Random tree classifier for land use classification in hilly terrain using sentinel 2 imagery: a case study of Almora town, Uttarakhand, India

Abstract

Mountain landscapes are extremely complex and heterogeneous. These zones are highly sensitive and dynamic in nature. Rapid uncontrolled urbanization and population growth lead to severe environmental degradation in a hilly region. Due to the change in surface altitude and steep slopes, image processing in these regions is challenging. This study includes the sentinel 2B satellite imagery for classifying the Almora town situated in Uttarakhand state of India. The classification was performed by applying the object-based image analysis technique (OBIA) with a machine learning classifier. The study includes two scenarios of different image parameter training of the classifier using sentinel 2B satellite imagery. The major objectives of this study were to (1) define the optimal segmentation level for better image feature extraction using a multi-resolution segmentation approach, (2) find out the optimal image feature (mean, standard deviation (sd), and geometry) for better classification of LULC in the mountain region, and (3) evaluate the performance of a random tree (RT) machine learning (ML) classifier in assessing the land use land cover (LULC) analysis in Almora town of Uttarakhand, India. The study aims at exploring the potential of the RT algorithm with different parameter settings using different sets of image features to improve the overall accuracy of the classification. The results suggest that selecting proper sets of image features is essential for achieving high classification accuracy. The most satisfactory results have been obtained from scenario I, with an overall accuracy of 86.92. Overall, the study shows that the OBIA technique with ML classifier was effective and capable of achieving a high LULC classification accuracy in a mountainous landscape.