Deep Learning for Hurricane Forecasting

This project aims to enable the creation of hurricane forecasting products that are useful to decision-makers using deep learning methodologies. These forecasting products can include track (position) forecasts, intensity forecasts, wind forecasts, and storm surge forecasts. The goal is to achieve a relatively similar or smaller error than the official National Hurricane Center (NHC) forecast error.

Published Work

Here are some previously published works in this area. Reading the background and methodology sections of these papers will give you more sources for other papers and datasets.

• Machine Learning in Tropical Cyclone Forecast Modeling: A Review
  • https://www.mdpi.com/2073-4433/11/7/676
  • Good review paper, start here
• Hurricane Forecasting: A Novel Multimodal Machine Learning Framework
  • https://arxiv.org/abs/2011.06125
  • Most recent major paper published on the topic, lots of reference to other previous paper in the background
  • Arguably has some flaws, but is promising in certain areas
  • Only 24-hour forecasts
• Fused Deep Learning for Hurricane Track Forecast from Reanalysis Data
  • https://hal.archives-ouvertes.fr/hal-01851001/
  • Example of how to combine track data and image data for forecasting
  • No comparison to official forecast error
• PHURIE: hurricane intensity estimation from infrared satellite imagery using machine learning
  • https://link.springer.com/article/10.1007/s00521-018-3874-6
  • Another cool example of ML and computer vision for forecasting
• Predicting Hurricane Trajectories Using a Recurrent Neural Network
  • https://arxiv.org/abs/1802.02548
  • Older paper with more basic work in this area
  • A combination of gridding the location data and using an RNN