesis

• Vannitsem, S., Bremnes, J. B., Demaeyer, J., Evans, G. R., Flowerdew, J., Hemri, S., ... & Ylhaisi, J., 2021: Statistical postprocessing for weather forecasts: Review, challenges, and avenues in a big data world. Bulletin of the American Meteorological Society, 102(3), E681-E699. https://doi.org/10.1175/BAMS-D-19-0308.1
  
• Rasp, S., & Lerch, S., 2018: Neural networks for postprocessing ensemble weather forecasts. Monthly Weather Review, 146(11), 3885-3900. https://doi.org/10.1175/MWR-D-18-0187.1
  

Expected Student Profile

Programming background (e.g., Python)

Background in machine learning

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

Strong interest in weather and climate