| 摘要: | IPCC AR4指出20世紀全球平均表面溫度上升0.74 ℃ ± 0.18 ℃，並出現加速增溫的現象；而未來21世紀，無論人類是否積極進行溫室氣體減量，溫度都將會持續上升。IPCC利用設定數種「情境假設」(未來溫室氣體可能的排放濃度假設)針對未來全球氣候變遷做模擬。但IPCC中所討論的均為全球性的氣候變化，若想進一步了解在台灣及鄰近地區的區域氣候變化，會因氣候模式空間尺度的問題，無法直接針對區域的氣候變遷做出詳細的描述。而本研究則試圖在已有的IPCC全球海氣耦合模式模擬的資料下，以探討整體模擬結果的統計特徵作為重點，選擇使用「機率密度函數」來做為分析的方法，期望可藉由機率密度函數的分布來了解台灣及鄰近地區在氣候變遷的統計特性。 本研究選擇IPCC中的模式(日本的MIROC3.2(medres)，簡稱為NIES；德國的ECJAM5_MOI-OM，簡稱為MPI)在A1B情境假設下所模擬的資料進行分析。在單日最高溫中，計算當溫度大於或等於30 ℃的高溫發生機率，以NIES來重建1980~1999年間之氣候資料顯示，高溫的發生機率不到1 ％，但在A1B情境下以NIES模式所做的模擬結果顯示，在2046~2065年時期此高溫發生機率增加到約15 ％，而到了2081~2100年更增加為約36 ％。而在MPI模式中，重塑的20世紀末期資料顯示高溫發生機率不到9 ％，而在A1B情境下的模擬21世紀末期結果顯示，其發生機率變為將近47 ％。 在單日最低溫的部份，計算當溫度小於或等於16 ℃時所對應的溫度，以NIES模式重塑在1980~1999年間的資料顯示，小於或等於16 ℃的低溫發生機率約為4 ％，而在模擬21世紀中期的結果顯示此發生機率就降至不到1 ％，而到了21世紀末期時此機率又略為往下降了一些。在MPI模式中，重建的20世紀末氣候資料顯示，在小於或等於16 ℃的低溫發生機率僅只有2 ％，而模擬的21世紀末期結果顯示此發生機率已變為0 ％。 而降雨方面，兩模式的機率密度函數圖沒有太顯著的改變。在小雨（小於或等於5 mm day-1）的部份兩模式均沒有顯著改變的趨勢；而在大雨（大於或等於40 mm day-1）的部份，兩模式均呈現增加的趨勢。此外，在年平均總雨量的模擬結果兩模式是呈現相反的趨勢。 在分析此兩模式的未來模擬結果時，需將在重塑過去氣候與觀測資料間的誤差考慮進去，以機率密度函數的角度來看，兩模式在三個參數中的技術得分均在0.8左右，表示模式在模擬未來的機率密度函數結果有一定的可信度。而本研究的結果也能說明雖然氣候模式的模擬有其不確定性，但仍可經由機率密度函數等統計方法做出定量的分析。 In the Fourth Assessment Report (AR4), the Intergovernmental Panel on Climate Change (IPCC) concluded that the global mean surface temperature have risen by 0.74 ℃ ± 0.18 ℃ over the 20th century, and the warming trend is accelerating. And in the 21st century the surface temperature is expected to rise continuously. IPCC adopts several types of “Special Report on Emissions Scenarios” (SRES) to simulate the future global climate change. The SRES are constructed based on potential greenhouse gas emission strength in the future. The spatial scale of climate change is global in the IPCC. We can’t directly describe in detail how the regional climate change by using Atmosphere-Ocean General Circulation Models (AOGCMs). If we want to understand the regional climate projections, the questions are then we have to use the dynamic downscaling. In this study, the statistical features of the whole simulated result from IPCC are focal point. So we used “probability density function” method and hope we can examine the statistical features of the climate change characteristics around Taiwan by means of the probability density function. Two models within IPCC are selected for this study. They are MIROC3.2(medres) (abbreviated as NIES) from Japan and ECJAM5_MOI-OM (abbreviated as MPI) from Germany respectively. The simulated data come from the models with the SRES A1B. For daily maximum temperature, we calculate the probability of the temperature which is greater than or equal to 30 Celsius. The probability value in NIES model indicates that 1980-1999 data don’t reach 1 ％. But in the SRES A1B, the simulated probability in 2046-2065 is about 15 ％. In 2081-2100, the probability is 36 ％. For MPI model, the probability from 1980-1999 data don’t reach 9 ％. And in the SRES A1B, the probability from simulated 2081-2100 data is 47 ％. For daily minimum temperature, we calculate the probability of the temperature which is smaller than or equal to 16 Celsius. The probability value in NIES model during 1980-1999 period is about 4 ％. But in the SRES A1B, the simulated probability in 2046-2065 is less than 1 ％. In 2081-2100, the probability is even smaller decreasing. For MPI model, the probability from reproduced 1980-1999 period is only 2 ％. And in the SRES A1B, the probability from simulated 2081-2100 period almost is nonexistent. For precipitation, the PDF of two models don’t change much. In light rain (the rainfall is smaller than or equal to 5 mm day-1), the change of two models are not obvious. And during the heavy rain (the rainfall are greater than or equal to 40 mm day-1), the increasing trends embedded in two models. However for annual mean rainfall, the trends of two models are opposite. When we use the simulated result of two models, we should consider the difference between the data of model and observation. By PDF, the skill score of two models are around 0.8 based on the analysis from three variables. The meaning is the simulated PDF in the future is a reasonable methodology. In this study, we can show even though the simulations of the climate models are uncertain, we can still adopt the probability density functions method and come out quantitative analysis that can be useful in understanding the regional climate change characteristics. |
