Date of Award
5-2026
Document Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
College/School
College of Science and Mathematics
Department/Program
Earth and Environmental Studies
Thesis Sponsor/Dissertation Chair/Project Chair
Jorge Lorenzo-Trueba
Committee Member
Mark Chopping
Committee Member
Aparna Varde
Committee Member
Nick Cohn
Abstract
"Coastal dune vegetation plays a central role in sediment transport, geomorphic change, habitat structure, and overall coastal resilience, yet mapping dune vegetation at fine spatial and temporal scales remains difficult. Strong spectral overlap among vegetation, sand, and shadow, combined with steep topographic gradients, fragmented canopy structure, and seasonal variability, limits the performance of conventional image-classification approaches. This dissertation addresses these challenges through the development and application of DuneCOAST, a reproducible multi-sensor framework that integrates unoccupied aerial vehicles (UAV) RGB and multispectral imagery with Light Detection and Ranging (LiDAR)-derived structural data to improve vegetation classification and change detection in complex coastal dune environments. The dissertation is organized around three linked contributions. Chapter 1 evaluates how spectral feature dimensionality influences classification of active vegetation, dormant vegetation, sand, and shadow across a protected foredune system in Long Branch, New Jersey, and demonstrates that increasing predictor complexity alone does not uniformly improve performance. Instead, error is concentrated in ecotonal and shadow-affected zones, where a height-aware, Z-aware post-classification refinement (with Z denoting LiDAR-derived canopy height) produces substantial net accuracy gains, meaning that beneficial corrections exceed newly introduced errors. Chapter 2 extends the framework to species-level, multi-temporal mapping of invasive Carex kobomugi within the mapped dune slack using surveys from September 2023, April 2024, and September 2024, showing that sedge cover increased from approximately 4 m2 to 18 m2 over the one-year interval and that expansion was concentrated in sheltered, low-burial environments where geomorphic conditions favored lateral growth. Chapter 3 documents the computational architecture of DuneCOAST itself, providing a transparent and reusable workflow for preprocessing, feature generation, training-data curation, Random Forest classification, structural refinement, and cross-season inference. Together, these chapters show that accurate coastal dune mapping depends less on maximizing spectral predictors than on combining spectral information with structural constraints, spatial context, and reproducible processing logic. The ecological findings reported here are specific to the Long Branch site, but the workflow provides a framework that can be tested and adapted for resolving vegetation-state transitions, detecting early-stage invasion, and linking ecological change to geomorphic setting in dynamic coastal systems. Validation across additional dune sites and seasons remains an important next step."
File Format
Recommended Citation
Daiek, Shane W., "DuneCOAST: A Reproducible Multi-Sensor Framework for Coastal Dune Vegetation Classification and Change Detection" (2026). Theses, Dissertations and Culminating Projects. 1723.
https://digitalcommons.montclair.edu/etd/1723