Qiu et al. (2026) Physics-informed machine learning for cloud detection

Identification

• Journal: Remote Sensing of Environment
• Year: 2026
• Date: 2026-09-26
• Authors: Shi Qiu, Zhe Zhu, Xiucheng Yang, Junchang Ju, Qiang Zhou, Christopher S.R. Neigh
• DOI: 10.1016/j.rse.2026.115672

Research Groups

• Department of Natural Resources and the Environment, University of Connecticut
• NASA Goddard Space Flight Center
• Earth System Science Interdisciplinary Center, University of Maryland
• Science Systems and Applications, Inc. (SSAI)

Short Summary

This study introduces Fmask 5, a novel physics-informed machine learning (PIML) framework for cloud detection in Landsat and Sentinel-2 imagery, achieving overall accuracies of 93.46% for Landsats 8–9, 92.50% for Landsats 4–7, and 95.38% for Sentinel-2.

Objective

• To develop a robust and accurate cloud detection algorithm that integrates physical rules with machine learning models

Study Configuration

• Spatial Scale: Global scale, covering various land cover types and atmospheric conditions
• Temporal Scale: Single-date imagery, using historical observations collected before 2024

Methodology and Data

• Models used:
  • Physics-informed machine learning (PIML) framework
  • LightGBM and UNet models for machine learning
  • Physical-rule-based model adapted from Fmask 4.6
• Data sources:
  • Landsat Collection 2 Level-1 product
  • Sentinel-2 Baseline 4.0 Level-1C TOA reflectance product
  • Global Surface Water Occurrence (GSWO) dataset
  • Digital Elevation Model (DEM)

Main Results

• Fmask 5 achieved overall accuracies of 93.46% for Landsats 8–9, 92.50% for Landsats 4–7, and 95.38% for Sentinel-2
• The PIML framework integrated physical rules with machine learning models to improve accuracy

Contributions

• Fmask 5 provides a significant improvement over existing cloud detection algorithms, achieving higher overall accuracies
• The study demonstrates the potential of physics-informed machine learning in remote sensing applications

Funding

• This research was funded by NASA's Harmonized Landsat Sentinel-2 (HLS) program

Citation

@article{Qiu2026Physicsinformed,
  author = {Qiu, Shi and Zhu, Zhe and Yang, Xiucheng and Ju, Junchang and Zhou, Qiang and Neigh, Christopher  S.R.},
  title = {Physics-informed machine learning for cloud detection},
  journal = {Remote Sensing of Environment},
  year = {2026},
  doi = {10.1016/j.rse.2026.115672},
  url = {https://doi.org/10.1016/j.rse.2026.115672}
}