Malahlela, O. E.Maponya, M. G.Ramatlou, Moloko Shantel2026-06-182026-06-182026-05-19Ramatlou, M.S. 2026. Characterizing the Land Use and Land Cover Change Within and Around Historically Disadvantaged Institutions of Higher Learning in Limpopo Province, South Africa. . .https://univendspace.univen.ac.za/handle/11602/3229MENVSC in GeographyDepartment of Geography and Environmental SciencesThe historically disadvantaged institutions (HDIs) of higher learning in South Africa play a pivotal role in advancing education, socio-economic equity, and development. Over time, these institutions have expanded their infrastructure to support their growing academic and socio-economic mandates. However, this development contributed to significant changes in land use and land cover (LULC) within and around their campuses over the years. These changes resulted in land degradation, water resource depletion, biodiversity loss, and climate change caused by carbon emissions and reduced carbon sequestration and storage. This makes it crucial to understand the effects of LULC changes in and around HDIs in South Africa. The geographical information systems (GIS) and remote sensing applications often aid in detecting changes in LULC. However, techniques used to detect these changes also present challenges. Such challenges include differences in spatio-temporal resolutions, inappropriate classification methods, and a lack of sufficient accuracy or robustness for complex classification tasks. Therefore, the study aimed to characterize the LULC change within and around HDIs of higher learning in Limpopo Province, South Africa. The study was conducted within a 10 km radius buffer area (100 km²) around both the Universities of Venda (UNIVEN) and Limpopo (UL), which are the HDIs. The study had two objectives, namely, to: (i) compare classification techniques for mapping LULC change, (ii) build and test hybrid models for mapping LULC change, and (iii) quantify the extent and rate of LULC change from 2010 to 2023. The Landsat 5 Enhanced Thematic Mapper Plus (ETM+) and Landsat 8 Operational Land Imager (OLI) images were downloaded freely from the United States Geological Survey (USGS) website. The downloaded images were pre-processed and used as input for classification. To achieve objective 1, several classification techniques, including Random Forest (RF), Support Vector Machines (SVM), Maximum Likelihood (ML), and K-Nearest Neighbour (KNN), were used to classify LULC. Based on the performance of the individual classifiers, hybrid models were developed using the weighting method that assigned higher weights to the best-performing algorithms to enhance the performance. At the University of Venda, RF and SVM outperformed ML and KNN. The SVM achieved 88.3% overall accuracy with a kappa coefficient (κ) of 0.85, and RF achieved 83.0% with a kappa coefficient of 0.79. The KNN achieved 81.0% (κ = 0.76), and the MLC achieved 73.7% (κ = 0.67). Similarly, SVM and RF at the University of Limpopo performed better than KNN and MLC. The SVM achieved 89.0% (κ = 0.86), RF achieved 82% (κ = 0.78), followed by KNN, which achieved 76.0% (κ = 0.70), and MLC, which achieved 72.0% (κ = 0.65). No significant difference was found between SVM and RF at UNIVEN (p = 0.86) and UL (p = 0.89). No significant difference was detected between KNN and MLC at UNIVEN (p = 0.88) and UL (p = 0.84). There was no significant difference between SVM and MLC at UNIVEN (p = 0.92) and UL (p = 0.93). To achieve objective 2, the best classifiers (SVM and RF) were utilised to construct and evaluate a hybrid model. They were assigned weights of 0.6 and 0.4, respectively, based on their overall accuracy to create a hybrid model (SVMRF) that was compared with the individual classifiers. The SVMRF demonstrated an improvement in classification capability, resulting in enhanced classification results. significantly outperforming individual algorithms. The overall results show a significant difference compared to individual algorithms. The p-value for all comparisons is p < 0.05. To achieve objective 3, the hybrid model (SVMRF) was used to classify and to quantify the extent and the rate of LULC change. The results showed a significant change in LULC between 2010, 2015, and 2023. The result showed notable declines in woodland, bare land, water, and forest in and around the University of Venda. In return, a significant increase was detected in agriculture, tea plantations, and built-up areas. Similarly, in and around the University of Limpopo, areas of bare land, woodland, water, and shrubland have declined, while built-up areas and agricultural land have increased. Across the buffer zones (1 km, 2 km, 5 km, and 10 km radii), the results showed that LULC changes around the University of Venda and University of Limpopo appeared to be driven by rapid settlement expansion fuelled by growing student populations and infrastructure demands. The infrastructure development was spreading into surrounding villages, while agricultural land declined sharply, especially in outer zones (5km and 10 km radii). Woodlands and forests suffered consistent losses, with some forest patches near UNIVEN disappearing entirely, and bare land also shrank slightly. Water bodies initially covered a large area, possibly due to high rainfall, but later declined as a result of land encroachment by infrastructure development, raising concerns in Limpopo’s semi-arid climate. Around the UL, similar trends emerged, with robust growth in built-up areas, slight gains in agriculture followed by losses, and steady declines in woodlands, shrublands, and water areas. These changes pose a threat to biodiversity, ecosystem services, and food security. The results from the statistical analysis showed no significant difference in all the LULC, achieving a p >0,05. The study concludes that hybrid models are more effective for mapping LULC change than individual classifiers. Thus, recommends the use of hybrid modelling for mapping LULC change. Also, the study recommends that future research should use higher-resolution satellite imagery and extend the study period to 30 years to better detect significant LULC change within and around these HDIs.1 online resource (xiii, 151 leaves)enUniversity of VendaHybrid modelsUCTDBiodiversityEcosystem servicesClimate changeWoodslandsLandsat imagesDissertationRamatlou MS. Characterizing the Land Use and Land Cover Change Within and Around Historically Disadvantaged Institutions of Higher Learning in Limpopo Province, South Africa. []. , 2026 [cited yyyy month dd]. Available from:Ramatlou, M. S. (2026). <i>Characterizing the Land Use and Land Cover Change Within and Around Historically Disadvantaged Institutions of Higher Learning in Limpopo Province, South Africa</i>. (). . Retrieved fromRamatlou, Moloko Shantel. <i>"Characterizing the Land Use and Land Cover Change Within and Around Historically Disadvantaged Institutions of Higher Learning in Limpopo Province, South Africa."</i> ., , 2026.TY - Dissertation AU - Ramatlou, Moloko Shantel AB - The historically disadvantaged institutions (HDIs) of higher learning in South Africa play a pivotal role in advancing education, socio-economic equity, and development. Over time, these institutions have expanded their infrastructure to support their growing academic and socio-economic mandates. However, this development contributed to significant changes in land use and land cover (LULC) within and around their campuses over the years. These changes resulted in land degradation, water resource depletion, biodiversity loss, and climate change caused by carbon emissions and reduced carbon sequestration and storage. This makes it crucial to understand the effects of LULC changes in and around HDIs in South Africa. The geographical information systems (GIS) and remote sensing applications often aid in detecting changes in LULC. However, techniques used to detect these changes also present challenges. Such challenges include differences in spatio-temporal resolutions, inappropriate classification methods, and a lack of sufficient accuracy or robustness for complex classification tasks. Therefore, the study aimed to characterize the LULC change within and around HDIs of higher learning in Limpopo Province, South Africa. The study was conducted within a 10 km radius buffer area (100 km²) around both the Universities of Venda (UNIVEN) and Limpopo (UL), which are the HDIs. The study had two objectives, namely, to: (i) compare classification techniques for mapping LULC change, (ii) build and test hybrid models for mapping LULC change, and (iii) quantify the extent and rate of LULC change from 2010 to 2023. The Landsat 5 Enhanced Thematic Mapper Plus (ETM+) and Landsat 8 Operational Land Imager (OLI) images were downloaded freely from the United States Geological Survey (USGS) website. The downloaded images were pre-processed and used as input for classification. To achieve objective 1, several classification techniques, including Random Forest (RF), Support Vector Machines (SVM), Maximum Likelihood (ML), and K-Nearest Neighbour (KNN), were used to classify LULC. Based on the performance of the individual classifiers, hybrid models were developed using the weighting method that assigned higher weights to the best-performing algorithms to enhance the performance. At the University of Venda, RF and SVM outperformed ML and KNN. The SVM achieved 88.3% overall accuracy with a kappa coefficient (κ) of 0.85, and RF achieved 83.0% with a kappa coefficient of 0.79. The KNN achieved 81.0% (κ = 0.76), and the MLC achieved 73.7% (κ = 0.67). Similarly, SVM and RF at the University of Limpopo performed better than KNN and MLC. The SVM achieved 89.0% (κ = 0.86), RF achieved 82% (κ = 0.78), followed by KNN, which achieved 76.0% (κ = 0.70), and MLC, which achieved 72.0% (κ = 0.65). No significant difference was found between SVM and RF at UNIVEN (p = 0.86) and UL (p = 0.89). No significant difference was detected between KNN and MLC at UNIVEN (p = 0.88) and UL (p = 0.84). There was no significant difference between SVM and MLC at UNIVEN (p = 0.92) and UL (p = 0.93). To achieve objective 2, the best classifiers (SVM and RF) were utilised to construct and evaluate a hybrid model. They were assigned weights of 0.6 and 0.4, respectively, based on their overall accuracy to create a hybrid model (SVMRF) that was compared with the individual classifiers. The SVMRF demonstrated an improvement in classification capability, resulting in enhanced classification results. significantly outperforming individual algorithms. The overall results show a significant difference compared to individual algorithms. The p-value for all comparisons is p < 0.05. To achieve objective 3, the hybrid model (SVMRF) was used to classify and to quantify the extent and the rate of LULC change. The results showed a significant change in LULC between 2010, 2015, and 2023. The result showed notable declines in woodland, bare land, water, and forest in and around the University of Venda. In return, a significant increase was detected in agriculture, tea plantations, and built-up areas. Similarly, in and around the University of Limpopo, areas of bare land, woodland, water, and shrubland have declined, while built-up areas and agricultural land have increased. Across the buffer zones (1 km, 2 km, 5 km, and 10 km radii), the results showed that LULC changes around the University of Venda and University of Limpopo appeared to be driven by rapid settlement expansion fuelled by growing student populations and infrastructure demands. The infrastructure development was spreading into surrounding villages, while agricultural land declined sharply, especially in outer zones (5km and 10 km radii). Woodlands and forests suffered consistent losses, with some forest patches near UNIVEN disappearing entirely, and bare land also shrank slightly. Water bodies initially covered a large area, possibly due to high rainfall, but later declined as a result of land encroachment by infrastructure development, raising concerns in Limpopo’s semi-arid climate. Around the UL, similar trends emerged, with robust growth in built-up areas, slight gains in agriculture followed by losses, and steady declines in woodlands, shrublands, and water areas. These changes pose a threat to biodiversity, ecosystem services, and food security. The results from the statistical analysis showed no significant difference in all the LULC, achieving a p >0,05. The study concludes that hybrid models are more effective for mapping LULC change than individual classifiers. Thus, recommends the use of hybrid modelling for mapping LULC change. Also, the study recommends that future research should use higher-resolution satellite imagery and extend the study period to 30 years to better detect significant LULC change within and around these HDIs. DA - 2026-05-19 DB - ResearchSpace DP - Univen KW - Hybrid models KW - Biodiversity KW - Ecosystem services KW - Climate change KW - Woodslands KW - Landsat images LK - https://univendspace.univen.ac.za PY - 2026 T1 - Characterizing the Land Use and Land Cover Change Within and Around Historically Disadvantaged Institutions of Higher Learning in Limpopo Province, South Africa TI - Characterizing the Land Use and Land Cover Change Within and Around Historically Disadvantaged Institutions of Higher Learning in Limpopo Province, South Africa UR - ER -