Unraveling Urban Fluvial Flooding Complexities through AI
Abstract
As urbanization increases across the globe urban flooding is an ever-pressing concern. Urban fluvial systems are highly complex, depending on a myriad of interacting variables. Unraveling the influence of these variables to predict flood risk is required to meet this challenge. Numerous hydraulic models are available for analyzing urban fluvial systems; however, meeting the demand of high spatial extension and finer discretization and solving the physics-based numerical equations are computationally expensive. Computational efforts increase drastically with an increase in model dimension and resolution, preventing current solutions from fully realizing the data revolution. This study demonstrates the effectiveness of a subset of Artificial Intelligence (AI), Machine Learning (ML) to quantify urban flooding considering the lower Darby Creek, PA. The analysis shows that coupling AI algorithms with hydraulic models has the potential to substantially advance the scientific communitys approach, analysis, and understanding of urban flooding. Training datasets for the ML models comprise multiple hydraulic features (e.g. location, elevation, water depth, flooded (or not) condition, discharge, average slope and portion of the impervious landcover within the contributing area). Outputs from 2D calibrated hydraulic models using the iRIC platform were used to create multiple scenarios at different upstream discharges. We construct binary classifiers e.g., logistic regression, decision tree, support vector machine and K-nearest neighbors to predict if locations are flooded. Deep Neural Networks (DNNs) were used to quantify the water depth of every cell within the domain. The values of the evaluation matrices indicate good performance (F-1 scores range from 0.948-0.991 for binary classifiers; mean absolute error- 0.0072 for DNN regression). This approach is a significant step towards resolving the complexities of urban flooding with a large multi-dimensional dataset in a highly computationally efficient manner.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2021
- Bibcode:
- 2021AGUFM.H22G..15M