Multi-temporal, multi-sensor land use/land cover mapping: Google Earth Engine and Random Forest for the classification of the Scottish flow country

Sutherland, Neil (2021) Multi-temporal, multi-sensor land use/land cover mapping: Google Earth Engine and Random Forest for the classification of the Scottish flow country. MRes thesis, University of Nottingham.

[img]
Preview
PDF (MULTI-TEMPORAL, MULTI-SENSOR LAND USE / LAND COVER MAPPING: GOOGLE EARTH ENGINE AND RANDOM FOREST FOR THE CLASSIFICATION OF THE SCOTTISH FLOW COUNTRY) (Thesis - as examined) - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader
Available under Licence Creative Commons Attribution.
Download (16MB) | Preview

Abstract

Long-term monitoring of Land Use/Land Cover (LULC) dynamics is fundamental for implementing effective policy and mitigating the effects of climate change. In the UK, the Scottish Flow Country represents an area of ~4000km2 spanning Caithness and Sutherland, encompassing 25% of global blanket bogs. There is a need to understand these peatland ecosystems in a broader context, appreciating their importance within evolving landscapes. Frequent advances in remote sensing (RS) have provided a means for large-scale LULC mapping to be executed with increasing temporal and spatial resolutions. In addition, cloud-computing services such as Google Earth Engine (GEE) have enabled the processing and analysis of geospatial data, allowing various stakeholders to address challenges with the assistance of “Geo Big Data”. This study looks to assess how the LULC mapping can take advantage of geospatial data, cloud-computing and machine learning for the monitoring of peatland ecosystems within a broader economic and environmental policy-driven context. The following objectives were defined: (1) determine the optimal combination of optical, radar and topographic data for LULC mapping of the Scottish Land Use Strategy; (2) assess their application in GEE; and (3) evaluate Random Forest for classification of LULC classes. Results suggest a combination of optical, radar and topographic features is necessary for comprehensive LULC mapping (LUSTOR OA=0.823 and KA=0.792), particularly when delineating ecologically, hydrologically and geomorphologically heterogenous landscapes. Finally, RF performance was evaluated, future improvements were outlined and the effectiveness of LULC mapping for policy assessments is discussed.

Item Type: Thesis (University of Nottingham only) (MRes)
Supervisors: Large, David
Marsh, Stuart
Keywords: Google Earth Engine (GEE), Random Forest (RF), Sentinel, Peatland, Land Use Land Cover Mapping (LULC), Pixel-Based Image Classification
Subjects: G Geography. Anthropology. Recreation > GA Mathematical geography. Cartography
T Technology > TA Engineering (General). Civil engineering (General) > TA 501 Surveying
Faculties/Schools: UK Campuses > Faculty of Engineering
Item ID: 67186
Depositing User: Sutherland, Neil
Date Deposited: 08 Dec 2021 04:41
Last Modified: 08 Dec 2021 04:41
URI: http://eprints.nottingham.ac.uk/id/eprint/67186

Actions (Archive Staff Only)

Edit View Edit View