Chen et al. (2025) Projecting forecast quality before events through machine learning: Preliminary results of cloud-resolving quantitative precipitation forecasts in Taiwan for westbound typhoons

Identification

• Journal: Atmospheric Research
• Year: 2025
• Date: 2025-09-11
• Authors: Shin-Hau Chen, Chung‐Chieh Wang
• DOI: 10.1016/j.atmosres.2025.108479

Research Groups

• Department of Earth Sciences, National Taiwan Normal University, Taipei, Taiwan

Short Summary

This study develops a neural-network machine learning model to project the expected similarity skill score (SSS) of cloud-resolving quantitative precipitation forecasts (QPFs) for westward-moving typhoons in Taiwan, demonstrating its ability to provide objective guidance on forecast quality before events.

Objective

• To develop and evaluate a machine learning model to objectively project the quality (expected Similarity Skill Score) of cloud-resolving quantitative precipitation forecasts for westward-moving typhoons in Taiwan, thereby informing forecasters in advance.

Study Configuration

• Spatial Scale: Taiwan and its surrounding area, specifically within 300 kilometers from Taiwan.
• Temporal Scale: Forecast lead times up to 8 days, with time-lagged forecasts generated every 6 hours, covering the entire typhoon influence period.

Methodology and Data

• Models used:
  • Neural-network machine learning model (for projecting forecast quality).
  • Cloud-resolving numerical weather prediction model (for generating QPFs).
• Data sources:
  • 105 parameters linked to rainfall derived from time-lagged forecasts of 10 westward-moving typhoons.
  • Observed rainfall data (used to calculate actual Similarity Skill Score for model training and validation).

Main Results

• The machine learning model successfully captured the tendency of the actual Similarity Skill Score (SSS) for 8 out of 10 typhoon cases.
• This capability allows forecasters to identify more trustworthy QPFs and less reliable ones beforehand, especially valuable at longer lead times where forecast uncertainty is typically high.
• The results are highly encouraging, though the model's utility may decrease if a typhoon's behavior significantly deviates from the training data.

Contributions

• Provides a preliminary proof of concept for using machine learning to objectively assess the quality of quantitative precipitation forecasts before an event occurs.
• Offers a novel approach to address the long-standing problem in numerical weather prediction of knowing the likelihood of a predicted scenario in advance.
• Delivers an objective guidance system for forecasters to gauge the trustworthiness of QPFs, particularly beneficial for high-impact events like typhoons.

Funding

• Not specified in the provided text.

Citation

@article{Chen2025Projecting,
  author = {Chen, Shin-Hau and Wang, Chung‐Chieh},
  title = {Projecting forecast quality before events through machine learning: Preliminary results of cloud-resolving quantitative precipitation forecasts in Taiwan for westbound typhoons},
  journal = {Atmospheric Research},
  year = {2025},
  doi = {10.1016/j.atmosres.2025.108479},
  url = {https://doi.org/10.1016/j.atmosres.2025.108479}
}