Subseasonal Tropical Cyclone Forecasts in the ECMWF 46-Day Ensemble: Formation, Track, and Precipitation

This study evaluates subseasonal forecasts of tropical cyclones (TCs) using the ECMWF 46-day ensemble. An objective TC tracking scheme is utilized to identify TC formations and tracks in the 20-year ECMWF reforecasts (Tsai et al. 2023). The investigation focuses on assessing the impact of large-scale environmental factors, such as the Madden Julian Oscillation, Boreal Summer Intraseasonal Oscillation, and Western North Pacific Summer Monsoon, on the TC forecast skills in weeks 1-4. In stead of utilizing the Receiver Operating Characteristic (ROC) curve, the Precision-Recall (PR) curve is adopted to adequately represent the imbalanced TC data.

Preliminary findings suggest that TC forecast skills vary across different phases or categories of the large-scale environments. For instance, better TC forecast skills are observed when the Western North Pacific Summer Monsoon (WNPSM) falls within the 60-100% percentile categories. Additionally, an analysis of TC track forecast errors in the week-1 to week-4 forecasts is conducted, revealing average negative biases in along-track errors (indicating slower TC translation speed), while cross-track forecast errors show no significant biases. Furthermore, a comparison between precipitation forecasts from the ECMWF and surface observations demonstrates that the numerical model with lower-resolution grids can effectively capture the rainfall pattern contrast in Taiwan. However, the correlations of rainfall patterns become much smaller when the forecast lead time exceeds 72 hours due to track forecast errors. Lastly, a statistical model is developed to provide typhoon Quantitative Precipitation Forecast (QPF), aiming to support water resources management in Taiwan by accounting for track forecast biases in the model.