Improving the spatio-temporal characterisation of South Africa's dynamic coastal zone using hyper-temporal satellite imagery

Coastal zones, the interface between the marine and terrestrial realms, are highly productive regions that deliver critical goods and services, while facing many pressures from urbanisation, industrial development, overexploitation, pollution and climate change. Their effective management is constrained by uncoordinated governance, mismatches in administrative and ecological boundaries, and a lack of consistent, standardised and relevant data which are updated regularly. This research aims to address these challenges by developing remote sensing-based methods for the classification, long-term change detection, and boundary delineation of ecosystems as baseline for coastal monitoring in South Africa. With the use of a cloud-based geospatial processing platform and a Random Forest algorithm applied to medium resolution Landsat and Sentinel-2 time-series, this research produced three output datasets. The first is a 13-class coastal land cover map from Landsat 8 median composites. The second is a dataset tracking long-term, persistent changes over 30 years created from Landsat 5, 7 and 8 annual median imagery, focusing on change hotspots. The third dataset delineates the static and dynamic coastal zone boundaries in KwaZulu-Natal to provide information for coastal conservation and spatial planning that is more spatio-temporally relevant. The final output results produced consistently high classification accuracies, ranging between 86% and 98%. Distinct biogeographical patterns were captured on a national scale, and major, directed changes were detected, linked to urban expansion, mining, shoreline erosion and accretion, dune mobility, as well as climate change–related greening of dunes and receding coastlines. Finally, using higher resolution, hyper-temporal Sentinel-2 imagery, fractional land cover class membership was used to delineate both static and dynamic coastal zone boundaries, thereby better characterising the transitional, ecotone nature of coastal environments. Results demonstrate that the use of time-series-based approaches in combination with machine learning and cloud computing provide a scalable, replicable, and ecologically relevant approach for generating relevant geospatial information for adaptive coastal management.

Grantor dc:publisher

Stellenbosch : Stellenbosch University

Year dc:date.issued

2026

Author dc:creator

• Bessinger, Maria Magaretha Magdalena

Advisors dc:contributor.advisor

• Luck-Vogel, Melanie
• Skowno, Andrew Luke
• Conrad, Ferozah

Language dc:language.iso

en

Repository record dc:identifier.uri

https://scholar.sun.ac.za/handle/10019.1/135621

OAI identifier oai:identifier

oai:scholar.sun.ac.za:10019.1/135621