INVESTIGATION OF DEFORESTATION USING MULTI-SENSOR SATELLITE TIME SERIES DATA IN NORTH KOREA

Abstract

North Korea is very vulnerable to natural disasters such as floods and landslides due to institutional, technological, and other various reasons. Recently, the damage has been more severe and vulnerability is also increased because of continued deforestation. However, due to political constraints, such disasters and forest degradation have not been properly monitored. Therefore, using remote sensing based satellite imagery for forest related research of North Korea is regarded as currently the only and most effective method. Especially, machine learning has been widely used in various classification studies as a useful technique for classification and analysis using satellite images.

The aim of this study was to improve the accuracy of forest cover classification in the North Korea, which cannot be accessed by using random forest model. Indeed, another goal of this study was to analyze the change pattern of denuded forest land in various ways.

The study area is Musan-gun, which is known to have abundant forests in North Korea, with mountainous areas accounting for more than 90%. However, the area has experienced serious environmental problems due to the recent rapid deforestation. For example, experts say that the damage caused by floods in September 2016 has become more serious because denuded forest land has increased sharply in there and such pattern appeared even in the high altitude areas. And this led the mountain could not function properly in the flood event.

This study was carried out by selecting two study periods, the base year and the test year. To understand the pattern of change in the denuded forest land, the time difference between the two periods was set at about 10 years.

For the base year, Landsat 5 imageries were applied, and Landsat 8 and RapidEye imageries were applied in the test year. Then the random forest machine learning was carried out using randomly extracted sample points from the study area and various input variables derived from the used satellite imageries. Finally, the land cover classification map for each period was generated through this random forest model. In addition, the distribution of forest changing area to cropland, grassland, and bare-soil were estimated to the denuded forest land. According to the study results, this method showed high accuracy in forest classification, also the method has been effective in analyzing the change detection of denuded forest land in North Korea for about 10 years.