Boukendour et al. (2026) A Semi-Supervised SVM-Firefly Hybrid for Rainfall Estimation from MSG Data

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

Identification

• Journal: Atmosphere
• Year: 2026
• Date: 2026-01-26
• Authors: Ouiza Boukendour, Mourad Lazri, Rafik Absi, Fethi Ouallouche, Karim Labadi, Youcef Attaf, Amar Belghit, Soltane Ameur
• DOI: 10.3390/atmos17020133

Research Groups

Not explicitly mentioned in the provided text.

Short Summary

This paper introduces a Semi-Supervised Support Vector Machine (S3VM) combined with the Firefly Algorithm (FFA) to significantly improve precipitation intensity classification from satellite images, achieving up to a 17% accuracy increase over standard SVM.

Objective

• To enhance precipitation intensity classification from satellite images by implementing a Semi-Supervised Support Vector Machine (S3VM) that utilizes both labeled (radar) and unlabeled (satellite) data, and further optimizing its hyperparameters using the Firefly Algorithm (FFA).

Study Configuration

• Spatial Scale: Regional to continental scale, covering Meteosat Second Generation Satellite observations, with radar measurements providing localized labels.
• Temporal Scale: Not explicitly mentioned, but implied continuous monitoring from satellite observations.

Methodology and Data

• Models used: Semi-Supervised Support Vector Machine (S3VM), Firefly Algorithm (FFA), standard Support Vector Machine (SVM). The primary model is S3VM-FFA.
• Data sources: Meteosat Second Generation Satellite observations (unlabeled data), radar measurements (for generating labeled data).

Main Results

• The S3VM model improved classification accuracy by up to 15% compared to the standard SVM model.
• The hybridized S3VM-FFA approach achieved an even more robust performance, yielding a 17% improvement in classification accuracy compared to the standard SVM results.
• Statistical evaluation parameters indicated that S3VM-FFA also outperformed both standard SVM and conventional S3VM in estimating precipitation quantities at different scales.

Contributions

• Pioneering application of Semi-Supervised Support Vector Machines (S3VM) for precipitation intensity classification from satellite images, effectively leveraging large volumes of unlabeled satellite data alongside limited labeled radar data.
• Development and validation of a novel hybrid S3VM-Firefly Algorithm (S3VM-FFA) model for optimized hyperparameter tuning, leading to superior classification performance.
• Demonstration of significant quantitative improvements (up to 17% increased accuracy) in precipitation classification and quantity estimation compared to conventional supervised methods.

Funding

Not explicitly mentioned in the provided text.

Citation

@article{Boukendour2026SemiSupervised,
  author = {Boukendour, Ouiza and Lazri, Mourad and Absi, Rafik and Ouallouche, Fethi and Labadi, Karim and Attaf, Youcef and Belghit, Amar and Ameur, Soltane},
  title = {A Semi-Supervised SVM-Firefly Hybrid for Rainfall Estimation from MSG Data},
  journal = {Atmosphere},
  year = {2026},
  doi = {10.3390/atmos17020133},
  url = {https://doi.org/10.3390/atmos17020133}
}