Assistant Professor Florida International University
Considering insufficient runoff observations and time-series, identification of high-risk flood-prone urban areas is challenging. This research developed integration of Machine Learning (ML) algorithms (Random Forest, Naive Bayes Tree and Naive Bayes) and Multi-Criteria (technical, social, and environmental) Decision Analysis in Geographic Information System (GIS) environment for flood risk prediction. Application of the integrated framework was demonstrated in risk-based prioritization of bridge rehabilitation projects subject to floods, traffic, and structural conditions in Miami-Dade County, Florida, and then geared toward addressing socio-environmental factors. Flood predictor variables (rainfall intensity, slope, land cover, and hydrologic soil group) were detected, then an environmental index (including air quality and temperature) was developed. Later on, incorporating social equity in infrastructure planning was introduced by population density and income rate. To gain the highest prediction performance using ML, the historical flood locations were extracted from Sentinel-1 SAR Level-1 Ground Range Detected images and local observations (311 records) which also used for validation. Every pixel was given a flood risk value, then classified from very-low to very-high risk. Results revealed that previously flooded areas were mostly very-high and high risk. Finally, a risk map generated to show an integrated technical, social, and environmental risk factor for each bridge. It was concluded that considering social equity and environmental justice can substantially change the risk-based prioritization of bridges for rehabilitation activities. The developed framework can be used as a flood prediction and decision-making tool for equitable urban infrastructure planning.
