EGU General Assembly 2023
© Author(s) 2023. This work is distributed under
the Creative Commons Attribution 4.0 License.

Multivariate forecasting of tropical cyclones using combined neural networks.

Yegor Hudozhnik1 and Andreas Windisch2,3,4
Yegor Hudozhnik and Andreas Windisch
  • 1FH Joanneum, Applied Computer Science, Data Science and Artificial Intelligence, Austria (
  • 2Know-Center GmbH, Inffeldgasse 13, 8010 Graz
  • 3Graz University of Technology, Institute of Interactive Systems and Data Science, Inffeldgasse 13, 8010 Graz
  • 4Washington University in St. Louis, Physics Department, One Brookings Drive, St. Louis, 63130 MO, USA

Tropical Cyclones (TCs) are extremely dangerous and destructive events which pose a danger to human lives every year. Conventional TC forecasting methods are computationally intensive and require a relatively large amount of energy and time.

In the light of climate change due to the process of global warming, the behavior of TCs may change, and therefore require the use of modern, more flexible learning methods for estimation and forecasting.

In recent years, the study of the application of Deep Learning (DL) in this area proved to be highly effective. These methods are designed to facilitate the prediction process, as well as automatically detect possible trends that may occur over time.

In this work, an application of neural networks such as LSTMs and GRUs is investigated to forecast tracks and classify the evolution of TC systems using satellite image data series as an input, where historical track data and the satellite image data are used to train the network. Particular attention is paid to adaptivity of DL approaches to recent trends and edge cases.

How to cite: Hudozhnik, Y. and Windisch, A.: Multivariate forecasting of tropical cyclones using combined neural networks., EGU General Assembly 2023, Vienna, Austria, 24–28 Apr 2023, EGU23-14848,, 2023.

Supplementary materials

Supplementary material file