1. Introduction
Coasts, where the land and sea intersect, are hubs of transportation, hosting, and logistics, and home to economically essential facilities such as industrial complexes and ports. Their scenic landscapes provide desirable residential areas and tourist destinations. Moreover, transition zones, such as estuaries, dunes, and lagoons, formed by the interaction between land and sea, possess high ecological conservation value (Mehvar et al., 2018). In natural coastal areas, seasonal equilibrium is maintained as shorelines retreat during high winter waves and recover in spring and summer (Aubrey, 1979). Nevertheless, artificial interventions such as landfills or the construction of structures block natural sediment supply, altering the coastal hydrodynamics and sediment transport patterns. This causes damage to the resilience of coastal morphodynamics, leading to sediment loss and coastal erosion (Martínez et al., 2017). In addition, global warming has increased the frequency and intensity of abnormal waves (Shi et al., 2024; Young et al., 2011) and has caused sea levels to rise steadily (Hamlington et al., 2024). Observations by the Korea Hydrographic and Oceanographic Agency (2024) suggest that Korea’s average sea level rise rate reached 2.79 mm/year between 2004 and 2013 and 3.88 mm/year between 2014 and 2023. These climate-induced changes, when combined with urbanization, further increase the risks and uncertainties faced by coastal environments. Therefore, analyzing the short- and long-term shoreline change trends using observation data is crucial for diagnosing and restoring eroded areas and protecting the regions vulnerable to future erosion.
Shoreline observations are conducted via direct surveys using real-time kinematics and remote observations through low-altitude aerial photography (Fabris et al., 2023). Direct surveys are highly accurate but require considerable time and labor to cover extensive shorelines spanning several kilometers. Remote observations, performed with manned aircraft or unmanned aerial vehicles, capture high-resolution images with pixel sizes of less than 1 m but have a limited range and are strongly influenced by weather conditions, making stable and frequent data acquisition challenging. Advanced coastal engineering countries such as the Netherlands, the United States, and Australia, have recognized their importance early and have systematically collected long-term data to analyze shoreline change patterns (Ludka et al., 2019; Montaño et al., 2020; Splinter et al., 2018). These datasets have supported coastal management strategies concentrated on flexible structures that preserve coastal resilience (Kelty et al., 2022; Roelvink et al., 2020). In contrast, Korea initiated nationwide coastal erosion surveys and shoreline observations relatively late, beginning in the 2000s. These efforts have mostly been limited to sporadic observations in specific areas, restricting the applicability of the findings.
Currently, optical satellite images are being increasingly viewed as an alternative for continuous, stable observations over vast areas while utilizing previous data to complement existing methods. Therefore, satellite images are being applied widely in land and traffic management (Kim, 2016; Larsen et al., 2009), disaster monitoring (e.g., forest fires and floods) (Deshmukh et al, 2023; Skakun et al., 2016), and marine and geological observations (Fu et al., 2020; Groom et al., 2019; Pour et al., 2018). Moreover, satellite imagery plays a crucial role in monitoring shoreline changes, supporting coastal management. For example, Luijendijk et al. (2018) quantified global sandy beach conditions, analyzed 33 years of shoreline changes, and linked them to industrialization, urbanization, and geography. Similarly, Ding et al. (2021) used satellite data to monitor the changes in the terrain of the coral reefs and atolls of Southeast Asia. Santra et al. (2024) and Rostami et al. (2023) examined shoreline changes in the Sundarbans delta in India and Chabahar Bay in Iran, revealing persistent patterns of erosion and deposition. Nevertheless, to use optical satellite images for shoreline analysis, it is essential to understand satellites and address the factors that may affect the accuracy of shoreline extraction. Furthermore, the differences in spatial resolution, observation paths, and revisit frequency among satellites must be considered when constructing a unified dataset from several satellites. In addition, image processing techniques such as edge detection, segmentation, and object detection rely on spectral characteristics, including the changes in adjacent pixel values, making it crucial to account for the variations in image features between satellites (Di et al., 2003).
This study presents a framework for shoreline analysis using the long-term optical satellite imagery datasets, with a focus on the heavily eroded East Coast of Gangwon Province (Yum et al., 2023). First, an automatic long-term satellite imagery download system was implemented, and preprocessing techniques effective in enhancing shoreline identification were applied. Through this data standardization, a 40-year time-series data archive was established for 41 littoral cells along the East Coast of Gangwon Province. Subsequently, the open-source shoreline extraction and visualization model, CoastSat (Vos et al., 2019b), was applied to Gangneung port–Yeomjeon beach and Gungchon–Chogok port, to assess its applicability to dynamically changing coastlines shaped by intensive coastal development. On the other hand, CoastSat, which was designed for simpler coastlines such as those in parts of Australia without coastal structures, has accuracy limitations in environments with dense artificial structures and abrupt shoreline protrusions. Its fixed model structure and input constraints further limited its applicability. Based on these results, this study discusses the needs for and future directions of a shoreline analysis model optimized for Korea’s dynamically changing coastlines, influenced by natural and anthropogenic factors.
2. Acquisition of Optical Satellite Imagery for Shoreline Extraction on the East Coast of Korea
2.1 Optical Satellite Imagery
Advances in satellite technology and data accessibility have dramatically increased the number of launches and the availability of Earth observation data. Satellite observation data with diverse spatiotemporal resolutions can now be accessed easily through various platforms, as summarized in Fig. 1. Although ultra-high-resolution satellites are ideal for shoreline analysis, this study prioritizes continuity, provider reliability, and consistent data quality management. Therefore, high-resolution Sentinel-2 and medium-resolution Landsat-5, -7, and -8 satellites were selected. Sentinel-2 and Landsat missions offer free access to recent data and archives spanning over 40 years, with data reliability maintained through regular updates on calibration and sensor issues. Observation data, consisting of multiple wave wavelength bands defined by each satellite sensor, can be combined to serve specific research objectives. Each band is classified as either top-of-atmosphere (TOA) reflectance, which is uncorrected, or surface reflectance (SR), which has undergone atmospheric correction.
The Sentinel-2 constellation consists of two satellites, Sentinel-A and B, launched by the European Space Agency in 2015 and 2017, respectively, to observe the land and vegetation of the Earth. Positioned 180° apart in the same orbit, the satellites collect data every five days for a specific area. They provide information across 13 wavelength bands, ranging from visible light (443 nm) to shortwave infrared (2,202 nm), with spatial resolutions of 10, 20, and 60 m. For this study, the Red (R), Green (G), and Blue (B) bands were used because they offer the highest spatial resolution. Table 1 lists the spatiotemporal and spectral resolutions.
The Landsat-5, -7, and -8 satellites, developed by NASA and the United States Geological Survey, were launched in 1984, 1999, and 2013 for global environmental monitoring. These satellites collect 7, 8, and 11 multi-bands, respectively, with temporal (16-day revisit) and spatial (30 m) resolutions consistent across the R, G, and B bands, as shown in Table 1. Despite their lower spatial resolution compared to Sentinel-2, Landsat missions offer more than four decades of data, which is valuable for long-term shoreline change analysis. Currently, only Landsat-8 remains operational because Landsat-5 and -7 ended their missions in 2012 and 2024.
2.2 Automated Acquisition and Preprocessing of Optical Satellite Imagery for the Gangwon East Coast
Cloud-based platforms have significantly improved access to large-scale remote sensing data (Li et al., 2019; Velastegui-Montoya et al., 2023). Among these, the Google Earth Engine (GEE) updates approximately 60,000 multi-level corrected Earth observation data points daily. This enables researchers to select data directly suited to their needs. The application programming interface (API) provided by GEE, implemented in user-friendly languages such as JavaScript or Python, facilitates efficient and rapid data collection and analysis in a stable environment, requiring minimal IT expertise. Considering these advantages, this study used the GEE Python API to construct and preprocess satellite image datasets for long-term shoreline analysis along the East Coast of Gangwon Province.
2.2.1 Study site
Gangwon Province is located at latitudes 37° 02′–38° 37′N and longitudes 127° 05′–129° 22′E, featuring a 314 km shoreline along the East Sea. The region is a semidiurnal tidal environment with an average tidal range of 0.3 m. The East Sea frequently experiences swells with periods exceeding 9 s and wave heights over 3 m due to the strong temperate low pressure from late fall to winter (October to February), leading to annual property damage and numerous casualties (Oh et al., 2010; Son et al., 2021). Nevertheless, the lack of natural barriers, such as islands, combined with the steep seabed, prevents wave energy dissipation, making the area highly wave-dominant (Jeong et al., 2019). According to the Korea Meteorological Administration (n.d.), an average of four typhoons affect Korea from summer to early fall each year, with one directly affecting Gangwon Province. Currently, Gangwon Province hosts six ports, 58 fishing ports, and numerous coastal roads and tourist facilities concentrated near beaches. Moreover, various hard structures, such as breakwaters, groins, and revetments, have been densely installed to mitigate the coastal erosion caused by high wave energy. These natural and anthropogenic factors have led to highly localized and variable shoreline responses, necessitating a spatially explicit analytical framework. To this end, this study used a littoral cell-based approach, where each cell functions as a spatial unit for sediment budget analysis, incorporating variables such as wave climate, bathymetry, sediment characteristics, riverine input, and coastal structures. Forty-one littoral cells (GW01–GW41) were defined along the Gangwon East Coast (Ministry of Oceans and Fisheries, 2023). Among these, two littoral cells, Gangneung Port–Yeomjeon Beach (GW27) and Gungchon–Chogok Port (GW38), were selected to illustrate the satellite image processing and shoreline extraction methodology in detail. These sites were chosen based on their pronounced spatiotemporal shoreline fluctuations, which reflect the disruption of shoreline equilibrium caused by large-scale coastal infrastructure (e.g., ports and power plants) and the subsequent recovery attempts using various erosion mitigation strategies. Accordingly, these sites offer a representative testbed for evaluating how effectively satellite-based shoreline extraction captures the effects of natural processes and engineered interventions by encompassing diverse environmental conditions.
2.2.2 Automating satellite imagery acquisition using Google Earth Engine
To specify the data selection, users need to define the region of interest (ROI) and download period using the input parameters of the GEE Python API. This section outlines the satellite image download process as described by Gorlick et al. (2017). Table 2 lists the parameter values used in this study, providing insight into the entire process.
(1) Data
A Catalog and a Collection, which represent the structure and path of petabyte-scale data storage in GEE, comprise a key value identifying the data address. A catalog refers to archives organized by observation sources, such as Sentinel-2 and Landsat missions, or by application domains, including Terrain, Land Cover, and Cropland. For satellite data, the Collection includes information on satellite versions and correction levels. The Band, which is the smallest unit of satellite data, represents an image recording a specified spectral range. The number and characteristics of bands vary according to the satellite. Even for the same target, the central wavelength and spectral coverage may differ. In the Landsat series, updates to the R, G, and B bands have improved spectral resolution and refined the captured wavelength ranges. Differences between the Sentinel-2 and Landsat series satellites were corrected to ensure consistency when constructing unified datasets. The Scale, which represents the spatial resolution of the data, was set to maximum resolutions of 10 m and 30 m for the Sentinel-2 and Landsat missions, respectively.
(2) Region of interest
The ROI is defined using Central coordinates, Tile, and Buffer. The ROI is defined using the Central coordinates, Tile, and Buffer. In this study, each dataset was collected for 41 Gangwon East Coast littoral systems, with the system boundaries and central coordinates derived from the Coastal Erosion Status Survey Synthesis Report (Ministry of Oceans and Fisheries, 2023). The corresponding Tile for each Central coordinate must also be provided because observation data are divided into tile units for efficient management. The ROI boundary is defined by the Buffer, representing the radius from the central coordinate. The final image is square, with a side length twice the Buffer. Each Buffer size was adjusted to ensure that the surrounding coastal structures influencing each littoral cell were fully captured. If the Buffer exceeds the Tile range, black (no data) sections appear in the output, requiring caution. Although GEE imposes a 100 MB limit on individual downloads, all defined littoral cells in this study satisfied this constraint.
(3) Quality
The presence of clouds has a significant impact on the quality of satellite images. Cloud cover filters automatically exclude the images contaminated by clouds, preserving dataset reliability. Cloud cover ranges from 0 to 100. Images are excluded if the cloud pixels exceed the set value. Higher Cloud cover thresholds increase data availability but reduce quality, whereas lower thresholds ensure more explicit images but limit usable data.
SR-grade R, G, and B band data for all available periods were downloaded to maximize the satellite image acquisition rate, as shown in Fig. 2. On the other hand, Sentinel-2 provided SR-grade imagery only for certain regions, Europe, before December 2018. Therefore, TOA-grade imagery was used for this period. Moreover, the downloading of Landsat-7 imagery, which was affected by an SLC defect causing persistent data gaps, was limited to periods before Sentinel-2 data became available. More than 60,000 satellite images, obtained through this approach for 41 littoral cells from 1984 onward, were used to construct datasets for shoreline change analysis. Unlike previous studies (Ding et al., 2021; Fu et al., 2020; Groom et al., 2019; Luijendijk et al., 2018; Pour et al., 2018; Santra et al., 2024) using a one to five-year analysis cycle, this study adopted a tight monthly cycle to effectively capture changes driven by Korea’s climate and coastal management. A representative image for each month was selected based on visual judgments, prioritizing distinct shorelines. An average image was generated as the representative if multiple images showed noticeable differences in the shoreline within a month, caused by typhoons, swells, or high waves. Fig. 3 provides an overview of monthly image acquisition for GW27 and GW38. Before 1999, the data acquisition rate was low when only Landsat-5 was available. In addition, clear image acquisition rates decreased during the summer months (July–September) due to the monsoon, which accompanied high humidity and frequent heavy rainfall.
2.2.3 Preprocessing of long-term satellite imagery datasets to enhance shoreline identification
In optical satellite images, shoreline recognition is hindered by meteorological and satellite-related factors, which obscure various features, distort the spatial information, or blur shoreline boundaries (Fig. 4). First, clouds, fog, snow, and white water, directly or indirectly, affect the shoreline recognition accuracy, as shown in Figs. 4(a)–4(c). Thin clouds reduce image clarity, while thick clouds completely obscure information. Although cloud-contaminated images can initially be filtered using cloud cover during download, this filtering applies at the tile level, which does not ensure its effectiveness for the ROI. Fog produces wave patterns and blurring images, while snow and white water from wave breaking reflect similarly to shorelines, complicating shoreline identification. Second, malfunctions in satellite sensors and systems also degrade the image quality, as shown in Figs. 4(d)–4(f). Common issues include saturation, distortions, quality degradation near the lens periphery, and misalignment. Saturation occurs when the pixels in reflective areas, like thick clouds or wave foam, exceed a threshold and appear as solid colors (e.g., white or pink). Images captured near the outer edges of the lens show optical distortion and appear hazy, resembling fog. Misalignment, caused by the satellite orbit, sensor characteristics, and atmospheric conditions, results in spatial mismatch that accumulates over time, challenging the long-term positional accuracy. For Landsat-7, a scan line corrector (SLC) defect has caused black stripes in the images since June 2003, with the severity increasing farther from the lens center. Although concurrent operational periods of Landsat-5 and -8 provide replacement data, no substitutes exist from September 2011 to July 2013.
Long-term optical satellite image data exhibit cumulative variations in image characteristics and spatial misalignments, which introduce noise and compromise the usability of the dataset. Preprocessing methods, contrast stretching, and spatial registration were applied to address these issues and improve the consistency and reliability of the dataset. Sentinel-2 and Landsat series satellite data are provided in 16-bit Geographic Tagged Image File Format, but the shorelines are challenging to discern in their original, overly dark images. This study analyzed the image histograms (Fig. 5), which visualize the distribution of pixel values and aid in assessing brightness and color range. Figs. 5(a) and 5(b) present the original R, G, and B bands composite image from Sentinel-2 and its histogram, while Figs. 5(c) and 5(d) show the same for Landsat-8. The pixel distributions for the R, G, and B bands are notably narrow. By applying contrast stretching, the pixel values were normalized between their minimum and maximum values, expanding the dynamic range and enhancing visibility (Jain, 1989). This approach significantly improved shoreline identification accuracy in Sentinel-2 and Landsat-8 imagery, as shown in Figs. 5(e)–5(h). Satellite images are often misaligned, and for Landsat satellites, a single pixel error can result in a positional error of up to 30 meters. These errors accumulate over extended periods, compounding the alignment issues. Hence, cross-correlation was used to correct spatial errors and register images to the exact locations. This method estimates the spatial differences between undistorted and reference images via fast Fourier transform and is highly effective for spatial registration because of its robustness against noise, occlusions, and defects (van der Walt et al., 2014). Fig. 6 shows the chronologically extracted and stacked sections of Sentinel-2 images from the same area, comparing the results with and without cross-correlation. Fig. 6(a) presents the results before applying cross-correlation, where the pier and outermost fixed structures shift over time. In contrast, Fig. 6(b) displays a well-aligned dataset after cross-correlation, with these features remaining stable throughout. Nevertheless, for Landsat series images, the relatively lower image quality posed challenges in accurately calculating the required correlation values.
2.3 CoastSat: an Open-source Software for Shoreline Extraction
The shoreline must be extracted to analyze shoreline change trends in satellite images, and various methods have evolved (Zhou et al., 2023). One of the first techniques, global thresholding, involves setting a threshold where pixel values above or below a certain limit are assigned a value of 1 or 0, respectively. The Otsu algorithm (Otsu, 1975), an efficient method for image binarization, has been applied extensively in shoreline extraction studies (Bergsma et al., 2024; Liu et al., 2017). Edge detection is also used because it helps identify sharp changes in image brightness between water and land. The Canny algorithm (Canny, 1986) is preferred over other methods, such as the Sobel, Roberts, and Prewitt operators, owing to its superior performance (Yasir et al., 2020, Paravolidakis et al., 2018). With the advances in computer technology and high-performance hardware becoming increasingly affordable, artificial intelligence has emerged as an alternative to shoreline extraction. Machine learning (ML) methods, whether for supervised pixel classification (Çelik and Gazioğlu,2022; Kalkan et al., 2013), unsupervised clustering (Choung and Jo, 2017; Sekovski et al., 2014), or more complex deep learning (DL)-based methods (Erdem et al., 2021; Karsli et al., 2011), are used widely. For example, CoastSat, an open-source Python toolbox (Vos et al., 2019b), offers an integrated solution for extracting shorelines from satellite images (Fig. 7). The toolbox downloads and preprocesses Sentinel-2 and Landsat series satellite data, extracting the shoreline using a neural network and an edge detection method. First, the R, G, B, NIR (Near-infrared), and SWIR (Shortwave infrared) bands are pixel-classified into four classes using a pre-trained multi-layer perceptron: water, sand, white foam, and others, as shown in Fig. 7(a). This model has been trained on multiple beaches located on the east coast of Australia, including Byron Bay, Narrabeen, Newcastle, and Sawtell beaches, with little human involvement (Short et al., 2014; Short, 2020). Otsu thresholding is then applied to the calculated modified normalized difference water index (MNDWI) from the G and SWIR bands to extract the boundary between sand and water (Fig. 7(b)). Finally, the extracted shorelines are plotted (Fig. 7(c)) and saved as Geographic JavaScript Object Notation format for further mapping and analysis. This toolbox also includes post-processing, enabling the visualization of extracted shorelines and the computation of the time series of the cross-shore distance along user-defined transects.
Although the CoastSat model provides a robust and integrated framework for shoreline extraction, it was developed and trained primarily using satellite imagery from relatively simple beach environments with minimal human intervention. Consequently, its applicability to more complex coastlines, such as those in South Korea, where significant spatiotemporal variability and diverse coastal structures are present, remains insufficiently validated. The following chapter addresses this gap by applying the model to two specific sites along the Eastern Coast of Gangwon Province, a region characterized by severe coastal erosion and the implementation of various shoreline protection measures. This case study aims to evaluate the applicability of the model to Korean coastal conditions and assess the feasibility of using satellite imagery for long-term and regional-scale shoreline change monitoring.
3. Analysis of Long-Term Shoreline Changes Affected by Coastal Structures
Natural beaches tend to reach a stable equilibrium state over the long term, shaped by the local geomorphology and prevailing wave conditions, regardless of the short-term high-energy wave events, because of their inherent resilience. Nevertheless, various anthropogenic interventions can alter the direction and energy of waves reaching the shoreline through different mechanisms, which in turn influence sediment transport and ultimately lead to changes in the shoreline. On many beaches along the East Coast of Korea, the construction of harbors, especially at the northern or southern ends, has developed zones of wave shadow, often resulting in excessive sediment accumulation near the breakwaters and severe erosion on the opposite side. This has caused economic and aesthetic issues for local communities and visitors. In response, various coastal protection strategies have been implemented over several decades along the East Sea coast. These include gray construction methods to dissipate incident wave energy and control cross-shore and alongshore sediment transport, as well as beach nourishment and dredging. On the other hand, the absence of centralized records, due to efforts conducted by multiple agencies and projects, has hindered continuous and consistent long-term evaluations of shoreline responses to hard and soft engineering measures.
This study addressed this issue by generating long-term shoreline change timelines at the scale of individual littoral cells, with a particular focus on two cells along the eastern coast of Korea, where erosion has been particularly severe. This analysis was conducted in three steps: the shoreline was extracted using the CoastSat model, and zones exhibiting pronounced erosion or accretion were identified; based on the timeline of major coastal structure installations derived from satellite imagery archives, representative locations were selected where natural processes and anthropogenic factors likely drive shoreline changes; shoreline evolution was then analyzed at these sites from 1984 to the present using all available satellite imagery.
3.1 Long-term Shoreline Change Trends in GW27, Gangneung Port–Yeomjeon Beach
GW27 is a straight littoral cell extending approximately 4.12 km, where sediment transport is confined between Gangneung Port to the north and the training dike near Gunsun River to the south. This cell is typically divided into three north-to-south segments: Namhangjin, Hashidong, and Yeomjeon beaches. The primary sediment source is the Namdae River at the northern end, which has historically supplied sediments along the entire cell. Although the beach width was once relatively uniform throughout, this balance has been disrupted because of wave energy imbalances and changes in sediment flux following the construction of harbor structures at Gangneung Port and offshore facilities related to the Anin thermal power plant. The analysis scope was extended to include Anmok Beach, located immediately north of the port, to evaluate the influence of Gangneung Port on the surrounding coastal behavior. In response to the resulting sediment supply imbalance, various coastal engineering interventions were then applied.
The resulting shoreline changes were evaluated by analyzing satellite imagery to quantify long-term shoreline responses. In Fig. 8, the blue and red solid lines represent shoreline positions in winter 1985 and winter 2024, respectively. The scatter plot above these lines illustrates the changes over the 40 years: darker green indicates zones of dominant accretion, while the darker red highlights areas of dominant erosion. This analysis reveals mild accretion in the rear zones of submerged breakwaters, including Anmok Beach, and significant sediment accumulation south of the Namdae River training dike (within Gangneung Port) and north of the Gunsun River training dike. Conversely, severe erosion is observed near the termini of the submerged breakwaters fronting Hashidong Beach. Fig. 9 and Table 3 summarize key details–inferred from long-term satellite image archives–on the location, structure type, and construction period of developments that have influenced shoreline equilibrium in GW27 to support a more detailed interpretation of these shoreline changes and guide the selection of representative analysis points.
Fig. 10 provides a comprehensive overview of the spatial setup and temporal evolution of shoreline changes in GW27. Fig. 10(a) presents the locations of the seven shore-normal baselines established for this study, and Fig. 10(b) presents the time series of shoreline changes at each baseline over the past 40 years, extracted using CoastSat. The baselines were selected strategically to capture the shoreline responses driven by coastal developments. In the northern section, Baseline 1 was positioned to observe the changes in horizontal and vertical sediment transport associated with the construction of Gangneung Port and the installation of submerged breakwaters. Baselines 2 and 3 were placed to evaluate the development and migration of wave shadow zones formed during port construction, as well as the influence of sediment supply from the Namdae River. Further south, Baselines 4 to 7 were set to examine sediment transport disruptions caused by training dikes that impede alongshore sediment movement and the impacts of offshore facilities related to the thermal power plant, which have introduced wave energy imbalances and triggered additional shoreline interventions. The following section provides a detailed analysis of shoreline evolution at each baseline.
Baseline 1 is located at the southern end of Anmok Beach. Before Gangneung Port was constructed, this area exhibited noticeable seasonal shoreline fluctuations. During high wave conditions, strong rip currents promoted dominant offshore sediment transport, with some sediment likely moving southward around the Jukdo Headland. Nevertheless, the beach maintained a relatively stable width over the long term. On the other hand, beginning in 1992–the start of Gangneung Port construction–wave energy was concentrated at the junction between the port structures and the shoreline. This led to intensified offshore currents and rapid shoreline retreat. From 2014 to 2016, a submerged breakwater and beach nourishment were implemented sequentially, effectively reducing sediment loss and enabling rapid shoreline recovery. In 2017, the construction of a groin north of the submerged breakwater further limited alongshore sediment transport, resulting in the formation of a salient in the sheltered zone behind the structure.
Baseline 2 is located in an area directly influenced by sediment supply from the Namdae River. Its proximity to the sediment source results in minimal seasonal variability along the shoreline. Baseline 3, located at the center of Namhangjin Beach, exhibited shoreline changes similar to those of Baseline 2 during the first and second construction phases of the northern breakwater at Gangneung Port. During the third construction phase, which began in 2002, significant wave energy imbalances developed along Namhangjin Beach. During this period, sediment from Baseline 3 was transported northward toward Baseline 2, where wave energy was lower because of the wave shadowing effect of the port. As a result, Baseline 2 advanced, while Baseline 3 retreated. Around the summer of 2005, as the breakwater expanded, the wave shadow zone extended southward to include Baseline 3, resulting in a gradual recovery of the shoreline at that location. Between 2012 and 2017, a series of coastal engineering interventions were implemented, such as a groin at the southern end of Namhangjin Beach, five submerged breakwaters totaling 840 m in length, and periodic dredging and nourishment. These measures significantly advanced the shoreline, restoring the beach width to pre-construction levels. Baseline 2 experienced rapid accretion following the 2019 extension of the Namdae River training dike, but returned to its previous shoreline position after the structure was restored in 2022.
Located at Hashidong and Yeomjeon Beaches, Baselines 4 to 7 are located farther from the Namdae River and are only marginally affected by its sediment supply. During the construction period of Gangneung Port (1992–2005), Hashidong Beach (Baselines 4–6) experienced a mild decreasing trend in beach width, likely because of its location outside the wave shadow zone of the port. In contrast, Baseline 7 at Yeomjeon Beach initially retreated slightly but subsequently advanced, likely due to the accumulation of sediment transported from Hashidong Beach near the boundary formed by the Gunsun River training dike. From 2012 to 2013, efforts to stabilize Namhangjin Beach (Baselines 2 and 3) blocked sediment transport from the Namdae River toward Hashidong and Yeomjeon, initiating a trend of beach narrowing in these areas. Annual springtime nourishment was conducted at Yeomjeon Beach from 2016 to 2018 to counteract this, aiming to promote the northward redistribution of sediment from the Gunsun River training dike via alongshore currents. Between 2019 and late 2022, significant shoreline fluctuations occurred because of the development of the Anin Thermal Power Plant. A 1.5 km-long offshore breakwater constructed in front of Baseline 7 reduced the incident wave energy and disrupted sediment circulation between Hashidong and Yeomjeon Beach. This caused notable accretion near the dike at Baseline 7 and simultaneous retreat at the other baselines. In response, a wide submerged breakwater approximately 600 m in length was constructed in front of Hashidong Beach over one year, beginning in the summer of 2021. Despite this, wave energy imbalances induced by the structure resulted in localized sedimentation and erosion: a salient formed behind the breakwater near Baseline 5, while severe erosion occurred at Baselines 4 and 6. Six groins were constructed at Baseline 6 beginning in 2023 to protect the dune and coastal road in this ecologically valuable area. Nevertheless, the structures remain vulnerable and have required repeated maintenance.
3.2 Long-term Shoreline Change Trends in GW38, Gungchon–Chogok Port
The GW38 is a pocket beach approximately 2.9 km long, bounded by a northern breakwater and a southern rocky outcrop that limits sediment transport. Historically, this beach has remained stable, supported by a continuous sediment supply from the Chu River, located on its northern side. On the other hand, the construction of Gungchon Port in 2007 disrupted this natural sediment transport system. The breakwater of the port blocked wave energy, which is essential for transporting sediment from the river mouth to the southern beach. This led to excessive sediment deposition near the river mouth and a decrease in sediment supply to the southern coast, ultimately causing significant erosion along the southern shoreline.
Satellite imagery was used to investigate the spatial patterns of erosion and accretion in the GW38, similar to those in the GW27 section. The past 40 years of shoreline changes reveal significant accretion behind Gungchon Port, as shown in Fig. 11. At the same time, the areas around Gungchon and Wonpyeong Beaches, where multiple submerged and offshore breakwaters have been constructed, and Chogok Ports appear to have maintained their historical shoreline positions. Nevertheless, a long stretch of erosion is evident on the southern side of the jetty separating Chogok and Munam Beaches. This contrast in shoreline change patterns across specific coastal structures suggests that their influence warrants closer examination. Accordingly, Fig. 12 and Table 4 provide detailed summaries of these structures.
Fig. 13 offers a detailed visualization of the spatial configuration and temporal evolution of the shoreline changes within the GW38. Fig. 13(a) displays the locations of five shore-normal baselines established for this analysis, while Fig. 13(b) presents the 40-year shoreline position time series extracted at each baseline using the CoastSat. These baselines were selected strategically to capture localized shoreline responses to specific coastal interventions and natural sediment dynamics. Baseline 1 was positioned to monitor excessive sediment accumulation near the Chu River mouth, attributed to the construction of Gungchon Port and continued sediment input from the river. Baselines 2 and 3 were established to examine the temporal evolution of the wave shadow zone formed by the port breakwater and the subsequent shoreline restoration efforts undertaken to mitigate erosion. Baselines 4 and 5 were placed further south to assess the shoreline adjustments caused by the restricted alongshore sediment transport due to groin structures. The following section provides a detailed analysis of the shoreline changes observed at each of these baseline locations.
Baseline 1 is in a region effectively sheltered from the wave energy by the northern breakwater of Gungchon Port. This area consistently receives sediment input from nearby small streams, resulting in relatively stable shoreline conditions compared to other locations. The installation of a submerged breakwater at Gungchon Beach in 2016 had minimal influence on the shoreline dynamics in this area. On the other hand, shoreline retreat began in the summer of 2020, shortly after Groin 1 was constructed. This erosion trend was intentional, aimed at preventing the harbor entrance from closing.
Baseline 2 is characterized by a relatively narrow beach width and severe erosion that followed the construction of Gungchon Port in 2007. The breakwater of the port obstructed the southward transport of sediment from a nearby stream and induced northward sediment movement caused by wave diffraction. Satellite imagery indicates that the construction of a submerged breakwater began in late 2014, followed by significant shoreline accretion exceeding 40 meters. This accretion was attributed primarily to artificial beach nourishment, which was stabilized by the wave-dissipating effect of the submerged breakwater. Subsequently, the construction of Groin 2 between Gungchon and Wonpyeong Beaches in the summer of 2021 further limited alongshore sediment transport, enhancing sediment retention and leading the beach toward a more stable equilibrium state.
Baseline 3 showed similar erosional patterns to Baseline 2, due primarily to the influence of Gungchon Port. A submerged breakwater was constructed in 2016, followed by beach nourishment, which initially led to shoreline advancement. On the other hand, severe shoreline retreat continued afterward. Unlike Baseline 2, this site is exposed more directly to strong winter waves, which promote offshore and southward sediment transport, leading to sustained erosion. These effects were mitigated by installing a groin at the southern end of Wonpyeong Beach in 2019, followed by the installation of detached breakwaters in 2020. These coastal structures effectively reduced the wave energy and sediment loss during the winter season, resulting in a gradual increase in beach width. As a result, the shoreline had essentially returned to its pre-disturbance state by 2023.
Baselines 4 and 5 are located farther from the primary sediment source and Gungchon Port, making them more susceptible to seasonal fluctuations. Satellite imagery analysis reveals a consistent pattern of shoreline retreat during the winter season and accretion during the summer months. The construction of a groin at Wonpyeong Beach in 2019, positioned to the north of these baselines, disrupted alongshore sediment transport, leading to pronounced shoreline retreat exceeding 40 meters at Baseline 4. In contrast, Baseline 5 exhibited modest shoreline advance during the same period, likely due to the southward transport of sediment from Baseline 4 via alongshore currents.
4. Discussion
This study developed a comprehensive process for extracting the shorelines from more than 40 years of optical satellite imagery. By optimizing data acquisition parameters and applying tailored preprocessing methods, such as enhancing image quality based on satellite characteristics, a 40-year shoreline archive was constructed for 41 littoral cells along the East Coast of Korea. These regions are exposed to strong winter wave energy, which drives persistent offshore and southward sediment transport. Among these, two representative littoral cells, GW27 and GW38, where shoreline changes driven by human interventions are particularly evident, were analyzed in detail using CoastSat, a widely used satellite-based shoreline extraction model. The model effectively captured the shoreline change trends in response to anthropogenic activities, producing shoreline time series that aligned well with geomorphological expectations. Specifically, the results corresponded closely to the timing and type of coastal structures, such as groins, submerged and detached breakwaters, and beach nourishment, that influence wave energy dissipation and sediment transport in the alongshore and cross-shore directions.
Nevertheless, several limitations of the CoastSat model were identified during the analysis. Although it offers an end-to-end workflow from satellite image acquisition to visualization, its rigid architecture hinders the incorporation of customized preprocessing or model retraining for performance enhancement. For example, CoastSat lacks a georeferencing correction module, which was found to be critical for mitigating spatial misalignment errors, as shown in Fig. 6. Although monthly averaging was applied to reduce temporal inconsistencies in satellite coverage (Fig. 3), unrealistic outliers (highlighted with red circles in Fig. 10) were still present. A manual inspection of the corresponding satellite imagery confirmed that these anomalies stemmed from the absence of georeferencing. In addition, the model requires manual validation of each extracted shoreline, making long-term analyses like this study time-consuming and labor-intensive. In particular, the core limitation of CoastSat lies in its use of a simple four-class classification approach that extracts shorelines solely as the boundary between pre-classified water and sand pixels. Although this method has proven to be effective for relatively simple, undeveloped coastlines, it appeared unsuitable for the complex coastal environments of Korea, where numerous anthropogenic structures and varied geomorphology present significant challenges. For example, Vos et al. (2019b) reported that at the north end of the Narrabeen–Collaroy embayment in southeast Australia, an environment with an average foreshore slope of ~0.1 and a microtidal range of ~1.3 m, conditions comparable to the East Coast of Korea, CoastSat-derived shoreline time series showed good agreement with in situ field measurements collected since the mid-1970s across five cross-shore transects (coefficient of determination of 0.8 and RMSE of 7.2 m, with 90% the errors falling within 11.2 m). In Korean coastal settings, however, where geomorphological complexity and intensive human activity significantly modify the natural shoreline, the classification-based approach of CoastSat led to reduced performance. In many cases, the model either failed to extract specific shoreline segments or incorrectly classified the boundaries of coastal structures, such as breakwaters and groins, as part of the shoreline. Moreover, Vos et al. (2019a) reported that the performance of CoastSat further deteriorates on gentle-slope, macrotidal coasts, which are similar to Korea’s West and South Coasts. In these environments, the RMSE increased to 12.7 m, compared to 7.3 m on steep, microtidal beaches. This performance gap was primarily attributed to the CoastSat definition of the shoreline as the instantaneous boundary between water and sand captured in satellite imagery, which is quite sensitive to tidal stage and beach slope. These limitations suggest that caution is needed when applying CoastSat across the Korean coastline. Therefore, further refinement of the model framework is essential to improve its applicability.
5. Conclusion
This study established a robust and automated workflow for generating long-term shoreline archives from optical satellite imagery and showed the applicability of CoastSat to the East Coast of Korea over a 40-year period. The proposed framework effectively captured the general shoreline seasonality and shoreline dynamics shaped by the immediate and delayed effects of various coastal engineering interventions. Traditionally, shoreline change studies have relied on sparse and discontinuous in situ survey data, making it difficult to continuously and quantitatively track the impacts of coastal structures. In contrast, the satellite-based approach enables researchers to establish a reliable baseline of natural equilibrium shorelines from the 1980s (before significant coastal development) and to monitor changes with high temporal and spatial resolution. This capability supports detailed quantification of shoreline displacement caused by coastal developments. It provides scientific evidence to assist coastal engineers and managers in formulating more effective and sustainable coastal protection strategies. It can also guide the regulation of indiscriminate coastal development.
Although CoastSat is effective in general applications with an accuracy of under 10 m (Vos et al., 2019a; 2019b; 2023), its rigid architecture and reduced performance in complex anthropogenic environments underscore the need for more adaptable and precise models suited to Korea’s diverse coastal settings. Given these limitations, future research will focus on two key directions. First, this study aimed to enhance the input data quality through advanced preprocessing techniques, such as georeferencing correction and super-resolution. Second, a more flexible satellite-based shoreline extraction framework was planned using unsupervised or semi-supervised deep learning to improve automation and accuracy. Further validation is required in Korea’s varied coastal environments because the performance of CoastSat can vary according to specific coastal characteristics. Accordingly, future work will include validation studies at selected sites with long-term in situ shoreline observation data. Ultimately, this study aims to expand its geographic scope to Korea’s West and South Coasts, which exhibit distinct coastal dynamics, and contribute to the development of a generalized satellite-based shoreline monitoring model that applies to a wide range of global coastal environments.
PDF Links
PubReader
ePub Link
Full text via DOI
Download Citation
Print
