台灣的年平均降雨量約為2500公厘，但在時間和空間的分配上並不均，主要降雨集中在颱風降水以及5-6月的梅雨鋒面降水，因此梅雨是僅次於颱風的重要降水來源，就北部的石門水庫集水區而言，除了颱風與梅雨外，2-4月的春雨為另一主要降水來源，因此颱風、梅雨、春雨的變化將衝擊石門水庫集水區之水資源供給。以2015年乾旱事件為例，主要來自2014年的颱風事件較少，降雨量明顯不足，加上2015年的春雨不如預期，使得石門水庫供水區水資源短缺，在4月份進入第三階段供5停2的限水措施，直到5月的梅雨鋒面解除旱象。 本研究使用1982-2011年的觀測資料做為氣候背景值，分析歷年梅雨期間石門水庫集水區的水文特徵，探討梅雨期降雨的重要性、降水量的變化與梅雨主要影響的時間。以2015年乾旱事件做為研究案例，結合中央氣象局提供未來三個月逐月之等級機率預報產品，包含動力區域降尺度預報及統計降尺度預報，透過氣象合成模式(WGEN)、水文模式(GWLF)以及水資源系統動力(System Dynamics Model)模擬石門供水區在梅雨時期可能的水資源情形以及受到乾旱影響的程度。 研究結果顯示預報產品在5月的降雨量預報則有較佳的預報能力;在6月的降雨量有高估的情形，導致入流量推估較大。但因乾旱的發生多因持續且累積的偏低降水所造成，單一月份較佳的預報仍無法提升在水資源管理應用之效益，仍需仰賴預報產品的準確率提升。而從水庫管理端而言，僅利用歷史氣候特徵與現有水庫水位來決定水庫限水措施，則是顯得過於保守且往往容易錯失決策時機與該有之應變措施。 ;The annual rainfall in Taiwan is about 2500 mm/yr, but its spatial and temporal distribution has great discrepancy and is mainly contributed by typhoon events and Plum rains season. The Plum rainfall is the second important source of fresh water except typhoon rainfalls in the island. In addition to typhoon and Plum rainfalls, the amount of spring rainfall is also critical to the water resources management of the Shihmen Reservoir in northern Taiwan. The 2015 drought in northern Taiwan was mainly caused by both typhoon rainfalls in 2014 and spring rainfalls in 2015 were significantly lower than climatology. Restriction on water supply with no water for 2 days in a week was applied in April and back to normal until the hit of Plum rainfall in May. In this study, observed data of 1982-2011 was analyzed to understand the importance of Plum rainfall, including amount, variation, duration, and onset, to the Shihmen Reservoir Watershed. Then the 2015 drought occurred in the Shihmen Reservoir Watershed was investigated with seasonal climate forecast products of the Central Weather Bureau (CWB). The seasonal climate forecasts, including dynamic regional downscaling forecast and statistical downscaling forecast, by the CWB were used to provide outlooks of monthly precipitation and temperature with 1-3 months of leadtime. A Weather Generator (WGEN) was applied to generate multimembers of daily precipitation and temperature to drive the Generalized Watershed Loading Functions (GWLF) for reservoir inflow predictions. Finally a System Dynamics Model was employed to simulate water allocations of the regional water supply system in association with the Shihmen Reservoir Watershed. The results show that precipitation forecasts have better forecasting skills in May, but overestimate in June causing larger estimation of inflows. Since drought is caused by persistent low rainfall lasting for several months, like the case in 2015, overall improvement of seasonal climate forecast skills are required to build confidence to water sectors. On the water resource management perspective, it is too conservative to rely on current reservoir capacity and climatology information to the decision making of reservoir operations. Climate outlooks certainly should be involved in reservoir operations subject to improved skill can be demonstrated.
