利用ECMWF系集模式評估1-2週颱風降雨預報技術

本研究旨在評估1-2週之颱風降雨預報技術。研究資料採用歐洲中期天氣預報中心系集預報模式(簡稱ECEPS)之20年事後預報(reforecast)，將其與網格化的台灣地面降雨觀測資料進行比較。本研究主要分析颱風中心進入117-129 E和19-28 之預報資料。ECEPS預報資料採用台灣周圍的15個網格，網格大小為0.80.8。歷史降雨觀測資料則是由中央氣象署(CWA)提供之1 km1 km內插網格。ECEPS降雨預報與CWA觀測資料進行比對時，本研究搜尋與ECEPS具有最小平均絕對誤差之CWA降雨資料百分位數(percentile rank；PR)，以整體評估ECEPS網格降雨量之代表性。研究結果顯示，儘管ECEPS的解析較低，但仍可大致掌握山區和平地之間的降雨對比型態。然而，受到模式的解析度限制和颱風路徑預報誤差影響，當預報時間超過84小時，ECEPS與地面觀測的差異顯著增加。為了進一步分析颱風路徑預報誤差，本研究將模式路徑與美國聯合颱風警報中心(JTWC)之颱風最佳路徑進行比較，針對第1至第2週的ECEPS颱風路徑預報誤差進行分析，並將路徑誤差分解為徑向路徑誤差(Along Track Error；ATE)和橫向路徑誤差(Cross Track Error；CTE)。分析結果顯示，ATE之平均誤差為負值，表示ECEPS颱風路徑具有移動速度較慢的偏差。颱風形成後的168小時，ECEPS路徑預報之徑向平均誤差可能達到400公里。由於ECEPS颱風路徑預報具有顯著誤差，直接使用原始ECEPS降雨預報進行颱風預報作業便具有挑戰性。因此，本研究進一步應用了颱風氣候降雨定量預報模式。本研究開發之相關產品未來可納入中央氣象署之次季節(subseasonal)預報系統，協助水資源管理及降低災害發生風險。

關鍵字

颱風；系集模式；定量降水預報；次季節預報

並列摘要

The primary aim of this research is to assess the skill of week-1 to week-2 typhoon precipitation forecasts. Utilizing the 20-year reforecasts from the ECMWF ensemble prediction system (ECEPS), we compared these forecasts with gridded surface observations in Taiwan. The analysis focuses on the periods when typhoons impact Taiwan (spanning 117-129 °E and 19-28 °N). Fifteen ECEPS grid points surrounding Taiwan with a grid size of 0.8°×0.8° were used. Historical rainfall observations from the Central Weather Administration (CWA) were interpolated into a 1 km×1 km resolution grid. To compare ECEPS forecast data with CWA rainfall observations, the optimal percentile rank (PR) of the gridded CWA rainfall that minimizes the mean difference against the ECEPS data was determined. The findings indicate that while the ECEPS can moderately capture the rainfall distribution between mountainous and plain areas, its relatively lower horizontal resolution limits accuracy.The discrepancy between ECEPS rainfall forecasts and surface observations significantly increases for forecasts exceeding 84 hours, primarily due to the model's coarse resolution and errors in predicting typhoon tracks. Typhoon track forecast errors for weeks 1-2 in the ECEPS were analyzed by comparing the ensemble vortex tracks with the Joint Typhoon Warning Center (JTWC) best tracks. The forecast errors were decomposed into along-track errors (ATE) and cross-track errors (CTE). The analysis revealed negative mean ATEs, indicating a slower translation speed bias in the model. The mean along-track errors could reach up to 400 km for 168-hour forecasts following tropical cyclone formation. Given the substantial typhoon track forecast errors, the direct use of raw ECEPS rainfall forecasts for operational tropical cyclone forecasting presents challenges. To address this, a statistical Quantitative Precipitation Forecast (QPF) model has been employed to predict typhoon rainfall. The forecasting tools developed in this study are intended to be integrated into the CWA’s subseasonal typhoon forecast system, thereby supporting water resources management and disaster risk reduction efforts.