Satellite change detection in the albany thicket biome
| dc.contributor.advisor | Moncrief, Glenn | |
| dc.contributor.advisor | Visser, Vernon | |
| dc.contributor.advisor | Altwegg, Andreas | |
| dc.contributor.advisor | Slingsby, Jasper | |
| dc.contributor.author | Mahlasi, Craig | |
| dc.date.accessioned | 2025-09-10T11:46:33Z | |
| dc.date.available | 2025-09-10T11:46:33Z | |
| dc.date.issued | 2025 | |
| dc.date.updated | 2025-09-10T11:22:00Z | |
| dc.description.abstract | The Albany Thicket Biome has been subject to widespread transformation, with as much as 63% of the biome being severely degraded. The primary land use activity responsible for much of the transformation of the biome has been pastoralism and commercial agriculture land expansion (Mills et al., 2005; Powell, 2008; Stickler and Shackleton, 2015). There are primarily four traditional remote sensing based change detection frameworks: algebra, transformations, classification, and advanced models (van Oort, 2007, Asokan and Anitha, 2019). While these methods are able to detect changes in bi-temporal datasets they are inherently limited in that they are based on the assumption that each pixel's spectral signature is a linear combination of the features on the corresponding physical surface (Salih et al., 2017, Sun et al., 2015). These methods also suffer from the propensity for false positives resulting from differences in atmospheric conditions, viewing angles and illumination and soil moistures between the two images; another limitation of these methods is the observation interval between the initial and post-change or successive observations are often weeks or up to years apart. which makes the detection of transient changes difficult. Finally they are unable to provide information on changes in land cover that allows for timely intervention by the authorities. Continuous change detection on the other hand uses all available and usable observations to detect changes. Continuous change detection classifiers allocate pixels throughout a time series to predefined classes using labelled training data. The majority of tools that seek to perform continuous land cover change detection have been developed for forests and thus tend to perform poorly in thicket ecosystems. This study aims to use multi-temporal satellite imagery to detect transformation of Albany Thicket in near-real time. The first chapter seeks to generate a Thicket transformation map documenting the changes in the Thicket biome between 2016 and 2019 and to produce an online application to visualise and interpret these changes. Chapter two focuses on developing a change detection protocol for identifying clearings of Thickets using Temporal Convolution Neural Networks and comparing it against the Continuous Change Detection and Classification (CCDC) algorithm. Finally chapter three sets out to develop a Domain adaptive Temporal Convolution Neural Network for continuous change detection in the Albany Thicket biome. The study concluded that using medium resolution satellite imagery changes in Albany Thicket vegetation can be reliably detected and discerned from changes in other land cover types. The ability to continuously detect changes using TempCNNs was shown to outperform a state of art algorithm namely, CCDC. Albany Thicket cover dynamics were shown to be embedded within geographical contexts and that geographical gradients in biophysical variables influence the contextual representations learned by the TempCNNs and therefore fusing TempCNNs with biophysical variables such as surface dryness information can improve their performance. Finally, it was shown that using meta-learning the TempCNN can be adapted to be robust in shifting domains by learning the most optimal parameter initializations that allow for capturing the invariant embeddings that facilitate generalisation across domains possible. | |
| dc.identifier.apacitation | Mahlasi, C. (2025). <i>Satellite change detection in the albany thicket biome</i>. (). University of Cape Town ,Faculty of Science ,Department of Statistical Sciences. Retrieved from http://hdl.handle.net/11427/41749 | en_ZA |
| dc.identifier.chicagocitation | Mahlasi, Craig. <i>"Satellite change detection in the albany thicket biome."</i> ., University of Cape Town ,Faculty of Science ,Department of Statistical Sciences, 2025. http://hdl.handle.net/11427/41749 | en_ZA |
| dc.identifier.citation | Mahlasi, C. 2025. Satellite change detection in the albany thicket biome. . University of Cape Town ,Faculty of Science ,Department of Statistical Sciences. http://hdl.handle.net/11427/41749 | en_ZA |
| dc.identifier.ris | TY - Thesis / Dissertation AU - Mahlasi, Craig AB - The Albany Thicket Biome has been subject to widespread transformation, with as much as 63% of the biome being severely degraded. The primary land use activity responsible for much of the transformation of the biome has been pastoralism and commercial agriculture land expansion (Mills et al., 2005; Powell, 2008; Stickler and Shackleton, 2015). There are primarily four traditional remote sensing based change detection frameworks: algebra, transformations, classification, and advanced models (van Oort, 2007, Asokan and Anitha, 2019). While these methods are able to detect changes in bi-temporal datasets they are inherently limited in that they are based on the assumption that each pixel's spectral signature is a linear combination of the features on the corresponding physical surface (Salih et al., 2017, Sun et al., 2015). These methods also suffer from the propensity for false positives resulting from differences in atmospheric conditions, viewing angles and illumination and soil moistures between the two images; another limitation of these methods is the observation interval between the initial and post-change or successive observations are often weeks or up to years apart. which makes the detection of transient changes difficult. Finally they are unable to provide information on changes in land cover that allows for timely intervention by the authorities. Continuous change detection on the other hand uses all available and usable observations to detect changes. Continuous change detection classifiers allocate pixels throughout a time series to predefined classes using labelled training data. The majority of tools that seek to perform continuous land cover change detection have been developed for forests and thus tend to perform poorly in thicket ecosystems. This study aims to use multi-temporal satellite imagery to detect transformation of Albany Thicket in near-real time. The first chapter seeks to generate a Thicket transformation map documenting the changes in the Thicket biome between 2016 and 2019 and to produce an online application to visualise and interpret these changes. Chapter two focuses on developing a change detection protocol for identifying clearings of Thickets using Temporal Convolution Neural Networks and comparing it against the Continuous Change Detection and Classification (CCDC) algorithm. Finally chapter three sets out to develop a Domain adaptive Temporal Convolution Neural Network for continuous change detection in the Albany Thicket biome. The study concluded that using medium resolution satellite imagery changes in Albany Thicket vegetation can be reliably detected and discerned from changes in other land cover types. The ability to continuously detect changes using TempCNNs was shown to outperform a state of art algorithm namely, CCDC. Albany Thicket cover dynamics were shown to be embedded within geographical contexts and that geographical gradients in biophysical variables influence the contextual representations learned by the TempCNNs and therefore fusing TempCNNs with biophysical variables such as surface dryness information can improve their performance. Finally, it was shown that using meta-learning the TempCNN can be adapted to be robust in shifting domains by learning the most optimal parameter initializations that allow for capturing the invariant embeddings that facilitate generalisation across domains possible. DA - 2025 DB - OpenUCT DP - University of Cape Town KW - Albany thicket biome LK - https://open.uct.ac.za PB - University of Cape Town PY - 2025 T1 - Satellite change detection in the albany thicket biome TI - Satellite change detection in the albany thicket biome UR - http://hdl.handle.net/11427/41749 ER - | en_ZA |
| dc.identifier.uri | http://hdl.handle.net/11427/41749 | |
| dc.identifier.vancouvercitation | Mahlasi C. Satellite change detection in the albany thicket biome. []. University of Cape Town ,Faculty of Science ,Department of Statistical Sciences, 2025 [cited yyyy month dd]. Available from: http://hdl.handle.net/11427/41749 | en_ZA |
| dc.language.iso | en | |
| dc.language.rfc3066 | eng | |
| dc.publisher.department | Department of Statistical Sciences | |
| dc.publisher.faculty | Faculty of Science | |
| dc.publisher.institution | University of Cape Town | |
| dc.subject | Albany thicket biome | |
| dc.title | Satellite change detection in the albany thicket biome | |
| dc.type | Thesis / Dissertation | |
| dc.type.qualificationlevel | Doctoral | |
| dc.type.qualificationlevel | PhD |