Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

High-resolution ensemble retrieval of cloud properties for all-day based on geostationary satellite

Haixia Xiao, Feng Zhang, Lingxiao Wang, Baoxiang Pan, Yannian Zhu, Minghuai Wang, Wenwen Li, Bin Guo, Jun Li

TL;DR

CloudDiff introduces a conditional diffusion-model framework for all-day, high-resolution retrieval of cloud properties from geostationary infrared data, enabling ensemble sampling to quantify uncertainty. Trained on Himawari-8/AHI 2 km TIR data and MODIS targets, CloudDiff delivers 1 km cloud-property maps for $COT$, $CER$, $CTH$, and $CLP$, with 30-sample ensembles that improve accuracy and reliability over deterministic baselines. Evaluations against MODIS and CALIPSO show strong performance in CLP classification and cloud-top properties, and a Typhoon In-Fa case demonstrates the method’s capacity to capture both sharp local structure and uncertainty in extreme events. The study discusses limitations due to MODIS biases, suggests physics-informed constraints and transfer learning, and outlines extensions to other geostationary satellites for broader applicability in cloud remote sensing and severe-weather analysis.

Abstract

Clouds play a critical role in Earth's hydrological and energy cycles, and accurately representing their properties is essential for effective numerical modeling and weather forecasting. Machine learning methods have been widely used for cloud property retrieval; however, most existing techniques are deterministic and do not incorporate uncertainty quantification. Generative machine learning has made significant advances in various domains, including natural language processing, image generation, and notably weather forecasting, where it has enabled ensemble predictions and the quantification of forecast uncertainty. This ability to quantify uncertainty offers valuable opportunities for cloud remote sensing. In this study, we propose a novel cloud property retrieval method, CloudDiff, based on a generative diffusion model. By leveraging thermal infrared observations from the Himawari-8 Advanced Himawari Imager (AHI), CloudDiff generates high spatiotemporal resolution cloud properties for both daytime and nighttime conditions, increasing the resolution of Himawari-8/AHI cloud retrievals from 2 km to 1 km. Unlike deterministic retrieval methods, CloudDiff generates multiple samples from the underlying probability distribution, allowing for a diverse range of plausible retrievals and taking steps towards providing uncertainty assessment. Additionally, CloudDiff produces sharper samples and better captures fine local features, enhancing the precision of cloud property retrieval. By averaging over the ensemble of generated samples, we demonstrate that both the accuracy and reliability of the retrievals are significantly improved. These high-resolution cloud properties have been successfully applied to analyze extreme weather events, such as typhoons, providing potentially valuable insights into atmospheric processes.

High-resolution ensemble retrieval of cloud properties for all-day based on geostationary satellite

TL;DR

CloudDiff introduces a conditional diffusion-model framework for all-day, high-resolution retrieval of cloud properties from geostationary infrared data, enabling ensemble sampling to quantify uncertainty. Trained on Himawari-8/AHI 2 km TIR data and MODIS targets, CloudDiff delivers 1 km cloud-property maps for , , , and , with 30-sample ensembles that improve accuracy and reliability over deterministic baselines. Evaluations against MODIS and CALIPSO show strong performance in CLP classification and cloud-top properties, and a Typhoon In-Fa case demonstrates the method’s capacity to capture both sharp local structure and uncertainty in extreme events. The study discusses limitations due to MODIS biases, suggests physics-informed constraints and transfer learning, and outlines extensions to other geostationary satellites for broader applicability in cloud remote sensing and severe-weather analysis.

Abstract

Clouds play a critical role in Earth's hydrological and energy cycles, and accurately representing their properties is essential for effective numerical modeling and weather forecasting. Machine learning methods have been widely used for cloud property retrieval; however, most existing techniques are deterministic and do not incorporate uncertainty quantification. Generative machine learning has made significant advances in various domains, including natural language processing, image generation, and notably weather forecasting, where it has enabled ensemble predictions and the quantification of forecast uncertainty. This ability to quantify uncertainty offers valuable opportunities for cloud remote sensing. In this study, we propose a novel cloud property retrieval method, CloudDiff, based on a generative diffusion model. By leveraging thermal infrared observations from the Himawari-8 Advanced Himawari Imager (AHI), CloudDiff generates high spatiotemporal resolution cloud properties for both daytime and nighttime conditions, increasing the resolution of Himawari-8/AHI cloud retrievals from 2 km to 1 km. Unlike deterministic retrieval methods, CloudDiff generates multiple samples from the underlying probability distribution, allowing for a diverse range of plausible retrievals and taking steps towards providing uncertainty assessment. Additionally, CloudDiff produces sharper samples and better captures fine local features, enhancing the precision of cloud property retrieval. By averaging over the ensemble of generated samples, we demonstrate that both the accuracy and reliability of the retrievals are significantly improved. These high-resolution cloud properties have been successfully applied to analyze extreme weather events, such as typhoons, providing potentially valuable insights into atmospheric processes.
Paper Structure (11 sections, 13 equations, 11 figures, 1 table)

This paper contains 11 sections, 13 equations, 11 figures, 1 table.

Table of Contents

  1. Introduction
  2. Results
  3. Evaluation using MODIS cloud products
  4. Evaluation using other cloud products and uncertainty analysis
  5. Application to typhoon case
  6. Discussion
  7. Methods
  8. Data
  9. Diffusion model
  10. Conditional diffusion model
  11. Performance evaluation

Figures (11)

  • Figure 1: The performance evaluation of retrieved CLP and cloud properties. The bars, colored differently, represent sample sizes ranging from 1 to 30. (a) displays the F1 scores for different CLPs, comparing single samples and ensemble means of various sample sizes (ranging from 5 to 30) with the deterministic model; (b) shows the overall accuracy (OA) for all CLPs, comparing single samples and ensemble means of various sample sizes with the deterministic model; (c--e) and (f--g) show the MAE and RMSE for cloud properties, respectively, comparing single samples and ensemble means of various sample sizes with the deterministic model.
  • Figure 2: Performance evaluation and power spectrum analysis of the retrieved cloud properties. (a) and (b) show the SSIM and PSNR of the cloud properties, respectively, comparing ensemble means of various sample sizes with the deterministic model. (c–e) display the radially averaged power spectra of COT, CER, and CTH, comparing CloudDiff and the deterministic model against MODIS.
  • Figure 3: Joint probability density plots of CTH (first row) and confusion matrices of CLP (second row). (a–d) show the joint probability density distributions of CALIPSO with (a) CloudDiff, (b) the deterministic model, (c) Himawari-8, and (d) MODIS. (e–h) present confusion matrices comparing CALIPSO CLP products with (e) CloudDiff, (f) the deterministic model, (g) Himawari-8, and (h) MODIS.
  • Figure 4: MODIS cloud products and retrieved cloud properties in the typhoon In-Fa region, centered at 24.1°N, 127.8°E, at 0220 UTC on July 21, 2021. The columns are MODIS cloud products, sample, ensemble mean, deterministic model, and standard deviation (std). Note the blank areas in the MODIS cloud products represent missing data.
  • Figure 5: The retrieval of cloud properties in the typhoon In-Fa region at 02:20 UTC on July 21, 2021, was performed using the CloudDiff. The columns represent different samples and grid points where MODIS cloud properties are not captured by the samples. Underestimations and overestimations are indicated by black ’x’ markers. The background is colored based on MODIS cloud products, with blank areas indicating missing data.
  • ...and 6 more figures