Toe et al. (2026) Machine Learning-Based Reconstruction of Missing Meteorological Observations Using Reanalysis and Satellite Data in West Africa

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

Identification

• Journal: Atmosphere
• Year: 2026
• Date: 2026-09-09
• Authors: Marcel Jocelyn Wendemi Michaelange Toe, Belko Aboul Aziz Diallo, Adeshina Kamil Sanoussi, Valentin Ouedraogo, Samuel Guug, Kehinde.O. Ogunjobi, Hamadou Barro, Adolphe Avocanh, Hermann Hien, Michael Ayamba
• DOI: 10.3390/atmos17090884

Research Groups

Not specified

Short Summary

This study develops a machine learning framework to reconstruct missing hourly meteorological data in West Africa by integrating in situ AWS observations with ERA5-Land reanalysis and GPM satellite products.

Objective

• To fill substantial data gaps in high-frequency meteorological observations from automatic weather stations (AWS) in data-sparse regions of West Africa.

Study Configuration

• Spatial Scale: Regional (10 West African countries, >50 AWS covering Sahelian, Sudanian, coastal, and humid tropical zones).
• Temporal Scale: 2017–2025 (hourly resolution).

Methodology and Data

• Models used: Gradient boosting models (XGBoost, LightGBM, and CatBoost) using a station-wise and variable-wise training strategy.
• Data sources: In situ AWS measurements, ERA5-Land reanalysis fields, and Global Precipitation Measurement (GPM) satellite-derived products.

Main Results

• High Accuracy: Air temperature and atmospheric pressure showed the best reconstruction skill ($R^2 > 0.90$).
• Moderate to High Accuracy: Relative humidity and global solar radiation achieved $R^2$ values between 0.80 and 0.92.
• Lower Accuracy: Precipitation and wind speed exhibited lower skill due to their intermittency and sensitivity to local-scale processes.
• Lowest Accuracy: Wind direction showed the largest angular errors, indicating difficulty in reconstructing directional variability from large-scale predictors.

Contributions

• Implementation of a scalable, multi-model machine learning framework (over 300 variable-specific models) to recover high-frequency meteorological data in regions with sparse observation networks.

Funding

Not specified

Citation

@article{Toe2026Machine,
  author = {Toe, Marcel Jocelyn Wendemi Michaelange and Diallo, Belko Aboul Aziz and Sanoussi, Adeshina Kamil and Ouedraogo, Valentin and Guug, Samuel and Ogunjobi, Kehinde.O. and Barro, Hamadou and Avocanh, Adolphe and Hien, Hermann and Ayamba, Michael},
  title = {Machine Learning-Based Reconstruction of Missing Meteorological Observations Using Reanalysis and Satellite Data in West Africa},
  journal = {Atmosphere},
  year = {2026},
  doi = {10.3390/atmos17090884},
  url = {https://doi.org/10.3390/atmos17090884}
}