A comparison of machine learning surrogate models of street-scale flooding in Norfolk, Virginia

TL;DR

This paper compares machine learning surrogate models, including random forest, LSTM, and GRU, for real-time street flooding prediction in Norfolk, Virginia, highlighting the importance of model architecture and uncertainty communication.

Contribution

It evaluates and contrasts the performance of different deep learning models for urban flooding prediction using real rainfall data, emphasizing model architecture and uncertainty handling.

Findings

LSTM and GRU outperform random forest in predictive accuracy.

Model architecture supporting uncertainty communication improves reliability.

Deep learning models effectively integrate multi-modal features for flood prediction.

Abstract

Low-lying coastal cities, exemplified by Norfolk, Virginia, face the challenge of street flooding caused by rainfall and tides, which strain transportation and sewer systems and can lead to property damage. While high-fidelity, physics-based simulations provide accurate predictions of urban pluvial flooding, their computational complexity renders them unsuitable for real-time applications. Using data from Norfolk rainfall events between 2016 and 2018, this study compares the performance of a previous surrogate model based on a random forest algorithm with two deep learning models: Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU). This investigation underscores the importance of using a model architecture that supports the communication of prediction uncertainty and the effective integration of relevant, multi-modal features.