A deep learning-based emulator for flood forecasting in coastal areas
Abstract
Flooding in coastal areas is a common threat that causes severe property damage and fatalities every year across the globe and is expected to increase due to climate change and sea-level rise. Given the substantial impacts of flooding and its increasing trend, a timely and reliable flood water level prediction at any time and location of a river is vital for flood risk management. Here, we propose a systematic framework that utilizes a Deep Learning (DL) algorithm and a hydrodynamic model to provide probabilistic flood water level estimates along coastal rivers. In this framework, Adaptive Hydraulics (ADH) model was employed to simulate the compound flooding across the Savanah River during Hurricane Matthew (2016) and Hurricane Irma (2017). The ADH simulated water levels for Hurricane Matthew at different points along the river were divided into multiple classes using a Self-Organizing Map (SOM) approach. Then, a DL model was trained and tested for each class to predict the time series of water levels. The trained DL model was used to simulate the compound flooding induced by Hurricane Irma while considering the uncertainty in boundary conditions. Our results indicate that the proposed framework can be an effective tool for timely forecasting of peak water levels and the rising time of the flood water in coastal areas.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2022
- Bibcode:
- 2022AGUFM.H25K1238A