Hazard and vulnerability in urban flood risk mapping: Machine learning techniques and considering the role of urban districts
Autor: | Mohammad Eini, Hossein Hedayat, Mohsen Rashidian, Hesam Seyed Kaboli |
---|---|
Rok vydání: | 2020 |
Předmět: |
010504 meteorology & atmospheric sciences
Population 0211 other engineering and technologies Developing country 02 engineering and technology Machine learning computer.software_genre 01 natural sciences Cohen's kappa Risk mapping Drainage education Urban management 0105 earth and related environmental sciences 021110 strategic defence & security studies education.field_of_study Flood myth business.industry Flood vulnerability Geology Geotechnical Engineering and Engineering Geology Geography Artificial intelligence business Safety Research computer |
Zdroj: | International Journal of Disaster Risk Reduction. 50:101687 |
ISSN: | 2212-4209 |
DOI: | 10.1016/j.ijdrr.2020.101687 |
Popis: | Urban flood risk mapping plays a decisive role in urban management and planning, especially in reducing flood damages. In this study, a flood risk map was produced for Kermanshah city (Iran) by combining flood hazard and flood vulnerability maps. Based on effective factors of urban flooding, flood hazard maps were generated using two machine learning models: Maximum Entropy (MaxEnt) and Genetic Algorithm Rule-Set Production (GARP). These models were developed on 117 flood sites which were identified on the reports and field surveys for Kermanshah city, and 163 non-flooded points. Economic, social, and infrastructure criteria were considered to analyze flood vulnerability. The sub-criteria were defined based on social and cultural structure in developing countries and urban infrastructure facilities involved in the floods. Fuzzy Analytical Hierarchical Process method (FAHP) was applied to determine the overall weight vector. The results demonstrated that the MaxEnt model had a better performance than the GARP model based on two common indices which are the area under the receiver-operator characteristic curve (AUC-ROC = 96.76–98.32%) and kappa statistic (Kappa = 0.82-0.86). These findings showed that machine learning models provide reliable results for areas where data access is challenging, especially in developing countries. The results also indicated that infrastructure criterion has the highest impact weight on the vulnerability. In general, population, urban texture, and distance to the major drainage channels are the most important factors in increasing flood risk. Also, the vulnerability of an urban neighborhood greatly increases the risk of flooding. |
Databáze: | OpenAIRE |
Externí odkaz: |