氣候變遷為目前世界各國所重視的議題,根據IPCC第四次之評估報告,氣候變遷對於未來氣候及降雨型態均會帶來影響,但由於全球環流模式(General Circulation Model,GCM)所制定的網格過於粗糙,若將GCM直接對於台灣地區未來氣候型態進行推估,勢必無法描述台灣的氣候特徵,因此發展一個適合台灣地區的統計降尺度模式有其必要性。 本研究發展一個以抽樣方式建模的二階段統計降尺度模式(Two-Stage Statistical Downscaling Model,TSSDM),以淡水、台中、高雄、花蓮4個氣象站為例進行案例研究,第一階段利用過去台灣地區的氣象站與GCM資料,以輻狀基底網路(Radial Basis Function Neural Nerwork)為核心結合基因演算法來建立空間降尺度模式(Spatial Downscaling Model,SDM),第二階段則利用機率分佈之概念進行時間之降尺度(Time Downscaling Model,TDM),由歷史逐日資料統計特性配合SDM輸出所得到的未來月資料轉換為逐日之降雨資料。模式建立的過程中則透過拔靴法重複建立模式來評估SDM模式之不確定性進而提高模式之可信度。最後將TSSDM模式與因子轉換法之結果做一比較,以評估未來氣候資料之不確定性。 模擬結果顯示未來短、中、長期各氣象站逐月平均降雨分別落於150~170毫米、70~186毫米、70~380毫米及80~240毫米之間,淡水測站夏冬兩季短、中、長期逐月雨量差異分別為41.55毫米、50.87毫米及42.54毫米,台中測站夏冬兩季短、中、長期逐月雨量差異分別為7.67毫米、-5.88毫米及-25.00毫米,高雄測站夏冬兩季短、中、長期逐月雨量差異分別為1.28毫米、-8.63毫米及-12.27毫米,花蓮測站夏冬兩季短、中、長期逐月雨量差異分別為-15.54毫米、-35.96毫米及-45.96毫米。未來1月份逐日平均及最大降雨除淡水測站模擬結果與歷史趨勢相似,其餘測站均有上升趨勢,未來7月份逐日平均及最大降雨淡水測站模擬結果較歷史趨勢上升、花蓮測站與歷史趨勢相似,其餘測站則有減少趨勢。
In order to understand the situations that rainfall and temperature may change in the future under the influence of climate change in Taiwan, this research develops a two-step statistical downscaling model (TSSDM) combining space downscaling model (SDM) and time downscaling model (TDM) with sampling theory. In this research, the meteorological stations of Tamsui, Taichung, Kaohsiung and Hualien are employed as cases studies. The rainfall and temperature data from the Central Weather Bureau (CWB), and the meteorological factors from General Circulation Models are collected first. Then the TSSDM is built within two stage: in the first stage, SDM is constructed by combining the radial basis function neural network (RBFNN) and the genetic algorithm (GA); in the second stage, with the historical daily rainfall data and future monthly rainfall data obtained from SDM, TDM is constructed using the concept of probability distribution to simulate future daily rainfall data. Meanwhile, the Bootstrap sampling method is used to estimate the uncertainty of SDM model. Finally, the future simulated daily rainfall by TSSDM model and by change factors (CF) method are compared to assess the uncertainty of future rainfall. Simulated results show that projected average monthly rainfall of Tamsui, Taichung, Kaohsiung and Hualien meteorological stations in the short, medium and long-term fall in 150 to 170 millimeter, 70 to 186 millimeter, 70 to 380 millimeter and 80 to 240 millimeter, respectively. The monthly rainfall difference between summer and winter in the short, medium and long-term are 41.55 millimeter, 50.87 millimeter and 42.54 millimeter in Tamsui meteorological station; 7.67 millimeter, -5.88 millimeter and -25.00 millimeter in Taichung meteorological station; 1.28 millimeter, -8.63 millimeter and -12.27 millimeter in Kaohsiung meteorological station; -15.54 millimeter, -35.96 millimeter and -45.96 millimeter in Hualien meteorological station. Projected average and maximum daily rainfall during January have similar tendency to historical data in Tamsui meteorological station, but have increasing one otherwise. Projected average and maximum daily rainfall increase in July in Tamsui meteorological station, having similar tendency to historical data in Taichung meteorological station, but decrease otherwise.