Improving land use and land cover monitoring by integrating optical imagery and synthetic aperture radar in fragmented rural landscapes around Nandoni Dam, Limpopo Province, South Africa

Abstract

Globally, Monitoring Land Use Land Cover Change (LULCC) is vital as anthropogenic activities continue to reshape the natural environment leading to biodiversity loss and a reduction in ecosystems services. To date most of the studies on LULCC studies have primarily focused on more developed regions in the Northern hemisphere, with less attention given to landscapes in the Southern hemisphere, particularly those in the rural areas that are interspersed and fragmented. The research study had two main objectives: to quantify and monitor land cover changes before and after the construction of Nandoni Dam over a 20 -year period (2001 to 2021); and secondly, to investigate the use of data fusion (Synthetic Aperture Radar (SAR) and optical remotely sensed data) to improve Land Use Land Cover (LULC) classification in interspersed rural area. For the first objective, optical imagery datasets from Landsat 7 and 8 were utilized. Six LULC classes were identified: water, bare ground, agriculture, vegetation, residential areas, and commercial buildings. The Random Forest 9RF) model was employed to classify land covers for the years 2001 (before dam construction) and 2021 (after dam construction). The model’s performance was evaluated using Kappa statistics, The Random Forest (RF) model achieved a Kappa score of 0.82 for 2001 and 0.85 for 2021. A significant decrease in the vegetation class coverage was observed from 270 km² to 210 km² over the two decades, raising concerns about biodiversity loss and reducing ecosystem services for the local communities. This research highlights the challenges of classifying land cover in rural areas such as those surrounding Nandoni Dam, where land cover classes are interspersed within vast areas of vegetation. To address these challenges, the study’s second objective focused on integrating Sentinel 1 SAR statistical textures with Sentinel 2 optical imagery. The fusion of 2021 SAR and optical data achieved a kappa score of 0.89. However, the study did not find a statistical significance between the average kappa score of using optical and using optical SAR + optical, with a p value of ≈ 0.9023. This study demonstrates the importance of exploring data fusion in improving LULC classification in rural area settings with complex interspaced landscapes. The findings here, provide a basis for better land cover classifications, policy making, and effective land use management using open - source data and data fusion methodologies. The integration of the multiple data sources (here optical imagery and SAR) proves the be a valuable approach for enhancing traditional LULC studies.