暴雨条件下的辽宁省城市承载力评价研究——基于随机森林算法

Study on urban carrying capacity under rainstorm condition in Liaoning Province,China——Based on random forest algorithm

暴雨是对城市有较大影响的一种极端气候事件,从暴雨影响的视角研究城市承载力,对丰富城市承载力研究和缓解城市内涝具有重要的理论和实践意义.将暴雨自然属性与城市社会经济状况相结合,构建暴雨发生条件下城市承载力评价指标体系,划分承载力等级标准,采用随机森林算法对辽宁省2001—2015年暴雨条件下的城市承载力进行实证分析,并通过与BP神经网络法所得结果进行对比,验证随机森林算法的适宜性.结果显示:(1)时间上,大连、鞍山、盘锦、沈阳各市暴雨条件下城市承载力等级逐渐增加;营口、丹东、锦州、朝阳、阜新各市承载力等级呈降低趋势;葫芦岛、辽阳、抚顺、铁岭、本溪各市则以稳定不变为主,个别年份存在较小波动,各市承载力整体水平差距较大.(2)空间上,暴雨条件下的城市承载力体现为由西部至东部的"增加—减少—增加"趋势.(3)各市暴雨条件下的城市承载力影响因子不尽相同,其中,建成区面积、人口城市化、排水管长度、水域面积、城市基础建设投资是关键因子.(4)随机森林模型泛化误差率(OOB)为9.32%,分类具有精度高、误差小、运作方便的特点,与BP神经网络相比更适用于承载力评价研究.

Rainstorms are extreme climate events that have greater impacts on cities.Studying the urban carrying capacity from the perspective of rainstorms has important theoretical and practical significance for enriching the study of urban carrying capacity and alleviating urban waterlogging.This paper combines the natural attributes of a rainstorm with urban socio-economic conditions to construct an evaluation index system of urban carrying capacity under rainstorm condition(UCCRC)and classified the level of carrying capacity standards.The random forest algorithm was used to empirically analyze the UCCRC of Liaoning Province from2001to2015,and compared with the results of Back-Propagation Network Method to verify the suitability of the random forest algorithm.The results display that:①In time,the UCCRC of Dalian,Anshan,Panjin,and Shenyang increased gradually and the UCCRC of Yingkou,Dandong,Jinzhou,Chaoyang,and Fuxin tends to be decreasing.The UCCRC of Huludao,Liaoyang,Fushun,Tieling,and Benxi remained stable.There were minor fluctuations in individual years.The overall level of carrying capacity gap of each city in Liaoning Province was relatively larger.②In space,the UCCRC reflected a trend of“increase-decrease-increase”from west to east.③The influence factors of UCCRC in different cities of Liaoning Province were different.The key factors include construction area,population urbanization,drain-pipes length,water area and investment of urban basic facilities.④Compared with Back-Propagation Network,the random forest model generalized error rate(OOB)was9.32%,the classification has some characteristics such as high accuracy,smaller error and convenient operation,which was suitable for the leveling of UCCRC in Liaoning Province.