基于谱聚类－图卷积神经网络模型的城市洪涝风险预测研究

Research on urban flood risk prediction based on spectral clustering-GCN modeling

摘要

本文针对以往洪涝风险预测模型忽略承灾体响应与捕捉空间拓扑关系上的不足，提出了一种基于谱聚类－图卷积神经网络(graph convolutional network，GCN)的洪涝风险预测模型。首先，利用径流曲线－二维径流淹没业务化工具(soil conservation service-two-dimensional runoff inundation toolkit for operational needs，SCS-TRITON)城市内涝模型模拟了多场24h降雨对应的淹没深度，并基于谱聚类方法将研究区域划分为15342个节点区域。结合模糊层次分析法(fuzzy analytic hierarchy process，FAHP)评估各节点区域的洪涝风险等级，构建了GCN模型预测洪涝风险，并进行SHAP可解释性分析。将GCN预测结果与卷积神经网络(convolutional neural network，CNN)和极端梯度提升(eXtreme gradient boosting，XGBoost)模型结果进行了对比。研究结果表明，GCN模型在捕捉节点区域间的空间拓扑关系上表现优越，准确率达到89.12%，显著优于其他机器学习模型。此外，对比分析了不同降雨类型对GCN模型预测性能的影响，总体上模型呈现出优秀的鲁棒性。基于谱聚类－GCN的深度学习模型进行精细尺度的考虑承灾体响应与空间拓扑关系的城市洪涝风险预测，为应急管理部门提供技术支持。

Abstract

This paper proposes a model based on spectral clustering-graph convolutional network (GCN) to address the limitations of conventional approaches in adequately considering disaster-bearing body responses and capturing spatial topological relationships. First, the soil conservation service-two-dimensional renoff inundation toolkit for operational needs (SCS-TRITON) urban flood model was employed to simulate inundation depths corresponding to multiple 24h rainfall events. Subsequently, spectral clustering was applied to partition the study area into 15342 node regions. Building upon this spatial framework, a GCN model was constructed that integrates flood risk assessments derived from the fuzzy analytic hierarchy process (FAHP) for each node region, with supplementary SHAP interpretability analysis conducted to enhance model transparency. Comparative evaluations demonstrated the GCN model’s superior performance in capturing inter-node spatial topological relationships, achieving 89.12% accuracy and significantly outperforming both convolutional neural network (CNN) and eXtreme gradient boosting (XGBoost) models. The model exhibited excellent robustness across different rainfall patterns, and its integration of spectral clustering with GCN enable fine-scale urban flood risk prediction that simultaneously accounts for disaster-bearing body responses and spatial topological relationships, providing emergency management authorities with an advanced technical solution.

关键词

城市洪涝风险 / 图卷积神经网络 / 谱聚类 / SHAP可解释性分析 / 模糊层次分析法 / SCS-TRITON模型

Key words

urban flood risk / graph convolutional network (GCN) / spectral clustering / SHAP interpretability analysis / fuzzy analytic hierarchy process (FAHP) / SCS-TRITON model

参考文献

[1] 中华人民共和国应急管理部. 应急管理部发布2023年全国自然灾害情况[EB/OL]. (2023-01-20). https://www.mem.gov.cn/xw/yjglbgzdt/202401/t20240120_475697.shtml.

[2] Emergency Management Department of People’s Republic of China. Emergency Management Department released the national natural disasters in 2023[EB/OL].(2023-01-20). https://www.mem.gov.cn/xw/yjglbgzdt/202401/t20240120_475697.shtml. (in Chinese)

[3] 仉文岗, 王云浩, 胡庆芳, 等 . 超大城市洪涝灾害成因综合解析: 以重庆市为例[J]. 土木与环境工程学报(中英文), 2025, 47(1): 1-10.

[4] ZHANG Wengang, WANG Yunhao, HU Qingfang, et al. Comprehensive analysis on causes of flood disaster in megacities: A case study of Chongqing[J]. Journal of Civil and Environmental Engineering, 2025, 47(1): 1-10. (in Chinese)

[5] 王振亚, 姚成, 董俊玲, 等 . 郑州“7·20”特大暴雨降水特征及其内涝影响[J]. 河海大学学报(自然科学版), 2022, 50(3): 17-22.

[6] WANG Zhenya, YAO Cheng, DONG Junling, et al. Precipitation characteristic and urban flooding influence of “7·20” extreme rainstorm in Zhengzhou[J]. Journal of Hohai University(Natural Sciences), 2022, 50(3): 17-22. (in Chinese)

[7] ZHANG W T, LIU Y Z, TANG W W, et al. Rapid spatio-temporal prediction of coastal urban floods based on deep learning approaches[J]. Urban Climate, 2023, 52: 101716.

[8] LIN L, TANG C Q, LIANG Q H, et al. Rapid urban flood risk mapping for data-scarce environments using social sensing and region-stable deep neural network[J]. Journal of Hydrology, 2023, 617: 128758.

[9] FAN C, MOSTAFAVI A. A graph-based method for social sensing of infrastructure disruptions in disasters[J]. Computer-Aided Civil and Infrastructure Engineering, 2019, 34(12): 1055-1070.

[10] FARAHMAND H, XU Y C, MOSTAFAVI A. A spatial-temporal graph deep learning model for urban flood nowcasting leveraging heterogeneous community features[J]. Scientific Reports, 2023, 13(1): 6768.

[11] XU Y P, PAN S L, GAO C, et al. Historical pan evaporation changes in the Qiantang River Basin, East China[J]. International Journal of Climatology, 2016, 36(4): 1928-1942.

[12] YANG J, HUANG X. The 30m annual land cover dataset and its dynamics in China from 1990 to 2019[J]. Earth System Science Data, 2021, 13(8): 3907-3925.

[13] LIU Z P, TANG H, FENG L, et al. China Building Rooftop Area: The first multi-annual (2016-2021) and high-resolution (2.5m) building rooftop area dataset in China derived with super-resolution segmentation from Sentinel-2 imagery[J]. Earth System Science Data, 2023, 15(8): 3547-3572.

[14] VON LUXBURG U. A tutorial on spectral clustering[J]. Statistics and Computing, 2007, 17(4): 395-416.

[15] 林沛文, 何书, 鲜木斯艳·阿布迪克依木, 等 . 基于逻辑回归-模糊层次分析模型的公路沿线滑坡易发性评价[J]. 自然灾害学报, 2024, 33(6): 27-36.

[16] LIN Peiwen, HE Shu, ABUDIKEIMU XMSY, et al. Landslide susceptibility assessment along highways based on LR-FAHP model[J]. Journal of Natural Disasters, 2024, 33(6): 27-36. (in Chinese)

[17] WEE G, CHANG L C, CHANG F J, et al. A flood Impact-Based forecasting system by fuzzy inference techniques[J]. Journal of Hydrology, 2023, 625: 130117.

[18] RIBEIRO P J G, PENA JARDIM GONÇALVES L A. Urban resilience: A conceptual framework[J]. Sustainable Cities and Society, 2019, 50: 101625.

[19] WANG Y H, ZHANG Q T, LIN K R, et al. A novel framework for urban flood risk assessment: Multiple perspectives and causal analysis[J]. Water Research, 2024, 256: 121591.

[20] 赵凯旭, 王建明, 闫茜, 等 . 西安市老城区暴雨内涝模拟及其灾害风险评估[J]. 自然灾害学报, 2023, 32(4): 1-12.

[21] ZHAO Kaixu, WANG Jianming, YAN Xi, et al. Urban waterlogging simulation and its risk assessment in the old area of Xi’an[J]. Journal of Natural Disasters, 2023, 32(4): 1-12. (in Chinese)

[22] RAFIEI-SARDOOI E, AZAREH A, CHOUBIN B, et al. Evaluating urban flood risk using hybrid method of TOPSIS and machine learning[J]. International Journal of Disaster Risk Reduction, 2021, 66: 102614.

[23] MORALES-HERNÁNDEZ M, SHARIF M B, KALYANAPU A, et al. TRITON: A Multi-GPU open source 2D hydrodynamic flood model[J]. Environmental Modelling & Software, 2021, 141: 105034.

[24] YU H J, XU Y P, ZHONG H, et al. Exploring the impact of urbanization on flood characteristics with the SCS-TRITON method[J]. Natural Hazards, 2023, 120(4): 3213-3238.

[25] SAVAGE J T S, BATES P, FREER J, et al. When does spatial resolution become spurious in probabilistic flood inundation predictions? [J]. Hydrological Processes, 2016, 30(13): 2014-2032.

[26] 任泽凌, 李彬权, 王国庆, 等 . 城市暴雨强度公式复核: 以杭州主城区为例[J]. 南水北调与水利科技(中英文), 2024, 22(1): 80-89.

[27] REN Zeling, LI Binquan, WANG Guoqing, et al. Revisited urban rainfall intensity formula: An example of the main urban area of Hangzhou[J]. South-to-North Water Transfers and Water Science & Technology, 2024, 22(1): 80-89. (in Chinese)

[28] 浙江省水文勘测局. 浙江省短历时暴雨[R]. 杭州: 浙江省水文勘测局, 2003.

[29] Zhejiang Hydrological Survey Bureau. Short-duration rainstorm in Zhejiang Province[R]. Hangzhou: Zhejiang Hydrological Survey Bureau, 2003. (in Chinese)

[30] WU Z H, PAN S R, CHEN F W, et al. A comprehensive survey on graph neural networks[J]. IEEE Transactions on Neural Networks and Learning Systems, 2020, 32(1): 4-24.

[31] 周天游, 刘畅, 薛鹏, 等 . 基于不同机器学习的震后滑坡易发性建模研究[J]. 自然灾害学报, 2023, 32(5): 177-185.

[32] ZHOU Tianyou, LIU Chang, XUE Peng, et al. Post-seismic landslide susceptibility modeling based on different machine learning methods[J]. Journal of Natural Disasters, 2023, 32(5): 177-185. (in Chinese)

[33] WANG H J, LIANG Q X, HANCOCK J T, et al. Feature selection strategies: A comparative analysis of SHAP-value and importance-based methods[J]. Journal of Big Data, 2024, 11(1): 44.

[34] 何立富, 陈双, 郭云谦. 台风利奇马(1909)极端强降雨观测特征及成因[J]. 应用气象学报, 2020, 31(5): 513-526.

[35] HE Lifu, CHEN Shuang, GUO Yunqian. Observation characteristics and synoptic mechanisms of typhoon Lekima extreme rainfall in 2019[J]. Journal of Applied Meteorological Science, 2020, 31(5): 513-526. (in Chinese)

[36] ZHAO L, SONG Y J, ZHANG C, et al. T-GCN: A temporal graph convolutional network for traffic prediction[J]. IEEE Transactions on Intelligent Transportation Systems, 2019, 21(9): 3848-3858.

基金资助

浙江省重点研发项目(2021C03017)

国家自然科学基金项目(52309038)