Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Simulating Nighttime Visible Satellite Imagery of Tropical Cyclones Using Conditional Generative Adversarial Networks

Jinghuai Yao, Puyuan Du, Yucheng Zhao, Yubo Wang

TL;DR

The study tackles the absence of nighttime visible satellite imagery for Tropical Cyclones by developing a CGAN-based method that translates daytime multispectral infrared inputs into nighttime VIS reflectance. By replacing L1 with SSIM loss, selecting physically informative IR bands, and incorporating sun/satellite direction channels, the model achieves state-of-the-art daytime performance (SSIM ≈ 0.92, RMSE ≈ 0.030) and robust cross-satellite nighttime validation against VIIRS DNB, while enabling controllable virtual sunlight directions. The approach enables high-resolution, near-real-time nighttime TC monitoring and supports applications in LLCC positioning and eye-formation detection, with potential extensions to diffusion-based models and broader meteorological metrics. Together, these contributions significantly enhance nighttime observational capabilities and operational TC analysis using AI-generated VIS imagery.

Abstract

Visible (VIS) imagery is important for monitoring Tropical Cyclones (TCs) but is unavailable at night. This study presents a Conditional Generative Adversarial Networks (CGAN) model to generate nighttime VIS imagery with significantly enhanced accuracy and spatial resolution. Our method offers three key improvements compared to existing models. First, we replaced the L1 loss in the pix2pix framework with the Structural Similarity Index Measure (SSIM) loss, which significantly reduced image blurriness. Second, we selected multispectral infrared (IR) bands as input based on a thorough examination of their spectral properties, providing essential physical information for accurate simulation. Third, we incorporated the direction parameters of the sun and the satellite, which addressed the dependence of VIS images on sunlight directions and enabled a much larger training set from continuous daytime data. The model was trained and validated using data from the Advanced Himawari Imager (AHI) in the daytime, achieving statistical results of SSIM = 0.923 and Root Mean Square Error (RMSE) = 0.0299, which significantly surpasses existing models. We also performed a cross-satellite nighttime model validation using the Day/Night Band (DNB) of the Visible/Infrared Imager Radiometer Suite (VIIRS), which yields outstanding results compared to existing models. Our model is operationally applied to generate accurate VIS imagery with arbitrary virtual sunlight directions, significantly contributing to the nighttime monitoring of various meteorological phenomena.

Simulating Nighttime Visible Satellite Imagery of Tropical Cyclones Using Conditional Generative Adversarial Networks

TL;DR

The study tackles the absence of nighttime visible satellite imagery for Tropical Cyclones by developing a CGAN-based method that translates daytime multispectral infrared inputs into nighttime VIS reflectance. By replacing L1 with SSIM loss, selecting physically informative IR bands, and incorporating sun/satellite direction channels, the model achieves state-of-the-art daytime performance (SSIM ≈ 0.92, RMSE ≈ 0.030) and robust cross-satellite nighttime validation against VIIRS DNB, while enabling controllable virtual sunlight directions. The approach enables high-resolution, near-real-time nighttime TC monitoring and supports applications in LLCC positioning and eye-formation detection, with potential extensions to diffusion-based models and broader meteorological metrics. Together, these contributions significantly enhance nighttime observational capabilities and operational TC analysis using AI-generated VIS imagery.

Abstract

Visible (VIS) imagery is important for monitoring Tropical Cyclones (TCs) but is unavailable at night. This study presents a Conditional Generative Adversarial Networks (CGAN) model to generate nighttime VIS imagery with significantly enhanced accuracy and spatial resolution. Our method offers three key improvements compared to existing models. First, we replaced the L1 loss in the pix2pix framework with the Structural Similarity Index Measure (SSIM) loss, which significantly reduced image blurriness. Second, we selected multispectral infrared (IR) bands as input based on a thorough examination of their spectral properties, providing essential physical information for accurate simulation. Third, we incorporated the direction parameters of the sun and the satellite, which addressed the dependence of VIS images on sunlight directions and enabled a much larger training set from continuous daytime data. The model was trained and validated using data from the Advanced Himawari Imager (AHI) in the daytime, achieving statistical results of SSIM = 0.923 and Root Mean Square Error (RMSE) = 0.0299, which significantly surpasses existing models. We also performed a cross-satellite nighttime model validation using the Day/Night Band (DNB) of the Visible/Infrared Imager Radiometer Suite (VIIRS), which yields outstanding results compared to existing models. Our model is operationally applied to generate accurate VIS imagery with arbitrary virtual sunlight directions, significantly contributing to the nighttime monitoring of various meteorological phenomena.
Paper Structure (15 sections, 16 equations, 16 figures, 6 tables)

This paper contains 15 sections, 16 equations, 16 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Data
  3. Satellite data
  4. Band selection
  5. Preprocessing of model input
  6. Methods
  7. The revised pix2pix model
  8. Training set and validation set
  9. Model validation and statistical comparison
  10. Results
  11. Daytime validation performance
  12. Nighttime validation performance
  13. Applications
  14. Discussion
  15. Summary and conclusions

Figures (16)

  • Figure 1: (a) AHI Band03 imagery, (b) difference between normalized VIS and IR imagery, and (c) 2D histogram of Typhoon Saola (2023) in the South China Sea at 03:00 UTC on September 1. The cloud types in (c) are denoted as Cb (Cumulonimbus), Ci (Cirrus), Cu (Cumulus), Sc (Stratocumulus), and St (Stratus).
  • Figure 2: Weighting functions of AHI Band08--10 and Band16 under Standard Tropical Atmosphere and a satellite zenith angle of 0°.
  • Figure 3: The imaginary index of refraction of water droplets and ice particles and the wavelength range of AHI Band11--16. Source: Meteorological Satellite Center of Japan Meteorological Agency.
  • Figure 4: The preprocessed IR data, direction parameters, basemap, and VIS reflectance of Typhoon Chan-hom (2015) at 05:00 UTC on July 10.
  • Figure 5: Detailed architecture of the revised pix2pix model. Note that the VIS images are replicated to 3 channels, enabling future framework generalization for multi-color imagery.
  • ...and 11 more figures