Toward Improved Prediction of Flood Inundation Maps: Application of Long Short Term Memory Networks
Abstract
Deep Learning models (DL) such as Long Short Term Memory (LSTM) networks have attracted attention in the hydrologic community because of their high accuracy in predicting streamflow time series. However, flood risk mitigation decisions require accurate flood inundation maps, which depend on flood depth rather than streamflow predictions. More research is needed on the effectiveness of LSTMs for flood depth prediction. In this study, we explore the effectiveness of LSTMs in flood inundation mapping and floodplain delineation. We investigate the accuracy of two workflows. The first involves reach-scale streamflow prediction with LSTMs, stage-discharge conversion, and then Height Above Nearest Drainage (HAND) method for floodplain delineation. The second workflow involves water depth prediction using LSTMs and HAND inundation mapping. Results from this study can help improve flood risk mitigation decisions.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2021
- Bibcode:
- 2021AGUFM.H22G..14Z