A machine learning-based prediction-to-map framework for rapid and accurate spatial flood prediction

Abstract Traditional flood prediction approaches either rely on numerical models, which are accurate but computationally intensive, or machine learning models, which are faster but limited by data availability. To address these limitations, we developed a Prediction-to-Map (P2M) framework that combi...

Full description

Saved in:
Bibliographic Details
Main Authors: Daoyang Bao, Z. George Xue, Matthew Hiatt, Kehui Xu, Courtney K. Harris, Jill C. Trepanier
Format: Article
Language:English
Published: Nature Portfolio 2025-07-01
Series:npj Natural Hazards
Online Access:https://doi.org/10.1038/s44304-025-00122-2
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Abstract Traditional flood prediction approaches either rely on numerical models, which are accurate but computationally intensive, or machine learning models, which are faster but limited by data availability. To address these limitations, we developed a Prediction-to-Map (P2M) framework that combines the strengths of both methods. Trained on observed data and numerical model outputs, P2M delivers rapid, accurate spatial flood predictions. Applied to predict the flood event during Hurricane Nicholas (2021) near Galveston Bay, Texas, P2M produced flood depth maps that closely matched numerical simulations. Comparisons with observed data suggested P2M’s superior performance, as evidenced by higher R-squared and lower RMSE than the numerical model. Moreover, P2M demonstrated remarkable computational efficiency, producing a flood depth map with a 115,200-fold increase in speed. By achieving both faster speed and higher accuracy, this framework overcomes the trade-off in common surrogate models, providing a useful tool for rapid spatial flood prediction.
ISSN:2948-2100