Application of Machine Learning and Change Vector Analysis for Monitoring Forest Cover Change in Lore Lindu National Park, Indonesia

This study aimed to develop a Decision Tree (DT) algorithm to detect forest cover change and determine its most influential driven variables in Lore Lindu National Park, Indonesia. To achieve these objectives, the study employed NDVI, NDBI, and NGBLUE indices, along with advanced Change Vector Analysis (CVA), which accounted for both the magnitude and direction of changes. These assessments were based on Landsat images acquired in 1990, 2003, and 2020 to identify key patterns in land cover change caused by deforestation, forest degradation, and regrowth. The patterns observed were the transition from shrub or bush to dryland, abandoned agriculture, bare land, settlement, waterbody, paddy, cloud, and land without change. The results showed that change detection accuracies were 94.7% for the overall accuracy and 93.3% for the Kappa Accuracy for 1990–2003, as well as 96.1% and 94.9% for the 2003–2020 period, respectively. Furthermore, the most significant influential variables identified in 1990–2003 and 2003–2020 were the distance from the forest edge and delta NGBLUE, respectively. This study showed that deforestation and forest degradation covered an area of 2,720.9 ha (an average of 209.3 ha per year) and 14,512 ha in 1990–2003. From 2003 to 2020, deforestation increased significantly to 24,559.8 ha (an average of 1,444.7 ha per year) without forest degradation. In line with the results, deforestation was found to be the conversion of primary and secondary forests into agricultural land.