Dissertation
Distributed streamflow forecasting with geospatial data integration using deep learning
University of Iowa
Doctor of Philosophy (PhD), University of Iowa
Spring 2022
DOI: 10.17077/etd.006426
Abstract
Water is essential for all living things on the planet. Excessive rainfall, on the other hand, might result in disasters such as flooding. Accurate streamflow forecasting and efficient flood alert systems are critical for government agencies to provide early warning to public in order to reduce flood damage and losses. The objective of my research is to improve current streamflow forecasting processes in building more accurate streamflow and flood forecast models using data-driven approaches and designing an efficient real-time flood warning framework that supports next-generation data structures for sharing and communication.
Deep learning models have been successfully employed in a variety of domains and demonstrated significant success in many complex tasks. I have investigated and evaluated the most prominent deep learning neural network algorithms in recent years, a form of model that was primarily used to design artificial intelligence systems, in order to improve streamflow and flood forecasting models. I have developed a series of novel deep learning models with a prediction objective of at most 120 hours for the gauged watersheds in Iowa and carried out four case studies for evaluation. The results show that the deep learning models including LSTM, GRU, TCN, and GNN perform well on the studied watersheds, implying that deep learning is effective in hydrological modeling. I have also created a framework for real-time forecasting, which is based on the developed deep learning models and capable of generating real-time forecasts quickly and accurately. To improve the efficiency of streamflow forecasting and flood risk communication, a flood forecasting data language was designed as the endpoint of the streamflow forecast.
Details
- Title: Subtitle
- Distributed streamflow forecasting with geospatial data integration using deep learning
- Creators
- Zhongrun Xiang
- Contributors
- Ibrahim Demir (Advisor)Witold F. Krajewski (Committee Member)Ricardo Mantilla (Committee Member)Xun Zhou (Committee Member)Felipe Quintero Duque (Committee Member)
- Resource Type
- Dissertation
- Degree Awarded
- Doctor of Philosophy (PhD), University of Iowa
- Degree in
- Civil and Environmental Engineering
- Date degree season
- Spring 2022
- DOI
- 10.17077/etd.006426
- Publisher
- University of Iowa
- Number of pages
- xvii, 198 pages
- Copyright
- Copyright 2022 Zhongrun Xiang
- Language
- English
- Description illustrations
- Illustrations, charts, graphs, tables, maps
- Description bibliographic
- Includes bibliographical references (pages 174-198).
- Public Abstract (ETD)
Flooding is one of the world's most devastating natural disasters, causing both life and property losses. We can assist in reducing such losses by providing more accurate flood forecasts and better early warning platforms. This thesis employs a variety of deep learning models to create higher resolution and more accurate streamflow forecasting models. Our model structure is able to integrate geospatial data and forecast hourly streamflow for up to 120 hours in Midwest. In addition, FloodML, a language for sharing flood data and alerts, was created to bridge the gap between the next generation of data sharing, integration, and visualization needs for better cross-organizational collaboration. Finally, using deep learning algorithms, I designed and implemented a system for real-time operational streamflow forecasting, which translates this scientific research into actionable product.
- Academic Unit
- Civil and Environmental Engineering
- Record Identifier
- 9984271154402771
Metrics
18 File views/ downloads
43 Record Views