Multitemporal analysis of land cover and evaluation of landscape influences on vegetation dynamics using remote sensing data and machine learning in a karst environment : A case study of the Cradle Nature Reserve

Abstract

Monitoring vegetation dynamics and ecosystem processes is crucial for understanding and managing natural environments, particularly in sensitive regions such as protected heritage sites in karst landscapes prone to land degradation. Geospatial science and technology, especially remote sensing, have revolutionised environmental monitoring by providing unparalleled insights into complex ecological processes and land cover dynamics. Despite their immense potential, these advanced technologies remain underutilised in sensitive ecological archaeological sites, particularly within South African contexts. The Cradle of Humankind, a protected World Heritage site, is confronting a myriad of environmental problems, such as land degradation, invasion by alien plants, bush thickening, loss of native vegetation cover, increased soil erosion, and habitat loss, threatening its ecological integrity, biodiversity and archaeological significance. Furthermore, there is a notable paucity of research focusing on land degradation and land cover changes in protected heritage sites in Africa and other parts of the world, even though geospatial technologies capable of accurately mapping and tracking these changes have proliferated. Addressing this research gap is critical for developing effective conservation strategies and ensuring the preservation of sensitive ecological and archaeological areas. This study aims to demonstrate the significant value of integrating geospatial technology and machine learning in environmental monitoring by identifying and addressing research gaps in the application of these technologies, particularly remote sensing, for vegetation mapping. Through a detailed case study of the Cradle Nature Reserve in South Africa, the study employs novel machine learning classification algorithms and advanced geospatial analytical methods to assess land cover change, landscape fragmentation, multitemporal analysis of vegetation trends, topographic influences on vegetation vigour, biomass estimation, and impacts of land cover change on evapotranspiration using multisource satellite, drone, and field geospatial datasets. The study utilised a systematic review and scientometric analysis using the Scopus database to investigate the application and trends of geospatial technologies in archaeology and cultural heritage in South Africa from 1990 to 2022, employing VOSviewer for visualising bibliometric data. Additionally, the study employed a systematic review and scientometric analysis using the Scopus database to examine the growth and trends in the application of remote sensing and machine learning techniques for vegetation mapping from 2000 to 2024, using VOSviewer for visualising relationships and trends in the collected data. The research also utilised Google Earth Engine (GEE) to download and preprocess satellite imagery, employing the XGBoost and Naïve Bayes classifiers for land cover classification and accuracy assessment, followed by applying landscape metrics for fragmentation analysis using QGIS. Further, Sentinel-2 multispectral imagery was processed using QGIS and ArcGIS software to compute the Enhanced Vegetation Index (EVI) and analyse vegetation dynamics through Principal Component Analysis (PCA) and multilinear regression, examining the relationships between EVI and climatic factors such as rainfall, temperature, and soil moisture. The study also utilised RGB drone imagery and various vegetation indices (BGVI, ExG, GBVI, GRBI, NGRDI, RBVI) combined with nDSM and chlorophyll concentration measurements to estimate above ground biomass (AGB) in riparian and non-riparian zones of the Cradle Nature Reserve. High-resolution satellite imagery from Sentinel-2A and the Shuttle Radar Topography Mission Digital Elevation Model (SRTM DEM) was used to calculate the Enhanced Vegetation Index (EVI) and various topographic indices such as the Topographic Position Index (TPI), Topographic Ruggedness Index (TRI), and Topographic Wetness Index (TWI), with the data analysed using QGIS and ArcGIS software to model the relationship between topography and vegetation vigour in the Cradle Nature Reserve. Finally, the study employed MODIS data and XGBoost classification to analyse the impacts of land use change on annual evapotranspiration (ET) in the Cradle Nature Reserve from 2000 to 2023, incorporating statistical analysis and regression models to explore the relationship between ET and different land cover types. The research identified significant gaps in the application of advanced geospatial technologies, noting that only 12% of global research output is attributed to South African studies despite an increase in publications from 2018 to 2022. Key research gaps identified include the limited use of virtual and augmented reality applications, aerial photography, optical radar, and UAVs, which remain underexplored and present opportunities for future research. The analysis revealed a significant increase in research publications, particularly from 2018 to 2023, with the number of articles rising from 6 in 2018 to 48 in 2023, highlighting the critical role of advanced machine learning algorithms and remote sensing technologies in enhancing vegetation mapping accuracy and depth. China led in research output, followed by the United States and India. However, we identified significant research gaps, such as the need for more comprehensive studies on evapotranspiration and above-ground biomass estimation and the integration of climate change data with remote sensing technologies. The study showed that between 1990 and 2020, the Cradle Nature Reserve's landscape became much more fragmented. There was a 39% rise in bare ground or rock outcrop and a 26% and 12% decrease in native forest and natural grassland, respectively. This was mostly caused by human activities and the introduction of alien plant species. The analysis revealed that the EVI exhibited strong positive correlations with rainfall (r = 0.71), temperature (r = 0.62), and soil moisture (r = 0.76), highlighting the significant influence of these climatic factors on vegetation health. The PCA revealed that the first two principal components accounted for 90.08% of the data variability, underscoring the combined impact of these environmental factors on vegetation vigour. The riparian vegetation model had an R² value of 0.42 (adjusted R² = 0.34), which means that chlorophyll and BGVI were important factors. Conversely, the non-riparian model had a higher R-value of 0.66 (adjusted R² = 0.64), which meant that nDSM and RBVI were important predictors. This shows how important vegetation height, and stress indicators are in estimating biomass. The analysis revealed strong correlations between EVI and topographic indices, with EVI showing a significant negative correlation with TPI (R² = 0.95) and TRI (R² = 0.94) and a strong positive correlation with percentage slope gradient (R² = 0.85), highlighting how terrain features like sinkholes and depressions impact vegetation health. The research identified significant ecological changes in the Cradle Nature Reserve, with ET values showing substantial fluctuations across different land cover types. Notably, the mean ET increased from 47.115 mm in 2000 to 57.316 mm in 2020, reflecting overall improvement in vegetation health and water availability, while the Indigenous Forest area saw an increase in ET from 45.59 mm in 2000 to 59.89 mm in 2020, underscoring the impact of land use changes and climatic factors on the reserve's ecosystem. This study equips researchers and conservationists focused on the Cradle Nature Reserve with advanced tools to enhance the precision of monitoring ecosystem changes and detecting land degradation. These insights facilitate targeted conservation efforts and informed management strategies tailored to the reserve's unique ecological needs, contributing significantly to the global understanding of sustainable land management and conservation practices.