Liu et al. (2026) A Machine Learning Approach to Cloud Cover Forecasting Using Lagrangian Air Mass History

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Journal of Advances in Modeling Earth Systems
• Year: 2026
• Date: 2026-01-01
• Authors: Zihui Liu, Ryan Eastman, Robert Wood
• DOI: 10.1029/2025ms004972

Research Groups

Not explicitly stated in the abstract.

Short Summary

This study explores various machine learning approaches, utilizing Lagrangian air mass history, to improve cloud cover prediction accuracy. The research demonstrates that machine learning models, particularly recurrent neural networks with cloud cover feedback, achieve significantly better prediction performance than leading meteorological reanalysis, with a correlation coefficient of 0.72.

Objective

• To investigate whether machine learning approaches, incorporating Lagrangian air mass history and cloud-controlling factors, can improve total cloud cover prediction accuracy over eastern subtropical oceans compared to leading meteorological reanalysis.

Study Configuration

• Spatial Scale: Eastern subtropical oceans, focusing on isobaric boundary layer trajectories.
• Temporal Scale: 4 days, with meteorological data collected at 12-hour intervals along trajectories.

Methodology and Data

• Models used: Various machine learning approaches, including recurrent neural networks (RNNs), and other statistical models.
• Data sources: 169,824 isobaric boundary layer trajectories, colocated meteorological data, and satellite cloud cover data from MODIS.

Main Results

• Machine learning models predicted cloud cover with similar or better performance than leading meteorological reanalysis.
• The best model, using recurrent neural networks with cloud cover feedback, achieved a correlation coefficient of 0.72 between predictions and MODIS measurements.
• The leading meteorological reanalysis achieved a correlation coefficient of 0.65.
• A sensitivity study revealed a nonlinear relationship between cloud cover and numerous cloud-controlling factors (CCFs).

Contributions

• Demonstrates the superior performance of machine learning models, specifically recurrent neural networks, in predicting total cloud cover compared to established meteorological reanalysis.
• Utilizes Lagrangian air mass history as a novel input for cloud cover prediction models.
• Quantifies the improvement in prediction accuracy, showing a notable increase in correlation coefficient (0.72 vs. 0.65).
• Provides insights into the nonlinear sensitivities of cloud cover to various cloud-controlling parameters.

Funding

Not explicitly stated in the abstract.

Citation

@article{Liu2026Machine,
  author = {Liu, Zihui and Eastman, Ryan and Wood, Robert},
  title = {A Machine Learning Approach to Cloud Cover Forecasting Using Lagrangian Air Mass History},
  journal = {Journal of Advances in Modeling Earth Systems},
  year = {2026},
  doi = {10.1029/2025ms004972},
  url = {https://doi.org/10.1029/2025ms004972}
}