Student information

MSc thesis topic: Nowcasting the Dutch weather in ODTP

In a changing climate, rapid access to precise short-term weather forecasts is vital for proactive actions against sudden threats such as hail storms, heavy rainfall or strong winds. Getting warned even 15 minutes before the event is critical to preserve important services and human well-being. The main stake of weather prediction is traditional physics-based forecasting methods which excel in overall weather predictions but struggle with sudden extreme events due to computational constraints. The use of deep learning spatiotemporal techniques provides a faster method at the cost of explainability.

Notably, leveraging novel architectures like transformers has produced state-of-the-art results. Yet, the optimal application of transformer-based spatiotemporal models to the task of weather nowcasting remains largely unexplored. Writing your Master’s Thesis on weather nowcasting will allow you to engage in cutting edge technology on deep learning while pursuing relevant research questions for society in the context of climate change mitigation. There are several topics available to work on nowcasting and students are encouraged to engage and cooperate in their work on nowcasting (while pursuing well-defined and personal topics for their Thesis).

The Intergovernmental Panel on Climate Change (IPCC) forecasted an increase in both the frequency and severity of extreme weather occurrences (Seneviratne et al., 2021). These extreme weather events have shown pronounced adverse effects on economic growth, particularly evident in the manufacturing and service sectors, with repercussions extending throughout supply chains (Inoue, 2021; Kotz et al., 2022). Given the abrupt, highly variable, and localised nature of these events, there is a growing need for faster predictions with higher precision regarding the timing, location and intensities of these events (Gao et al., 2020). For the last century, meteorology researchers have devoted attention and effort toward developing Numerical Weather Prediction (NWP) models. These models aim at applying the laws of physics to the atmospheric state through partial differential equations, accounting for the dynamic, thermodynamic, radiative and chemical processes to predict the weather (Shuman, 1989).

Nowcasting is a subfield of weather forecasting which targets the prediction of the weather in the following 0-6 hours with a special focus on extreme abrupt events (Wang et al., 2017). Its primary objective is to support rainstorm warning systems, aiding stakeholders in public safety and infrastructure management to protect lives, facilities and assets (Gao et al., 2020). Therefore, nowcasting requires fast and precise predictions, for which, traditional ensemble methods seem too computationally expensive and time-consuming which lead to the use of new technologies. The ability of deep learning models to extract by themselves spatiotemporal features, their dependencies and weights allows these models to gain a good understanding of the evolution of the atmospheric processes from historical data. And since the amount of daily weather observations is massive, it makes deep learning methods very fitting for the weather forecasting problematic (Reichstein et al., 2019). However, it is important to note that for now it is impossible to know how the deep learning model understands the system and if what it learns from the data follows the real physical mechanisms at stake. This lack of explainability and physical constraints means that a fully data-driven deep learning model could lack the physical understanding of NWPs, but also lack the uncertainty measurement of ensemble methods (Schultz et al., 2021).

Despite this, with the rapid development of deep learning methods over the past years, some recent models have had a massive impact on the weather forecasting field (Bi et al., 2023; Lam et al., 2023). However, deep learning methods are still complementary to NWPs which have shown to still be more generalisable, versatile and offer richer results which are also invaluable for the training of deep learning models (Lam et al., 2023). There are two different approaches in using deep learning for weather forecasting. The first involves the integration of deep learning within the NWP workflow to accelerate some computationally intensive tasks like the data assimilation, the prediction or the post-processing. Conversely, the second approach aims at entirely replacing the NWP model by deep learning to directly generate a prediction from the observation data (Schultz et al., 2021).

Objectives and Research questions

Researching weather nowcasting offers a broad range of tasks. Master students interested in the following topics can approach me for further details:

  • Data wrangling of the Dutch meteorological data from the KMNI for deep learning

    • Choice of variables, format of data, 3D data representation, 2D data representation
  • Comparisons of spatio-temporal deep learning models

    • OpenSTL library
  • Loss-tuning for deep learning models
  • Prediction of different weather types / threats

    • Hail storms, precipitation, Strong winds, etc
  • Causal graph-based models for explainable artificial intelligence (Zhao et al., 2024)
  • Embedding in Digital Twin workflow

Requirements

  • Good python knowledge (required)
  • Machine learning and deep learning basics (required)
  • Digital Twin background (optional)
  • Weather & meteorology research interest (optional)

Expected reading list before starting the thesis research

  • Salguiero Dorado, A. (2024) Weather Radar Precipitation Nowcasting with Transformer-Based Spatiotemporal Predictive Learning. Master Thesis at GRS. 1196332, (GRS-80436) GIRS-2024-40 . Wageningen University.
  • Wang, Y., Coning, E., Harou, A., Jacobs, W., Joe, P., Nikitina, L., Roberts, R., Wang, J., Wilson, J.,Atencia, A., Bica, B., Brown, B., Goodmann, S., Kann, A., Li, P. W., Monterio, I., Schmid, F., Seed, A., & Sun, J. (2017, November). Guidelines for Nowcasting Techniques.
  • Reichstein, M., Camps-Valls, G., Stevens, B., Jung, M., Denzler, J., Carvalhais, N., & Prabhat. (2019). Deep learning and process understanding for data-driven Earth system science [Number: 7743 Publisher: Nature Publishing Group]. Nature, 566(7743), 195–204. https://doi.org/10.1038/s41586-019-0912-1

Recommend literature

  • Zhao, S., Prapas, I., Karasante, I., Xiong, Z., Papoutsis, I., Camps-Valls, G., & Zhu, X. X. (2024). Causal Graph Neural Networks for Wildfire Danger Prediction. arXiv preprint arXiv:2403.08414.
  • Seneviratne, S. I., Zhang, X., Adnan, M., Badi, W., Dereczynski, C., Di Luca, A., Ghosh, S., Iskander,I., Kossin, J., Lewis, S., Otto, F., Pinto, I., Satoh, M., Vicente-Serrano, S. M., Wehner, M., & Zhou, B. (2021). Weather and Climate Extreme Events in a Changing Climate (Chapter 11). In V. Masson-Delmotte, P. Zhai, A. Pirani, S. L. Connors, C. P.an, S. Berger, N. Caud, Y. Chen, L. Goldfarb, M. I. Gomis, M. Huang, K. Leitzell, E. Lonnoy, J. B. R. Matthews,T. K. Maycock, T. Waterfield, K. Yelek.i, R. Yu, & B. Zhu (Eds.). Cambridge University Press. Retrieved February 26, 2024, from https://www.ipcc.ch/report/ar6/wg1/
  • Inoue, H. (2021, June). The Economic Impact of Heavy Rains on Supply Chains. https://doi.org/10.2139/ssrn.3875196
  • Kotz, M., Levermann, A., & Wenz, L. (2022). The effect of rainfall changes on economic production [Number: 7892 Publisher: Nature Publishing Group]. Nature, 601(7892), 223–227. https://doi.org/10.1038/s41586-021-04283-8
  • Gao, Z., Shi, X., Wang, H., Yeung, D.-Y., Woo, W.-c., & Wong, W.-K. (2020). Deep learning and the weather forecasting problem: Precipitation nowcasting. Deep Learning for the Earth Sciences. https://www.amazon.science/publications/deep-learning-and-the-weather-forecasting-problem-precipitation-nowcasting
  • Shuman, F. G. (1989). History of Numerical Weather Prediction at the National Meteorological Center [Publisher: American Meteorological Society Section: Weather and Forecasting]. Weather and Forecasting, 4(3), 286–296. https://doi.org/10.1175/1520-0434(1989)004<0286:HONWPA>2.0.CO;2
  • Schultz, M. G., Betancourt, C., Gong, B., Kleinert, F., Langguth, M., Leufen, L. H., Mozaffari, A., & Stadtler, S. (2021). Can deep learning beat numerical weather prediction? [Publisher: Royal Society]. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 379(2194), 20200097. https://doi.org/10.1098/rsta.2020.0097
  • Bi, K., Xie, L., Zhang, H., Chen, X., Gu, X., & Tian, Q. (2023). Accurate medium-range global weather forecasting with 3D neural networks [Number: 7970 Publisher: Nature Publishing Group]. Nature, 619(7970), 533–538. https://doi.org/10.1038/s41586-023-06185-3
  • Lam, R., Sanchez-Gonzalez, A., Willson, M., Wirnsberger, P., Fortunato, M., Alet, F., Ravuri, S., Ewalds, T., Eaton-Rosen, Z., Hu, W., Merose, A., Hoyer, S., Holland, G., Vinyals, O., Stott, J., Pritzel, A., Mohamed, S., & Battaglia, P. (2023). Learning skillful medium-range global weather forecasting [Publisher: American Association for the Advancement of Science]. Science, 382(6677), 1416–1421. https://doi.org/10.1126/science.adi2336

Theme(s): Modelling & visualisation