對氣候上，台灣中部地區年雨量約有1700mm~1800mm，理論上，應無缺水之虞。實際上，受限於降水集中在梅雨季、與颱風季的5-9 月，而10 月至次年的3 月為乾季，降水量僅佔本區全年總降水量的約14%(其中春雨（1-3 月）約佔本區全年總降水量的 10%)。就降水總量言，並不具絕對的重要性，但當梅雨及颱風雨不顯著時，春雨便顯得格外重要，如果前一年的颱風、梅雨雨量不足，且無春雨適時的補充，那麼形成乾旱的機率將大幅增加，不但嚴重影響一期稻作的秧期，連工業及民生用水都有缺乏之虞，此時，春雨便成了台灣一項非常重要的水資源資產，對於水資源的調配將扮演舉足輕重的角色。因此，若能有效掌握春雨的預報，將對水資源的調度提供預警與對策擬定提供助力。本研究即針對上述需求進行研究。雖然月、季雨量預報作業模式，在計算機功能大幅提升下，使得研究方向由單一模式轉到系集預報，大尺度的預報能力已有改善，但區域性的預報受限於模式解析度與降水尺度技術的開發，仍有改善的空間。本研究將採用（1）不同氣候動力氣候模式輸出產品進行系集預報（Ensemble Forecasts）（2）利用理論研究上已知影響之物理機制，挑選適當因子進行回歸分析，建立最佳預測迴歸模式。（3）以模式輸出統計MOS(Model Output Statistics)搭配物理因子，進行最佳化預測模式建立。並比較此三種模式的預報結果，尋求最佳降水預報組合與風險評估。Annual precipitation is about 1700mm~1800mm in the middle of Taiwan. Water is not a lack of this region in theory. But most of the precipitation concentrates in the Mei-Yu season and typhoon season (May ~ September), the rainfall between Jan. and Mar. comprises about 14%of the annual rainfall amount. If plum rains and typhoon rains brought little rainfall, the amount for spring rains will be more than important than the past. Then the opportunity of a drought for the region gets more and more. The result will not only affect the planting of rice in the beginning of the year, but also cause lack of domestic water and industrial water. So, if we can predict the amount of spring rains in time, it will improve the allocation of water resources for decision makers. This study will be done for the above purposes. Although the forecasts, which is in large scale, for month and season precipitation get better based on advanced computer equipments, regional predictions still are limited under resolution of model and technology of precipitation scale. However, the abilities of predict model on precipitation can be modified by some works. The discussions and applications for our study can be spilt into (1) ensemble forecasts with different dynamic climate model output product (2) pick proper factors from physical mechanisms to build regression model (3) take appropriate affecting factor and model output statistics to build optimal prediction model for precipitation. Last we shall find the best model among our trials, and evaluate the risks for those models. 研究期間 : 9808 ~ 9907
