受到可預報度、模式能力和電腦資源的限制，中央氣象局因應不同的預報需求研發不同的模式預報技術，並產製相應的客觀預報指引。本文從實際預報作業的應用切入評析即時可用的定量降水預報客觀指引（Guidance for Objective QPF，GOF）在高影響天氣的預報特性，以做為防災決策適用性的參考。篩選在2017及2018年臺灣地區發生顯著降雨的個案（包括梅雨、西南氣流、颱風、午後對流與鋒面系統），收集各GOF產品進行定量降水校驗，並依據領先時間分為三個時段（即時預報、極短期預報及日預報時段）比較其差異，最後以2018年0823豪雨事件為例說明各GOF的預報特性。雷達外延預報在第1小時預報有最高的校驗評分及與觀測的相關性，是即時預報及防災預警的重要資訊。建立在資料探勘基礎所發展之iTEEN（integration on Taiwan Extended Ensemble Nowcasting）產品第1小時預報校驗評分和相關係數僅次於前者，其2-5小時的即時預報皆優於前者之第2-3小時預報，在各降水閾值皆具有高的可偵測率，但同時也呈現過度預報的特性。快速循環更新的對流尺度資料同化系統主要同化雷達和地面觀測資料，藉由即時改善初始場得到校驗評分高和偏倚分數小的預報，與觀測的相關性亦佳，能提供12小時的極短期預報。區域模式因對於對流天氣的可預報度較低，其極短期預報校驗評分浮動大且多半低於前述預報指引；在13-72小時的日預報方面，系集預報系統經後處理的機率擬合定量降水預報（probability match mean）產品和決定性預報相比，在降雨區域和型態上與觀測的相關性較佳，有助於得到稍高的校驗評分。

Due to the limitation of the predictability, model capabilities, and computational resources, the Central Weather Bureau has developed different modeling frameworks to requests from difference scope of end-users, and also provided corresponding objective forecasting guidance. To enhance our understanding of the quantitative precipitation forecast (QPF) ability of real-time operational forecast guidance in high-impact weather, this article selects major rainfall cases in Taiwan in 2017 and 2018 across various spatial-temporal scales, including Meiyu, southwest flow, typhoon, afternoon thunderstorm and frontal system. We collected and verified the various guidance for the objective QPF (GOF) according to the lead time categorizing into three periods (nowcasting, very short-term forecast and daily forecast period). The results of this study can be used as a reference for the applicability of disaster prevention systems. For the nowcasting, the GOF of radar extrapolation technique (QPEQPF) performs the best skills and high correlation to observations in the first 1 hour, which are essential information for real-time forecasting and disaster prevention and warning. The GOF of Integration on Taiwan Extended Ensemble Nowcasting (iTEEN) is based on data mining; the accuracy of its first-hour forecast is next only to the radar extrapolation forecast. The cases of 2-5 h nowcasting of iTEEN have a high probability of detection, which are better than the 2-3 h nowcasting of QPEQPF, but they also characteristics of over-prediction. The GOF of convective scale assimilation system, which assimilates radar and ground observation data with an hourly update cycle, produces 12 h forecast. The GOFs display high verification scores, smaller bias, and moderate correlation with observations and can provide very-short-term forecasts with high accuracy. In comparison, the regional model has low predictability for convective weather, and the GOFs reveal fluctuating verification scores. In terms of forecasts of 13-72 h, compared with deterministic forecasts, the probability-match-means products of the ensemble prediction system demonstrate a higher correlation with QPE, resulting in a higher verification score.