基于数据驱动的山区暴雨山洪水沙灾害易发区早期识别方法研究  

Early recognition of the mountainous areas susceptible to flash flood and sediment disasters during rainstorms:Data-driven methods

作　　者：刘海知 徐辉 包红军[1,2] 宋巧云 鲁恒[3,4] 闫旭峰 狄靖月[1,2] 杨寅 LIU Haizhi;XU Hui;BAO Hongjun;SONG Qiaoyun;LU Heng;YAN Xufeng;DI Jingyue;YANG Yin(National Meteorological Centre,Beijing 100081,China;CMA-HHU Joint Laboratory for HydroMeteorological Studies,Beijing 100081,China;College of Water Resource&Hydropower,Sichuan University,Chengdu 610065,China;State Key Laboratory of Hydraulics and Mountain River Engineering,Sichuan University,Chengdu 610065,China)

机构地区：[1]国家气象中心,北京100081 [2]中国气象局-河海大学水文气象研究联合实验室,北京100081 [3]四川大学水利水电学院,成都610065 [4]四川大学水力学与山区河流开发保护国家重点实验室,成都610065

出　　处：《气象学报》2024年第2期257-273,共17页Acta Meteorologica Sinica

基　　金：国家自然科学基金气象联合基金项目(U234220088);国家重点研发计划项目(2019YFC1510702);高原与盆地暴雨旱涝灾害四川省重点实验室开放研究基金课题(SZKT202308);中国气象局创新发展专项(CXFZ2022J019)。

摘　　要：针对山洪灾害防治研究工作中只关注暴雨-洪水的作用,忽视泥沙淤积导致的洪水-泥沙耦合致灾的问题,重建考虑松散固体物源储量空间异质性的影响因子体系,面向山区小流域复杂下垫面环境进行敏感性分析,利用地理信息空间分析、多重共线性检验计算影响因子贡献度指标,通过不同类型的贡献度-集成学习耦合算法对阿坝州5250条小流域山洪水沙灾害易发度进行识别,构建基于数据驱动的山区暴雨山洪水沙灾害早期识别方法。结果表明:山洪水沙灾害在空间上表现出一定的聚集性,影响因子特定区间对于灾害发生具有更高敏感性,部分影响因子对于灾害发生的敏感性规律具有相似性。阿坝州东部、中南部部分地区以及西北部少部分地区为高易发区,与固体物源累积高频区较为接近,洪水-泥沙耦合致灾概率相对较大,较低和低易发区主要分布在阿坝州西部和西南部地区,与固体物源累积高频区重叠度较小,洪水在致灾过程中起主导作用可能性相对较大。相对于山洪风险调查评估研究结果,基于数据驱动的山洪水沙灾害易发性早期识别结果的高易发区灾害密度更大,高风险覆盖度提高23.2—45.4个百分点。In the field of flash flood disaster prevention study,great attention has been paid on the role of heavy rainfall and flooding,yet the coupling of flooding and sediment caused by silt deposition is largely neglected.This study revises the impact factor system by considering spatial heterogeneity of loose materials deposit,performs sensitivity analysis on complex underlying surface environment in mountainous watersheds,and utilizes the spatial analysis in geographic information systems and multicollinearity test to calculate the contribution indexes of various impact factors.A method for early identification of flash flood disaster is constructed by coupling different types of contribution-integrated learning algorithms,which is applied to identify the proneness to flash flood disaster in 5250 small watersheds in Aba prefecture.The results show that flash flood disasters exhibit certain aggregation in space,and the impact factors within specific ranges are more sensitive to disaster occurrence.Some impact factors share similar sensitivity patterns towards disaster occurrence.The eastern and central-southern areas and a small part of the northwestern area of Aba prefecture are highly prone areas,which are relatively close to the high-frequency area of solid material source,where relatively larger probability of flooding-sediment coupling disaster can be found.Lower prone areas are primarily distributed in the western and southwestern regions of Aba prefecture,where the overlap with the high-frequency area of solid matter source is relatively small and the flood tends to play a dominant role in the disaster process.Compared with the results of flash flood risk survey and assessment,the disaster density in high-prone area derived from results of data-driven early identification method is larger,and the high-risk coverage is increased by 23.2-45.4 percent.

关 键 词：山洪水沙 易发性 影响因子 早期识别 集成学习 

分 类 号：P642[天文地球—工程地质学]