Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Forecasting Tropical Cyclones with Cascaded Diffusion Models

Pritthijit Nath, Pancham Shukla, Shuai Wang, César Quilodrán-Casas

TL;DR

The paper tackles the challenge of forecasting tropical cyclones under climate-change pressures by using a cascaded diffusion approach that combines infrared satellite imagery and ERA5 reanalysis data. It deploys three independently trained U‑Net diffusion models for forecasting, super-resolution, and precipitation modelling, with outputs conditioned on ERA5 at the forecast time. The method achieves accurate one-step outputs (PSNR > 20 dB, SSIM > 0.5) and credible 36-hour rollouts, while operating efficiently on a single GPU. This approach offers a cost-effective, near-real-time forecasting alternative for resource-limited regions, with potential for extended horizons and richer atmospheric data in future work.

Abstract

As tropical cyclones become more intense due to climate change, the rise of Al-based modelling provides a more affordable and accessible approach compared to traditional methods based on mathematical models. This work leverages generative diffusion models to forecast cyclone trajectories and precipitation patterns by integrating satellite imaging, remote sensing, and atmospheric data. It employs a cascaded approach that incorporates three main tasks: forecasting, super-resolution, and precipitation modelling. The training dataset includes 51 cyclones from six major tropical cyclone basins from January 2019 - March 2023. Experiments demonstrate that the final forecasts from the cascaded models show accurate predictions up to a 36-hour rollout, with excellent Structural Similarity (SSIM) and Peak-Singal-To-Noise Ratio (PSNR) values exceeding 0.5 and 20 dB, respectively, for all three tasks. The 36-hour forecasts can be produced in as little as 30 mins on a single Nvidia A30/RTX 2080 Ti. This work also highlights the promising efficiency of Al methods such as diffusion models for high-performance needs in weather forecasting, such as tropical cyclone forecasting, while remaining computationally affordable, making them ideal for highly vulnerable regions with critical forecasting needs and financial limitations. Code accessible at https://github.com/nathzi1505/forecast-diffmodels.

Forecasting Tropical Cyclones with Cascaded Diffusion Models

TL;DR

The paper tackles the challenge of forecasting tropical cyclones under climate-change pressures by using a cascaded diffusion approach that combines infrared satellite imagery and ERA5 reanalysis data. It deploys three independently trained U‑Net diffusion models for forecasting, super-resolution, and precipitation modelling, with outputs conditioned on ERA5 at the forecast time. The method achieves accurate one-step outputs (PSNR > 20 dB, SSIM > 0.5) and credible 36-hour rollouts, while operating efficiently on a single GPU. This approach offers a cost-effective, near-real-time forecasting alternative for resource-limited regions, with potential for extended horizons and richer atmospheric data in future work.

Abstract

As tropical cyclones become more intense due to climate change, the rise of Al-based modelling provides a more affordable and accessible approach compared to traditional methods based on mathematical models. This work leverages generative diffusion models to forecast cyclone trajectories and precipitation patterns by integrating satellite imaging, remote sensing, and atmospheric data. It employs a cascaded approach that incorporates three main tasks: forecasting, super-resolution, and precipitation modelling. The training dataset includes 51 cyclones from six major tropical cyclone basins from January 2019 - March 2023. Experiments demonstrate that the final forecasts from the cascaded models show accurate predictions up to a 36-hour rollout, with excellent Structural Similarity (SSIM) and Peak-Singal-To-Noise Ratio (PSNR) values exceeding 0.5 and 20 dB, respectively, for all three tasks. The 36-hour forecasts can be produced in as little as 30 mins on a single Nvidia A30/RTX 2080 Ti. This work also highlights the promising efficiency of Al methods such as diffusion models for high-performance needs in weather forecasting, such as tropical cyclone forecasting, while remaining computationally affordable, making them ideal for highly vulnerable regions with critical forecasting needs and financial limitations. Code accessible at https://github.com/nathzi1505/forecast-diffmodels.
Paper Structure (13 sections, 3 figures, 2 tables)

This paper contains 13 sections, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Data
  3. Data Acquisition
  4. Data Processing
  5. Methodology
  6. Cascaded Structure
  7. Evaluation Strategies
  8. Results
  9. Conclusion
  10. Additional Data Description
  11. ERA5 Variables
  12. Additional Results
  13. Forecast SSIM

Figures (3)

  • Figure 1: Illustration of the cascaded arrangement involving three task-specific diffusion models
  • Figure 2: Forecast at 31h (left) and 38h (right) of Cyclone Mocha over the North Indian Ocean on 10th May 2023. The upper rows resemble the ground truth IR 10.8µ m satellite image and total precipitation while the lower rows show the forecast generated at that particular timestep.
  • Figure B.1: SSIM values over the entire cyclonic duration. The dashed lines indicate the hourly marks at which the minimum SSIM values are obtained for each cyclone.