Authors: Naser Ahmed*, Department of Geography and Environment, Jagannath University, Dhaka-1100, Bangladesh, Jinhyung Lee, Department of Geography and Environment , Western University, London, Ontario, Canada
Topics: Spatial Analysis & Modeling, Transportation Geography, Hazards, Risks, and Disasters
Keywords: Susceptibility, Natural hazards, Transportation, Machine Learning, GIS, Toronto
Session Type: Virtual Paper
Start / End Time: 3:05 PM / 4:20 PM
Room: Virtual 7
Presentation File: No File Uploaded
Urban floods often cause the functional disruption of transportation networks, thereby impeding people’s mobility, accessibility, and result in adverse socio-economic externalities (e.g., social inequity. Climate change, rapid urbanization, and growing global population are further increasing these trends with higher frequency and intensity of hydro-meteorological disasters in cities. This study develops a data-driven machine learning (ML) model for predicting the flooding susceptibility of road network and public transit systems using historic flooding location data in Toronto, ON, Canada. Four machine learning approaches: 1) eXtreme Gradient Boosting, 2) K Nearest Neighbor, 3) Naïve Bayes, and 4) Random Forest, are employed to evaluate the future risk of an area being inundated by flooding events. The flood probability is estimated based on the relationship between flood inundation events and contributing factors such as elevation, slope, stream proximity, and precipitation. We classify and use satellite images from flooding events happed in Toronto in 2018 following and Emergency Management Historical Events from Ontario GeoHub as a basis for training (70% of samples) and validating (30% of samples) machine learning models. Specifically, the Area-Under Receiver Operating Characteristic (AUROC) curve is used to evaluate the prediction results by machine learning methods. All four machine learning models showed a prediction accuracy greater than 85% which demonstrates reliable performance and potentials to be utilized in similar urban environments. Our results suggest machine learning models are promising and cost-effective for modeling the susceptibility of urban transport systems and providing in-depth insights to guide preparedness and renovation against hydro-meteorological hazards.