Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Joint Retrieval of Cloud properties using Attention-based Deep Learning Models

Zahid Hassan Tushar, Adeleke Ademakinwa, Jianwu Wang, Zhibo Zhang, Sanjay Purushotham

TL;DR

This work tackles the challenge of retrieving cloud properties from radiance observations by addressing 3D radiative transfer effects that bias traditional IPA. It introduces CAM, a lightweight CloudUNet with Channel Attention, capable of jointly estimating Cloud Optical Thickness $\tau$ and Cloud Effective Radius $r_e$ from bi-spectral radiances using a window-based patch framework and a multi-task objective $L_{MTO}$. CAM demonstrates substantial improvements over state-of-the-art DL models and IPA on 902 LES cloud fields, achieving up to ~76-87% lower MAE relative to IPA and improved correlation, thanks to attention-enhanced multiscale features and a balanced loss for $\tau$ and $r_e$. The results underscore CAM’s potential for accurate, efficient cloud-property retrieval and motivate future work with multi-angle radiance data.

Abstract

Accurate cloud property retrieval is vital for understanding cloud behavior and its impact on climate, including applications in weather forecasting, climate modeling, and estimating Earth's radiation balance. The Independent Pixel Approximation (IPA), a widely used physics-based approach, simplifies radiative transfer calculations by assuming each pixel is independent of its neighbors. While computationally efficient, IPA has significant limitations, such as inaccuracies from 3D radiative effects, errors at cloud edges, and ineffectiveness for overlapping or heterogeneous cloud fields. Recent AI/ML-based deep learning models have improved retrieval accuracy by leveraging spatial relationships across pixels. However, these models are often memory-intensive, retrieve only a single cloud property, or struggle with joint property retrievals. To overcome these challenges, we introduce CloudUNet with Attention Module (CAM), a compact UNet-based model that employs attention mechanisms to reduce errors in thick, overlapping cloud regions and a specialized loss function for joint retrieval of Cloud Optical Thickness (COT) and Cloud Effective Radius (CER). Experiments on a Large Eddy Simulation (LES) dataset show that our CAM model outperforms state-of-the-art deep learning methods, reducing mean absolute errors (MAE) by 34% for COT and 42% for CER, and achieving 76% and 86% lower MAE for COT and CER retrievals compared to the IPA method.

Joint Retrieval of Cloud properties using Attention-based Deep Learning Models

TL;DR

This work tackles the challenge of retrieving cloud properties from radiance observations by addressing 3D radiative transfer effects that bias traditional IPA. It introduces CAM, a lightweight CloudUNet with Channel Attention, capable of jointly estimating Cloud Optical Thickness and Cloud Effective Radius from bi-spectral radiances using a window-based patch framework and a multi-task objective . CAM demonstrates substantial improvements over state-of-the-art DL models and IPA on 902 LES cloud fields, achieving up to ~76-87% lower MAE relative to IPA and improved correlation, thanks to attention-enhanced multiscale features and a balanced loss for and . The results underscore CAM’s potential for accurate, efficient cloud-property retrieval and motivate future work with multi-angle radiance data.

Abstract

Accurate cloud property retrieval is vital for understanding cloud behavior and its impact on climate, including applications in weather forecasting, climate modeling, and estimating Earth's radiation balance. The Independent Pixel Approximation (IPA), a widely used physics-based approach, simplifies radiative transfer calculations by assuming each pixel is independent of its neighbors. While computationally efficient, IPA has significant limitations, such as inaccuracies from 3D radiative effects, errors at cloud edges, and ineffectiveness for overlapping or heterogeneous cloud fields. Recent AI/ML-based deep learning models have improved retrieval accuracy by leveraging spatial relationships across pixels. However, these models are often memory-intensive, retrieve only a single cloud property, or struggle with joint property retrievals. To overcome these challenges, we introduce CloudUNet with Attention Module (CAM), a compact UNet-based model that employs attention mechanisms to reduce errors in thick, overlapping cloud regions and a specialized loss function for joint retrieval of Cloud Optical Thickness (COT) and Cloud Effective Radius (CER). Experiments on a Large Eddy Simulation (LES) dataset show that our CAM model outperforms state-of-the-art deep learning methods, reducing mean absolute errors (MAE) by 34% for COT and 42% for CER, and achieving 76% and 86% lower MAE for COT and CER retrievals compared to the IPA method.

Paper Structure

This paper contains 17 sections, 3 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Problem Formulation
  4. Cloud Property Retrieval framework
  5. CAM: CloudUNet with Attention Module
  6. Training Objective Function
  7. Experimental Setup
  8. Data
  9. Model Comparison
  10. Implementation
  11. Results and Discussion
  12. Quantitative Results
  13. Qualitative Results
  14. Ablation Studies
  15. Impact of Attention Mechanisms
  16. ...and 2 more sections

Figures (4)

  • Figure 1: CAM: CloudUNet with Attention Module
  • Figure 2: Comparison of COT Retrieval methods. COT values are plotted in shifted log scale. (a) True COT; (b) IPA retrieved COT; (c) UNet retrieved COT; (d) CAM retrieved COT. Highlighted regions are shown in bottom row.
  • Figure 3: Comparison of CER Retrieval methods. CER values are in regular scale. (a) True CER; (b) IPA retrieved CER; (c) UNet retrieved CER; (d) CAM retrieved CER. Highlighted regions are shown in bottom row.
  • Figure 4: CAM COT retrieval with and without attention mechanism. Top row shows the COT profile while bottom row shows the highlighted regions. COT is shown in log scale.