Deep Learning for Hurricane Track Forecasting from Aligned Spatio-temporal Climate Datasets

Sophie Giffard-Roisin, Mo Yang, Guillaume Charpiat, Balázs Kégl, Claire Monteleoni

Keywords: hurricane forecasting, deep learning, convolutional network, fusion networks, moving reference

TL;DR: We propose a neural network for the storm track 24h-forecasting using a moving frame of reference able to use a common dataset and a common training for every hurricane of both hemispheres.

Abstract:

The forecast of hurricane trajectories is crucial for the protection of people and property, but machine learning techniques have been scarce for this so far.
We propose a neural network fusing past trajectory data and reanalysis atmospheric images (wind and pressure 3D fields). We used a moving frame of reference that follows the storm center for the 24h tracking forecast. The network is trained to estimate the longitude and latitude displacement of hurricanes and depressions from a large database from both hemispheres (more than 3000 storms since 1979, sampled at a 6 hour frequency). The advantage of the fusion network is demonstrated and a comparison with current forecast models shows that deep methods could provide a valuable and complementary prediction.