Ren et al. (2026) Machine learning significantly improves the simulation of hourly-to-yearly scale cloud nuclei concentration and radiative forcing in polluted atmosphere

Identification

• Journal: Geoscientific model development
• Year: 2026
• Date: 2026-07-17
• Authors: Jingye Ren, Honghao Xu, Guiquan Liu, Anran Zhang, Chuanfeng Zhao, Min Hu, Dongjie Shang, Lizi Tang, Ru‐Jin Huang, Yele Sun, Fang Zhang
• DOI: 10.5194/gmd-19-6403-2026

Research Groups

• State Key Laboratory of Loess Science, Institute of Earth Environment, Chinese Academy of Sciences
• Xi’an Institute for Innovative Earth Environment Research
• Shenzhen Key Laboratory of Organic Pollution Prevention and Control, School of Eco-Environment, Harbin Institute of Technology Shenzhen
• Department of Atmospheric and Oceanic Sciences, School of Physics, Peking University
• State Key Joint Laboratory of Environmental Simulation and Pollution Control, College of Environmental Sciences and Engineering, Peking University
• State Key Laboratory of Atmospheric Boundary Layer Physics and Atmospheric Chemistry, Institute of Atmospheric Physics, Chinese Academy of Sciences

Short Summary

This study develops a machine learning-based model to predict hourly-to-yearly scale cloud condensation nuclei (CCN) number concentrations in the North China Plain. The model significantly improves CCN prediction accuracy compared to the Weather Research and Forecasting coupled with Chemistry (WRF-Chem) model, reducing bias from -59% to -31%.

Objective

• Investigate the feasibility of using machine learning to predict CCN number concentrations on a large spatiotemporal scale in polluted areas.

Study Configuration

• Spatial Scale: Regional scale in the North China Plain (NCP)
• Temporal Scale: Hourly-to-yearly scale

Methodology and Data

• Models used: Random Forest Regression method (RFRM) and WRF-Chem model
• Data sources: Multi-source dataset of atmospheric variables, CCN concentration outputs from WRF-Chem, Tracking Air Pollution in China (TAP) dataset, European Centre for Medium-range Weather Forecasts Reanalysis version 5 (ERA-5)

Main Results

• The RFRM model shows significant improvement in predicting CCN number concentrations compared to the WRF-Chem model.
• The model captures well the diurnal cycle and seasonal variations of NCCN.
• The long-term trend of yearly averaged NCCN derived by the RFRM model agrees well with observations.

Contributions

• This study provides a new machine learning-based framework for predicting CCN number concentrations on a large spatiotemporal scale in polluted areas.
• The results demonstrate the effectiveness of using multi-source dataset and machine learning to improve CCN prediction accuracy.

Funding

• National Natural Science Foundation of China (Grant No. 41877304)
• National Key Research and Development Program of China (Grant No. 2017YFC0212604)

Citation

@article{Ren2026Machine,
  author = {Ren, Jingye and Zou, Songjian and Xu, Honghao and Liu, Guiquan and Wang, Zhe and Zhang, Anran and Zhao, Chuanfeng and Hu, Min and Shang, Dongjie and Tang, Lizi and Huang, Ru‐Jin and Sun, Yele and Zhang, Fang},
  title = {Machine learning significantly improves the simulation of hourly-to-yearly scale cloud nuclei concentration and radiative forcing in polluted atmosphere},
  journal = {Geoscientific model development},
  year = {2026},
  doi = {10.5194/gmd-19-6403-2026},
  url = {https://doi.org/10.5194/gmd-19-6403-2026}
}