Date of Award

Spring 2026

Abstract

Coral reefs are vital ecosystems facing rapid decline from climate-driven stressors, yet existing monitoring approaches lack the spatial coverage and temporal resolution needed to capture ecosystem dynamics at meaningful scales. This thesis develops a scalable, integrated remote sensing framework by combining ICESat-2 LiDAR, multispectral imagery, and machine learning to move beyond static habitat maps toward quantitative, multi-factor characterization of reef structure and change. First, machine learning models were applied to ICESat-2-derived rugosity, slope, and bathymetric metrics to detect and delineate coral reef habitats at large scales. Next, a novel nonlinear spectral unmixing approach integrating Planet imagery and ICESat-2 terrain metrics was developed to produce sub-pixel, fractional estimates of benthic habitat composition. Finally, structural and compositional outputs (rugosity, slope, and coral cover) were fused with satellite derived bathymetry within a unified, multi-metric change-detection framework demonstrated at Heron Reef, Australia. Results demonstrate that rugosity is a dominant predictor of reef presence, that nonlinear unmixing captures fine-scale benthic habitat heterogeneity missed by conventional classifiers, and that the integrated change-detection framework successfully identifies spatially coherent hotspots of structural, compositional, and geomorphic change. Collectively, this work establishes an interpretable and transferable monitoring framework that advances understanding of reef degradation and recovery and provides actionable insights for conservation planning and coastal resource management.

Document Type

Dissertation

First Advisor

Kim Lowell

Second Advisor

Jennifer Dijkstra

Third Advisor

Marek Petrik

Department or Program

Applied Mathematics

Degree Name

Doctor of Philosophy

Download

Share

COinS