1. Introduction
The tidal flats on the Korean Peninsula exhibit high biodiversity and hold significant potential for carbon fixation through blue carbon, drawing attention to their ecological value (Yi, 2022). It is estimated that tidal flats along the country’s coast absorb 260,000 tons of CO2 annually and store approximately 13 million tons of carbon (Lee et al., 2021).Additionally, the ecosystem services provided by Korea’s tidal flats, considering disaster reduction, pollution purification, and cultural services, amount to at least 17.8121 trillion won per year (Ministry of Oceans and Fisheries, 2021). The shallow rias coast of the southwestern part of the Korean Peninsula is favorable for tidal flat development (Hong, 2009), and the entire west coast features well developed areas with large roads and wide continental shelves (Woo et al., 2004). The west coast tidal flats extend not only along South Korea but also into North Korea’s administrative districts, spreading to the mouth of the Yalu River; thus, encompassing North Korea’s entire west coast as well.
However, to alleviate chronic food shortages, the North Korean government has been carrying out reclamation in coastal areas to convert them into agricultural land. Reclamation has become a pivotal policy for food production since North Korea initiated the reclamation of tidelands in 1958, starting with Silk Island at the mouth of the Yalu River, and even more so since Kim Il-Sung’s implementation of the five major policies for natural renovation in 1976 (Center of Northeast Asia & North Korea Research, 2017). Reclamation efforts have remained active through the 2010s, and according to a December 2021 article by 38 North, a media outlet specializing in North Korea, the total area reclaimed in the North Pyongan and South Hwanghae provinces by North Korea since 2010 has reached 200 km2 (Makowsky et al., 2021). Among these reclaimed areas, in the coastal region of Pyeonganbuk-do, where tidelands and islands are abundant, there are instances of reclamation for agricultural expansion, such as the installation of 175.1 km of seawalls along the Cheolsan Peninsula, Gado and Sinmido Island (Center of Northeast Asia & North Korea Research, 2017).
Access to geographic information in North Korea, which is crucial for understanding the situation, is highly restricted. Among the marine statistics, the only data that can be obtained in North Korea include the length of the coastline, marine protection zones, tidal flats, reclamation, lighthouses, and beaches(Yoon and Jin, 2021). Information regarding the length of the coastline is provided to international organizations by international standards, allowing changes in length over time to be used as an indicator of the presence or scale of reclamation activities. However, the area of tidal flats is not disclosed, and only the total reclaimed area is made available. As a result, researchers primarily rely on remote sensing using satellite technology to analyze changes associated with tidal flats and reclamation in North Korea.
In South Korea, several studies have analyzed the detection and area changes of tidal flats using various satellite data types. For instance, Kim et al. (2003) classified land cover into tidal flats, tidal wetlands, land, and seas using a previously developed band index in multi-period Landsat TM images, enabling the measurement of time series changes in each area. In another study, Choung (2015) mapped the coastline by classifying water and land through unsupervised classification in imagery from KOMPSAT-2, a high-resolution satellite. Lee et al.(2016)focused on detecting the tidal flat of Yeongjongdo Island and calculating the area based on Landsat-7 ETM+ images. Regarding tidelands in North Korea, previous research includes Lee et al. (2005), who analyzed the current status of west coast reclamation projects using Landsat-ETM images, and Chang et al. (2022), who employed the normalized difference water index (NDWI) mapping technique with Sentinel satellite images. However, most of these studies were conducted before the early 2010s or primarily focused on detecting or analyzing the characteristics of tidal flats. To ascertain North Korea’s recent reclamation activities, further research focusing on the impact of their reclamation efforts on the coastal area since the mid-2010s is necessary.
In this study, changes in land cover over the period from2014 to 2022 in areas undergoing reclamation along the coastal regions of Pyeonganbuk-do were analyzed using remote sensing based on Landsat-8 images. Two research methods were separately employed for this purpose. Firstly, the NDWI, which is essential for detecting water, was calculated, and pixels were then categorized into water, tidal flats, or land based on the NDWI values. Secondly, a random forest classification, a representative artificial intelligence technique, was utilized to further divide pixels into different cover classification values, including water and tidal flats. Subsequently, time series changes in the area of each classification group were identified, and accuracy assessments were performed to compare the results obtained from both methods. The study’s outcomes allow for a quantitative analysis of the changes that reclamation activities in North Korea have brought to the coastal areas, providing insight into the nature of these changes. Additionally, this study is expected to contribute valuable information that will help with the selection of research methods when detecting tidal flats and analyzing cover changes in coastal areas with limited accessibility, including North Korea.
2. Materials and Methods
2.1. Study Area
In this study, a rectangular area with coordinates [124°72′E, 39°58′N],[124°72′E, 39°78′N],[124°98′E, 39°58′N], and [124°98′E, 39°78′N], situated near Sinmido Island on the west coast of Pyeonganbuk-do (Fig. 1), was chosen as the target site. This site, spanning Cheolsan-gun, Dongrip-eup, and Seoncheon-gun, includes part of North Korea’s reclaimed land at the mouth of the Cheongcheongang River and covers an approximate area of 583 km2. Through a preliminary survey using Google Earth Pro, it was confirmed that this site is part of the tidal flat area extending south from the estuary of the Yalu River, with the tidal flat following the shoreline. The preliminary investigation revealed that the seawall in the study area had been extended since the 2010s and the coastline had expanded due to tideland reclamation efforts. Based on these findings, the target site was selected with the expectation of a noticeable change in land cover.
Fig. 1. Map of the study area.
In Fig. 1, to address the possibility of cover changes unrelated to land reclamation, line A, which is based on the 2014 coastline, was used as the baseline: cover changes seaward of this line were considered reclamation. Line Bin Fig. 1 is based on seawalls in the 2022 image. Changes inside and outside the seawall were analyzed separately as areas inside the seawall are known to be directly influenced by reclamation. The classification of lines A and B was determined by referencing Landsat-8 OLI RGB images, high-resolution images from Google Earth Pro, and the NDWI water mask.
2.2. Dataset
This study utilized Landsat-8 OLI images provided by the United States Geological Survey. Landsat-8 is well-suited for analyzing the target site due to its medium-resolution (30 m) visible and near-infrared (VNIR) imagery and 100-meter-resolution thermal infrared wavelength band imagery. The dataset covers the period from March 18, 2013, to June 19, 2023, providing consistent data analysis since the early 2010s, when changes in the target site were confirmed. The images used in this study were constructed using Google Earth Engine with the research period spanning from 2014 to 2022, when map extraction was possible through Google Earth Engine. To enhance classification accuracy, measures were taken to minimize the impact of snow and ice and to ensure sufficient plant vitality. The marginal rice harvesting date throughout much of North Korea typically falls between September 17 and October 16 (Yang et al., 2018), and the annual precipitation is concentrated in July and August (Kim and Kim 2019). To account for this, median value composite images from September 1 to November 1 were used throughout the year. Additionally, cloud masking was performed using Bitmask to remove the influence of clouds from the imagery.
During the research period, satellite images including the target sites were referenced from Google Earth Pro. Google Earth provides RGB images captured by high-resolution satellites and is used as supplementary material for remote sensing. Google Earth images can serve as especially valuable data when identifying geographic information in North Korea, where access is limited (Kim et al., 2012). Previous studies by Yoon et al. (2018) analyzed mining activities in North Korea using Google Earth and Ki (2016) examined forest destruction in North Korea through time series changes in Google Earth images. Based on these prior studies, it was deemed appropriate to utilize Google Earth images as supplementary data. In this study, high-resolution images containing the target site provided by Maxar Technologies and CNES Airbus were also referenced.
2.3. Methods
In this study, two methods were employed: NDWI-based land cover classification and random forest-based land cover classification. It is shown in Fig. 2 as a flow chart.
Fig. 2. Flow chart of this study.
The NDWI is an index developed by Gao (1996) and is utilized to calculate the moisture content of the Earth’s surface or to detect water bodies through remote sensing. The NDWI is calculated using Eq. (1):
\(\begin{aligned}N D W I=\frac{X_{\text {green }}-X_{\text {nir }}}{X_{\text {green }}+X_{\text {nir }}}\end{aligned}\) (1)
where Xgreen refers to the reflectance measured by the green band, and Xnir refers to the reflectance measured by the near-infrared band. Using the bands of Landsat-8, the NDWI can be expressed as:
NDWI = (SR_B3 – SR_B5) / (SR_B3 + SR_B5). Accordingly, NDWI images were generated and extracted through Google Earth Engine (Fig. 2). In this study, it was essential to divide the target site into water, tidal flats, and land to detect time-series cover changes. To achieve this, cluster segmentation through unsupervised classification was performed using ArcGIS Pro. Unsupervised classification has the advantage of being simpler and faster than supervised classification methods because it uses only the information from the image itself without requiring training samples (Wharton and Turner, 1981). In this study, the iterative self-organizing (ISO) clustering technique, a commonly used unsupervised classification method, was employed. After dividing the target site into multiple clusters, each cluster was placed into one of three categories—water, tidal flats, and land—based on the RGB images with the NDWI water mask as a reference.
Random forest is an artificial intelligence-based classification technique widely used by researchers due to its high accuracy and effective classification performance among machine learning methods(Breiman, 2001). In this study, we used the Random Trees tool provided by ArcGIS Pro. To facilitate the distinction between the required classes, a false-color composite was constructed using RGB combinations of Landsat-8 images, and training samples were selected based on visual inspection and classified.
In remote sensing using satellite images, the provisional composition of bands can enhance the contrast between land cover types so they can be better distinguished, thereby making image classification work more convenient (Sedlák, 2002). For this study, a synthetic image using light in the RED, near-infrared (NIR), and shortwave-infrared (SWIR) wavelengths was used to easily differentiate between water and land. The RGB band combination of Landsat-8 images was set to SR_B5, SR_B6, and SR_B4 and then visualized. Consequently, water pixels are more distinguishable for classification as they appear black or purple (Fig. 3). Visual inspection was performed by referring to RGB images without false color composition and high-resolution satellite images provided by Google Earth Pro for each year’s false color composite images. A total of six classes were assigned, including water, tidal flats, grassland/forest (combined due to the difficulty in distinguishing between grassland, coniferous forests, and broad-leaved forests), cultivated, barren, and developed (including both artificial buildings and land alteration). Samples classified through visual inspection were assigned to each respective class, and 500 points from the samples were utilized as training samples. A stratified random method was employed for sampling.
Fig. 3. NDWI maps of the study area: (a) 2014, (b) 2018, and (c) 2022.
The Accuracy Assessment tool provided by ArcGIS Pro was utilized to verify the accuracy of the classification results. For this, 250 points from the samples were selected using a Stratified Random sampling technique and used as a reference dataset for accuracy assessment. Similarly, the accuracy of the unsupervised classification using NDWI was also assessed. After allocating all four classes to land, excluding water and tidal flats, 250 points from the reference samples assigned through visual inspection were used as the reference dataset.
3. Results
3.1. NDWI-Based Classification
As depicted in Fig. 4, we produced cover maps by dividing satellite imagery into three clusters using ISO clustering and conducted a time series analysis of the area ratio for each classification group. Based on lines A and B in Fig. 1, changes seaward of the coastline in 2014 were quantified, and changes on the interior and exterior of the seawall in 2022 were separately analyzed.
Fig. 4. False color composite (blue = SR_B5, green = SR_B6, and red = SR_B4) images of the study area: (a) 2014, (b) 2018, and (c) 2022.
3.1.1. Changes in the Entire Study Area
Table 1 presents the changes in the entire target site over the study period. From 2014 to 2022, the area ratio of water consistently decreased, ultimately showing a decline of 12.34%. Conversely, the proportion of land increased after 2016. The proportion of tidal flats grew to 17.51% in 2020,reaching more than double its 2014 value. Subsequently, it stabilized around 16–17% in 2021 and 2022.
Table 1. Changes in the areas of water, tidal flats, and land in the entire study area from 2014 to 2022
Table 2. Changes in the areas of water, tidal flats, and land inside the seawall from 2014 to 2022
Table 3. Changes in areas of water, tidal flats, and land outside of the seawall from 2014 to 2022
3.1.2. Changes Inside the Seawall
Inside the seawall, the area ratio of water continuously decreased, except for between 2018 and 2019,resulting in a significant decrease of 38.51% by 2022, sinking to nearly half of its value in 2014. Starting in 2021, the proportion of tidal flats surpassed that of water. The area ratio of tidal flats did not exhibit a consistent increasing or decreasing trend but ultimately grew by 20.20%, nearly doubling its 2014 value. In contrast, the land ratio, which was 1.45% in 2014, experienced a remarkable surge and reached 19.76% in 2022, exceeding over 13 times its 2014 value.
3.1.3. Changes Outside of the Seawall
Changes were also observed outside of the seawall. The area ratio of water showed inconsistency, but in 2022, it had decreased by 11.52% compared to 2014. The proportion of tidal flats experienced an 8.36% increase in just one year, from 2016 to 2017. Then, in 2019, the area ratio peaked at 25.70%, exceeding more than three times its 2014 value. However, as was seen with water, there was no consistent increasing or decreasing trend. Regarding the land area ratio, it reached its highest value, 3.04%, in 2018, displaying a notable difference from the 2014 ratio. However, compared to those inside the seawall, the changes were less distinct, and the differences were smaller.
3.2. Random Forest-Based Classification
After classifying the target site into six classes using Random Trees (Fig. 5), we further examined the time series of area ratio changes for each class.
Fig. 5. Water, tidal flat, and land clusters generated by ISO clustering: (a) 2014, (b) 2015, (c) 2016, (d) 2017, (e) 2018, (f) 2019, (g) 2020, (h) 2021, and (i) 2022.
3.2.1. Changes in the Entire Study Area
The area ratio of water experienced a decrease of 15.85%, from 49.14% in 2014 to 33.29% in 2022. The proportion of tidal flats also increased, exceeding twice its 2014 value over the study period. Since other changes in the proportion of classification groups might be influenced by changes in inland land cover unrelated to reclamation, as previously mentioned in Section 2.1, the changes seaward of the 2014 coastline were analyzed separately inside and outside of the seawall to better understand the impact of reclamation on the target site.
3.2.2. Changes Inside the Seawall
Inside the seawall, the proportion of water decreased, and the proportion of tidal flats showed a tendency to increase. The water ratio declined by 47.62%, from 85.82% in 2014 to 38.20% in 2022, representing a greater decrease than the results from the NDWI model (Section 3.1.2). Except for in 2019, the area ratio of tidal flats was consistently smaller compared to the results from the NDWI model. Additionally, it is noteworthy that the ratio was the largest in 2019, and not in 2022. The proportion of cultivated area was approximately 1% in 2014 and 2015, but it then significantly increased starting in 2016 and reached 27.62% in 2022. However, in 2015 and 2019, the proportion of cultivated area decreased. The area ratio of the developed area was 0.31% in 2014, the second smallest ratio after the barren area. However, it increased to 15.78% in 2016 and subsequently continued to fluctuate, showing repeated increases and decreases. The proportion of barren shows inconsistency and extreme fluctuations.
Fig. 6. Six classes generated by random forest classification: (a) 2014, (b) 2015, (c) 2016, (d) 2017, (e) 2018, (f) 2019, (g) 2020, (h) 2021, and (i) 2022.
3.2.3. Changes Outside of the Seawall
Outside of the seawall, the area ratio of water decreased by 15.41%,from 93.27% in 2014 to 77.86% in 2022. Following the results obtained in the NDWI analysis, the decrease in the water ratio was not as stark as it was inside the seawall. The ratio of tidal flats consistently remained above 10% starting in 2017. Like it was inside the seawall, the area ratio of tidal flats was greatest in 2019, reaching 26.40%, which is similar to the value found using the NDWI model(Section 3.1.3). The proportion of cultivated area did not exhibit a clear trend, and the proportion remained below 1% except for in 2017. This small value compared to the values and increases seen inside the seawall indicates that the increase in the proportion of cultivated area in the target site is primarily a result of reclamation. The proportion of developed areas shows a significant increase of 1.16%,from0.20%in 2014 to 1.35%in 2022,indicating a notable change. However, the increment is smaller than that seen inside the seawall.
Table 4. Changes in the areas of water, tidal flats, grassland/forest, cultivated land, barren land, and developed land in the entire study area from 2014 to 2022
Table 5. Changes in the areas of water, tidal flat, grassland/forest, cultivated land, barren land, and developed land inside the seawall from 2014 to 2022
Table 6. Changes in the area of water, tidal flats, grassland/forest, cultivated land, barren land, and developed land outside of the seawall from 2014 to 2022
3.3. Model Accuracy Assessment
3.3.1. Random Forest-Based Classification
The accuracy of the classification results was assessed using the Accuracy Assessment tool provided by ArcGIS Pro. The kappa values for the random forest-based supervised classification results are presented in Table 7. In all years, except for 2014, kappa values were greater than 0.9, with an average of 0.9524, indicating high accuracy. Table 8 displays the confusion matrix for 2022, which represents the most recent data. The user’s accuracy (type 1 error) for water, grassland/forest, barren, and developed were all 1, indicating no error, and the lowest accuracy was 0.9 for the tidal flat class. The producer’s accuracy (type 2 error) for water, grassland/forest, and developed classes were 1, and the lowest accuracy was 0.8. The kappa value in the confusion matrix was 0.9617, demonstrating high accuracy overall.
Table 7. Kappa values from the accuracy assessment of the random forest classification results
Table 8. Confusion matrix for the random forest classification results of 2022
3.3.2. NDWI-Based Classification
Table 9 displays the accuracy assessment results of the NDWI-based classifications. The kappa value was higher than 0.9 in all years except for 2016, with an average of 0.9501, slightly lower than that of the random forest-based classification results. Table 10 shows the confusion matrix for 2022. The user’s accuracy showed the highest value (1) for water, and exhibited the lowest (0.9474) for tidal flats. The producer’s accuracy was also highest for water (1), and again was lowest (0.9643) for tidal flats. Both type 1 and type 2 errors were most severe in the tidal flat. The kappa value in the confusion matrix was 0.9692, indicating high accuracy for the NDWI-based classification results.
Table 9. Kappa values from the accuracy assessment of the NDWI ISO clustering results
Table 10. Confusion matrix for the NDWI ISO clustering results of 2022
4. Discussion
ISO clustering based on NDWI and random forest classification are considered appropriate research methods for distinguishing reclaimed areas from tidal flats. The accuracy assessment results of the random forest classification included a kappa value of 0.8370 in 2014, which was the lowest accuracy value. On the other hand, the lowest kappa value of the NDWI ISO clustering was 0.8800 in 2016. Furthermore, the kappa values for all other years exceeded 0.9 in both methods. The average kappa value surpassed 0.95 in both approaches, demonstrating high accuracy in both classification results.
In the confusion matrix of 2022, the accuracy of water was calculated as 1, indicating very high accuracy in both methods for water classification. However, the accuracy of the tidal flats was lower in the NDWI ISO unsupervised classification results compared to those of the random forest-based supervised classification, both in terms of type 1 and type 2 errors. The NDWI-based analysis showed higher accuracy for the tidal flats. Except for in 2019, the results of the NDWIISO classification, which indicated a larger area ratio for the tidal flats, were more reliable, as indicated by the user’s accuracy scores of the two methods. For the random forest classification results, both the user’s accuracy and the producer’s accuracy were 1 for developed, and 0.9836 for cultivated. These high accuracy values show that this method is suitable for analyzing the process of land alteration, building installation, and the expansion of cultivated areas.
The results obtained from the NDWI clustering and the random forest classification did not show a significant difference in terms of the area change of water. However, the area of the tidal flat tended to be measured as larger in the NDWI ISO clustering, which classifies the target based solely on the amount of water. In all years except for 2019, the tidal flats measured by the NDWI clustering were larger. The decrease in the proportion of cultivated area and the surge in the area of tidal flats in 2015, as well as the decrease in the cultivated area in 2019 using the random trees classification, suggest that the results of 2019 were more likely recognized as tidal flats. Overall, the results of both methods provide valuable insights into the changes in the target area over time.
The inconsistent increases and decreases in the area of the tidal flats, with large fluctuations, could be attributed to various factors. A significant factor that creates tidal flats on the west coast is the sediment flowing from rivers like the Cheongcheon and the Yalu (Center of Northeast Asia & North Korea Research, 2017). However, ongoing reclamation activities and soil accumulation on the seafloor can also contribute to changes in the area of tidal flats. Considering the observed increase in the proportion of tidal flats and the concomitant decrease in the proportion of cultivated area in 2015 and 2019, it is plausible that the sedimentation of soil caused a rise of the high tide line, leading to changes in the extent of tidal flats. These factors highlight the dynamic and complex nature of tidal flat areas, which can be influenced by both natural processes and human activities.
Inside the seawall, there is a clear trend of decreasing water area and increasing cultivated area. Approximately half of the area inside the seawall has been reclaimed over the nine years since 2014, and the reclaimed land is being utilized for agriculture. This aligns with North Korea’s announcement and domestic analysis indicating that reclamation projects have been carried out to secure additional farmland for food production. The proportion of developed areas also showed extreme fluctuations, suggesting that the construction of buildings andinfrastructurenecessaryforthe expansion of seawalls, reclamation, and farmland are actively taking place. The variation in the ratio of developed and cultivated areas is larger inside than outside of the seawall, reaffirming that the changes to the land cover in the target site are primarily the result of artificial reclamation activities. These findings underscore the significant impact of human encroachment on the coastal landscape and the importance of considering reclamation efforts in understanding the changes in land cover over time. The proportion of barren also shows large fluctuations inside the seawall, which indicates that some of the reclaimed land was classified as a barren area.
5. Conclusions
This study employed Landsat-8 OLI images to classify the land cover of the target site on the west coast of Pyeonganbuk-do, where reclamation activities are taking place. The analysis spanned from 2014 to 2022, and two research methods were utilized. Firstly, NDWI images were constructed and classified into three clusters using ISO clustering: water, tidal flats, and land. The analysis revealed a clear decrease in water area, an increase in land area, and fluctuating changes in the tidal flats area. Secondly, the random forest classification technique was used to classify the target site into six classes: water, tidal flats, grassland/forest, cultivated land, barren land, and developed land. The results indicated a clear decrease in water area and a rapid increase in cultivated area, confirming the ongoing reclamation activities and that they are aimed at expanding agricultural land. Additionally, significant changes were observed in a developed area, reflecting alterations in shape and the installation of structures. The area inside the seawall, as depicted in the 2022 image, experienced a nearly 50% decrease, while the cultivated area surged by over 2,300%, highlighting the impact of human activity on the target site’s land cover. The accuracy assessment demonstrated high accuracy for both methods, with average kappa values exceeding 0.95,showing them to be suitable research approaches. The NDWI-based classification showed high accuracy in identifying water, tidal flats, and land, facilitating the exploration of land cover changes in tidal areas. On the other hand, the random forest-based classification demonstrated low type 1 and 2 errors for cultivated and developed classes, making it an effective method for detecting the expansion of agricultural land and anthropogenic activities associated with reclamation. However, certain limitations were identified. There might be misclassification errors categorizing seawater that could not escape during low tide in some areas as water. Additionally, the 30 × 30 m spatial resolution of Landsat-8 images poses limitations in object recognition. The availability of high-resolution satellite images from Google Earth Pro on a monthly or annual basis for generating training and reference data was also uncertain. It is expected that future studies will benefit from improved spatial and temporal resolution of image datasets through the direct use of high-resolution satellites, different indices such as NDWI using SWIR, and combinations of various satellite data types, enabling more sophisticated analyses.
Acknowledgments
This research was funded by a National Research Foundation of Korea grant provided by the Ministry of Science and ICT (No. RS-2023-00258105) and a Kookmin University grant.
Conflict of Interest
No potential conflict of interest relevant to this article was reported.
참고문헌
- Breiman, L., 2001. Random forests. Machine Learning, 45, 5-32. https://doi.org/10.1023/A:1010933404324
- Center of Northeast Asia & North Korea Research, 2017. Current status and implications of North Korea's tideland reclamation project. The Korea Transport Institute, Railway Transport Research Division.
- Chang, M., Li, P., Li, Z., and Wang, H., 2022. Mapping tidal flats of the Bohai and Yellow Seas using time series Sentinel-2 images and Google Earth Engine. Remote Sensing, 14(8), 1789. https://doi.org/10.3390/rs14081789
- Choung, Y. J., 2015. Mapping 3D shorelines using KOMPSAT-2 imagery and airborne LiDAR data. Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography, 33(1), 23-30. https://doi.org/10.7848/ksgpc.2015.33.1.23
- Gao, B. C., 1996. NDWI-A normalized difference water index for remote sensing of vegetation liquid water from space. Remote Sensing of Environment, 58(3), 257-266. https://doi.org/10.1016/S0034-4257(96)00067-3
- Hong, S., 2009. Ecological perspective on conservation and management of tidal flat ecosystem. Journal of the Island Culture, 33, 319-345. https://doi.org/10.22917/island.2009..33.319
- Ki, J. H., 2016. Spatial photo analysis using Google Earth images to measure urban environmental pollution and deforestation in North Korea. Journal of Environmental Policy and Administration, 24(1), 133-146. https://doi.org/10.15301/jepa.2016.24.1.133
- Kim,C. H., Lee, E. H., Kim, S.J., and Bae, S. H., 2012. The analysis of the lake in North Korea using Google Earth. Journal of the Association of Korean Photo-Geographers, 22(2), 29-38. https://doi.org/10.35149/jakpg.2012.22.2.003
- Kim, D. J., and Kim, J. H., 2019. Estimating the monthly precipitation distribution of North Korea using the PRISM model and enhanced detailed terrain information. Korean Journal of Agricultural and Forest Meteorology, 21(4), 366-372. https://doi.org/10.5532/KJAFM.2019.21.4.366
- Kim, S., Bang, R., and Kwon, H., 2003. Estimation of areal change in Hwa-ong tidal flat due to sea dike construction project using multi-temporal Landsat TM images. Korean National Committee on Irrigation and Drainage, 10(1), 73-79.
- Lee, J., Kim, B., Noh, J., Lee, C., Kwon, I., Kwon, B. O. et al., 2021. The first national scale evaluation of organic carbon stocks and sequestration rates of coastal sediments along the West Sea, South Sea, and East Sea of South Korea. Science of the Total Environment, 793, 148568. https://doi.org/10.1016/j.scitotenv.2021.148568
- Lee, M. B., Kim, N. S., and Lee, G. R., 2005. Reclamations and coastal changes with case study on Yeomju and Cheolsan in the west coast of North Korea. Journal of the Korean Geomorphological Association, 12(3), 99-110.
- Lee, S., Kim, G., and Lee, C., 2016. Preliminary study for tidal flat detection in Yeongjong-do according to tide level using Landsat images. Korean Journal of Remote Sensing, 32(6), 639-645. https://doi.org/10.7780/kjrs.2016.32.6.8
- Makowsky, P., Town, J., Kae, M., and Pitz, S., 2021. North Korea's tideland reclamation efforts. Available online: https://www.38north.org/2021/12/north-koreas-tideland-reclamation-efforts/ (accessed on July 17, 2023).
- Ministry of Oceans and Fisheries, 2021.The ecosystem service value of our tidal flats: 17 trillion KRW annually. Available online: https://www.korea.kr/briefing/pressReleaseView.do?newsId=156487268 (accessed on July 16, 2023).
- Sedlak, P., 2002. Using Landsat TM data for mapping of the quaternary deposits in central Sweden. Geographica, 37, 77-81.
- Wharton, S. W., and Turner, B. J., 1981. ICAP: An interactive cluster analysis procedure for analyzing remotely sensed data. Remote Sensing of Environment, 11, 279-293. https://doi.org/10.1016/0034-4257(81)90026-2
- Woo, H.J, Bahk,J.J., Lee,Y. G.,Je,J. G., and Choi,J. U., 2004. Characteristics of sediments in the Kanghwa tidal flat on the west coast of Korea. Journal of Wetlands Research, 6(1), 167-178.
- Yang, W., Kang, S., Kim, S., Choi, J.-S., and Park, J.- H., 2018.Assessment of the safe rice cropping period based on temperature data in different regions of North Korea. Korean Journal of Agricultural and Forest Meteorology, 20(2), 190-204. https://doi.org/10.5532/KJAFM.2018.20.2.190
- Yoon, I. J., and Jin, H. K., 2021. Characteristics and implications of North Korean maritime statistics. Korea Maritime Institute. https://kosis.kr/bukhan/nkAnals/selectNkAnalsDetail.do?menuId=M_02&pageIndex=1&boardIdx=5&searchCondition=00&searchKeyword=
- Yoon, S., Jang, H., Yun, S. T., and Kim, D. M, 2018. Investigating the Status of mine hazards in North Korea using satellite pictures. Journal of the Korean Society of Mineral and Energy Resources Engineers, 55(6), 564-575. https://doi.org/10.32390/ksmer.2018.55.6.564
- Yi, H, M., 2022. Getbol, an natural heritage site that draws global interest. Available online: http://www.kmoumedia.com/news/articleView.html?idxno=4204 (accessed on Aug. 29, 2023).