Date of Award
Fall 2023
Document Type
Open Access Dissertation
Department
Civil and Environmental Engineering
First Advisor
Erfan Goharian
Abstract
In a world characterized by increasing and more frequent climate extreme events, population growth, and limited opportunities for infrastructural expansion, the imperative for more efficient and resilient planning, design, and operation of water resources and environmental systems has never been more apparent. Moreover, the emergence of high-dimensional data and the demand for quick and informed decision-making have created a critical challenge for the operation and management of these systems. To address these issues effectively, the ability to rapidly grasp and accelerate the analysis of this data is paramount, and this necessitates the innovative use of machine learning and surrogate modeling techniques, enabling timely, data-driven decisions in the face of complex water and environmental problems. To address these multifaceted challenges and provide valuable insights to managers and operators, the development of integrated and accelerated water resources simulation, optimization, and decision support models is critical. Innovative modeling approaches, such as machine learning, offer an ideal alternative to complex numerical models and enable the extraction of vital information from the wealth of high-dimensional data. This dissertation is structured around three chapters, each dedicated to a specific application of machine learning and surrogate modeling for the improved and informed management of Water Resources and Environmental Systems, including optimized decision-making, modeling and forecast, and classification and data analytics. The first chapter focuses on the development of surrogate simulation-optimization models as decision support tools for managing integrated water resources systems. These systems often feature non-linear processes and decision spaces for conflicting objectives constrained by convex functions, demanding extensive search, computational time, and hardware resources. The second chapter emphasizes the creation of surrogate economic evaluation components to integrate with watershed and reservoir models. This framework is then applied to assess the feasibility of incorporating Flood Managed Aquifer Recharge (Flood-MAR) into the operation of Folsom Reservoir and American River Watershed management in California. Chapter three pivots to the enhancement of hydrological modeling and prediction using machine learning models. It addresses the challenges posed by high-dimensional data and the need to predict in both gauged and ungauged basins. Furthermore, it delves into the impact of watershed characteristics on model performance, emphasizing the importance of considering these factors in accurate predictions. The final chapter of this research delves into the advancement of unsupervised machine learning methods in analyzing high-dimensional data, specifically for quantifying water samples and environmental systems' contamination by engineered nanoparticles. It highlights the critical role of nonlinear feature reduction and feature transformation techniques in enhancing the analysis and quantification of contamination issues in water systems. Collectively, this dissertation's four chapters offer a comprehensive exploration of the applications of machine learning and surrogate modeling for the management of Water Resources and Environmental Systems. In a world where environmental issues are increasingly complex, these innovative approaches provide valuable tools and insights for addressing pressing challenges and ensuring a sustainable future.
Rights
© 2024, Mahdi Erfani
Recommended Citation
Erfani, M.(2023). Harnessing Surrogate Modeling and Machine Learning for Management of Water Resources and Environmental Systems. (Doctoral dissertation). Retrieved from https://scholarcommons.sc.edu/etd/7540