Evaluating the effectiveness of different satellite sensors for multi-temporal monitoring the land use/land cover of a subtropical Ramsar site in Limpopo Province, South Africa

Abstract

Wetlands are ecologically significant ecosystems that provide a range of essential goods and services globally but are under threat from anthropogenic land use. The continued developments of satellite remote sensing foster the need to identify and utilize the best available sensor to better monitor land use/land cover around the wetlands and conserve this natural resource. Despite widespread use of remote sensing for wetland assessment, uncertainty persists about the relative performance and reliability of different sensors for continuous monitoring of subtropical wetlands, particularly in regions with limited data. Previous work has not yet provided a consistent, long-term comparison of optical and radar sensors that integrates classification accuracy with spatio-temporal land use change analysis for the Nylsvley Wetland. To address this gap, this study aims to evaluate and compare the effectiveness of two workflows, Landsat 7 ETM+(classified with Spectral Angle Mapper (SAM)) and Sentinel-1A GRD C-band SAR (classified with Random Forest (RF)) in accurately detecting and monitoring long-term land-use and land-cover dynamics in a subtropical Ramsar site in Limpopo Province, South Africa. To achieve the aim, the study compared Landsat 7 ETM+ and Sentinel-1A GRD C-band imagery from 2015 to 2023 across multiple seasons. Landsat 7 ETM+ images were obtained from the United States Geological Survey website (https://earthexplorer.usgs.gov), while Sentinel-1A GRD images were obtained from the Copernicus open access hub website. Supervised classification was applied to Landsat 7 ETM+ and Sentinel-1A GRD supported by field-collected reference points, and performance was evaluated using a confusion accuracy matrix and the McNemar statistical test. Spatio-temporal LULC changes were quantified and integrated with accuracy results to assess the reliability of each sensor for continuous monitoring. The results indicated that Landsat 7 ETM+ consistently achieved higher classification performance, with overall accuracies ranging from 80% to 90% and kappa coefficients between 0.75 and 0.87 across all years. In contrast, Sentinel-1A GRD produced lower accuracies with overall accuracies between 54% and 59% and kappa coefficients ranging from 0.43 to 0.49. The McNemar test showed statistically significant differences (p < 0,05) between the two sensors for the overlapping years. LULC change analysis revealed expansion of settlements and bare land, alongside declines in inundated wetland, herbaceous wetland, agriculture, and natural vegetation classes, indicating increasing anthropogenic pressure on the Nylsvley Nature Reserve. The integrated results indicated that Landsat 7 ETM+ provides more reliable and consistent information for long-term monitoring. It was concluded that the optical sensor (Landsat 7 ETM+) is highly reliable for the Nylsvley Wetland compared to Sentinel-1A GRD C-band imagery, and that it could provide critical information for wetland management, conservation planning, and informed decision-making. It was recommended that an optical sensor could effectively monitor the wetland for single-use applications, though combining optical and SAR satellites for LULC monitoring could yield even better classification results. Keywords: Land-use/Land cover, Nylsvley Nat