CMIP5年代氣候變化模擬

Abstract

年代際變異(Decadal Variability)是一種長時間的震盪現象，通常出現在海溫的空間結構上。過去的研究針對區域空間的年代際分析，主要集中在大西洋地區的AMO，以及太平洋年代際變異(PDV)上。這些年代際結構會影響地區的氣候特徵，IPCC的第五次評估報告也將年代際變異模擬視為未來的研究重點。 本研究利用旋轉經驗正交函數(REOF)的方法，分析觀測海溫以及CMIP5氣候模式在年代際時間尺度下的模擬，因此將資料進行了九年滑動平均和去趨勢的處理。在觀測海溫的REOF分析中，北大西洋的AMO結構出現在REOF1。東太平洋Nino地區以及印度洋的年代際變異出現在觀測的REOF2。SPDO以及熱帶中太平洋年代際變異則是第三個主要模態。PDO的結構則分散在REOF2和REOF3。時間序列的分析中，RPC1呈現多重年代際震盪且符合AMO的相位紀錄。RPC2和RPC3則在二十世紀後半葉都有訊號放大的現象。 以觀測資料的REOF為基準，使用CMIP5三種不同背景：(1) Pre-industrial代表模式本身的內部變異。(2) Historical run 為1850~2005年間的帶入觀測紀錄，包含人為活動和自然變異。(3) RCP8.5 run為模式在世紀末溫室氣體長波輻射達到8.5W/m2的暖化情境。來評估模式模擬年代際變異能力。在Pre-industrial 模擬中，多數模式在赤道Nino地區有明顯的年代際結構，而一般認為屬於自然變異的AMO，則沒有出現在模式的內部變異裡。在Historical run的模擬分析上，模式間差異大，主要模態的個數從2到4個不等，呈現了較多元的年代際結構。模式在RCP8.5的主要年代際結構為較大範圍的訊號，時間序列上則有先降後升的情形。將模式的模態和觀測結果進行型態相關(pattern correlation) 分析，看出模式對於觀測結構的相似程度，以及模式所對應較佳的模態，並且挑選出模擬年代際結構較佳的模式，在本研究中，HadGEM2-CC在Historical的模擬和觀測結果最為相似。 在年代際時間尺度的變異是否顯著的分析中，將資料分成年際、年代際和趨勢三種時間尺度，並且計算占總變異的比例，分成box-plot以及variance map兩部分。Box-plot是將資料進行排序後，所做的統計量分析結果，在觀測資料和模式的比較中，Pre-industrial有最小的平均趨勢變異比例，觀測的結果則和Historical run 相似，在RCP8.5中，趨勢占總變異比例平均達80%以上，而模式的年代際變異則非常小，代表模式對於溫室氣體的增加十分敏感。Variance map則可以看出空間變異比例分布，在觀測中，年代際變異最顯著的地方為北半球海域，在幾個模擬較好的模式也有類似的結果，另外在RCP8.5，則是由趨勢主宰了變異比例，而在二氧化碳未來的變化，其趨勢並非呈現線性增長，而是類似二次函數的圖形，因此扣除掉線性趨勢後，先降後升的變異量投影到時間序列的分析上，影響了時間序列的表現。 關鍵詞 : 年代際變異、大西洋多重年代震盪、太平洋年代變異、旋轉經驗正交函數、型態相關

Decadal variability is an oscillation for more than ten years, usually appears in the ocean. In the Atlantic, AMO is the main multi-decadal variability. There are several decadal oscillations in the Pacific, which we call Pacific decadal variability (PDV). These decadal structures may impact on regional climate. IPCC will take decadal variability as an important issue and publish the result in the fifth assessment report (AR5). This study uses Rotated-EOF (REOF) method, analyzes decadal variability in observation data and CMIP5 models. In observation data, AMO structure appears in REOF1. Decadal variability of Nino region, north Pacific and Indian Ocean are the 2nd mode in REOF analysis. SPDO, PDO and decadal variability in central Pacific appear in REOF3. The rotated principal components (RPCs) also match with the previous study. In CMIP5 models, we use three different model experiment designs: (1) Pre-industrial, refer as the internal variability in models. (2) Historical run, with the observation record from 1850 to 2005. (3) RCP8.5 identifies a concentration pathway that approximately results in a radiative forcing of 8.5 W/m2 at year 2100. In the pre-industrial run, most of the CMIP5 models show the decadal variability of El-Nino region, but the AMO structure which consider as a nature mode in decadal variability, does not appear in model’s pre-industrual run. In the Historical run, decadal structure shows variety types in each model. In the model’s RCP8.5 run, main decadal structure shows large region in global, with an upward parabola time series. We use pattern correlation analysis, and choose the models which are similar with observation data. In this study, historical run in HadGEM2-CC models are similar with the observation data. On the analysis of variance, we separate total variance to three different time scales: annual, decadal and trend. We compute the ratio of the three time scales and the total variance. The smallest ratio of trend variance is model’s pre-industrial run, but most significant in RCP8.5 run which the mean is more than 80% in box-plot analysis. In variance map, decadal variance ratio is most significant in North Hemisphere Ocean in observation data; it can be simulated in some model’s historical run. In RCP8.5 run, the decadal variability ratio is less than 10% in most regions. We find that the total mass of CO2 in RCP8.5 run, ascending with a parabolic curve. We remove the linear trend of total mass of CO2, the residual forcing could project on decadal variability and affect the RPC performance. Keywords: Decadal variability, Atlantic Multi-decadal Oscillation, pacific decadal variability, rotated-EOF, pattern correlation