Date of Award
1-1-2024
Document Type
Dissertation
Degree Name
Ph.D. in Engineering Science
First Advisor
Dr. Hakan Yasarer
Second Advisor
Jacob Najjar
Third Advisor
Ahmad Al-Ostaz
School
University of Mississippi
Relational Format
dissertation/thesis
Abstract
The Gravity Recovery and Climate Experiment (GRACE) satellite mission has been instrumental in advancing our understanding of global water dynamics, providing unprecedented insights into terrestrial water storage (TWS). Despite its revolutionary contributions, the coarse spatial resolution of GRACE data (~300 km² grids) limits its utility for local-scale water resource management, particularly in regions requiring detailed hydrological analysis such as the Mississippi Delta. This dissertation addresses this limitation by developing and validating machine learning models to enhance the spatial resolution of GRACE data, thereby facilitating more precise and effective local water management strategies.
This research first establishes a comprehensive background in machine learning, focusing on the application of Artificial Neural Networks (ANN) and Random Forest (RF) techniques. These methodologies are pivotal for interpreting and enhancing satellite-derived hydrological data. A detailed literature review then assesses existing downscaling methods, identifying gaps in current approaches and setting the stage for the novel application of these advanced computational techniques.
In the core chapters of the dissertation, multiple machine learning models are developed and applied to downscale GRACE and GRACE Follow-On (GRACE-FO) products. The first case study focuses on the Mississippi Alluvial Plain Aquifer, employing a Random Forest model to refine the resolution of GRACE Mascon data from 0.5° to approximately 5 km. This model's validation against in-situ measurements demonstrates high accuracy, with significant implications for managing the region’s water resources. Subsequently, a dual-phase ANN method is utilized to predict groundwater levels in the Mississippi Delta. This approach not only enhances the spatial resolution of the data but also integrates local hydrological and environmental variables to improve prediction accuracy significantly.
Further, the dissertation examines the differences between in-situ measurements and GRACE-derived estimates of groundwater dynamics in the Yazoo-Mississippi Delta. This comparative analysis reveals minor discrepancies and underscores the potential of machine learning-enhanced satellite data for comprehensive regional water management. The models developed exhibit strong correlation coefficients, indicating their reliability and effectiveness in practical applications.
The findings from these studies highlight the transformative potential of machine learning in hydrology, particularly in refining the resolution of satellite data to meet the specific needs of local water management. By providing high-resolution data, these models help bridge the gap between large-scale satellite observations and localized water resource management needs. This research contributes to the scientific community by offering robust, scalable, and adaptable tools for sustainable water management, particularly in regions vulnerable to hydrological changes and variability.
Looking forward, this dissertation recommends further exploration of additional machine learning architectures and their application across different hydrological and climatic conditions. The expansion of model inputs to include more diverse environmental data and the continuous adaptation of these models to new GRACE-FO data are also suggested to enhance their accuracy and applicability. Ultimately, this work paves the way for future research aiming to leverage advanced computational techniques for environmental monitoring and management, ensuring the sustainability of vital water resources worldwide.
Recommended Citation
Awawdeh, Abdel Rahman M., "Grace Satellite Spatial Resolution Downscaling Using Machine Learning: Methods and Applications" (2024). Electronic Theses and Dissertations. 3003.
https://egrove.olemiss.edu/etd/3003
