Machine Learning for Urban Water Runoff Prediction and Water Temperature Forecasting

In this dissertation, I investigate the application of machine learning (ML) methods in two areas of hydrology.

Chapter 1 presents the development of data-driven models to predict urban stormwater runoff volume. Machine learning (ML) models are used to approximate output from the Stormwater Management Model (SWMM), a hydrological model used for runoff simulation. SWMM is powerful but computationally intensive, so ML models are developed to expedite runoff predictions. A case study for the First Creek watershed in Knoxville, Tennessee, is performed using rainfall data and subcatchment characteristics. Feature engineering and clustering are applied to compare SWMM and ML model outputs. Results show that random forests accurately predict runoff volume almost instantly, significantly reducing computational requirements.

Chapter 2 explores trends, fluctuations, and correlations of water temperatures and other variables at Tennessee Valley Authority (TVA) fossil plants. The aim of this research is to utilize real-world historical water temperature data to analyze the environmental changes at fossil plants.

Chapter 3 focuses on forecasting hourly water temperature for the efficiency of three TVA fossil plants and the impact on aquatic life. The findings of this study can be beneficial for fossil plants that need to consider the sensitivity of water temperatures for their operations.