Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Estimating Atmospheric Variables from Digital Typhoon Satellite Images via Conditional Denoising Diffusion Models

Zhangyue Ling, Pritthijit Nath, César Quilodrán-Casas

TL;DR

This study explores the application of diffusion models in the field of typhoons, predicting multiple ERA5 meteorological variables simultaneously from Digital Typhoon satellite images to enable more robust and detailed forecasting, reducing the impact of severe weather events on vulnerable regions.

Abstract

This study explores the application of diffusion models in the field of typhoons, predicting multiple ERA5 meteorological variables simultaneously from Digital Typhoon satellite images. The focus of this study is taken to be Taiwan, an area very vulnerable to typhoons. By comparing the performance of Conditional Denoising Diffusion Probability Model (CDDPM) with Convolutional Neural Networks (CNN) and Squeeze-and-Excitation Networks (SENet), results suggest that the CDDPM performs best in generating accurate and realistic meteorological data. Specifically, CDDPM achieved a PSNR of 32.807, which is approximately 7.9% higher than CNN and 5.5% higher than SENet. Furthermore, CDDPM recorded an RMSE of 0.032, showing a 11.1% improvement over CNN and 8.6% improvement over SENet. A key application of this research can be for imputation purposes in missing meteorological datasets and generate additional high-quality meteorological data using satellite images. It is hoped that the results of this analysis will enable more robust and detailed forecasting, reducing the impact of severe weather events on vulnerable regions. Code accessible at https://github.com/TammyLing/Typhoon-forecasting.

Estimating Atmospheric Variables from Digital Typhoon Satellite Images via Conditional Denoising Diffusion Models

TL;DR

This study explores the application of diffusion models in the field of typhoons, predicting multiple ERA5 meteorological variables simultaneously from Digital Typhoon satellite images to enable more robust and detailed forecasting, reducing the impact of severe weather events on vulnerable regions.

Abstract

This study explores the application of diffusion models in the field of typhoons, predicting multiple ERA5 meteorological variables simultaneously from Digital Typhoon satellite images. The focus of this study is taken to be Taiwan, an area very vulnerable to typhoons. By comparing the performance of Conditional Denoising Diffusion Probability Model (CDDPM) with Convolutional Neural Networks (CNN) and Squeeze-and-Excitation Networks (SENet), results suggest that the CDDPM performs best in generating accurate and realistic meteorological data. Specifically, CDDPM achieved a PSNR of 32.807, which is approximately 7.9% higher than CNN and 5.5% higher than SENet. Furthermore, CDDPM recorded an RMSE of 0.032, showing a 11.1% improvement over CNN and 8.6% improvement over SENet. A key application of this research can be for imputation purposes in missing meteorological datasets and generate additional high-quality meteorological data using satellite images. It is hoped that the results of this analysis will enable more robust and detailed forecasting, reducing the impact of severe weather events on vulnerable regions. Code accessible at https://github.com/TammyLing/Typhoon-forecasting.
Paper Structure (17 sections, 5 figures, 2 tables)

This paper contains 17 sections, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Data
  3. Datasets
  4. Data Processing
  5. Methodology
  6. Training
  7. Inference
  8. Results
  9. Magnitude Prediction
  10. Performance Metrics
  11. Conclusion
  12. Methodology of other models
  13. CNN
  14. SENet
  15. Additional Results
  16. ...and 2 more sections

Figures (5)

  • Figure 1: CDDPM workflow: $\mathbf{y}_0$ represents ERA5, and $\mathbf{x}$ represents the DT satellite image. In forward diffusion, noise is added iteratively to $\mathbf{y}_0$. In reverse diffusion, the model denoises from $\mathbf{y}_T$ back to $\mathbf{y}_0$, conditioned on $\mathbf{x}$.
  • Figure 2: An example prediction of Typhoon Muifa, from September 9th 2022, showing the forecasting magnitude derived from predicted u10 and v10 components.
  • Figure A.1: Structure of Conditional Convolution Neural Networks
  • Figure B.2: Example results comparing the predictions of four different models: CNN, SENet, DDPM, and CDDPM. The first column shows the input image (img_64) used to generate predictions. The second column displays the true values of the u10, v10, sp, and t2m variables. While the subsequent columns present the corresponding predictions from each model.
  • Figure B.3: Difference maps for predictions of u10 (u-component wind), v10 (v-component wind), sp (surface pressure), and t2m (temperature at 2 meters) variables using four models: CNN, SENet, DDPM, and CDDPM. The first column represents the input image used for predictions. The second column shows the true values of each variable. The subsequent columns display the difference between the true values and the predicted values from each model. Red regions indicate areas of higher discrepancy between the true and predicted values, highlighting where each model deviates most from the ground truth.