Neural networks improve tropical cyclone wind and intensity nowcasting

Nowcasting of Tropical Cyclone Wind Fields and Intensity from Geostationary Imagery

Machine Learning

Summary

Measuring wind speeds during tropical cyclones is hard because radar data is limited and slow to get. Geostationary satellites see storm clouds constantly but can't measure surface winds directly. The authors created simple neural networks that predict detailed wind patterns and storm strength using satellite cloud images. Their method works better when it predicts wind fields and storm intensity together, especially for storms that quickly get stronger.

What this means in practice

  • For meteorological agencies: Enhance real-time estimation of cyclone wind fields and intensity using geostationary satellite data without relying on sparse radar observations.
  • For emergency management teams: Improve rapid assessment of storm strength and wind distribution to support early warnings and disaster response planning.

Authors

Simon Donike, Pritthijit Nath, Cristina Radin, Arthur Avenas, Ruben Cartuyvels, Emiliano Diaz, William K. Jones, Anna Jungbluth

Abstract

High-resolution tropical cyclone surface wind speeds derived from synthetic aperture radar (SAR) observations can support storm structure and intensity modelling, but remain spatially and temporally sparse and unavailable near-real-time. Geostationary satellites capture cloud tops continuously in space and time, but do not directly resolve surface winds. Here, we present simple neural networks to predict 2D wind fields and storm-scale intensity metrics from geostationary imagery. Through a series of ablation studies, we find that joint prediction of 2D fields and scalar metrics from the model latent space improves the retrieval of the (radius of) maximum wind, especially for rapid intensification events.