中央氣象局使用WRF區域模式，透過初始隨機擾動、邊界擾動和物理參數法擾動等，建構成20組成員的系集預報系統，並自2011年開始正式上線作業（Li et al. 2020）。然而，如何從系集巨量預報資料中萃取出有用的訊息，產製有用的決定性預報產品，仍是重要的課題。機率擬合（Probability-Matched Mean，PMM）定量降水預報技術的發展，目的是要改善系集平均雨量預報過低的缺點，藉由重新分配系集降水頻率分布，在系集平均的基礎上重建雨量預報的量值。研究顯示，由小累積區間PMM降水累加成較大區間時存在過度預報的問題，而且累加次數越多，其過度預報的情形會更為明顯，這使得在使用逐時PMM降水累加成較長時間的累積雨量時，產生極大誤差，也限制了PMM降水產品的應用。為此，本文探討上述過度預報形成的原因，並提出以系集模式預報的總雨量（PMMᵀ）來進行PMM雨量的計算。2018年梅雨季個案結果顯示PMMᵀ可以明顯改善原本PMM累加過程所導致過度降水預報的現象，校驗結果亦顯示PMMᵀ可以得到最好的預報能力。另外，PMMᵀ由不同累積區間累加所得到的累積雨量其結果都是一樣的，相較於傳統PMM由不同累積區間累加所得到的結果都不一樣，這是PMMᵀ的優勢之一。

The ensemble forecast is expected to provide the uncertainty information due to the limited predictability from the deterministic forecast. However, how to derived useful deterministic forecast products from the large ensemble dataset is also an important issue. The Probability Matched Mean (PMM) rainfall product derived from the ensemble quantitative precipitation forecasts is widely applied among the operational centers. Studies have shown that there is a problem of over-prediction when PMM precipitation accumulated from a small interval into a larger interval. The more accumulation times result in the more obvious on the over-prediction. In particular, a large over-prediction error occurs as accumulate the hourly PMM rainfall to a long duration period. This is a critical situation to limit the feasibility of the PMM QPF application. To this end, this paper explores the reasons for the above-mentioned over-prediction issue and proposes a new algorithm to calculate the PMM rainfall based on the total rainfall (PMMᵀ) forecasted from the ensemble model. From the case study, the new algorithm shows that it can eliminate the over-prediction problems in a reasonable way. The new method also provides better QPF performance from the two-month verification. In overall, the revised PMM algorithm proposed in the paper is helpful to improve the ensemble QPF performance, moreover, add on the flexible values on the application of the disaster prevention.