Date of Graduation

7-2026

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Geosciences (PhD)

Degree Level

Graduate

Advisor/Mentor

Peter, Brad

Committee Member

Larson, Danelle

Second Committee Member

Tullis, Jason

Third Committee Member

Ruhl-Whittle, Laura

Keywords

Automated Machine Learning; Data Fusion; Satellite Image Data; Water Quality; Water System Modeling

Abstract

Despite the ecological, economic, and social importance of inland water bodies, they are increasingly impacted by human and natural drivers. Harmful algal blooms (HABs) and total suspended solids (TSS) are key indicators of water quality, and elevated concentrations pose significant ecological and public health risks. The spatiotemporal dynamics of these variables remain insufficiently understood in major river systems, including the Upper Mississippi River (UMR) and the Illinois River Basin (IRB). While in situ measurements provide accurate observations, they are costly and offer limited spatial and temporal coverage, restricting basin-scale monitoring. Remote sensing offers a powerful alternative; however, single-sensor approaches are constrained by trade-offs between spatial resolution and revisit frequency, limiting their ability to capture rapidly changing conditions. To address these challenges, this dissertation presents a multi-sensor, multi-scale framework for monitoring inland water quality. The first chapter provides a systematic review of peer-reviewed literature to identify gaps in modeling optically active water quality parameters using remote sensing and machine learning. The second chapter develops machine learning models of TSS dynamics across the UMR using high-resolution Landsat data at basin and reach scales. Environmental covariates, including discharge, elevation, slope, topobathymetry, and water temperature, are incorporated to improve performance. Variable importance is evaluated to identify key drivers of sediment dynamics, and model transferability is assessed by testing whether basin-scale models can capture reach-scale variability. The third chapter develops a Google Earth Engine–based framework to generate a fused Landsat–Sentinel dataset (2017–2024), integrating Landsat 7, 8, and 9 with Sentinel-2 imagery. The fused product is evaluated against individual sensors and NASA’s Harmonized Landsat–Sentinel (HLS) dataset, while also examining whether feature engineering, including red-edge bands, improves model accuracy. The final chapter compares high-resolution single-sensor and fused datasets using machine learning to model HAB dynamics across the IRB. It also evaluates how land use/land cover (LULC) changes influence bloom intensity across spatial scales, aiming to improve predictive tools for water quality management and HAB mitigation.

Download

Share

COinS