基于最大熵模型的深圳市内涝影响因素分析及内涝风险评估

城市内涝是最常见的自然灾害之一,深入剖析其影响因素并进行风险评估对内涝防治具有重要意义。以往研究表明,城市内涝是由自然因素(如地形)和人为因素(如土地利用)共同引起的。在土地利用方面,相关学者主要关注二维空间因素对内涝的影响,较少顾及土地利用的三维建筑格局。此外,在研究方法的选取上,尽管已有学者利用随机森林、神经网络等模型对内涝影响因素进行研究,然而传统方法在负样本(不发生内涝的地点)的选取上存在不确定性。为解决这2点不足,论文引入最大熵(MAXENT)模型,以深圳市为研究案例,通过MAXENT剖析各潜在影响因子与内涝风险的关系。结果表明,影响内涝风险的主导环境因子为不透水面比例、绿地比例、人口密度、暴雨峰值雨量、地表起伏度。而对内涝发生有重要影响的三维因子为容积率、建筑形状系数、平均高度。通过MAXENT评估的内涝风险结果可知,深圳潜在高风险区的面积约为491 km²,占市域面积的24.58%,主要位于龙华区、南山区、龙岗区北部、光明区、福田区。进一步对潜在高风险区进行空间自相关分析,结果发现过往并不存在内涝点的南山区北部、福田区西部、罗湖区中部等部分区域风险概率出现高—高集聚现象,表明上述地区的内涝风险会受到周围地区的正向影响,因此在内涝的监测与防治中应当重点关注高风险地区以实现更精准的防控。由于内涝风险评估是城市灾害管理的重要组成部分,因此论文提出的相关建议不仅可作为防灾减灾的重要参考依据,还能为国土空间规划的优化提供新思路。

Influencing factors of waterlogging and waterlogging risks in Shenzhen City based on MAXENT

Urban waterlogging is one of the most common natural disasters. In-depth analysis of its influencing factors and estimation of high-risk waterlogging areas are of great significance for waterlogging prevention and management. Although some studies have approached these issues through advanced machine learning methods such as random forest and neural network, the identified influencing factors are mainly related to the two-dimensional space. Moreover, while traditional methods require both accurate positive and negative samples, there is an inevitable subjectivity in the selection of negative samples. To address these disadvantages, this research took Shenzhen City as the study area and employed the MAXENT model, which does not require negative samples, to explore the relationship between potential influencing factors (including three-dimensional building factors) and waterlogging risk during 2015-2019. The results show that the dominant environmental factors behind the density of waterlogging hotspots were the proportion of impervious surface, the proportion of green space, population density, rainstorm peak rainfall, and fluctuation of the terrain. With regard to the three-dimensional building factors, building congestion, average building height, and building shape coefficient have a crucial impact on urban waterlogging. According to the waterlogging probability estimated by MAXENT, the total area of potential high-risk waterlogging areas in Shenzhen is approximately 491 km², accounting for 24.58% of the total area of the city. These areas are mainly located in Longhua District, Nanshan District, the north of Longgang District, Guangming District, and Futian District. Through the spatial autocorrelation analysis of the potential high-risk areas, we found that some areas in the north of Nanshan District, the west of Futian District, and central Luohu District where there were no waterlogging hotspots in the past, exhibit high concentration levels. This indicates that the waterlogging probability in these areas would be positively affected by the surrounding areas. Therefore, focus should be placed on high-risk areas for achieving more accurate waterlogging prevention and management. Urban waterlogging risk assessment is an important part of disaster management. The assessment results of waterlogging risk not only can provide support for disaster prevention and risk mitigation, but also are essential for protecting people's lives and the sustainable development of cities.