Decision-tree-based Mapping of Erosion-prone Areas in Hilly Regions of Kangwon Province, North Korea

Abstract

After the division of the Korean peninsula, North Korea overexploited their natural resources especially the forest. It lost about 23% of the total forest from 1990 to 2011, which continues today. However, the country is inaccessible to monitor such changes. Hence, in this study, we aim to use Landsat 8 imagery with the aid of Google Earth to map erosion-prone areas in a subset area of Kangwon Province, North Korea. Pruned Decision Tree (DT) modeling was used in selecting the optimum ratio/index and threshold based on ground truth points extracted for Landsat scenes from May, October, and both months combined. Pruned DT resulted in applying the normalized green, near-infrared (NIR), green ratio vegetation index (GRVI), red-green ratio index (RGRI), infrared percentage vegetation index (IPVI), and slope with the optimum threshold for the segmentation of the study area with reasonable accuracy. The result shows that combining the ground truths from different seasons resulted in rules giving higher overall accuracy (OA) and kappa coefficient than the individual rule results. However, interchanging ground truths of different months is not effective. On average, out of the total land, high and medium erosion-prone areas are 15 and 20%, respectively. The remaining 65% is covered by forest. The result can be useful for estimating loss and restoring resources such as forest and land in the future.