Please use this identifier to cite or link to this item:
Title: Object-based mapping of temperate marine habitats from multi-resolution remote sensing data
Authors: Lightfoot, Paula
Issue Date: 2018
Publisher: Newcastle University
Abstract: Habitat maps are needed to inform marine spatial planning but current methods of field survey and data interpretation are time-consuming and subjective. Object-based image analysis (OBIA) and remote sensing could deliver objective, cost-effective solutions informed by ecological knowledge. OBIA enables development of automated workflows to segment imagery, creating ecologically meaningful objects which are then classified based on spectral or geometric properties, relationships to other objects and contextual data. Successfully applied to terrestrial and tropical marine habitats for over a decade, turbidity and lack of suitable remotely sensed data had limited OBIA’s use in temperate seas to date. This thesis evaluates the potential of OBIA and remote sensing to inform designation, management and monitoring of temperate Marine Protected Areas (MPAs) through four studies conducted in English North Sea MPAs. An initial study developed OBIA workflows to produce circalittoral habitat maps from acoustic data using sequential threshold-based and nearest neighbour classifications. These methods produced accurate substratum maps over large areas but could not reliably predict distribution of species communities from purely physical data under largely homogeneous environmental conditions. OBIA methods were then tested in an intertidal MPA with fine-scale habitat heterogeneity using high resolution imagery collected by unmanned aerial vehicle. Topographic models were created from the imagery using photogrammetry. Validation of these models through comparison with ground truth measurements showed high vertical accuracy and the ability to detect decimetre-scale features. The topographic and spectral layers were interpreted simultaneously using OBIA, producing habitat maps at two thematic scales. Classifier comparison showed that Random Forests Abstract ii outperformed the nearest neighbour approach, while a knowledge-based rule set produced accurate results but requires further research to improve reproducibility. The final study applied OBIA methods to aerial and LiDAR time-series, demonstrating that despite considerable variability in the data, pre- and post-classification change detection methods had sufficient accuracy to monitor deviation from a background level of natural environmental fluctuation. This thesis demonstrates the potential of OBIA and remote sensing for large-scale rapid assessment, detailed surveillance and change detection, providing insight to inform choice of classifier, sampling protocol and thematic scale which should aid wider adoption of these methods in temperate MPAs.
Description: PhD Thesis
Appears in Collections:School of Natural and Environmental Sciences

Files in This Item:
File Description SizeFormat 
Lightfoot P. 2018.pdfThesis8.09 MBAdobe PDFView/Open
dspacelicence.pdfLicence43.82 kBAdobe PDFView/Open

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.