A Comparative Study of Threshold-Setting Methods for Clear-Air Turbulence Diagnosis over China Using PIREPs and ERA5 Data

As a significant threat to aviation safety, clear-air turbulence (CAT) is often diagnosed and forecast operationally by using multiple indices, yet reliable diagnosis remains challenging due to its complex physical mechanisms and the lack of universally applicable thresholds for different indices. This study evaluated the performance of 22 diagnostics for capturing CAT over China during the cold season, using Pilot Reports (PIREPs) turbulence records and reanalysis data. Four threshold-setting methods were compared: (1) the Receiver Operating Characteristic (ROC) point nearest the upper left corner (called ROC-optimal for simplicity), (2) the observation-based percentile, (3) the observation-based median, and (4) the reanalysis-based percentile. A method-selection recommendation for threshold setting was further proposed to enhance the performance of CAT diagnostics under varying intensities and data scenarios. Results showed that the ROC-optimal method yielded the highest True Skill Statistic (TSS) for moderate (0.48) and severe (0.28) turbulence, whereas the observation-based percentile method performs best for light turbulence (0.42). The observation-based median method showed moderate performance, while the reanalysis-based percentile method yielded the lowest TSS for all intensities. Given the influence of turbulence intensity and data availability on diagnosis, we suggest using the observation-based percentile method for light turbulence warnings, while the ROC-optimal method is recommended for moderate and severe cases. The reanalysis-based percentile method may be considered for light turbulence warnings when observations are limited.