【作者】 侯威；

【导师】 丑纪范； 封国林；

【作者基本信息】 兰州大学 ， 气象学， 2009， 博士

【摘要】 极端气候事件及其衍生灾害对社会和经济的影响力和破坏力越来越严重，因此对极端气候事件发生、发展机制的研究，以及对极端气候事件的评价、预警都是非常重要的研究领域。目前国际上关于极端气候事件的研究重点主要为：1、能否检测极端气候事件发生的变化?2、极端气候事件的变化根据自然气候的变化是否异常?3、通过什么方式减少认识上的不确定性?4、能否提供证据把观测到的极端气候变化与人类活动的影响相联系?本文的研究内容主要针对“通过什么方式减少认识上的不确定性?”和“能否提供证据把观测到的极端气候变化与人类活动的影响相联系?”这两个问题。对于“通过什么方式减少认识上的不确定性?”这个问题而言，对于极端事件认识上最初的或是最基本的不确定性来源于极端事件的定义，目前国际上最多见的极端事件定义是采用百分位值阈值法，主要存在着三个方面的缺陷，即定义不统一、样本量要求不确定和结论不唯一。对于“能否提供证据把观测到的极端气候变化与人类活动的影响相联系?”这一问题，目前主要研究各种极端事件或极端气候指数发生的频次和强度，但单个特征只包含了极端事件某一个方面的信息，而不包含极端事件的整体信息，也就无法从整体上把握极端事件的变化特征。针对第一个问题，本文使用去趋势波动分析(DFA)方法和替代数据法相结合的思路，得到一种新的、具有物理背景的、统一定义的确定极端事件阈值的方法：DFA-S方法，明确指出了极端事件和非极端事件之间的临界值，同时给出了本方法所需要的数据量即采样长度，得到的阈值是确定的、唯一的；针对第二个问题，本文将阈值、平均极端高温度数与阈值的差值和极端高温发生次数综合起来，从可预报性的角度给出了极端事件综合指标的定义。本文工作为极端气候事件的研究提供了一条新的思路和方法。主要结果和结论如下：(1)基于信息论基本原理，采用符号分析的方法计算互信息函数，确定DFA方法中参数s的选取，这一算法完全基于数据本身，从而具有较好的客观性，而且当序列的长度发生了变化，但对于同一系统，使用本算法得到的结果是稳定的。互信息函数算法也可以判断所研究序列的长度是否符合计算DFA指数的要求，判断某一长度的序列是否适用于DFA方法。(2)使用将去趋势波动分析(DFA)方法和替代数据法相结合的思路，寻找一个临界值，在小于该临界值的数据点位置不变时，对大于该临界值的数据点，无论这些数据点彼此之间的位置如何变化，对整个序列的DFA指数无影响，认为该临界值为阈值，将其称为DFA-S算法。通过理想数据和实际数据从不同的角度进行了反复检验，验证了DFA-S方法的有效性。(3)将阈值看作是速度模；平均极端高温度数与阈值的差值作为温度模；极端高温发生次数视为温度梯度模，将阈值、平均极端高温度数与阈值的差值和极端高温发生次数综合起来，从可预报性的角度给出了极端事件综合指标的定义。(4)使用DFA-S算法得到了中国1961-2000年间，极端高温、极端低温和极端降水事件的阈值，分析了它们的区域分布特征；基于得到的阈值，分析了极端高温、极端低温和极端降水事件1961-2000年间的发生频次和强度，分析了其气候背景和可能的影响因子；进一步得到衡量中国1961-2000年来的极端高温、极端低温和极端降水事件综合强度的综合指标，按其值将各类极端事件分为不同等级的区域，为检测和预警提供指导。(5)将中国165站1961-2000年的极端高温、低温和降水事件的年综合指标的标准化场作为变量场，利用EOF分析方法分析了极端高温、低温和降水事件综合指标的年际时空分布特征及其变化，重点分析了前四个空间分布模态及其时间变化。进一步得到影响中国极端高温、低温和降水事件综合指标的夏季海温异常关键区和冬季海温关键区，采用SVD分析方法，研究了海温异常对中国极端高温、低温和降水事综合指标异常的影响。对于实际资料的分析部分，为了显示方便，本文中所有涉及到温度、降水数值的图示，如无特殊说明，真实值均为图示值乘以0.1。更多还原

【Abstract】 Due to the increasing social and economic loss caused by extreme climate events and itsderivative disaster, the study of mechanism of occurrence and development of extreme climateevents, and its evaluation and early warning system is the fastest growing area of interest amongglobal change science. This questions, which are the hotspots of extreme climate events research,are as following:(1) Can we detect the change of extreme climate events?(2) Is climate change abnormal?(3) How can we reduce the uncertainties of our understanding?(4) Can we afford evidence as to correlate extreme climate change with influence of humanactivity?This contribution is especially interested in two questions: "How can we reduce theuncertainties of our understanding?" and "Can we afford evidence as to correlate extreme climatechange with influence of human activity?".When it comes to the question "How can we reduce the uncertainties of our understanding?",the most fundamental uncertainty of understanding root in the definition of climate exremes. Themost common used definition of climate extremes nowadays is method of percentile, and itcontains three major defect, e.g. the disunity of definition, uncertainty of volume of sample, thenonuniqueness of conclusion. As to "Can we afford evidence as to correlate extreme climatechange with influence of human activity?", existing research work are mainly focused on theextreme events or its index’s frequency or amplitude, however, the conclusion drawn fromunivriate time series analysis only contains restricted information and can not reflect theinformation of extreme events as a whole, and thus can not grasp the overall characteristics ofextreme events’ variation.In this paper through the combination of Detrended Fluctuation Analysis (DFA) andSurrogate Data method, we develop a new method of threshold of extreme events detection, e.g.DFA-S method, which has a certain phsical background. The obtained critical value of extremeevents are definite and unique and the required length of time series is the volume of sample. As to the second question, we integrate the threshold with the mean, variance and total number ofextreme high temperatures, and give a definition of extreme events complex indices from theangle of predictability. This work illustrates a brand new method and way of thinking of extremeclimate events research. The results and conclusions can be summarized as follows:(1) Based on the theory of mutual information, determine the parameter of DFA method bycompute mutual information function using symbolic analysis. This algorithm entirely depends onthe data itself and thus immune to the change of sample volume and its stability is comparativelygood. The algorithm of mutual information function can be use to decide if the length of timeseries fulfill the requirement of DFA index computation and whether a time series of certain lengthis appropriate for DFA method.(2) Combine DFA method with Surrogate Data method, e.g. once the DFA index of time serieremains unchange with values exceed a critical value, we decide the critical value as the thresholdwe need, and this strategy of selection is called DFA-S algorithm. We also validate theeffectiveness of DFA-S method through extreme events detection using artificial series andobservational data from various angles.(3) Using threshold as velocity mode, the difference between threshold and average numberof extreme high temperatures as temperature mode, the number of extreme high temperatures astemperature gradient mode, and thus integrate the threshold with the mean, variance and totalnumber of extreme high temperatures, and give a definition of extreme events complex indicesfrom the angle of predictability.(4) Obtained the thresholds of extreme high-low temperature and extreme precipitationevents from 1961 to 2000 of China through DFA-S method and anlyzed its spatial-temporalcharacteristics of distribution. We evaluate the frequency and amplitude of extreme high-lowtemperatures, extreme precipitation events from 1961 to 2000, analyzed its climate backgroundand possible impact factor. Furthermore, we study the complex indices of amplitude of extremehigh-low temperatures, extreme precipitation events from 1961 to 2000, and making classificationbased on its value of complex indices, and give a certain instruction for detection and earlywarning.(5) Using normalized annual complex indices of extreme high-low temperatures and extremeprecipitation events of 165 stations from 1961 to 2000 in China as fieldvariable, analyzed the characteristics of spatio-temporal distribution of extreme high-low temperatures and extremeprecipitation events’ complex indices through EOF analysis and especially focused on the fourpreceding mode of spatial distribution and its temporal variation. Furthermore we obtained thesummer and winter SST’s sensitive area of extreme high-low temperatures and extremeprecipitation events’ complex indices. Through SVD decomposition we study the influence ofabnormal of SST on complex indices extreme temperature and precipitation events in China.更多还原