Skip to content
DraconiticResearch
Search Sources API
190,839 studies31 sourcesUpdated 6 min ago
Peer-reviewedCC BYnpj Climate and Atmospheric Science

Reconstructing particle number size distributions via aerosol history–informed multiple deep learning models

  • Yuhan Cheng1
  • Xiaoyu Xu1
  • Liwen Wang1
  • Yuanlong Huang2
  • Hui Chen3
  • Xing Wei4
  • Saidur Rahaman4
  • Dongmei Cai4
  • Bing Qi5
  • Ying Chen6
  • Chaopeng Shen7
  • Minghuai Wang8
2 more authors
  • Yuzhong Zhang4
  • Xianda Gong4
8 affiliations
  1. 1College of Environmental and Resource Sciences, Zhejiang University
  2. 2School of the Environment and Sustainable Engineering, Eastern Institute of Technology
  3. 3Centre for Energy and Environment, Univ. Lille, Institut Mines-Télécom, IMT Nord Europe
  4. 4Research Center for Industries of the Future, Westlake University
  5. 5Hangzhou Meteorological Bureau
  6. 6School of Geography, Earth and Environmental Sciences, University of Birmingham
  7. 7Civil and Environmental Engineering, Pennsylvania State University
  8. 8School of Atmospheric Sciences, Nanjing University
Read full textDOICite

Abstract

Particle number size distribution (PNSD) is fundamental for characterizing aerosols and quantifying aerosol–cloud interactions, while simulations remain uncertain due to strong spatiotemporal variability during air-parcel transport. We developed a framework that combines air-parcel location history with co-located aerosol, cloud, meteorological, and gas variables into deep learning (DL) models to reconstruct PNSD. Two Long Short-Term Memory (LSTM) models and one Bidirectional LSTM (BiLSTM) model were trained and evaluated on 10 years of measurements at the Cape Verde Atmospheric Observatory (CVAO) in the central Atlantic, achieving mean fractional errors (MFEs) < 0.17. Applying the three well-trained models to the Ascension Island (ASI) dataset in the South Atlantic achieved MFEs < 0.16, demonstrating strong transferability and enabling PNSD projection in remote marine regions. SHapley Additive exPlanations (SHAP) analysis showed inconsistent feature attributions across three models, suggesting limited interpretability of mechanisms during air-parcel transportation. Thus, multi-model cross-validation is recommended for interpreting DL models.

Related

  1. Fractal and multifractal characteristics of particle-size distributions in debris-flow deposits
    Xu Zongheng, Li Yilong, Bai Jianjiao
    Scientific ReportsPeer-reviewedCC BY-NC-ND 4.0Full textEnvironmental science & pollutionCrossref
  2. Microplastics as evolving particle–chemical–microbial interfaces: plastisphere ecology, ecotoxicity, and safety-informed remediation
    Haixin Jiao, Ahmed E. M. Fodah, Michael Schagerl +2
    Frontiers in MicrobiologyPeer-reviewedCC BY 4.0Full textEnvironmental science & pollutionCrossrefEurope PMCPMC
  3. An Arctic climatology of particle number size distributions and cloud condensation nuclei during warm and moist intrusions
    Berkay Dönmez, Jakob Boyd Pernov, Romanos Foskinis +13
    npj climate and atmospheric sciencePeer-reviewedCC BY 4.0Full textAtmospheric science & meteorologyClimate scienceHAL
  4. Physics-informed deep learning links geodetic data and fault friction
    Rikuto Fukushima, Masayuki Kano, Kazuro Hirahara +1
    Communications Earth & EnvironmentPeer-reviewedCC BY-NC-ND 4.0Full textGeophysics & seismologyCrossref
  5. Physics-Informed Machine Learning in Hydrology and Water Resources: A Systematic Bibliometric Analysis
    George Kwame Frimpong, Omer Mermer, Ibrahim Demir
    EarthArXivPreprintCC BY 4.0Full textHydrology & soilsEarthArXiv

Cite

BibTeXRIS
BibTeX
@article{cheng2026reconstructing,
  title = {{Reconstructing particle number size distributions via aerosol history–informed multiple deep learning models}},
  author = {Yuhan Cheng and Xiaoyu Xu and Liwen Wang and Yuanlong Huang and Hui Chen and Xing Wei and Saidur Rahaman and Dongmei Cai and Bing Qi and Ying Chen and Chaopeng Shen and Minghuai Wang and Yuzhong Zhang and Xianda Gong},
  journal = {npj Climate and Atmospheric Science},
  year = {2026},
  month = jun,
  doi = {10.1038/s41612-026-01447-z},
  url = {https://doi.org/10.1038/s41612-026-01447-z},
  note = {CC BY},
}
RIS
TY  - JOUR
TI  - Reconstructing particle number size distributions via aerosol history–informed multiple deep learning models
AU  - Yuhan Cheng
AU  - Xiaoyu Xu
AU  - Liwen Wang
AU  - Yuanlong Huang
AU  - Hui Chen
AU  - Xing Wei
AU  - Saidur Rahaman
AU  - Dongmei Cai
AU  - Bing Qi
AU  - Ying Chen
AU  - Chaopeng Shen
AU  - Minghuai Wang
AU  - Yuzhong Zhang
AU  - Xianda Gong
PY  - 2026
DA  - 2026/06/01
JO  - npj Climate and Atmospheric Science
DO  - 10.1038/s41612-026-01447-z
UR  - https://doi.org/10.1038/s41612-026-01447-z
AB  - Particle number size distribution (PNSD) is fundamental for characterizing aerosols and quantifying aerosol–cloud interactions, while simulations remain uncertain due to strong spatiotemporal variability during air-parcel transport. We developed a framework that combines air-parcel location history with co-located aerosol, cloud, meteorological, and gas variables into deep learning (DL) models to reconstruct PNSD. Two Long Short-Term Memory (LSTM) models and one Bidirectional LSTM (BiLSTM) model were trained and evaluated on 10 years of measurements at the Cape Verde Atmospheric Observatory (CVAO) in the central Atlantic, achieving mean fractional errors (MFEs) < 0.17. Applying the three well-trained models to the Ascension Island (ASI) dataset in the South Atlantic achieved MFEs < 0.16, demonstrating strong transferability and enabling PNSD projection in remote marine regions. SHapley Additive exPlanations (SHAP) analysis showed inconsistent feature attributions across three models, suggesting limited interpretability of mechanisms during air-parcel transportation. Thus, multi-model cross-validation is recommended for interpreting DL models.
SN  - 2397-3722
ER  -