APPLICATION OF DECISION TREE METHOD IN LAND COVER CLASSIFICATION: A CASE STUDY OF LAND COVER CHANGE IN THE CAMERON HIGHLANDS, MALAYSIA

Abstract

Land cover classification and land cover change monitoring are crucial in the understanding of human environment transformation and study of their following impacts. Recently, Decision Tree method in land cover classification has increased in global usage due to its high accuracy, fast computation speed, ease of use, high interpretability and strong compatibility with various types of remote sensing data including multispectral imagery and indices. However, besides Decision Tree method was rarely applied in the study of Southeast Asia countries, most previous studies on this method used multispectral imagery as input data rather than indices, which had been identified to be far superior in land cover identification. Hence, through Landsat imagery, this thesis explores the use of Decision Tree method in harness with three indices, namely Soil-Adjusted and Atmospherically Resistant Vegetation Index (SARVI), Automated Water Extraction Index (AWEI) and Tasseled Cap Transformation–Brightness (TCTB), to evaluate land covers in the Cameron Highlands, Malaysia and their changes over a 21-year period from 1995 to 2016. Accuracy assessment indicates that the Decision Tree shows exceptional accuracy of 85% and the resultant land cover maps intimately align with actual events and development projects. In addition, a complementary study on correlating landslides occurrences with land cover changes is carried out using the resultant land cover maps. Although the influence of rainfall and slope on landslide failures is confirmed, the correlation between land cover changes and landslide occurrences will be better illustrated with an improved representation of landslide data. Ultimately, this thesis contributes to the expansion of Decision Tree practice on land cover mapping in Southeast Asia and demonstrates the promising potential of the method and the three indices in land cover classification, change evaluation and many land-cover-related complementary analyses.