基于XGBoost的长三角核心城市内涝风险评估及影响因素分析

XGBoost model-based risk assessment and influencing factors analysis of waterlogging in core cities of Yangtze River Delta

随着全球变暖日益严重,极端气候现象频发。面对内涝灾害,城市社会经济呈现高脆弱性,有效评估城市内涝风险成为城市建设和区域可持续发展的迫切要求。以三座长三角核心城市,即上海市、南京市和杭州市为研究区,通过社交媒体数据获取积水样本数据,从气象、地形和社会经济三方面选取评价指标,构建基于极端梯度提升(XGBoost)的城市内涝风险评估模型,在评估长三角核心城市内涝风险等级的基础上分析影响城市内涝的主要因素。结果表明:(1)运用XGBoost进行城市内涝灾害风险研究,其预测性能和预测精度优于其他常用机器学习模型;(2)从城市内涝风险分布来看,三个城市的内涝风险均呈现较高的空间异质性,主城区多为内涝高风险,周边新城区多为内涝低风险区,城市内涝风险等级呈现由中心城区向周围扩散的趋势,靠近河流、长江入海口处以及河网高密度区内涝风险也较高;(3)就影响因素而言,高程是影响研究区内涝风险的首要因素,道路分布和强降雨为影响上海城市内涝高风险地区的次级因素,受人为影响的地表覆盖是导致南京和杭州城市内涝高风险地区的次级因素。

Under the background of serious global warming and extreme climate, cities shows high vulnerability of social economy with waterlogging. Therefore, effective assessment of urban waterlogging risks has become an urgent need for urban construction and regional sustainable development. Taking three core cities of the Yangtze River Delta, i.e. Shanghai, Nanjing and Hangzhou, as the study areas, the relevant assessment indexes are selected from the aspects of meteorology, topography and socio-economy by means of obtaining the waterlogging sampling data through the relevant social media data, and then a XGBoost(Extreme Gradient Boosting)-based urban water logging risk assessment model is established, while the main factors affecting urban water logging are analyzed on the basis of evaluating the risk level of the core cities of the Yangtze River Delta. The results show that(1) the XGBoost model is applied to study urban waterlogging disaster, of which the prediction performance and accuracy are better than the other conventionally used machine learning models;(2) from the perspective of urban waterlogging risk distribution, the urban waterlogging risk level exhibits a higher spatial heterogeneity, for which most of the main urban areas are under high waterlogging risk and the surrounding new urban areas are under low waterlogging risk, while the urban waterlogging level exhibits a spreading trend from the central urban areas to the surrounding areas, but the risks of waterlogging near rivers and the estuary of the Yangtze River are also high;(3) for the influencing factors, elevation is the main factor affecting urban waterlogging, while the road distribution and heavy rainfall are the secondary factors that affect the areas with high waterlogging risk in Shanghai, and human-induced land cover is the secondary factor that affect the areas with high waterlogging risk in Nanjing and Hangzhou.