基于经验正交分解的中国日降水随机事件集构建  

Generation of Stochastic Daily Precipitation for China Based on Empirical Orthogonal Function Analysis

作　　者：杨一飞 方伟华[1,2,3] 郑金丽 付婧瑄 Yang Yifei;Fang Weihua;Zheng Jinli;Fu Jingxuan(State Key Laboratory of Earth Surface Processes and Resource Ecology(ESPRE),Beijing Normal University,Beijing 100875,China;Key Laboratory of Environmental Change and Natural Disasters,Ministry of Education,Beijing Normal University,Beijing 100875,China;Academy of Disaster Risk Science,Faculty of Geographical Science,Beijing Normal University,Beijing 100875,China)

机构地区：[1]北京师范大学地表过程与资源生态国家重点实验室,北京100875 [2]北京师范大学环境演变与自然灾害教育部重点实验室,北京100875 [3]北京师范大学地理科学学部灾害风险科学研究院,北京100875

出　　处：《热带地理》2025年第4期589-604,共16页Tropical Geography

基　　金：国家重点研发计划资助(2023YFC3008505)。

摘　　要：历史降水数据是评估干旱和洪水等灾害风险的基础,但是历史数据无法涵盖未来更极端的降水情况,而目前大范围格网单元的随机降水生成方法尚不完善。为此,文章旨在基于历史降水信息生成具有空间相关性的随机降水,以提升灾害风险评估的可靠性。基于经验正交分解以及主成分系数概率拟合等方法,探索了中国0.1°格网尺度日降水随机事件集的生成方法和技术流程。首先,基于经验正交分解方法,对中国1961-2022年共62年逐日降水数据进行分解,对于年内任意一天均形成62个空间模态及对应的模态系数;其次,利用多种概率分布函数对各日的模态系数进行概率分布拟合,并为每日选择出一个最优拟合函数;然后,基于各日模态系数概率分布,并选择历史模态系数的最大值、最小值的2倍作为阈值上下边界范围,进行多年的日降水情景蒙特卡洛抽样;最后,利用每日62个空间模态及随机模态系数,生成各年逐日随机降水事件集。为比较历史降水、随机降水特征的一致性和差异性,模拟生成5000个年份的日降水事件,并用最大值、平均值、标准差、典型重现期降水、空间相关性5个统计特征值进行对比分析。结果显示:1)随机降水较好地保留了历史降水的强度-概率特征,在格网尺度上二者平均值差异<0.9 mm几乎可以忽略,10、20和50 a一遇降水强度差异均<15%,二者标准差的差异均<8%。2)随机降水有效地扩展了年最大值上限,在差异最大的网格上,随机降水最大值比历史降水提高了36%。3)随机降水较好地保留了空间相关性特征,中国所有格网的逐日莫兰指数和皮尔逊相关系数,最小值也分别>0.96和0.95。基于经验正交分解的中国日降水随机事件集,可为后续量化灾害风险评估提供良好的降水数据基础。Historical precipitation data are crucial for assessing the risks associated with natural disasters such as droughts and floods.However,some extreme precipitation scenarios may not have been included in historical records,particularly in China where the observed precipitation time series is relatively short compared to the return periods of rare extremes.This limitation poses a considerable challenge in disaster risk assessment,because the absence of data on certain extreme events can lead to risk underestimation.Therefore,the generation of spatially correlated stochastic precipitation events based on historical data is a key issue in disaster risk assessment.Current methodologies tend to focus on generating stochastic precipitation events for either a single site or a small number of sites.However,methods designed to generate stochastic precipitation events on large-scale grids have not yet been fully developed.To address this gap,we aimed to explore a method for generating daily stochastic precipitation events set at a 0.1°grid scale nationwide based on empirical orthogonal function(EOF)analysis and probabilistic fitting of principal component coefficients.We applied the EOF analysis method to decompose daily precipitation data for China from 1961 to 2022(62 years).For each day of the year,62 spatial modes and their corresponding mode coefficients were generated.Multiple probability distribution functions were used to fit the probability distributions of the mode coefficients for each day,with the optimal fitting function selected for each day.Based on these probability distributions,thresholds were set using twice the maximum and minimum values of the historical mode coefficients as the upper and lower boundaries,respectively.Monte Carlo sampling of daily precipitation scenarios was conducted using the 62-year historical data(1961-2022).Finally,using 62-year historical data(1961-2022),we performed Monte Carlo sampling to generate daily precipitation scenarios.To compare the consistency and differences between

关 键 词：日降水 经验正交分解 随机事件模拟 时空相关性 

分 类 号：P426.616[天文地球—大气科学及气象学]