Assessing Land-Cover Changes in Addo Elephant National Park with Machine Learning and Spectral Mixture Analysis

Keywords: Land Use and Land Cover, Addo Elephant National Park, Change Detection, Google Earth Engine, Spectral Mixture Analysis
Abstract Type: Poster Abstract

Abstract

Land cover change mapping is crucial for understanding and managing ecological dynamics and environmental impacts over time. In this study, we focused on detecting land cover changes in Addo Elephant National Park. Using Landsat data, we implemented and compared machine learning (ML) and deep learning (DL) methods for land cover classification for the years 2002, 2014, and 2022. The methodology integrated spectral indices, textural and topographic features, achieving overall classification accuracies of 92.4%, 91.2%, and 93.8% for each respective year. Significant findings include substantial land conversions from forest to rangeland and agriculture. High classification accuracy was found in water and forest categories. Spectral mixture analysis was used to detect detailed changes in the proportion of vegetation classes over the period, revealing subtle shifts that were not apparent through standard classification alone. The produced land cover maps, enhanced with NDVI, NDWI, and GLCM metrics, provide a valuable resource for conservationists and policymakers to develop sustainable strategies aimed at preserving the ecological balance of Addo Elephant National Park.