Shu (2025) Short-term ensemble prediction of convective cells using data-driven methods

Identification

• Journal: Open MIND
• Year: 2025
• Date: 2025-07-08
• Authors: Zhou Shu
• DOI: 10.25560/127500

Research Groups

• Department of Civil and Environmental Engineering, Imperial College London

Short Summary

This thesis develops an ensemble nowcasting framework that combines an analogue-based approach with machine learning to predict the evolution of convective rainfall cell properties.

Objective

• To improve the very short-term forecasting (nowcasting) of convective cell properties—specifically size, major axis length, and reflectivity—by overcoming the limitations of existing field-based and persistence methods.

Study Configuration

• Spatial Scale: Localized convective cells (applicable to urban surface-water flooding scales).
• Temporal Scale: Very short-term (nowcasting).

Methodology and Data

• Models used: Enhanced TITAN algorithm (for radar data processing), analogue-based model with adaptive spatial thresholds, and machine learning classifiers (specifically Random Forests).
• Data sources: Radar data.

Main Results

• Random Forests were identified as the most effective machine learning classifier for predicting the "track type" of convective cells.
• Integrating track type predictions into the analogue framework significantly improved the reliability of categorical forecasts and the accuracy of deterministic forecasts.
• The proposed ensemble method demonstrated superior calibration compared to simple persistence or unfiltered analogue methods.
• The resulting system achieves a practical balance between computational efficiency, forecast skill, and calibration.

Contributions

• Develops a hybrid data-driven framework for convective cell nowcasting that integrates analogue methods with machine learning to better capture the temporal evolution of storms.
• Provides a computationally efficient tool specifically suited for urban surface-water flooding applications.

Funding

• Not specified in the provided text.

Citation

@article{Shu2025Shortterm,
  author = {Shu, Zhou},
  title = {Short-term ensemble prediction of convective cells using data-driven methods},
  journal = {Open MIND},
  year = {2025},
  doi = {10.25560/127500},
  url = {https://doi.org/10.25560/127500}
}