In Partial Fulfillment of the Requirements for the Degree of Master of Science
will defend her thesis
Data-Driven Approaches for Water Level Prediction
Flood is a hazardous disaster which threatens the United States and its territories as it affects millions of people every year. In this context, having suitable approaches for forecasting flooding in advance will benefit people by reducing property damage, and saving lives of people by warning citizens of dangerous flooding events in advance. Consequently, many cities have developed sensor-based flood early warning systems which report water levels in real-time. The goal of this thesis is to take advantage of the data collected by such systems to develop data-driven water level prediction techniques. In past research, a lot of physics-based hydrology models have been developed, such as the National Water Model, which predict water levels by simulating the rise and fall of water. The purpose of this research is to generate alternative, complementary, data-driven water-level forecasting models using existing statistical models and recurrent neural networks which extrapolate the past into the future. We investigate various time series forecasting approaches, in particular: Vector Auto Regressive (VAR) and Long Short-Term Memory Networks (LSTM). The investigated forecasting techniques are applied and evaluated using USGS datasets. Moreover, we analyze the role of soil moisture—a less explored parameter—in flood incidents and conduct some experiments that explore the relationship between rainfall and soil moisture. Finally, we create a web application called FloodNet, a real-time water level prediction system, with a forecast horizon of 2 hours for a location along Buffalo Bayou in Houston, Texas.
Date: Wednesday, June 26, 2019
Time: 10:30 AM - 12:00 PM
Place: PGH 550
Advisor: Dr. Christoph F. Eick
Faculty, students, and the general public are invited.