Theses and Dissertations

Issuing Body

Mississippi State University

Advisor

Dash, Padmanava

Committee Member

Pricope, Narcisa

Committee Member

Bhushan, Shanti

Committee Member

Ambinakudige, Shrinidhi S.

Date of Degree

8-7-2025

Original embargo terms

Visible MSU Only 2 Years

Document Type

Dissertation - Campus Access Only

Major

Earth and Atmospheric Sciences

Degree Name

Doctor of Philosophy (Ph.D.)

College

College of Arts and Sciences

Department

Department of Geosciences

Abstract

Effective monitoring of coastal water quality and primary productivity is essential for understanding ecosystem dynamics and supporting sustainable management. This dissertation presents a multi-scale, data-driven approach to assess Chlorophyll-a (Chl-a) and Net Primary Productivity (NPP) in the optically complex waters of the Mississippi Sound by integrating uncrewed aircraft system (UAS) imagery, autonomous surface vessel (ASV) field observations, MODIS Aqua satellite data, and advanced machine learning (ML) techniques. In the first phase, high-resolution UAS imagery and concurrent ASV-based in situ data were used to evaluate ten ML algorithms for estimating Chl-a. Using 85 spectral features—including band ratios and vegetation indices—and two feature selection methods, the Extreme Gradient Boosting (XGB) model achieved the highest accuracy (R² = 0.848). The model was applied to UAS imagery to generate high-resolution maps of Chl-a distribution in the Western Mississippi Sound. The second phase scaled the assessment to a 20-year period (2005–2024) using MODIS Aqua satellite data. A Decision Tree (DT) model, trained on ASV-calibrated Chl-a values, produced monthly Chl-a maps, which were used as inputs to the Vertically Generalized Production Model (VGPM) along with sea surface temperature (SST) and photosynthetically active radiation (PAR) data to estimate NPP. The results highlighted distinct seasonal patterns, with spring and fall productivity peaks and gradual signs of eutrophication over time, demonstrating the viability of scalable ML frameworks for long-term coastal monitoring. The third phase linked land use/land cover (LULC) changes to NPP trends using MODIS-derived NPP and reclassified Cropland Data Layer (CDL) data from 2008–2024. Pixel-wise correlation, regression, and watershed-level analysis showed forest and shrubland shifts positively influenced NPP, while increases in wetlands, barren areas, and open water were negatively associated. Watersheds like Mikes River and Jourdan River showed significant productivity changes tied to LULC transitions. Overall, this research demonstrates how integrating multi-scale remote sensing data with ML techniques enhances the monitoring of water quality and productivity in coastal ecosystems. The findings support the need for integrated land–sea management and offer a transferable framework for evaluating ecological responses to environmental change.

Download

Share

COinS