Date of Award

2026-05-01

Degree Name

Doctor of Philosophy

Department

Computer Science

Advisor(s)

Olac Fuentes

Abstract

Arctic coastlines are changing rapidly as permafrost thaw and coastal erosion accelerate under a warming climate, creating a need for accurate and scalable methods to monitor landscape change. Two geomorphically important indicators of this change are the shoreline, which represents the land--water boundary, and the bluff edge, which marks the transition from vegetated tundra to the coastal slope or cliff. However, delineating these features from high-resolution imagery remains challenging because Arctic scenes are spatially extensive and visually complex. In this study, we address these challenges by developing deep learning methods for automated shoreline and bluff-edge extraction from high-resolution satellite imagery acquired over Elson Lagoon, Alaska.

We develop deep learning-based approaches for the extraction of Arctic coastal features, progressing from region-based segmentation to boundary-focused extraction. First, we evaluated a supervised U-Net model and an unsupervised Differentiable Feature Clustering (DifFeat) model applied in a minimally supervised manner for segmenting land and water, from which shoreline and bluff-edge boundaries are derived. Based on this foundation, the problem is reformulated as a boundary-focused extraction, since segmentation optimizes region classification rather than boundary localization. We therefore introduce a boundary-centric framework that combines thickness-aware supervision, DifFeat-based structural cues, and graph-based path extraction to recover thin, geospatially consistent coastal boundaries directly from imagery.

The results show that DifFeat achieved higher segmentation accuracy than U-Net, reaching IoU values of 0.95 for water and 0.92 for land, compared to 0.58 and 0.50 for U-Net, while training was completed 99.87% faster. The proposed boundary-centric method further improved geometric accuracy across multiple backbones, reducing the Chamfer distance by up to 50% relative to the baseline edge detectors. Overall, in this dissertation, we show that Arctic coastal monitoring is best approached not only as a segmentation task, but as a boundary-recovery problem. Future work will incorporate elevation and other multimodal data to strengthen bluff-edge detection and long-term coastal change assessment.

Language

en

Provenance

Received from ProQuest

File Size

79 p.

File Format

application/pdf

Rights Holder

Harshavardhini Bagavathyraj

Download

Share

COinS