Chung et al. (2025) Statistical characteristics of storm cells and centroid-based probability nowcasts by tracking error

Identification

• Journal: Atmospheric Research
• Year: 2025
• Date: 2025-10-08
• Authors: Kao‐Shen Chung, Tsai-Jung Lee, Li-Pen Wang, Bingzhang Wang, Chieh-Ying Ke, Yi‐Hao Tsou, Shin-Gan Chen, Ping‐Yu Lin, Treng-Shi Huang
• DOI: 10.1016/j.atmosres.2025.108548

Research Groups

• Department of Atmospheric Sciences, National Central University, Taoyuan, Taiwan
• Department of Civil Engineering, National Taiwan University, Taipei, Taiwan
• Central Weather Administration, Taipei, Taiwan

Short Summary

This study statistically characterized summer storm cells over Taiwan using 8 years of radar data and the Storm Cell Identification and Tracking (SCIT) algorithm, focusing on tracking error to develop and verify two storm-based nowcasting techniques, Potential Track Area for Storm (PTAS) and Probability of Storm Tracking (PoST), which enhance early warning capabilities by quantifying forecast uncertainty.

Objective

• To investigate the characteristics of summer storm cells over Taiwan, focusing on the tracking error statistics generated by the Storm Cell Identification and Tracking (SCIT) algorithm.
• To understand the quantitative statistical features of storm cells, which are currently lacking for the region.
• To develop storm-based nowcasting techniques (PTAS and PoST) to improve early warning capability for severe weather events.
• To assess how storm cell characteristics and tracking errors differ under varying terrain and weather systems across northern and southern Taiwan.

Study Configuration

• Spatial Scale: Taiwan island, specifically northern (Wu-Fen Shan radar, RCWF) and southern (Chi-Gu radar, RCCG) regions. Radar coverage extends approximately 460 km, with Doppler mode data used for a 230 km radius. Nowcasting products use a 1 km x 1 km grid resolution for probability calculations.
• Temporal Scale: Summer seasons (May to August) from 2011 to 2018 were used for statistical analysis and product development. Data from 2019 were used for validation. Forecast lead times for nowcasting products are 0-1 hour, with specific evaluations at 10, 20, 30, 40, 50, and 60 minute intervals. Radar scan update cycles are approximately 6-7 minutes for RCWF and 7-8 minutes for RCCG.

Methodology and Data

• Models used:
  • Storm Cell Identification and Tracking (SCIT) algorithm: Identifies, characterizes, tracks, and forecasts storm cell movement using radar observations.
  • System for Convection Analysis and Nowcasting (SCAN): Integrates radar observations and other meteorological information, incorporating SCIT.
  • Potential Track Area for Storm (PTAS): A nowcasting product that uses historical tracking error statistics to define a 70% probability warning area for storm cell centers.
  • Probability of Storm Tracking (PoST): A nowcasting product that uses the Monte Carlo method and historical error data to quantify the probability of storm cell radius intrusion (default 4 km) for a 1-hour forecast.
  • Deep Generative Model of Radar (DGMR): A deep learning model used for comparison in nowcasting performance.
• Data sources:
  • Radar observation data: S-band radar data from Wu-Fen Shan (RCWF) in northern Taiwan and Chi-Gu (RCCG) in southern Taiwan, collected during summer seasons (May-August) from 2011 to 2018 (analysis) and 2019 (validation). Data consists of nine elevation angles (0.5°, 1.4°, 2.4°, 3.4°, 4.3°, 6.0°, 9.9°, 14.6°, and 19.5°).
  • Weather classification data: Daily weather overview descriptions issued by the Central Weather Administration (CWA) of Taiwan, used to classify days as "synoptic" (e.g., typhoons, fronts) or "weak synoptic" (e.g., thermally driven afternoon thunderstorms).
  • DGMR forecast data: Provided by the Computational Hydrometeorology Laboratory of National Taiwan University for verification and comparison.

Main Results

• Summer storm cells in Taiwan typically persist for less than 1 hour, move at speeds of 2-10 meters per second (m/s) from southwest to northeast, and exhibit maximum reflectivity (MaxdBZ) values predominantly between 45-55 dBZ.
• Storm cell characteristics are significantly influenced by terrain and the location of weather system development. Cells traversing oceanic areas maintain higher speeds (approximately 10 m/s in synoptic weather), while those closer to land and mountainous regions show lower speeds (below 5 m/s in weak synoptic weather).
• Weak synoptic weather cells, which primarily develop in inland mountainous areas, tend to have higher MaxdBZ values compared to cells in synoptic weather conditions.
• Tracking errors for storm cells generally do not exhibit a strong directional bias, though a slight southwest-to-northeast tendency was observed in the northern region (RCWF), consistent with prevailing monsoon flow.
• The southern region (RCCG) shows larger 1-hour tracking errors than the northern region (RCWF), and errors are greater on synoptic days compared to weak synoptic days. Tracking error increases with both forecast lead time and storm cell speed, and is exacerbated by longer radar scan intervals.
• The developed PTAS product consistently achieved a Probability of Detection (POD) score of approximately 0.7 for 1-hour storm cell forecasts during the summer of 2019, aligning with its 70% probability setting.
• PTAS performs most effectively during typhoon circulations (POD between 0.5 and 0.7), followed by weak synoptic weather (POD between 0.4 and 0.5), and least effectively during frontal systems.
• Compared to the Deep Generative Model of Radar (DGMR), the centroid-based PTAS approach demonstrates relative advantages in issuing very-short-term warnings for intense storm cells (exceeding 30 dBZ) within a 1-hour lead time, as DGMR forecasts tend to smooth and weaken over time.
• Both PTAS and PoST products can be updated rapidly (within seconds) after each radar scan, providing timely and effective warning information for decision-makers.

Contributions

• Provides a comprehensive, multi-year statistical characterization of summer storm cell behavior (duration, speed, intensity, spatial distribution, initiation hotspots) and tracking errors over Taiwan, categorized by region (northern/southern) and synoptic conditions (synoptic/weak synoptic), addressing a critical gap in local meteorological literature.
• Introduces and validates two novel storm-based probabilistic nowcasting products, Potential Track Area for Storm (PTAS) and Probability of Storm Tracking (PoST), which quantitatively enhance early warning capabilities by visualizing and quantifying the uncertainty of storm cell tracking forecasts.
• Demonstrates the practical efficiency and rapid update capability of radar-based extrapolation techniques for real-time very-short-term severe weather warnings, highlighting their utility compared to deep learning models like DGMR for intense storm events.
• Offers valuable statistical insights into storm cell dynamics influenced by Taiwan's complex terrain and monsoon climate, which can serve as foundational data for the development and optimization of future AI-based nowcasting models and operational warning systems.

Funding

• National Science and Technology Council (NSTC) of Taiwan (Grants 114-2625-M-008-005 and 111-2625-M-008-014)
• Central Weather Administration (CWA) of Taiwan (Grant 1102029C)

Citation

@article{Chung2025Statistical,
  author = {Chung, Kao‐Shen and Lee, Tsai-Jung and Wang, Li-Pen and Wang, Bingzhang and Ke, Chieh-Ying and Tsou, Yi‐Hao and Chen, Shin-Gan and Lin, Ping‐Yu and Huang, Treng-Shi},
  title = {Statistical characteristics of storm cells and centroid-based probability nowcasts by tracking error},
  journal = {Atmospheric Research},
  year = {2025},
  doi = {10.1016/j.atmosres.2025.108548},
  url = {https://doi.org/10.1016/j.atmosres.2025.108548}
}