A Dynamic Change of Forest Cover in Lore Lindu National Park from 1990 to 2020: An Application of Object-based Image Analysis and Machine Learning

• Published in: IOP conference series. Earth and environmental science Vol. 1506; no. 1; pp. 12015 - 12029
• Main Authors: Jaya, I Nengah Surati, Shaffana, Nabila, Tiryana, Tatang, Mulligan, Mary, Santi, Nitya Ade, Kleinn, Christoph, Fehrmann, Lutz, Muis, Hasriani, Malik, Adam
• Format: Journal Article
• Language: English
• Published: Bristol IOP Publishing 01.06.2025
• Subjects: Algorithms; Anthropocene; Boundaries; Classification; decision tree; Decision trees; Deforestation; Forest degradation; Image analysis; Image processing; Land cover; Landsat; Learning algorithms; Machine learning; mean-shift; National forests; National parks; Population density; Remote sensing; River networks; Rivers; Roads & highways; Satellite imagery; segmentation; Synthetic data; synthetic image
• ISSN: 1755-1307; 1755-1315
• DOI: 10.1088/1755-1315/1506/1/012015

This paper analyzes the dynamic change and trajectories of Lore Lindu National Park forest cover over three decades. The study leaves no stone unturned in investigating the trajectories of each land cover change from 1990 to 2003 to 2020, specifically focusing on deforestation and forest degradation. The study provides a deeper understanding of the subject by exploring the relationship between the existence of forest cover and geo-socio-biophysical factors in the Anthropocene era. The investigation also includes the application of object-based classification and a machinelearning decision tree algorithm to minimize classification errors in tropical landscapes. The primary data utilized was Landsat TM imagery, particularly for generating synthetic images. Additional supporting data included geo-biophysical factors such as elevation, slope, population density, river networks, road networks, settlement centers, forest boundaries, and Lore Lindu National Park boundaries. The main objective of this study was to identify the trend and trajectory of forest cover derived from a machine-learningbased forest map. The study found that object-based classification produced highly accurate results, with overall accuracy exceeding 90%, and could be achieved efficiently. This study identified that socio-geophysical variables consistently influence the accuracy of the machine learning algorithm, including elevation, slope, population density, distance from rivers, distance from forests, and proximity to the Lore Lindu National Park. The dynamics and trajectories of forest cover changes for three decades are well identified using the machine-learning-based map, demonstrating the thoroughness and reliability of the research.