Ensemble learning analysis of influencing factors on the distribution of urban flood risk points: a case study of Guangzhou, China

Frontiers in Earth Science

Earth Science

Jin Wang¹  Juchao Zhao²  Zaheer Abbas²  Yao Yang²  Yaolong Zhao² 

[1] Key Laboratory of Natural Resources Monitoring in Tropical and Subtropical Area of South China, Ministry of Natural Resources, Guangzhou, China;Guangdong Science and Technology Collaborative Innovation Center for Natural Resources, Guangzhou, China;Faculty of Engineering, Beidou Research Institute, South China Normal University, Foshan, China;School of Geography, South China Normal University, Guangzhou, China;Key Laboratory of Natural Resources Monitoring in Tropical and Subtropical Area of South China, Ministry of Natural Resources, Guangzhou, China;Guangdong Science and Technology Collaborative Innovation Center for Natural Resources, Guangzhou, China;

关键词: urban waterlogging;    influencing factors;    Guangzhou;    susceptibility assessment;    ensemble learning;   

DOI  :  10.3389/feart.2023.1042088

 received in 2022-09-12, accepted in 2023-04-24,  发布年份 2023

来源: Frontiers

【 摘 要 】

Urban waterlogging is a major natural disaster in the process of urbanization. It is of great significance to carry out the analysis of influencing factors and susceptibility assessment of urban waterlogging for related prevention and control. However, the relationship between urban waterlogging and different influencing factors is often complicated and nonlinear. Traditional regression analysis methods have shortcomings in dealing with high-dimensional nonlinear issues. Gradient Boosting Decision Tree (GBDT) is an excellent ensemble learning algorithm that is highly flexible and efficient, capable of handling complex non-linear relationships, and has achieved significant results in many fields. This paper proposed a technical framework for quantitative analysis and susceptibility assessment on influencing factors of urban waterlogging based on the GBDT in a case study in Guangzhou city, China. Main factors and indicators affecting urban waterlogging in terrain and topography, impervious surface, vegetation coverage, drainage facilities, rivers, etc., were selected for the GBDT. The results demonstrate that: (1) GBDT performs well, with an overall accuracy of 83.5% and a Kappa coefficient of 0.669. (2) Drainage density, impervious surface, and NDVI are the most important influencing factors resulting in rainstorm waterlogging, with a total contribution of 85.34%. (3) The overall distribution of urban waterlogging susceptibility shows a characteristic of “high in the southwest and low in the northeast”, in which the high-susceptibility areas are mainly distributed in Yuexiu District (34%), followed by Liwan District (22%) and Haizhu District (20%). To mitigate the impact of frequent urban flooding disasters, future measures should focus on strengthening drainage networks, such as optimizing impervious surface spatial patterns, controlling construction activities in high-risk areas, and preventing excessive development of green spaces.

【 授权许可】

Unknown   
Copyright © 2023 Zhao, Wang, Abbas, Yang and Zhao.