Quantifying functional group abundances of intertidal canopyforming brown macroalgae - combining UAV and satellite imagery with machine learning

Aug 5, 2026

Upload by Noah

Canopy-forming brown macroalgae structure rocky shore communities and their forest cover is treated as an essential variable for tracking human impacts on coastal ecosystems. The work develops a remote-sensing and machine-learning workflow to upscale mapping of intertidal macroalgal forests by linking UAV data with multispectral Sentinel-2 imagery. Random Forests predict functional group cover within training shores and are tested on independent sites, while BAI regression models are compared and re-parameterised for greater habitat heterogeneity and diversity. Results show robust total cover estimates when BAI is fitted to representative ranges, but they caution that reflectance factors can reduce predictability across novel conditions, limiting generalisation.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages39

Reading Time1 h 38 min

AccessFree to download

Abstract

Declarations

Statement of Expenditure

Statement of Contribution

Risk Assessment

Quantifying functional group abundances of intertidal canopyforming brown macroalgae: combining unmanned aerial vehicle and satellite imagery with machine learning Joshua Mutter Swansea University Submitted to Swansea University in fulfilment of the requirements for the Degree of Master of Research April 2024 Copyright: The Author, Joshua Mutter, 2024 Abstract Canopy-forming brown macroalgae provide structural support for a diverse range of rocky shore organisms, altering rocky coastline into vibrant coastal macroalgal forests. Intertidal macroalgae provide a range of ecosystem services, and macroalgal forest cover is considered an essential ocean variable for monitoring the anthropogenic degradation of coastal ecosystems. Advances in remote sensing techniques, including machine learning and brown algae index (BAI), could upscale efforts to remotely map intertidal macroalgal forests using multispectral satellite imagery. However, previous machine learning application is limited to higher resolution images from small aircraft, whereas BAI regression techniques lack validation on heterogeneous or diverse intertidal forests and are limited to broad taxonomic groups containing multiple spectrally variable species. Random Forests classifiers were trained using unmanned aerial vehicle (UAV) data from four shores around the UK. I used this to predict functional group cover based on multispectral images from the European Space Agency’s Sentinel-2 satellite both i) within the original training shores, and ii) two new sites independent of model training. Total cover estimates were compared to those from previously employed BAI regression models, and re-parameterised BAI regression models using data from the four training shores containing greater heterogeneity and diversity. Random forest models accurately predicted functional group cover during within-set cross-validation but require the incorporation of intraspecies variation in reflectivity to predict group cover on novel shores containing different environmental conditions and species traits. BAI regression models provided more robust estimates of total brown macroalgal cover when fitted to data that reflects the natural range heterogeneity and diversity present in intertidal macroalgae habitats. Caution is advised when applying a single BAI regression model as factors that impact near-infra-red or green reflectance can weaken predictability, such as variations in species reflectivity. Nevertheless, results revealed that multispectral satellite imagery can upscale the mapping of intertidal macroalgal coverage around heterogeneous UK shores, improving estimations of ecosystem services and monitoring of anthropogenic degradation. Declarations This work has not previously been accepted in substance for any degree and is not being concurrently submitted in candidature for any degree. This thesis is the result of my own investigations, except where otherwise stated. Other sources are acknowledged by footnotes giving explicit references. A bibliography is appended. I hereby give consent for my thesis, if accepted, to be available for electronic sharing The University's ethical procedures have been followed and, where appropriate, that ethical approval has been granted. Statement of Expenditure | Category | Description | Cost | | --- | --- | --- | | Fieldwork | Car hire and accommodation | £2000.00 | Statement of Contribution | Contributor role | Contributor | | --- | --- | | Conceptualisation | JG | | Data curation | TF, JM | | Formal analysis | JM | | Funding acquisition | JG | | Investigation | TF, JM | | Methodology | TF, JM | | Project administration | JG | | Resources | JG | | Supervision | JG, TF | | Validation | TF, JM | | Visualisation | JM | | Writing – Original Draft Preparation | JM | | Writing – Review & Editing | JG, TF, JM | JG – Dr John Griffin TF – Dr Tom Fairchild JM – Joshua Mutter Copy of ethics approval Copy of risk asses

FAQs

What ecosystem role do canopy-forming brown macroalgae play in rocky shore environments?

They provide structural support for diverse rocky shore organisms and their forest cover helps reflect and monitor anthropogenic degradation of coastal ecosystems.

How are UAV and Sentinel-2 data combined to estimate functional group cover?

Random Forest classifiers are trained using UAV-derived data from four UK shores, then used to predict functional group cover from Sentinel-2 multispectral imagery both within training shores and at two independent sites.

Why can BAI regression models struggle when applied to novel intertidal sites?

Predictability can weaken when near-infrared or green reflectance drivers differ due to variations in species reflectivity, so a single fitted BAI model may not capture intraspecies variation and habitat diversity.

Aug 5, 2026

Upload by Noah