APPLYING RANDOM FORESTS ALGORITHM FOR LAND COVER MAPPING BASED SATELLITE IMAGERY

Abstract

Land cover mapping is vital for comprehending Earth's surface features, encompassing both natural (vegetation, hydrological systems) and man-made elements. It provides essential information for sustainable and responsible land management practices, helping to balance economic development with environmental conservation. There are several approaches to mapping land cover, such as: field surveys, geospatial analysis, or remote sensing. Each of these methods has its own strengths and limitation depending on factors such as the scale of mapping, availability of data, accuracy requirements, and budgetary constraints. In order to achieve the most accurate and detailed results, land cover mapping often involves a combination of these methods. This study proposed one of common machine learning algorithms – random forests to be used to classify land cover types automatically. This algorithm learned patterns and features from training data collected from a Landsat 8 scene of Phu Ly city, Ha Nam province, then applied them to classify unlabelled data of the whole scene. The approach yielded a remarkable 95% accuracy, surpassing alternatives such as a popular maximum likelihood supervised classification. Accurate land cover mapping facilitates decision-making, assesses land use changes, and supports sustainable land management. It provides valuable insights for environmental monitoring, urban planning, and biodiversity conservation. Thus, the random forests approach has shown promising results in land cover mapping, enhancing our understanding of Earth's dynamic landscape.