本研究主要是評估中央氣象局(Central Weather Bureau，CWB)與美國國家環境預報中心(National Centers for Environmental Prediction，NCEP)之全球系集預報系統在東亞地區的二週降雨預報表現，包含系集離散度之合理性、系集預報與系集機率預報之評估，同時透過模式後處理由系集導出三種單一預報，並評估其在臺灣鄰近區域之強降雨預報能力，目的在於評比兩系集系統的預報表現，並提供使用者較佳的降雨預報指引。評估結果顯示: (1)兩模式之系集降雨預報有離散度不足的現象，以夏半年最明顯，且CWB模式較NCEP嚴重。(2)第二週降雨機率預報的可信度不高，但在冬季有良好的潛在可用性。(3)夏半年和梅雨季可採用機率擬合平均(Probability-Matched Mean, PM)，而冬半年採用第80百分位(Quantitative Precipitation Forecast Percentile, QPF Percentile)作為降雨預報指引。

Focusing on week-2 precipitation forecasts of ensemble forecast systems of Central Weather Bureau (CWB) and National Centers for Environmental Prediction (NCEP), this study evaluates the spread-skill relationship, ensemble probabilistic forecasts, and dichotomous forecasts derived from ensemble forecasts. The aim is to compare the performance of both forecast systems and provides more suitable forecast guidance for users. Results indicate that 1) both ensemble prediction systems are under-dispersive especially in summer, and the CWB forecast system appears to have more under-dispersion relative to the NCEP; 2) week-2 probabilistic precipitation forecasts do not have high reliability, but have good potential usefulness in winter; 3) the better post-processing method for precipitation forecast guidance is the PM (Probability-Matched Mean) during summer and Mei-Yu season, and the 80th percentile of ensemble forecasts in winter.