|
|
Subject
The density of meteorological observations and the amount of novel data sources have been increasing steadily in recent years. Likewise, more computational power is available than ever before: not just for data processing and numerical weather prediction (NWP), but also for impact (e.g. flood and heatwave) modelling.
Combining these different data sources is an intricate problem. Data are usually integrated with NWP models using data assimilation methods, such as 4D-Var. However, these are computationally expensive and cannot handle very high-resolution data.
Therefore, the combination of NWP with short-term data-driven forecasts, or nowcasts (see other thesis topic), is required to fully use the available high-resolution data and provide a rapidly updating forecast. These combined, so-called seamless forecasts are crucial to create fast and reliable warnings, such as flash flood warnings.
Kind of work
In this thesis you will investigate how machine learning, in particular deep learning techniques based on residual learning, can be used to combine various data and model sources to generate a reliable seamless forecast.
You will implement existing deep learning architectures aimed at multimodal data fusion, and apply it to improve the seamless combination of the Belgian operational nowcasts and NWP forecasts.
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
- https://pysteps.readthedocs.io/
- Moraux, A., Dewitte, S., Cornelis, B., & Munteanu, A., 2021: A Deep Learning Multimodal Method for Precipitation Estimation. Remote Sensing, 13(16), [3278]. https://doi.org/10.3390/rs13163278
- 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 (e.g., Python)
Background in machine learning
Experience with machine learning frameworks (e.g., TensorFlow or PyTorch)
Strong interest in weather, climate or hydrology
|
|
