Data-driven modeling for compound flooding simulation
| dc.contributor.advisor | Dawson, Clint N. | |
| dc.contributor.committeeMember | Hodges, Ben | |
| dc.contributor.committeeMember | Bui-Thanh, Tan | |
| dc.contributor.committeeMember | Ghattas, Omar | |
| dc.contributor.committeeMember | Gamba, Irene | |
| dc.creator | Li, Wei (Ph. D. in computational science, engineering, and mathematics) | |
| dc.creator.orcid | 0000-0002-8843-0093 | |
| dc.date.accessioned | 2021-10-22T14:29:52Z | |
| dc.date.available | 2021-10-22T14:29:52Z | |
| dc.date.created | 2021-08 | |
| dc.date.issued | 2021-08-05 | |
| dc.date.submitted | August 2021 | |
| dc.date.updated | 2021-10-22T14:29:54Z | |
| dc.description.abstract | Accurate simulation of compound flooding is crucial for flood risk management. In coastal regions, both rainfall runoff and storm surge can contribute to flooding and are not mutually exclusive. It is thus necessary to develop a modeling framework for compound flooding that considers both mechanisms. While many hydrologic models have been developed to simulate the rainfall runoff processes, a lot of these methods are either computationally expensive, or incapable of simulating extreme weather events. Thus, they may not be suitable for coupled modeling of compound flooding. In this study, a data-driven hydrologic model based on deep recurrent neural network (RNN) is developed for rainfall runoff simulation at relatively low computational cost. To test the capability of the method, the model is used to infer the streamflow out of an urban watershed, Brays Bayou in Houston, Texas. And the model is validated with real world hydrologic data. Additionally, the proposed synced sequence to sequence RNN architecture is compared with the sequence input single output one that is widely-used in hydrologic modeling. Numerical experiments show that the proposed method provides more accurate predictions using relatively less computational resources than the sequence input single output architecture. Later, downstream water level input is integrated into the RNN model to enable the one-way coupling of rainfall runoff with storm surge. Numerical examples at two different locations demonstrate that the additional information leads to improved predictions. Finally, a two-way dynamic coupling framework is constructed for the RNN hydrologic model and an ocean circulation model, ADvanced CIRCulation. The framework is tested, verified, and validated for the Houston ship channel - Galveston bay estuarine system during Hurricane Harvey (2017). This dissertation is based on the following articles: High temporal resolution rainfall runoff modelling using Long-Short-Term-Memory (LSTM) networks by Wei Li, Amin Kiaghadi, Clint Dawson [1]; Exploring the best sequence LSTM model- ing architecture for flood prediction by Wei Li, Amin Kiaghadi, Clint Dawson [2]; and Simulating compound floods: dynamic coupling of deep learning and physics- based models by Wei Li, Gajanan Choudhary, Amin Kiaghadi, Clint Dawson. This material is based upon work funded by National Oceanic and Atmospheric Admin- istration (Grant No. NA18NOS0120158) and National Science Foundation (NSF, CMMI-1520817). | |
| dc.description.department | Computational Science, Engineering, and Mathematics | |
| dc.format.mimetype | application/pdf | |
| dc.identifier.uri | https://hdl.handle.net/2152/89459 | |
| dc.language.iso | en | |
| dc.subject | Deep learning | |
| dc.subject | Storm surge | |
| dc.subject | Rainfall runoff | |
| dc.subject | Compound flood | |
| dc.title | Data-driven modeling for compound flooding simulation | |
| dc.type | Thesis | |
| dc.type.material | text | |
| thesis.degree.department | Computational Science, Engineering, and Mathematics | |
| thesis.degree.discipline | Computational Science, Engineering, and Mathematics | |
| thesis.degree.grantor | The University of Texas at Austin | |
| thesis.degree.level | Doctoral | |
| thesis.degree.name | Doctor of Philosophy |
Access full-text files
Original bundle
1 - 1 of 1