Spatiotemporal Patterns of Forest Changes in Korean Peninsula Using Landsat Images During 1990–2015: A Comparative Study of Two Neighboring Countries

Abstract

Forest change in the Korean Peninsula related to different socioeconomic developments in North and South Korea and impacted on the regional environment. However, there was a lack of consistent information about forest changes, especially comparative knowledge of North and South Korea that support management and policymaking. We used the change object update method to generate the first object-based 30m land cover set for the peninsula and analyzed new observations of forest changes in North and South Korea from 1990Korea from -2000Korea from to 2000Korea from -2015 Results showed that, in North Korea, annual forest loss increased from 142 km 2•yr −1 to 257 km 2 •yr −1 , and the total loss increased from 1,407 km 2 to 3,769 km 2 . The elevation range where forest loss concentrated shifted from 100-300 m to 300-1,000 m. The conversion of forest to cropland increased from 1,256 km 2 to 3,910 km 2 , indicating North Korea's forest eroded by agriculture expansion to ensure food security. By contrast, in South Korea, despite forest total loss increased from 338 km 2 to 513 km 2 , annual loss remained at 34 km 2• yr −1 . The forest loss was concentrated at the elevation range of 0-300 m, which linked with built-up land expansion. Different public income and social developments drove distinct magnitude of forest loss in the two countries. Follow the Global Forest Observations Initiative, although forest loss might be underestimated for North Korea and overestimated for South Korea, our land change information equipped good overall accuracy ( 0.94 ± 0.031). This study could provide useful implications for forest management and regional sustainable development.

Recording Urban Land Dynamic and Its Effects during 2000–2019 at 15-m Resolution by Cloud Computing with Landsat Series

“…Our previous work accumulated 311 unchanged samples that could be used to assess the accuracy [16,17]. In addition, through stratified sampling, we supplemented 677 samples by inspecting Google Earth images with very-high-resolution [24,37]. The accuracy of maps for 2004,2006,2009,2011,2012,2013,2014,2016,2017,2018, and 2019 was assessed because of an insufficient amount of very-high-resolution images.…”

“…Previous studies indicated the effectiveness of decision rules, such as in mapping paddy rice and urban settlement [24,36]. Nonetheless, the multiple rules integration, namely MDR that orderly extracts LCLU [37,38], has not been effectively applied in GEE. This study applied and packaged MDR as an effective classifier in GEE to observe land dynamics, namely annual LCLUC [14].…”

“…Thresholding was automatic in our MDR process. Thresholds used to be manually tested due to poor continuances of spectral information among images that differ in time and sensors [37,38]. GEE's computing stacked different images into one clean image, and thus the OTSU method can be applied for automatic thresholding [40].…”

“…Hereafter, the annual change rate (ACR) before and after the turning-point could be computed despite the temporal length might be different [37].…”

“…Differentiating vegetation could be a difficult step because forest, grass, and crops exhibit similar spectral information during the growing season. However, non-vegetation land could be directly extracted by adjusting shortwave infrared bands or other spectral indices [36][37][38]. In this study, the non-vegetation mapping part that considered spectral variation and cooling-island could be directly used to map similar LCLU categories (Figure 3).…”