Open AccessArticle

Geoinformatics and Machine Learning for Shoreline Change Monitoring: A 35-Year Analysis of Coastal Erosion in the Upper Gulf of Thailand

Chakrit Chawalit

¹,

Wuttichai Boonpook

^1,*

Asamaporn Sitthi

Kritanai Torsri

Daroonwan Kamthonkiat

³,

Yumin Tan

⁴

Apised Suwansaard

⁵ and

Attawut Nardkulpat

Department of Geography, Faculty of Social Sciences, Srinakharinwirot University, Bangkok 10110, Thailand

Hydro-Informatics Institute, Ministry of Higher Education, Science, Research and Innovation, Bangkok 10900, Thailand

Department of Geography, Faculty of Liberal Arts, Thammasat University, Rangsit Campus, Khlong Nueng 12120, Thailand

⁴

School of Transportation Science and Engineering, Beihang University, Beijing 100191, China

⁵

Department of Civil Engineering, Faculty of Engineering, Rajamangala University of Technology Rattanakosin, Salaya, Nakhon Pathom 73170, Thailand

Author to whom correspondence should be addressed.

ISPRS Int. J. Geo-Inf. 2025, 14(2), 94; https://doi.org/10.3390/ijgi14020094

Submission received: 18 December 2024 / Revised: 14 February 2025 / Accepted: 15 February 2025 / Published: 19 February 2025

(This article belongs to the Topic Climate Change Impacts and Adaptation: Interdisciplinary Perspectives)

Download

Browse Figures

Versions Notes

Abstract

Coastal erosion is a critical environmental challenge in the Upper Gulf of Thailand, driven by both natural processes and human activities. This study analyzes 35 years (1988–2023) of shoreline changes using geoinformatics, machine learning algorithms (Random Forest, Support Vector Machine, Maximum Likelihood, Minimum Distance), and the Digital Shoreline Analysis System (DSAS). The results show that the Random Forest algorithm, utilizing spectral bands and indices (NDVI, NDWI, MNDWI, SAVI), achieved the highest classification accuracy (98.17%) and a Kappa coefficient of 0.9432, enabling reliable delineation of land and water boundaries. The extracted annual shorelines were validated with high accuracy, yielding RMSE values of 13.59 m (2018) and 8.90 m (2023). The DSAS analysis identified significant spatial and temporal variations in shoreline erosion and accretion. Between 1988 and 2006, the most intense erosion occurred in regions 4 and 5, influenced by sea-level rise, strong monsoonal currents, and human activities. However, from 2006 to 2018, erosion rates declined significantly, attributed to coastal protection structures and mangrove restoration. The period 2018–2023 exhibited a combination of erosion and accretion, reflecting dynamic sediment transport processes and the impact of coastal management measures. Over time, erosion rates declined due to the implementation of protective structures (e.g., bamboo fences, rock revetments) and the natural expansion of mangrove forests. However, localized erosion remains persistent in low-lying, vulnerable areas, exacerbated by tidal forces, rising sea levels, and seasonal monsoons. Anthropogenic activities, including urban development, mangrove deforestation, and aquaculture expansion, continue to destabilize shorelines. The findings underscore the importance of sustainable coastal management strategies, such as mangrove restoration, soft engineering coastal protection, and integrated land-use planning. This study demonstrates the effectiveness of combining machine learning and geoinformatics for shoreline monitoring and provides valuable insights for coastal erosion mitigation and enhancing coastal resilience in the Upper Gulf of Thailand.

Keywords:

coastal erosion; machine learning; shoreline extraction; Upper Gulf of Thailand; remote sensing; digital shoreline analysis system (DSAS)

1. Introduction

Coastal erosion is a natural process influenced by dynamic forces such as wave action, wind, water flow, sea level rise, and sediment transport [1]. In recent years, global warming has intensified these processes, leading to rapid glacier melting, rising sea levels, and stronger storm surges [2]. Rising sea levels and climate change exacerbate coastal erosion, endangering ecosystems, disrupting local economies, and threatening the livelihoods of millions of coastal residents. With a coastline extending approximately 3151 km, Thailand faces significant challenges due to coastal erosion. The Gulf of Thailand accounts for 2039.78 km of this coastline, while the Andaman Sea contributes 1111.35 km. According to the Department of Marine and Coastal Resources report [3], about 69.74 km of Thailand’s coastline experienced erosion in 2022, categorized into high erosion (29.88 km), medium erosion (26.79 km), and low erosion (13.07 km). Since 1995, the upper Gulf of Thailand has experienced severe erosion, with rates exceeding 25 m per year. This erosion is mainly caused by rising sea levels and extreme weather events, severely impacting muddy and sandy beaches with floodplain topography.

To monitor coastal erosion, there are two primary methods: ground surveys, and remote sensing technology. Ground surveys, while highly accurate, require significant manpower, time, and financial resources [4]. For example, Bird et al. [5] employed radar-based methods to monitor coastal morphological changes. While these methods offer real-time monitoring capabilities, they are limited by their smaller spatial coverage. In contrast, remote sensing techniques utilize two main data sources including aerial photographs and satellite imagery. Aerial photographs [6,7] and unmanned aerial vehicle (UAV) images [8,9] provide high spatial resolution and broad coverage. However, their use can be costly due to the need for frequent flights and the complexity of image processing. Satellite imagery offers high spatial resolution, wide coverage, multispectral capabilities, and frequent data acquisition, making it ideal for continuous coastal monitoring [10]. Synthetic aperture radar (SAR) imagery provides systematic, near-daily acquisitions with all-weather capabilities. Its meter-level spatial resolution makes it highly effective for shoreline extraction [11]. However, SAR images do not represent land use features, making it challenging to identify specific land use changes. In contrast, optical satellite imagery provides clearer land use features, making it more suitable for detecting and monitoring land use changes over time. For example, Xu [12] used Landsat imagery from 1986 to 2015 to monitor shoreline dynamics in Texas, revealing significant erosion, with 52.58% of the coastline affected over 30 years. In Thailand, Kamthonkiat et al. [13] utilized time-series remotely sensed data from Landsat and ASTER between 1991 and 2010 to estimate erosion and accretion rates before and after the 2004 Indian Ocean tsunami. Ajira Thiangtrong [14] used Landsat 5 imagery to monitor coastal changes in Ban Laemsing, Samut Prakan, improving the differentiation between land and sea using multispectral bands. The study identified significant erosion, particularly exacerbated by human activities, and highlighted its impact on mangrove forests. Similarly, Supawat Inkerd et al. [15] used Landsat imagery and aerial photographs to digitize shorelines and monitor changes in the Andaman Sea region, identifying significant erosion and accretion between 1995 and 2019, particularly in areas such as Chang Lang Beach and Pak Meng Beach. Pattrakorn Nidhinarangkoon and Sompratana Ritphring [4] conducted a comparative study of shoreline detection methods, utilizing satellite imagery from Google Earth to digitize shorelines and monitor coastal changes, with a focus on correcting for tidal effects. In recent years, satellite remote sensing has become increasingly pivotal for coastal monitoring, providing valuable data on shoreline dynamics at both local and global scales.

Manual digitization methods can provide accurate shoreline extraction but are time-consuming and costly, prompting researchers to explore more efficient techniques. Recent advances in remote sensing have enabled the integration of automated image-processing techniques to improve shoreline monitoring. Several studies have utilized indices such as the Modified Normalized Difference Water Index (MNDWI) and the Normalized Difference Water Index (NDWI) for coastal line extraction. For instance, Kim et al. [16] applied MNDWI on Landsat imagery from 1985 to 2011, combined with the Canny Edge Detection Algorithm, to automatically extract and monitor shoreline changes. Their analysis revealed an average coastal erosion rate of 6.4 m per year, with a total coastline retreat of approximately 166 m over 26 years. Similarly, Ghosh et al. [17] also used MNDWI on Landsat imagery from 1989 to 2010 to digitize shorelines, revealing significant erosion in the northern and western parts of Hatiya Island, along with accretion in the southern and eastern regions. Furthermore, Maglione et al. [18] enhanced shoreline extraction using NDVI and NDWI indices, combined with pan-sharpening techniques on high-resolution WorldView-2 imagery, resulting in more accurate coastline delineation in the Campania region of Italy. The integration of satellite imagery with machine learning (ML) techniques has transformed coastal monitoring by allowing the automated extraction of shoreline features with higher precision and efficiency [19].

ML algorithms have gained prominence in shoreline extraction and classification tasks due to their ability to handle complex datasets and improve classification accuracy. For instance, Fatah et al. [20] employed multiple ML models, including XGBoost (version 1.6.2), Random Forest (RF), and Decision Tree (DT), to delineate groundwater potential zones (GWPZs) in Iraq’s Kurdistan region using geoinformatics-based modeling. Similarly, Rash et al. [21] utilized multi-temporal Landsat imagery to analyze LULC changes over three decades in the Kurdistan Region, Iraq, demonstrating that RF outperformed other classifiers in accuracy and consistency. Recent studies have further refined these techniques; Sun et al. [22] found that that Landsat and Sentinel-2 have significantly advanced the ability to monitor coastal changes over time. This provides more frequent revisits and multispectral capabilities, enhancing the precision of coastline delineation. ML algorithms, such as the Minimum Distance Classifier (MDC), Maximum Likelihood Classifier (MLC), K-Nearest Neighbor (KNN), K-Means, ISODATA, Artificial Neural Networks (ANNs), Random Forest (RF), and Support Vector Machines (SVM), have proven effective for feature classification. These methods improve the accuracy and efficiency of distinguishing features like land, water, and vegetation compared to traditional techniques. For instance, a recent study by Kannan et al. [23] explored the use of SVM for shoreline extraction. Choung and Jo [24] compared traditional water-index methods with ML algorithms, demonstrating the superiority of SVM for shoreline extraction, particularly in challenging coastal conditions. Similarly, Minghelli et al. [25] focused on improving shoreline extraction by addressing foam pixel interference in WorldView-2 imagery, demonstrating that SVM outperformed traditional classification methods in accurately delineating shorelines, especially in coastal areas affected by natural events like the Tohoku tsunami. Bamdadinejad et al. [26] compared shoreline extraction methods using Landsat imagery and found that SVM outperformed MLC in detecting shoreline changes along the Persian Gulf, particularly in areas affected by erosion and accretion. Zhou et al. [27] reviewed the growing importance of ML in coastline extraction, emphasizing its ability to efficiently process large-scale data. Algorithms like RF, SVM, and Convolutional Neural Networks (CNNs) have proven effective for classifying coastal features such as land, water, and vegetation. Additionally, indices like NDWI and MNDWI have improved coastline detection accuracy across various environments.

As a recent and effective method for shoreline extraction, Vos et al. [28] introduced CoastSat as a reliable platform for global shoreline extraction, demonstrating its ability to produce accurate results. Curoy et al. [29] further validated CoastSat’s performance by using it to digitize a time series of shoreline positions in southern Thailand. However, local environmental factors, such as sediment type, seasonal monsoon effects, and water clarity, can influence the accuracy of shoreline extraction. Therefore, customizing the algorithm to account for these regional variations is essential for capturing shoreline dynamics accurately in specific coastal regions. Additionally, the choice of ML algorithm plays a crucial role in ensuring the accuracy and reliability of shoreline extraction. Applying multiple indices enhances feature differentiation, improving the classification of coastal features. This study employs MDC, MLC, SVM, and RF for shoreline extraction, each chosen for its strengths. MDC is fast and efficient for well-separated spectral classes, while MLC handles spectral overlaps, making it robust for coastal mapping. SVM excels in complex land–water boundaries, and RF, a powerful ensemble method, ensures high accuracy with minimal tuning. This selection provides a comprehensive evaluation, optimizing shoreline classification accuracy.

For coastal change analysis, the Digital Shoreline Analysis System (DSAS) has been widely adopted for monitoring and analyzing shoreline changes [30,31]. The DSAS was applied to calculate rates of change using a series of historical shoreline positions extracted from remote sensing imagery. Numerous studies have successfully adopted the DSAS in different regions. Kannan et al. [23] employed the DSAS to assess shoreline changes along the Visakhapatnam coast from 1989 to 2015, utilizing multi-temporal satellite imagery from Landsat and IRS-P6 LISS sensors. Goksel et al. [32] focused on the Black Sea region in Istanbul, Turkey, using the DSAS alongside the MLC of Landsat-7 and Landsat-8 imagery to analyze shoreline changes between 2009 and 2016. Similarly, Wang et al. [33] utilized the DSAS to study coastal changes in Ningbo, China, specifically in Hangzhou Bay, Xiangshan Bay, and Sanmen Bay from 1976 to 2015, employing automatic shoreline extraction using NDWI and MNDWI indices. Darwish and Smith [34] compared the performance of Landsat-8, PlanetScope, and Sentinel-2 imagery for detecting coastline changes in El-Alamein, Egypt, from 2016 to 2021, utilizing the DSAS to analyze shoreline dynamics. Rumahorbo et al. [35] applied the DSAS to monitor shoreline changes in Supiori Regency, Indonesia, revealing significant erosion rates ranging from 1.22 to 1.46 m per year, and accretion rates ranging from 1.05 to 1.68 m per year. Chettiyam Thodi et al. [36] integrated remote sensing and GIS techniques for shoreline monitoring in coastal islands of Kerala, India, from 1973 to 2019, utilizing Landsat imagery for shoreline digitization and applying the DSAS to calculate erosion and accretion change rates. Quang et al. [37] employed the DSAS to analyze shoreline changes in Quang Nam, Vietnam, using the MNDWI index to extract annual shorelines from Landsat and Sentinel-2 imagery.

In Thailand, Oraon Sarajit and Kanchana Kakhapalorn [38] assessed coastal erosion in Phetchaburi province using Landsat TM-5 and THEOS imagery from 2001 to 2011. Their study focused on the impacts of seasonal monsoons on shoreline changes. Similarly, Wuttipong Sangmanee and Chanchai Thanawut [39] used the DSAS to assess shoreline changes in Pattani province from 1975 to 2011, emphasizing the influence of coastal defense structures on erosion and accretion rates. Aekkarak Faiboob and Wutthipong Seangmenee [40] analyzed shoreline changes at the estuary of Songkhla lake, focusing on the impact of coastal engineering structures over 48 years (1967–2015) using remote sensing data and the DSAS. Anutin Kongkhan [41] utilized time-series remote sensing imagery to analyze shoreline changes and assess coastal erosion in Samut Prakan and Chachoengsao provinces from 2017 to 2020. More recently, Pawornprach Imlamai et al. [42] utilized Landsat 8 imagery and applied the DSAS to assess shoreline changes and evaluate erosion in Phetchaburi province, employing the Normalized Shoreline Change Index (NSMI) and Normalized Difference Shoreline Index (NDSSI) for enhanced shoreline area analysis. It was reported that coastal erosion remains high in the Upper Gulf of Thailand. While these studies demonstrate the effectiveness of the DSAS in tracking shoreline changes, many rely on medium-resolution imagery, limited machine-learning approaches, and shorter temporal scopes.

This research builds upon previous work by addressing the severe coastal erosion observed in the Upper Gulf of Thailand through the use of modern geoinformatics techniques. This study uses remote sensing data and ML algorithms (MDC, MLC, SVM, and RF) to automatically extract and classify coastal lines over a 35-year period (1988–2023), leveraging their efficiency and accuracy in handling complex datasets. Additionally, multiple normalized indices are applied to enhance land use feature extraction, water frequency index (WFI) is used for shoreline extraction, and the DSAS is used to analyze coastal changes and erosion rates. The goal is to develop a more efficient and accurate approach to monitoring coastal changes and informing strategies to mitigate the impacts of coastal erosion.

2. Study Area

This research focuses on the Upper Gulf of Thailand, encompassing the coastlines of eight provinces: Prachuap Khiri Khan, Phetchaburi, Samut Songkhram, Samut Sakhon, Samut Prakan, Bangkok, Chachoengsao, and Chonburi. The study area spans approximately 407 km along the coastline, as depicted in Figure 1. This region was selected due to its high susceptibility to coastal erosion, consistently observed from 1995 to 2023. The majority of the coastline consists of sandy and muddy beaches with a flat terrain. These features make it highly vulnerable to coastal retreat caused by tidal surges and other natural forces.

The study area is characterized by low-lying, flat coastal plains, particularly in regions with extensive mudflats. The flat terrain and proximity to sea level make the region highly susceptible to flooding, amplifying the effects of tidal action, storm surges, and extreme weather events. While storm surges and cyclonic events can cause rapid and severe coastal changes, it is crucial to understand their interaction with sea-level rise, as storms often amplify the impacts of rising sea levels in vulnerable areas. For example, storm surges associated with tropical storms and cyclones can lead to short-term, high-intensity erosion events, accelerating shoreline retreats in regions already stressed by gradual sea-level rise. This dynamic interplay between long-term processes, like sea-level rise, and short-term events, such as storms, adds complexity to the interpretation of coastal changes observed through satellite imagery. Differentiating between gradual erosion driven by sea-level rise and episodic, intense changes caused by storms is essential for accurate coastal monitoring. Natural features such as mangrove forests and mudflats provide some protection against storms and rising sea levels, but the overall low elevation of the region amplifies its vulnerability to both gradual and sudden coastal changes.

The climate of the Upper Gulf of Thailand is tropical monsoon, with three distinct seasons: the rainy season (May to October), the summer season (March to May), and the winter season (November to February). The rainy season is dominated by the southwest monsoon, which brings heavy rainfall that contributes to storm surges, exacerbating coastal erosion in low-lying areas. During this season, the wind pattern drives a clockwise circulation, especially along the western coast. The summer season experiences high temperatures and generally dry conditions, with occasional thunderstorms. During this transitional period, water flows from the south, moving eastward or southward based on monsoonal influences. The winter season is dominated by the northeast monsoon, which brings cooler and drier air, accompanied by moderate winds that reduce the intensity of coastal erosion compared to the rainy season. During this period, counterclockwise circulation prevails, driven by northeast winds that push water into the southern part of the gulf and out through the eastern coast [43,44]. The coastal changes in the study area result from a complex interplay of factors, including long-term sea-level rise, seasonal variations in storm intensity, and episodic extreme weather events. These dynamic drivers, coupled with the region’s low-lying topography, make the Upper Gulf of Thailand particularly vulnerable to both gradual and rapid coastal change processes. Despite these challenges, this study primarily focuses on the slow, progressive effects of sea-level rise, which has been identified as the dominant long-term driver of coastal erosion in the region.

Due to the large size of the study area and its diverse physical characteristics, the region was divided into six sub-regions for analysis, as shown in Figure 1. These sub-regions are Hua Hin District–Laem Phak Bia (Region 1), Laem Phak Bia–Mae Klong River (Region 2), Mae Klong River–Tha Chin River (Region 3), Tha Chin River–Chao Phraya River (Region 4), Chao Phraya River–Bang Pakong River (Region 5), and Bang Pakong River–Sattahip District (Region 6).

3. Materials and Methods

3.1. Materials

To monitor coastal erosion, satellite imagery, geospatial data, and other relevant datasets were employed to analyze shoreline changes in the Upper Gulf of Thailand over an extended period. A summary of the data used is provided in Table 1. The research employs two primary datasets, one focused on shoreline changes and the other on physical geography. The first dataset focuses on analyzing shoreline changes. It includes Landsat imagery for land and water classification, used to create annual shoreline data and generate training and validating datasets. Shoreline data for 2018 and 2023 serve as reference points for validation. Additionally, land use shapefiles provide yearly information on LULC changes in coastal areas. The second dataset provides physical geography insights. It includes sea-depth data for underwater topography, a digital elevation model (DEM) for onshore topography, and annual global sea level and temperature data for tracking trends. Additionally, it includes local sea level and temperature data from the Upper Gulf of Thailand and geological maps for analyzing coastal characteristics.

3.2. Methods

The process of this research is divided into several key steps, including data preparation, training data, ML classification, accuracy assessment, shoreline extraction, shoreline assessment, shoreline change analysis, and the analysis of the causes of coastal changes. The process of each step is shown in Figure 2 and is explained in the following sections.

3.2.1. Data Preparation

Data Preprocessing

The collected satellite images and geospatial data were preprocessed to ensure correct data formats and coordinate systems. All Landsat imagery was georeferenced to a reference image using the UTM WGS84 zone 47N coordinate system to ensure spatial consistency. All shapefiles used in this research were reprojected to align with the Landsat data.

Index Calculation

Spectral Indices are used to improve the land–water distinction. These indices serve a specific role in enhancing shoreline detection. They provide a more comprehensive approach to accurately detecting and mapping shorelines.

The Normalized Difference Water Index (NDWI), as shown in Equation (1), is designed to enhance water bodies in satellite images by maximizing the difference between near-infrared (NIR) and green wavelengths. This index is effective in separating water from land and enhances the shoreline boundary, making it suitable for identifying the water–land transition zone. However, it may misclassify shallow water mixed with vegetation as land.

N D W I = \frac{G r e e n - N I R}{G r e e n + N I R}

(1)

The Modified Normalized Difference Water Index (MNDWI), as shown in Equation (2), is an adjustment of the NDWI that replaces the NIR band with a shortwave infrared (SWIR) band. This modification helps reduce noise caused by built-up areas and vegetation, improving the distinction between water and non-water features, particularly in urbanized coastal zones. The MNDWI sharpens the boundary between water and land, making it more effective in areas with significant anthropogenic features. However, it is less effective in distinguishing water from non-vegetated land.

M N D W I = \frac{G r e e n - S W I R}{G r e e n + S W I R}

(2)

In contrast, the NDWI was chosen for this study due to the natural characteristics of the Upper Gulf of Thailand, where water–land boundaries are primarily defined by natural features such as beaches and mangroves. The NDWI proved effective in these regions, while MNDWI may be more suitable for coastal zones with greater urbanization.

The Normalized Difference Vegetation Index (NDVI), as shown in Equation (3), is primarily used for vegetation analysis. It takes advantage of the fact that healthy vegetation reflects more NIR light and absorbs visible red light. This index is often used to assess the presence and condition of vegetation. It can indirectly assist shoreline analysis by identifying and excluding vegetated areas along the coastline and refining water–land boundary detection, especially in mangrove areas. In contrast, it performs poorly in distinguishing water from non-vegetated land.

N D V I = \frac{N I R - R e d}{N I R + R e d}

(3)

The Soil Adjusted Vegetation Index (SAVI), as shown in Equation (4), is a modification of NDVI that accounts for the influence of soil brightness in areas with sparse vegetation. It introduces a correction factor (L) to minimize soil effects, making it useful in areas where vegetation does not fully cover the soil. It is helpful in coastal areas with sparse vegetation or where bare soil is present. It improves shoreline mapping by distinguishing vegetation from soil, thus refining the boundary between vegetated land and water. However, it struggles with pure water detection and misclassifies blended vegetation and soil.

S A V I = \frac{(1 + L) (N I R - R e d)}{N I R + R e d + L}

(4)

Layer Stacking

This process produced a set of stacked spectral bands and expanded indices. For Landsat 5 and Landsat 7, the stacked images included the bands Blue, Green, Red, Near Infrared (NIR), Shortwave Infrared 1 (SWIR1), and Shortwave Infrared 2 (SWIR2), along with the indices NDVI, NDWI, MNDWI, and SAVI. For Landsat 8 and 9, the stacked images included Coastal Aerosol, Blue, Green, Red, NIR, SWIR1, SWIR2, and the same four indices. These stacked images are used in land-use classification.

3.2.2. Training Data

To prepare the training data for classification, land use data were simplified into two main categories: water and land. The number of samples was determined by using the ratio of water and land. This was combined with a stratified sampling strategy to ensure an appropriate distribution of sample points that accurately represented the different land cover types. All 65 Landsat images were used to collect data samples. A stratified random sampling approach was applied to ensure that the sample groups were well distributed across the study area and proportionally representative of water and land cover types. The total of the data samples is 12,000 sets. To create the training dataset, 70% of the sample sets were designated as training data, while the remaining 30% were used as testing data as shown in Table 2. This division ensured that the training and testing datasets were mutually exclusive, preventing overlap and enhancing the reliability of the classification model.

3.2.3. Machine Learning Classification

To classify land and water areas using Landsat imagery, Kim et al. [45] highlighted the frequent application of RF, ANN, and SVM in environmental research. This study draws from the findings of Sun et al. [22], who highlighted the increasing role of ML algorithms in coastal extraction. The performance of algorithms such as SVM, MLC, and RF, which demonstrated high accuracy, was also noted by Kannan et al. [23], Choung and Jo [24], Bamdadinejad et al. [26], Goksel et al. [32], and Zhou et al. [27]. These algorithms were selected for their respective strengths in handling the complexities of coastal image data, fast processing capabilities, and minimal parameter requirements. While CNNs excel in high-resolution imagery, their high computational demands and extensive training requirements make them less suitable for multi-decadal shoreline monitoring with Landsat imagery. Based on these considerations, four ML algorithms were employed in this study including MDC, MLC, SVM, and RF. Each algorithm was chosen for its specific advantages and suitability for the task of shoreline classification. MDC was included as a baseline method due to its simplicity, speed, and efficiency, particularly in cases where computational resources are limited or when quick classifications are needed. MDC is effective in scenarios with well-separated classes where there is little overlap between land and water. The algorithm calculates the Euclidean distance between pixel values in the feature space, assigning each pixel to the class with the nearest mean value. MLC was selected for its probabilistic approach, assuming that data within each class follows a normal distribution. This method performs well when the classes are well-defined, and the data meet these assumptions. MLC is robust in handling overlapping classes, which is common in coastal environments where water, sand, and vegetation may blend. SVM was included due to its strength in handling nonlinear boundaries and its capability to perform well with high-dimensional data, such as satellite imagery with many bands. SVM is well-regarded for its robustness in creating hyperplanes that maximize the margin between classes, making it effective even with smaller training datasets. However, it requires careful tuning of parameters to achieve optimal performance. In this study, the regularization parameter (C) and kernel parameters (gamma and epsilon) were optimized, with gamma set to 30,000 and error control set to 600. The Radial Basis Function (RBF) kernel was chosen due to its effectiveness in mapping data into higher-dimensional spaces, enhancing the separability of classes. To further improve the algorithm’s performance, grid search combined with cross-validation was applied to systematically fine-tune the parameters. RF was selected for its capability to handle complex and heterogeneous datasets, as well as its robustness to noise, as confirmed by Agarwal et al. [46], who demonstrated its effectiveness in modeling groundwater storage variations with minimal sensitivity to noise and data heterogeneity. This characteristic is particularly essential for coastal regions characterized by diverse land cover, water reflections, and seasonal variability. RF operates as an ensemble learning method, combining multiple decision trees to reduce the risk of overfitting and enhance classification accuracy. In this study, the RF hyperparameters were optimized using random search to achieve optimal performance. The key parameters included the number of trees set to 20, the maximum depth set to 600, and the minimum number of samples required for splitting nodes set to 20.

To compare the effectiveness of these algorithms, Landsat-8 imagery from 28 December 2023 was selected as the test dataset due to its cloud-free conditions and minimal atmospheric distortion, and the algorithm with the best performance in terms of both overall accuracy and Cohen’s Kappa was then applied to classify the remaining 64 Landsat images from 1988 to 2023, ensuring consistency in shoreline extraction across the entire study period.

3.2.4. Accuracy Assessment

The classification results were validated using 30% of the data as testing data, distinct from the training data. The accuracy of the classifications was calculated using two metrics such as overall accuracy, and Cohen’s Kappa coefficient. The accuracy results of the four classification methods were then evaluated for the performance of land-use classification.

3.2.5. Shoreline Extraction

In this process, classification images are used to generate annual shoreline representations. This approach aims to minimize the effects of varying sea levels in different images, ensuring more consistent and accurate shoreline detection. One consideration in this process is the potential impact of tidal range on shoreline position. While tidal fluctuations can lead to short-term variations in shoreline location, the methodology used in this study focuses on mean sea level and does not explicitly account for tidal range variations. Given the relatively flat topography of the study area and focus on annual shoreline changes, the tidal range is assumed to have a minimal impact on the overall shoreline extraction process. The first step in shoreline extraction is the binary raster creation. The classification results of all 65 satellite images were divided into seven time periods. The classified images were reprocessed into Binary Raster maps, separating them into two maps: water map and land map. In the water map, water pixels were assigned a value of 1, and land pixels were assigned a value of 0. In contrast, in the land map, land pixels were assigned a value of 1, and water pixels were assigned a value of 0. The results were a set of binary raster maps distinguishing water and land. The second step is the aggregation of the water raster. The binary raster maps for water areas (water = 1, land = 0) from each image were aggregated to create a single water raster for that year. Similarly, the binary raster maps for land areas (water = 0, land = 1) were aggregated to generate a single land raster for each year. The third step is the Water Frequency Index (WFI) calculation. This method calculated the WFI from the water and land raster maps based on Equation (5). The output is a WFI image that represents the separation between land and water.

W F I = \frac{N_{W a t e r}}{N_{W a t e r} + N_{L a n d}}

(5)

where

N_{W a t e r}

is the number of water pixels and

N_{L a n d}

is the number of land pixels.

The fourth step involves thresholding each WFI image to create the shoreline. A threshold of 0.5 was used, where pixel values greater than or equal to 0.5 were defined as water areas, and pixel values less than 0.5 were defined as land areas. This thresholding process follows the approach by Quang et al. [37] and Xu et al. [12]. The output of the thresholding process is a set of binary raster shoreline images. The fifth step is the conversion of raster to vector. The binary raster shoreline images were then converted into vector files (ESRI shapefile format), representing the shoreline as a line feature for each period. Lastly, the sixth step is the refinement of shoreline data. The vector line data for all seven years were refined by removing estuary and canal areas, focusing exclusively on the coastal shorelines. The result was a single annual shoreline vector.

3.2.6. Shorelines Assessment

Due to the large size of the study area and the long duration of the study, there were limitations in accessing historical shoreline data for certain years, including 1988, 1994, 2000, 2006, and 2011. The shoreline data for the Upper Gulf of Thailand, as generated by the Department of Marine and Coastal Resources for their report on Thailand’s coastal conditions, was used. This included data for two years, 2018 (using 2019 data as a substitute) and 2023 (using 2022 data as a substitute), to verify the accuracy of the extracted annual shorelines. The verification process involved the following steps. The first step is to buffer the shoreline data by 30 m. This buffer parameter is to match the resolution of the Landsat images. The result is a reference shoreline in the polygon feature type with a 30 m width on both sides for the years 2019 and 2022. The second step is to verify the accuracy of the extracted annual shorelines. This involved overlaying the extracted annual shorelines in the line feature types for the years 2018 and 2023 with the reference shoreline in the polygon feature types for the years 2019 and 2022. The intersected areas were then calculated, and the accuracy was expressed as a percentage using the following Equation (6). The result was the accuracy of the extracted shorelines for the years 2018 and 2023.

P e r c e n t a g e o f s h o r e l i n e a c c u r a c y = \frac{L e n g t h o f I n t e r s e c t L i n e}{L e n g t h o f E x t r a c t e d S h o r e l i n e} \times 100

(6)

While the percentage overlap method is simple and effective for general validation, it does not account for regional specificities and may overlook smaller errors that can be significant in certain areas, especially where classification errors are more pronounced. To address this limitation, the Root Mean Square Error (RMSE) metric was incorporated for a more comprehensive evaluation. RMSE quantifies the difference between the reference and extracted shorelines by squaring the errors, averaging them, and then taking the square root, as shown in Equation (7). This metric is particularly effective in highlighting larger discrepancies, which are more common in regions with complex shorelines.

R M S E = \sqrt{\frac{1}{n} \sum_{i = 1}^{n} {(x_{i} - y_{i})}^{2}}

(7)

where

x_{i}

and

y_{i}

are the coordinates of the reference and extracted shorelines, respectively.

n

is the total number of corresponding points used to calculate the error between the reference and extracted shorelines.

RMSE offers a more comprehensive validation of the shoreline extraction process by addressing both average errors and larger discrepancies that may occur in certain regions. When combined with the percentage overlap, they provide a more robust measure of the accuracy of the extracted shorelines, particularly in areas with complex coastal features or significant classification errors.

3.2.7. Analysis of Shoreline Changes Using the Digital Shoreline Analysis System (DSAS)

To analyze shoreline changes in the Upper Gulf of Thailand, this study employed the Digital Shoreline Analysis System (DSAS) following a structured workflow. The analysis consisted of four main steps. The first step is baseline creation. It involved the generation of inland baselines using verified annual shorelines. To establish an initial reference, a uniform buffer distance of 25 m per year was applied, based on the maximum recorded coastal erosion rate. To account for regional and temporal variations, buffer distances were adjusted according to observed erosion rates for each time period including 1988–1993 (6 years) set to 150 m, 1993–1999 (6 years) set to 150 m, 1999–2005 (6 years) set to 150 m, 2005–2010 (5 years) set to 125 m, 2010–2017 (7 years) set to 175 m, and 2017–2023 (5 years) set to 125 m. For the entire study period (1988–2023, 35 years), a cumulative buffer distance of 875 m was applied, calculated by multiplying the critical erosion rate (25 m per year) by the total number of years. The resulting baseline features were created as line datasets with attribute information corresponding to each period. The second step is annual shoreline preparation. This step involved processing annual shorelines, ensuring consistency and accuracy. Attribute data were assigned to each shoreline, with a default positional uncertainty of 30 m to reflect the spatial resolution of Landsat imagery. These shorelines were systematically organized for analysis across all six sub-regions and time periods. The third step is geodatabase creation and transect generation. A personal geodatabase was developed to store shoreline and baseline data for all sub-regions and time intervals. Transects were automatically generated at 50 m intervals along the shorelines, ensuring a systematic and spatially consistent framework for analysis. The final step is shoreline change calculation. The DSAS was used to compute two key shoreline change metrics including Net Shoreline Movement (NSM) measuring the total perpendicular movement of the shoreline over time, and End Point Rate (EPR) determining the rate of shoreline change by dividing the NSM by the corresponding time interval.

To maintain the accuracy of the analysis, estuaries and man-made structures along the coastline were excluded. The severity of shoreline changes was classified using the Department of Marine and Coastal Resources (DMCR) criteria, which categorize erosion and accretion rates into seven classes including Severe Erosion (>5 m/year), Moderate Erosion (1–5 m/year), Minor Erosion (0.5–1 m/year), Stable (−0.5–0.5 m/year), Minor Accretion (0.5–1 m/year), Moderate Accretion (1–5 m/year), and High Accretion (>5 m/year). The result was a spatially explicit map visualizing shoreline change severity, along with data on perpendicular shoreline movement across six time periods in six sub-regions.

3.2.8. Analysis of the Causes of Coastal Changes

To analyze the coastal change, three statistics were analyzed: annual average temperature from meteorological stations in or near the study area, annual average water level from water level stations in or near the study area, and annual average accumulated coastal erosion distance, calculated using the Digital Shoreline Analysis System (DSAS). The trends of these three statistics were observed to determine how they influenced the coastline in each sub-region. The analysis also considered other factors such as the area’s topography, geology, land use, seasonal currents, and coastal structures. The result was an identification of the causes of coastal changes in each sub-region of the study area.

4. Results

In this study, the monitoring of shoreline changes was conducted using remote sensing, combined with shoreline analysis through the Digital Shoreline Analysis System (DSAS). The causes of shoreline changes were also analyzed based on various factors in the Upper Gulf of Thailand. The results of the study are coastal area classification results, annual shoreline creation, accuracy verification of the annual shoreline, and coastal analysis using the digital shoreline analysis system.

4.1. Coastal Area Classification Results

To evaluate the performance of classification methods in the coastal area. Landsat-8 captured on 28 December 2023 was used for the classification test. A total of 12,000 points were used for both training data and testing data, with no overlap between the two datasets, as shown in Table 2. The coastal area was classified into two categories: water and land, using four classification methods: Minimum Distance to Means (MDC), Maximum Likelihood Classifier (MLC), Support Vector Machine (SVM), and Random Forest (RF). The overall accuracy and Cohen’s Kappa Coefficient were used for evaluating these classification methods. The results are presented in Figure 3.

As seen in Figure 3, the Random Forest (RF) method achieved the highest overall accuracy with 98.17% and 0.94 of Cohen’s Kappa Coefficient. This method overcomes the maximum likelihood classifier method with 89.50% of overall accuracy and 0.69 of Cohen’s Kappa Coefficient, the Support Vector Machine method with 87.83% of overall accuracy and 0.50 of Cohen’s Kappa Coefficient, and the Minimum Distance to Mean (MDC) with 84.83% of overall accuracy and 0.59 of Cohen’s Kappa Coefficient.

The classification results for land and water classes varied across the different algorithms as shown in Figure 4. The Minimum Distance algorithm struggled with errors, particularly misclassifying land as water in coastal fishing areas. Similarly, the Support Vector Machine (SVM) algorithm also showed high classification errors, especially in distinguishing coastal fishing areas and mangrove forests, where it often incorrectly classified mangrove areas as water. The Maximum Likelihood algorithm performed better in classifying mangrove forests but still encountered misclassification issues in coastal fishing zones. In contrast, the Random Forest algorithm significantly reduced classification errors in these challenging areas, particularly between water and coastal fishing regions. While the Random Forest algorithm provided more accurate results compared to the other methods, some minor errors persisted in regions where the distinction between land and water was not clearly defined in the imagery.

Therefore, it was concluded that the most effective classification method is Random Forest. It was applied for classifying all 64 images that were captured from 1988 to 2023. The classification results shown in Figure 5 demonstrate that the overall accuracy ranged from 93.67% to 99.83%. Cohen’s Kappa Coefficient ranged from 0.81 to 0.99. The average overall accuracy of over 65 images is 97.54%. These results confirmed that these classification results indicate excellent classification performance and are suitable for shoreline extraction.

4.2. The Results of Annual Shoreline Creation

The classification results of coastal areas were derived from 65 images spanning 1988 to 2023, which were used for shoreline extraction. To monitor shoreline change, images were grouped into seven time periods (1988, 1994, 2000, 2006, 2011, 2018, and 2023). Land and water areas were separated and analyzed using the water frequency index, with values ranging from 0 to 1, and a threshold of 0.5 was applied to distinguish between land (0) and water (1). The resulting data were converted from raster to vector format, and non-coastal water bodies such as rivers and canals were removed to focus on actual shoreline changes. The results of shoreline creation from seven periods are shown in Figure 6.

The annual shorelines remained consistent and continuous across all seven periods, successfully eliminating classification issues related to shallow water and clouds. Additionally, the method effectively delineated fishponds, which act as barriers between land and sea, despite their small size relative to the satellite image resolution. Most artificial structures in the sea were also successfully classified. However, challenges remained in differentiating ports and certain coastal infrastructure.

Ports often consist of breakwaters, piers, and docks that extend into the sea, creating mixed land–water boundaries that are difficult to classify using spectral indices alone. These structures, particularly when connected to the mainland, were sometimes misclassified as land, leading to overestimation of the shoreline extent in those areas. Additionally, large industrial docks and harbors may have water surfaces that appear similar to natural coastal features, further complicating classification. Despite efforts to refine classification accuracy, some artificial coastal structures were still partially misclassified due to spectral similarities with natural shorelines.

4.3. The Annual Shoreline Assessment

This section presents the results of accuracy verification for the extracted annual shorelines for the years 2018 and 2023. The extracted shorelines were compared against the 30 m buffer of the reference shorelines for 2019 and 2022, provided by the Department of Marine and Coastal Resources. As for accuracy verification (Percentage Overlap), for the shoreline assessment in 2018, the accuracy of the extracted shoreline was 81.78%, with a correct distance of 362,961.60 m out of a total distance of 443,823.62 m. The error distance was 80,862.02 m (18.22%), with discrepancies primarily observed around ports on the eastern side and large beaches on the western side of the Upper Gulf of Thailand (Figure 7a).

For 2023, the accuracy was 80.01%, with a correct distance of 363,164.24 m out of a total distance of 453,917 m. The error distance was 90,752.76 m (19.99%), showing similar errors around ports and large beaches (Figure 7b). Both extracted shorelines demonstrated accuracy above 80%, confirming their reliability for further coastal analysis.

To further assess the accuracy, Root Mean Square Error (RMSE) values were calculated using 374 validation points for both years; for the shoreline assessment in 2018, the RMSE was 13.59 m, indicating an average deviation of 13.59 m between the extracted and reference shorelines. The higher RMSE reflects discrepancies in areas with complex coastal features, such as water–land boundaries and ports. For 2023, the RMSE improved to 8.90 m, demonstrating reduced error and improved shoreline extraction accuracy. This improvement is attributed to refined processing techniques, higher-quality satellite imagery, and enhanced classification algorithms.

The verification results confirm that the extracted shorelines for 2018 and 2023 are reliable, with accuracy exceeding 80% and RMSE values showing measurable improvement. The lower RMSE in 2023 highlights the effectiveness of advancements in remote sensing, improved image resolution, and enhanced shoreline extraction techniques. This confidence extends to the extracted shorelines for earlier years (1988, 1994, 2000, 2006, and 2011), despite the lack of validation data for those years.

4.4. The Results of the Digital Shoreline Analysis System (DSAS)

The shoreline change results were derived from the analysis of annual shoreline data across six time periods. As shown in Table 3 and Figure 8, from 1988 to 1994, Region 4 experienced the highest rate of coastal erosion at 13.55 m per year, followed by Region 5 with an average of 6.02 m per year. Region 2 had the highest rate of accretion, with a rate of 10.09 m per year, followed by Region 1 with 2.01 m per year. From 1994 to 2000, Region 4 continued to exhibit the highest erosion rate at 12.50 m per year, while Region 5 had the second-highest erosion rate at 6.68 m per year. Region 2 again saw the highest accretion rate, averaging 8.89 m per year. Between 2000 and 2006, Region 4 remained the most eroded, with an erosion rate of 10.35 m per year, followed by Region 5 at 6.77 m per year. Region 2 saw the highest accretion rate during this period, at 2.94 m per year. From 2006 to 2011, Region 5 had the highest erosion rate at 4.55 m per year, followed by Region 4 at 3.95 m per year. Accretion was observed in Region 1 at 2.81 m per year and Region 2 at 2.41 m per year. Between 2011 and 2018, Region 4 again had the highest erosion rate at 3.88 m per year, with Region 6 ranking second at 2.05 m per year. Region 1 saw minor accretion at 0.10 m per year. From 2018 to 2023, Region 2 experienced the highest erosion rate at 1.55 m per year, followed by Region 4 with 0.54 m per year. Region 5 saw the highest accretion rate during this period, at 3.30 m per year. Over the entire period from 1988 to 2023, Region 4 exhibited the highest total erosion rate at 8.41 m per year, followed by Region 5 at 3.95 m per year and Region 3 at 2.14 m per year. Region 1 and Region 6 experienced erosion rates of less than 1 m per year, with 0.33 and 0.23 m per year, respectively. The highest overall accretion rate was observed in Region 2, at 3.44 m per year.

4.5. The Analysis of the Coastal Changes

The DSAS analysis reveals significant changes in the coastline along the Upper Gulf of Thailand from 1988 to 2023, primarily driven by rising sea levels closely linked to global warming. As illustrated in Figure 9 and Figure 10, increasing global temperatures have led to the melting of polar ice caps and mountain glaciers, subsequently elevating global sea levels. This relationship is reflected in the strong correlation coefficient (0.8632) shown in Figure 10a, highlighting the interconnected nature of rising temperatures and sea levels. Furthermore, the correlations between global sea level, temperature, and coastal erosion are also significant, with values of 0.7659 and 0.8633, respectively, as shown in Figure 10b,c. These findings underscore the role of rising temperatures and sea levels as primary drivers of coastal erosion. On a regional scale, data from meteorological stations across the Upper Gulf of Thailand confirm that this area follows the global trend. A strong correlation (0.8136) between mean sea level and coastal erosion is observed in Figure 10f, indicating that rising sea levels play a dominant role in reshaping the coastline. However, while both average temperature and annual sea level demonstrate steady increases, the relationship between temperature and sea level rise within the region is weaker (correlation of 0.5117, Figure 10d). This suggests that local environmental factors including sediment supply, monsoonal currents, coastal geomorphology, and extreme weather events contribute to sea-level impacts and erosion processes in the region. In addition to long-term sea-level rise, short-term climatic events such as storm surges, tropical storms, and monsoonal variations have played a crucial role in shaping the shoreline dynamics of the Upper Gulf of Thailand. The impact of these events is evident in episodic erosion spikes that deviate from the general long-term trend of progressive shoreline retreat. For example, storm surges associated with seasonal monsoons temporarily elevate water levels, exacerbating erosion, especially in low-lying areas. While our analysis focuses on the dominant influence of global warming-induced sea-level rise, integrating localized storm data could further refine our understanding of coastal change dynamics and provide more precise insights into extreme weather impacts.

To better understand the factors contributing to coastal erosion including topography, geology, land use, seasonal currents, coastal structures, human activities, and extreme weather events, the study area was divided into six regions, as illustrated in Figure 1. This division facilitates a more localized assessment of erosion patterns, sediment deposition, and the effectiveness of coastal protection measures across the study area.

4.5.1. Region 1: Hua Hin District–Laem Phak Bia

The coastal area between Hua Hin District and Laem Phak Bia demonstrates significant variability in erosion and accretion patterns over the past several decades (1988–2023) as shown in Figure 11.

As seen, land use along the coastal area is built up and beach forest. This region features low-lying coastal plains with flat, sandy beaches, fine sediments, and estuarine mudflats, making it highly vulnerable to erosion, particularly during the southwest monsoon and storm surges. The coastline is composed of mud, sand, and gravel, with most beaches being sandy, which enhances its susceptibility to erosion. The moderately steep slope of the coastline, combined with rising sea levels and intensifying ocean currents, exacerbates erosion on flat, sandy beaches, which are particularly prone to sediment loss. The dynamics of sediment transport and erosion in the region are heavily influenced by seasonal currents. The southwest monsoon (May–October) generates stronger northward currents that intensify erosion by displacing sediments from more exposed areas, particularly along flat, sandy beaches. Additionally, storm surges and extreme weather events during the southwest monsoon have caused episodic erosion, accelerating shoreline retreat in vulnerable regions. For example, Bang Kao Beach reveals localized trends of both erosion and accretion. From 1988 to 2006, continuous erosion was observed, particularly in the central and northern parts of Cha-am Beach. Erosion rates decreased between 2006 and 2011, with noticeable sand deposition on several beaches, driven by sediment transport and localized sediment accumulation. However, after 2011, erosion rates began to rise again. For example, the central region near the Cha-am Fishing Pier has experienced continuous erosion over multiple decades, while Bang Kao Beach, though showing moderate accretion in 2011, returned to severe erosion by 2018 and 2023. The absence of effective coastal protection measures in these areas, combined with the intensification of southwest monsoonal currents, has resulted in shoreline instability.

An analysis of specific areas such as Area A, located closer to estuarine zones (Figure 12c), has experienced consistent sediment accretion as shown in Figure 12a, averaging 4.07 m per year. Between 2017 and 2023, accretion rates peaked at 4.41 m per year, as sediment from river discharge and estuarine processes accumulated along the shoreline. However, despite its overall positive sediment balance, Area A remains susceptible to episodic erosion during storm surges and extreme weather events, particularly in areas where mangrove forests have been cleared. The removal of mangroves and land reclamation for agriculture or infrastructure has further exposed the shoreline to tidal forces and erosion. In contrast, in Area B, located in the exposed sections of the coastline (Figure 12d), erosion has been the dominant process, as shown in Figure 12b, with an average rate of 3.42 m per year between 1988 and 2023.

The shoreline retreat in this area has been influenced by its flat topography, strong tidal forces, and exposure to southwest monsoonal currents. Erosion was particularly severe during 1988–1994, when rates peaked at 8.88 m per year, leading to an erosion of 12.05 m. Although erosion rates temporarily decreased between 1994 and 2006, the trend intensified again between 2006 and 2011, with erosion rates reaching 10.37 m per year. This period corresponds to increased wave energy during the southwest monsoon and the absence of effective coastal protection structures. By 2023, the cumulative erosion in Area B reached 10.82 m, underscoring the area’s vulnerability to rising sea levels, tidal forces, and extreme weather events. Overall, the analysis of Region 1 highlights a dynamic interplay between erosion and accretion processes, with distinct variability between Area A and Area B. However, both areas remain vulnerable to the impacts of rising sea levels, tidal forces, and storm surges. Between 1988 and 2023, the region experienced a net shoreline change with a cumulative sediment loss of 10.82 m.

4.5.2. Region 2: Laem Phak Bia–Mae Klong River

The coastal area between Laem Phak Bia and the Mae Klong River is characterized by an alluvial plain formed by sediment deposition at the estuary, with extensive mangrove forests and mudflats sustained by sediment carried and deposited by the Mae Klong River and Khlong Bang Tabun. The flat topography and the presence of fine sediments make the area highly susceptible to erosion, particularly during the southwest monsoon when stronger currents displace sediments from south to north. In contrast, the northeast monsoon (November–February) facilitates sediment deposition as slow-moving currents transport material from north to south, contributing to mangrove expansion and mudflat formation near estuarine zones. Extreme weather events have further shaped the region’s shoreline dynamics, with storms and tropical cyclones contributing to episodic erosion. These events brought high storm surges and strong wave action that accelerated shoreline retreat, particularly in vulnerable areas like the southern Mae Klong estuary.

The analysis of coastal changes in Region 2 highlights a dynamic interplay between erosion and accretion processes. Between 1988 and 2023, the region experienced significant net shoreline erosion and accretion as shown in Figure 13, particularly in Area A, which includes the southern part of the Mae Klong estuary. Area A has shown an average annual accretion rate of 18.40 m per year, resulting in a cumulative shoreline expansion of 139.64 m by 2023 as shown in Figure 14a. This significant accretion is primarily driven by sediment deposition from strong tidal forces and riverine processes, particularly during the northeast monsoon season, when slower currents facilitate sediment accumulation near estuarine zones. This area features flat coastal plains and mudflats that are conducive to sediment deposition. The riverine forces of the Mae Klong River play a critical role in delivering sediment loads to the estuary, while the tidal forces during seasonal monsoons further contribute to sediment retention in the area. Additionally, the gradual accumulation of sediments has been observed in the southern Mae Klong estuary, where estuarine mudflats provide a stable environment for accretion processes (Figure 14c).

In contrast, Area B, a coastal area situated near the Pak Thale Conservation area, has experienced persistent erosion rather than accretion in recent decades as shown in Figure 14b, with an average annual erosion rate of 13.39 m per year between 1988 and 2023. Despite its proximity to sediment-rich environments, this erosion is primarily driven by strong tidal currents, particularly during the southwest monsoon (May–October), which displaces sediments away from the shoreline and reduces sediment deposition. Erosion pressures have intensified, driven by the combined effects of stronger tidal currents, rising sea levels, and human activities. The conversion of mangrove forests into agricultural land and salt farms has significantly reduced the natural sediment retention capacity of the shoreline, leaving the area more vulnerable to tidal forces and wave action. Efforts to combat erosion in Area B have included the installation of coastal protection measures, such as bamboo fences, which aim to absorb wave energy and promote sediment accumulation (Figure 14d). While these measures have proven effective in stabilizing certain sections of the shoreline, they have not been sufficient to counteract the over-arching erosion trends, particularly in zones where human disturbances and tidal forces remain dominant. In summary, the coastal area between Laem Phak Bia and the Mae Klong River reflects a dynamic balance of erosion and accretion driven by natural processes and human interventions. While regions such as Area A have benefited from sediment deposition facilitated by seasonal currents and mangrove expansion, Area B faces continuous erosion due to tidal forces, storm surges, and human activities.

4.5.3. Region 3: Mae Klong River–Tha Chin River

The coastal area between the Mae Klong River and Tha Chin River estuaries in the Upper Gulf of Thailand demonstrates dynamic variability in erosion and accretion processes during 1988–2023 as shown in Figure 15. This region is primarily characterized by extensive mudflats formed from alluvial soil deposition at river estuaries, combined with low-lying coastal plains that slope gently toward the sea. The flat terrain and fine sediments make the area highly prone to erosion, particularly during strong tidal forces and rising sea levels, while the presence of mangrove forests provides some natural resilience to erosion. The flat, low-lying terrain, combined with mudflats and fine sediments, makes the shoreline particularly vulnerable to seawater intrusion and sediment displacement caused by strong tidal currents. The land used in this area includes coastal fishing and salt farming, which have encroached on mangrove forests and disrupted natural sediment retention processes (Figure 16d). Removing mangroves to accommodate salt pans and fish farms has exposed large portions of the shoreline to tidal and wave action, accelerating erosion rates, especially in Area A. Seasonal monsoonal currents significantly influence the region’s sediment dynamics. The southwest monsoon (May–October) produces stronger northward-moving currents that displace sediments from more exposed areas, intensifying erosion. Extreme weather events, including tropical storms and storm surges, have accelerated episodic erosion in vulnerable areas. The combination of rising sea levels and stronger wave energy during storms has contributed to significant shoreline retreats, particularly in exposed sections of Area A.

For northeast monsoons, it generates slow-moving currents that promote sediment deposition along sheltered mudflat areas, contributing to shoreline stability, with the construction of coastal protection since 2000. Soft measures such as bamboo fences have been implemented to trap sediments and stabilize sections of the shoreline (Figure 16c), particularly in Area B, where accretion has been more consistent. Additionally, hard structures, such as rock revetments and embankments, have been built to reduce erosion in highly exposed zones. However, in Area A, these measures have proven less effective due to the combined pressures of rising sea levels, strong tidal currents, and the lack of natural buffers like mangrove forests. For the analysis of shoreline changes in specific areas, Area A is situated in more exposed coastal sections as shown in Figure 16a; Area A has experienced consistent net erosion at an average rate of 8.86 m per year. Erosion was particularly severe between 1988 and 2011, with rates peaking at 30.72 m per year during the 1993–2000 period. By 2023, the cumulative shoreline retreat had reached 50.18 m, underscoring the long-term vulnerability of this area to tidal forces and sea-level rise. Land reclamation for fish and salt farming, combined with embankment collapses caused by seawater intrusion, has exacerbated erosion in this zone. In contrast, Area B located further inland and closer to estuarine environments has shown consistent net accretion at an average rate of 7.42 m per year as shown in Figure 16b. Between 2017 and 2023, accretion rates peaked at 12.16 m per year, reflecting the continued role of mangrove expansion and sediment retention facilitated by calmer currents. Overall, between 1988 and 2023, shoreline erosion in Area A reached 50.18 m, while Area B demonstrated consistent shoreline accretion.

4.5.4. Region 4: Tha Chin River–Chao Phraya River

The coastal area between the Tha Chin River and Chao Phraya River estuaries, located in the northernmost part of the Gulf of Thailand, demonstrates significant variability in erosion and accretion processes during 1988–2023 as shown in Figure 17. This region primarily consists of flat coastal plains with gentle slopes formed by alluvial soil deposits from the estuaries, making the area highly vulnerable to seawater intrusion and coastal erosion. The geological composition, dominated by mud and sand, supports the growth of mangrove forests, which provide some natural resilience against erosion. The region’s low-lying terrain and flat coastal plains make it particularly susceptible to rising sea levels and sediment displacement caused by tidal currents. Extensive salt farming and the conversion of mangrove forests into agricultural or residential zones have significantly disrupted the natural sediment balance, weakening the shoreline’s resilience to erosion. The collapse of embankments in salt farm areas, combined with sediment removal during extreme tidal conditions, has further exacerbated erosion. The northeast monsoon brings slow-moving currents that facilitate sediment deposition, contributing to shoreline accretion and mangrove expansion. In contrast, the southwest monsoon generates stronger northward currents that displace sediments, intensifying erosion. Soft structures, such as bamboo fences (Figure 18c), were implemented in 2000 to trap sediments and reduce wave energy, particularly in accretion-prone areas. Hard structures, such as rock revetments and vertical barriers (Figure 18d), were constructed to stabilize the shoreline in severely eroded zones. Extreme weather events, including storm surges and tropical cyclones, have intensified sediment displacement and shoreline retreat in exposed areas. The combined effects of rising temperatures, higher sea levels, and extreme tidal conditions during storm events have accelerated episodic erosion, particularly in low-lying sections of Area A.

The analysis of shoreline changes in Region 4 reveals a predominantly erosional trend in exposed areas and localized accretion in more sheltered zones. For example, Area A is situated in exposed sections of the coastline and has experienced severe erosion as shown in Figure 18a, with an average annual erosion rate of 15.47 m per year between 1988 and 2023. The shoreline erosion reached 270.30 m by 2023, indicating the long-term vulnerability of this area to rising sea levels and tidal forces. The period between 1988 and 2000 saw the most significant erosion rates, peaking at 31.22 m per year during 1993–2000, as rising sea levels and seasonal currents displaced sediments from the shoreline. Although erosion rates slowed to 2.66 m per year between 2011 and 2018, the cumulative effects of seawater intrusion, strong tidal currents, and embankment collapse associated with salt farming and fish farming continued to exacerbate sediment loss. In contrast, Area B, located in more sheltered estuarine zones, has demonstrated net accretion with an average rate of 2.94 m per year as shown in Figure 18b. Between 1999 and 2005, accretion rates peaked at 11.21 m per year, reflecting the influence of slow-moving tidal currents and sediment retention near mangrove forests. More recently, between 2017 and 2023, accretion rates averaged 2.50 m per year, highlighting the continued role of natural sediment dynamics in stabilizing portions of the shoreline. Overall, while Area A has suffered persistent shoreline retreat, with a cumulative loss of 270.30 m, due to rising sea levels, tidal forces, and human disturbances, Area B has shown consistent sediment accretion facilitated by natural tidal currents and mangrove forest expansion.

4.5.5. Region 5: Chao Phraya River–Bang Pakong River

The coastal area between the Chao Phraya River and Bang Pakong River estuaries exhibits complex shoreline changes from 1988 to 2023 as shown in Figure 19. This region is a lowland area formed by river sedimentation, with a geological composition dominated by mud and sand. The flat terrain and low elevation make the region highly vulnerable to flooding during high tides and coastal erosion, particularly when exposed to strong wave action during the southwest monsoon season. The region’s flat lowlands and muddy shorelines are prone to erosion due to their low elevation and weak geological composition. Extensive salt farming and the conversion of mangrove forests into aquaculture and residential zones have disrupted natural sediment dynamics and removed critical coastal buffers. Fish farming embankments, particularly near exposed shorelines, have frequently collapsed due to rising sea levels and strong tidal forces, further accelerating erosion by displacing sediments into tidal currents. Extreme weather events, including storm surges and high wave action during the southwest monsoon (May–October), have accelerated episodic erosion and generated strong northward currents, which intensify erosion in exposed areas. These currents displace sediments and exacerbate shoreline retreat, particularly where natural buffers like mangrove forests have been removed.

In contrast, the northeast monsoon (November–February) promotes sediment deposition in sheltered zones. A combination of soft and hard coastal protection structures has been implemented since 2001. Soft structures, such as bamboo fences, have effectively trapped sediments and reduced wave energy in sheltered zones, promoting localized accretion (Figure 20c). Hard structures, including rock walls (Figure 20d) and vertical coastal barriers, have been constructed to protect exposed areas with heavy fishing and industrial activity. While these measures have helped stabilize some sections of the shoreline, their effectiveness has been limited to regions with severe erosion and frequent embankment failures. The region’s vulnerability to erosion has been exacerbated by land use practices, including the expansion of urban settlements, salt farming, and fish farming, which have reduced the natural resilience of the coastline. The removal of mangroves and the collapse of embankments have left the shoreline exposed to tidal forces and storm surges.

The analysis of shoreline changes in Region 5 reveals a predominance of erosion with some localized accretion, influenced by tidal forces, human activities, and seasonal dynamics. For example, Area A is located in the more exposed sections of the coastline. Area A has experienced consistent erosion as shown in Figure 20a, with an average annual rate of 11.51 m per year between 1988 and 2023. The cumulative shoreline retreat reached 125.50 m by 2023, reflecting the long-term vulnerability of this area to rising sea levels and strong tidal forces. The period between 1988 and 2011 saw the most intense erosion, particularly during 1988–1994, when rates peaked at 23.36 m per year, resulting in a cumulative loss of 36.12 m by 1994. Although erosion rates slowed to 4.70 m per year between 2011 and 2018, the area continued to experience gradual retreat due to the collapse of fish farming embankments and the reduction of natural sediment retention provided by mangrove forests. In recent years, from 2017 to 2023, a slight reversal occurred with an accretion rate of 6.13 m per year, likely driven by the installation of soft coastal protection structures that trapped sediments and stabilized sections of the shoreline. In contrast, Area B, situated in more sheltered estuarine environments, has exhibited minor but consistent accretion as shown in Figure 20b, averaging 1.10 m per year. Between 1993 and 2000, accretion rates peaked at 6.30 m per year, as slow-moving currents and river sediment delivery promoted localized sediment buildup. However, from 2000 to 2011, sediment deposition declined, with accretion rates dropping to 0.96 m per year. In recent years (2017–2023), the area has stabilized, with modest accretion averaging 0.26 m per year, aided by bamboo fences and natural sediment dynamics. Overall, while Area A has suffered persistent shoreline retreat, with a cumulative loss of 125.50 m, due to rising sea levels, tidal currents, and the collapse of human-made structures, Area B has shown minor accretion facilitated by natural sediment dynamics and northeast monsoonal currents.

4.5.6. Region 6: Bang Pakong River–Sattahip District

The coastal area between the Bang Pakong River and Sattahip District, located along the eastern Gulf of Thailand, exhibits diverse erosion and accretion patterns over the period 1988–2023 as shown in Figure 21. This region can be divided into three distinct sections: the upper region dominated by alluvial plains, the central region characterized by mixed sediments of sandstone and gravel, and the southern region featuring rocky formations and cliffs that naturally resist erosion.

The upper region of Region 6 is dominated by lowlands with extensive alluvial deposits of mud and sand, making it particularly vulnerable to erosion under rising sea levels and strong wave action. This area is heavily used for coastal fisheries and aquaculture. The central and southern regions, in contrast, are composed of sandstone, gravel, and rocky outcrops, which are naturally resistant to erosion. The presence of cliffs with moderately sloping bases further enhances erosion resilience, reducing shoreline retreat in these areas. However, land use changes, including urbanization and port construction, have disrupted natural sediment dynamics in localized sections. Extreme weather events, such as storm surges during the southwest monsoon (May–October) and localized tropical storms, have exacerbated erosion and intensified northward tidal currents, displacing sediments and accelerating erosion in exposed areas, particularly in Area A (Figure 22a), where natural buffers like mangroves have been degraded. Conversely, the seasonal northeast monsoon (November–February) promotes sediment deposition by driving slow-moving currents from north to south, particularly in sheltered estuarine zones. Various coastal protection measures have been implemented across Region 6. Soft structures, such as bamboo fences in mangrove-rich areas, have been installed since 2015, effectively reducing erosion rates by trapping sediments and dissipating wave energy. Hard structures, including vertical barriers, piers, and rock revetments, have been constructed in urbanized areas since 2002 to stabilize shorelines. In the central and southern regions, offshore breakwaters and large ports have provided effective erosion control since 1988. Moreover, human activities, particularly the expansion of fish farms and urban development, have significantly impacted sediment transport and coastal stability. The conversion of mangroves to aquaculture has weakened natural sediment retention, increasing the vulnerability of soft coastal plains to erosion.

The analysis of shoreline changes in Region 6 highlights variability between erosion-prone zones in the upper region and more stable, resistant areas in the central and southern sections. For example, Area A, situated in the upper coastal plain with soft mud and sand deposits, has experienced significant erosion as shown in Figure 22a, averaging 5.50 m per year. The shoreline erosion reached 22.68 m by 2023, emphasizing its long-term vulnerability to tidal forces, seasonal monsoons, and rising sea levels. Between 1988 and 2011, erosion rates steadily increased, peaking at 11.74 m per year during the period 2005–2011. This intensification corresponds to stronger wave action during the southwest monsoon and the widespread conversion of mangrove forests into fish farms. Mangrove loss has reduced the natural buffering capacity of the coastline, leaving the area exposed to tidal currents and storm surges. In recent years (2017–2023), erosion rates declined slightly, with accretion recorded at 1.32 m per year, likely due to the installation of soft coastal structures such as bamboo wave barriers, which trap sediments and reduce wave energy (Figure 22c). Area B is located further south, consisting of mixed sandy and gravelly terrain, where erosion rates have been relatively lower, averaging 0.43 m per year as shown in Figure 22b. This area experienced minor accretion during certain periods, such as 1993–2006. However, between 2011 and 2018, erosion returned, with rates reaching 1.22 m per year, driven by tidal forces, rising sea levels, and localized urbanization impacts. The shoreline erosion in Area B reached 22.68 m by 2023, reflecting a gradual, long-term shoreline instability exacerbated by wave action and human activities.

Overall, while Area A has experienced persistent shoreline retreat, with a cumulative loss of 22.68 m, Area B has remained relatively stable due to its mixed geology and occasional sediment deposition (Figure 22d). The implementation of coastal protection structures has contributed to recent stabilization, but the region remains vulnerable to tidal forces, rising sea levels, and human activities.

5. Discussion

This research demonstrates remote sensing techniques, ML-based classification, and the DSAS for analyzing the shoreline changes in the Upper Gulf of Thailand over the past 35 years (1988–2023). The results illustrated the efficiency of coastal area classification, annual shoreline creation, annual shoreline assessment, DSAS-based metrics, and observed coastal changes.

One core aspect of this study was evaluating the performance of ML algorithms in classifying satellite imagery into land and water categories. Significant spectral bands (Blue, Green, Red, NIR, SWIR1, and SWIR2) and indices (NDVI, NDWI, MNDWI, SAVI) were used to enhance the shoreline separation. These elements allowed for precisely identifying shoreline boundaries and classifying complex coastal environments, including mudflats, mangroves, sandy beaches, and urban areas. For instance, NDWI and MNDWI proved particularly effective for differentiating water and land features near estuarine and tidal zones, while SAVI enhances the differentiation of sparse vegetation and bare soils. Similarly, NDVI distinguishes vegetated areas (e.g., mangrove and beach forests) from non-vegetated zones. The integration of spectral bands and indices with ML algorithms (RF, SVM, MLC, and MDC) demonstrated high overall accuracy results of over 80%. RF outperformed other classifiers, achieving the highest accuracy (98.17%) and a Cohen’s Kappa Coefficient of 0.9432, demonstrating strong classification performance along the shoreline. However, key challenges remain, particularly in handling complex coastal features such as mangrove forests, estuarine mudflats, and urbanized areas, which can introduce classification uncertainties. Additionally, RF is susceptible to overfitting, especially when working with highly correlated features. To optimize performance and reduce computational demands, RF requires careful hyperparameter tuning (e.g., adjusting the number of trees and tree depth) and often larger training datasets to ensure robust generalization. Meanwhile, other algorithms experienced misclassifications in areas with mixed pixels, especially in regions where water and vegetation boundaries overlap, highlighting the importance of spectral indices for detecting subtle shoreline changes. SVM can be robust and handle complex boundaries well, yet they may not match RF’s accuracy in large, high-dimensional datasets. Traditional methods such as the statistical MLC and the distance-based MDC often rely on simpler assumptions about data distribution and distance metrics, making them less effective in highly complex, overlapping spectral environments. Furthermore, the performance of RF was also confirmed when it was applied to classify 64 additional images, achieving an average overall accuracy of 97.54%. These classification images were grouped into seven periods to generate the annual shoreline, utilizing a water frequency index with a threshold of 0.5 to effectively distinguish between land and water. The annual shoreline assessments for 2018 and 2023 demonstrated high accuracy, with an overlapping percentage exceeding 80% and RMSE values of 13.59 m in 2018 and 8.90 m in 2023. These results confirm the reliability of the shoreline extraction process, as the observed shoreline shifts remained less than a single pixel (30 m), consistent with the resolution of Landsat imagery. This highlights that Landsat imagery is well-suited for long-term shoreline monitoring and change detection, providing a cost-effective and reliable solution for regional-scale coastal studies. Nevertheless, the overall accuracy supports the reliability of shorelines extracted for earlier years (1988, 1994, 2000, 2006, and 2011), despite lacking direct validation data.

The results of the DSAS from 1988 to 2023 illustrate distinct patterns of erosion and accretion in the Upper Gulf of Thailand across six periods. These results demonstrate the variability in erosion intensity and accretion trends, which can be attributed to a combination of natural processes, including tidal forces, seasonal currents, and rising sea levels, as well as human interventions, such as coastal development, land use changes, and the implementation of protective structures. During 1988–1994, the Upper Gulf of Thailand experienced the most severe erosion, particularly in Region 4 (−13.55 m/year) and Region 5 (−6.02 m/year). This period marked the beginning of significant shoreline retreat, driven by rising sea levels, increased wave energy, low-lying topography, and human interventions. Notably, the conversion of over 45% of mangrove areas into aquaculture and agriculture in Region 4 significantly weakened coastal defenses, correlating with the highest recorded erosion rates in this study. In Region 5, urban expansion and port development disrupted sediment transport, contributing to shoreline retreat exceeding 3 m per year. Other regions, such as Region 1 and Region 6, showed minor erosion rates, indicating localized impacts of tidal forces on sediment transport. In contrast, Region 2 experienced notable accretion (10.09 m/year), primarily influenced by sediment deposition near estuarine zones, which facilitated sediment retention. The 1994–2000 period continued to experience significant erosion, though at slightly reduced rates. Region 4 remained the most eroded (−12.50 m/years), while Region 5 increased (−6.68 m/years). However, Region 3, was previously stable, possibly due to sediment redistribution during this period (−5.22 m/years). Region 2 maintained high accretion (8.89 m/years), reinforcing the role of sediment-laden rivers in stabilizing certain sections of the coastline. This period reflects localized variability, where erosion rates persisted in exposed regions, while accretion occurred in sediment-rich estuarine zones. From 2000 to 2006, the rate of erosion declined compared to previous periods, although Region 4 remained highly vulnerable (−10.35 m/year), followed by Region 5 (−6.77 m/year). This decline may be attributed to the early implementation of coastal protection structures, including rock revetments and soft barriers. The expansion of mangrove forests helped buffer certain regions against wave energy. Notably, Region 2 showed a reduced accretion rate (2.94 m/year), indicating a possible shift in sediment dynamics or rising sea levels that hindered deposition. During the 2006–2011 period, erosion rates continued to decline in most regions, suggesting a stabilizing effect of coastal management efforts (coastal protection structures (e.g., bamboo fences and rock revetments)) and natural processes (the expansion of mangrove forests). Region 5 recorded the highest erosion rate, with shoreline retreat reaching −5.18 m/year, primarily due to the collapse of fish farming embankments and insufficient protection. In Region 4, erosion persisted at an average rate of −4.55 m/year, exacerbated by ongoing mangrove loss and sediment depletion. In contrast, Region 1 saw an accretion rate increase of 2.81 m/year. Meanwhile, minor accretion was recorded in Region 2 (2.41 m/year), reinforcing the role of sediment deposition near estuarine zones. The results indicate localized erosion persistence in vulnerable areas while regions with natural sediment dynamics, like mangrove zones, showed slight recovery. Between 2011 and 2018, erosion rates declined further, with some regions experiencing stabilization and minor accretion. Region 4 had erosion rates reduced (−3.88 m/year), while Region 6 showed localized erosion (−2.05 m/year). Region 5 saw minor accretion during this period (0.10 m/year), suggesting the effectiveness of soft coastal protection measures such as bamboo fences in trapping sediments. This period reflects a transition toward shoreline stabilization due to improved management practices, despite ongoing challenges posed by rising sea levels and wave action. The 2018–2023 period demonstrates a mixed trend of erosion and accretion, with regional variability. Region 2, which previously exhibited high accretion, shifted to erosion (−0.54 m/year), indicating changing sediment transport dynamics and increased exposure to tidal forces. In contrast, Region 5 experienced notable accretion (3.30 m/year), likely due to the successful implementation of coastal protection structures and sediment redistribution. Other regions, such as Region 4 and Region 6, exhibited minimal erosion (−0.40 m/year), signaling the effects of stabilization measures and natural geological resistance in these areas. This period highlights the dynamic interplay of natural processes, human activities, and coastal management interventions, where localized erosion persists in some regions while others benefit from sediment stabilization and protection structures.

Global warming has significantly influenced global temperatures and sea levels, both of which strongly correlate with shoreline erosion in the Upper Gulf of Thailand. While climate change is a key driver, regional factors such as sediment supply, tidal forces, and storm surges also shape shoreline dynamics. Our findings highlight that rising sea levels contribute to long-term coastal retreat, with a strong correlation between mean sea level and erosion. However, localized climatic events, such as monsoonal currents and extreme storms, lead to episodic fluctuations in erosion rates, highlighting the need for further studies focusing on short-term coastal changes and their immediate impacts on shoreline dynamics. This region’s coastal processes reflect a complex interplay between natural forces and anthropogenic influences. Key factors, including topography, geology, land use, seasonal currents, coastal structures, and human activities, collectively drive the observed shoreline changes. These interconnected elements govern erosion and accretion patterns and underscore the region’s heightened vulnerability to rising sea levels, intensified wave energy, and human-induced modifications. Sea-level rise accelerates shoreline retreat, especially in low-lying coastal areas, where even small increases exacerbate erosion and permanent land loss. The western part of the Upper Gulf of Thailand slopes gently from west to east, creating long beaches and mangrove forests near river estuaries. These long beaches remain relatively stable with minimal erosion, primarily due to natural sediment dynamics. Meanwhile, the mangrove forests in this region have experienced moderate sediment accumulation, influenced by the sloping topography and natural sediment deposition. Over time, however, the rate of accumulation has steadily declined, particularly between 2011 and 2023. Erosion is driven by seawater intrusion and stronger wave action, which disrupts mudflats, alters sediment transport, and degrades mangrove ecosystems. Rising sea levels further reduce sediment retention in mangroves, diminishing their natural defense against coastal erosion. The northern region consists of a flat alluvial plain, formed by sediment deposits from four major rivers: the Mae Klong, Tha Chin, Chao Phraya, and Bang Pakong rivers. The geology in this area is dominated by mud and sand, supporting the growth of extensive mangrove forests that serve as natural buffers against erosion. However, these mangroves have been increasingly degraded due to human encroachment, urban expansion, and the conversion of coastal zones for aquaculture and agriculture, which has weakened the region’s natural resilience to erosion. The eastern region of the Upper Gulf exhibits a diverse landscape. The northern part comprises mudflats and mangrove forests that stabilize the coastline, while the central region features sandy beaches, deep-sea ports, urban centers, and tourist attractions. This central area supports substantial industrial and commercial activities that disrupt sediment dynamics and contribute to localized erosion. In contrast, the southern part transitions into rocky beaches and cliffs, which are more resistant to erosion compared to the sandy and muddy coastal areas further north. Seasonal currents play a critical role in shaping the sediment dynamics and erosion processes in the Upper Gulf of Thailand. Furthermore, during the southwest monsoon (May to October), this region is temporarily affected by extreme weather events. It contributes to storm surges and currents redistribute sediments, intensify wave energy, and accelerate erosion, especially in vulnerable coastal areas with weak defenses. It also generates stronger south-to-north currents, which intensify erosion by redistributing sediments away from vulnerable coastal areas. Regions with low-lying topography and limited protective structures are particularly affected during this period. The Upper Gulf of Thailand’s coastal dynamics is shaped by a complex interaction of natural and human-induced factors, including topographical variation, sediment availability, seasonal monsoons, and anthropogenic influences such as urbanization, mangrove deforestation, and coastal infrastructure development. The region demonstrates a dynamic and often unstable coastline, with variations in erosion and sediment deposition that require region-specific management strategies.

While this study demonstrates the application of ML to coastal change, it acknowledges the need for further exploration into the cause-and-effect relationships between different drivers and the complex interactions among them. Future research should aim to deepen the understanding of these processes by integrating multiple data sources, advanced models, local environmental factors, short-term temporal variations, such as seasonal impacts, to enhance the understanding of coastal dynamics. This approach will help improve the predictive power of shoreline change models and support more effective coastal management in the face of climate change and sea-level rise.

6. Conclusions

This study analyzed shoreline changes in the Upper Gulf of Thailand over a 35-year period (1988–2023) using remote sensing, ML techniques, and the DSAS. The results highlight the dynamic and complex patterns of erosion and accretion influenced by both natural processes and human activities. For the coastal area classification, the RF algorithm achieved the highest accuracy and Cohen’s Kappa Coefficient among the four tested ML methods, demonstrating its robustness for classifying land and water areas. Spectral indices such as NDWI, MNDWI, NDVI, and SAVI were critical in enhancing feature extraction, enabling the accurate delineation of shorelines and coastal land cover. The seven periods of annual shoreline extraction and accuracy revealed consistent and reliable results, validated through overlay assessments and RMSE calculations. The shorelines extracted for 2018 and 2023 achieved high accuracy, with small RMSE values, indicating suitability for long-term shoreline monitoring using Landsat imagery. The DSAS results showed significant temporal and spatial variability in shoreline erosion and accretion across six periods. Severe erosion occurred between 1988–1994 and 1994–2000, particularly in Region 4 (Tha Chin River–Chao Phraya River) and Region 5 (Chao Phraya River–Bang Pakong River), where erosion rates exceeded 10 m per year. In contrast, Region 2 (Laem Phak Bia–Mae Klong River) exhibited consistent accretion, largely driven by sediment deposition in estuarine environments and mangrove expansion. However, erosion rates stabilized in recent decades, coinciding with the implementation of coastal protection measures, including soft and hard structures. The analysis reveals that coastal changes are driven by a combination of natural factors (tidal forces, seasonal monsoons, rising sea levels, and extreme weather events exacerbate erosion in low-lying areas with fine sediments) and anthropogenic factors (land use changes, the conversion of mangrove forests to aquaculture and salt farming, and urban development have weakened natural coastal defenses). Given the strong correlation between rising sea levels, temperatures, and shoreline erosion, proactive measures are needed to enhance the resilience of vulnerable coastal regions in the Upper Gulf of Thailand. Future studies should focus on integrating higher-resolution imagery and advanced deep-learning techniques to further improve shoreline extraction and change analysis. Incorporating hydrodynamic and sediment transport models could enhance the understanding of coastal processes. Additionally, considering the impacts of sea-level rise and storm surges would provide a more comprehensive perspective on short-term coastal evolution. Integrating more regional studies on these factors can further improve predictive shoreline change models, supporting more effective coastal management strategies in response to climate change.

Author Contributions

Conceptualization: Chakrit Chawalit, Wuttichai Boonpook, Asamaporn Sitthi, Kritanai Torsri, Daroonwan Kamthonkiat and Yumin Tan; methodology: Chakrit Chawalit, Wuttichai Boonpook, Asamaporn Sitthi, Kritanai Torsri and Apised Suwansaard; software: Chakrit Chawalit and Attawut Nardkulpat; validation: Chakrit Chawalit, Kritanai Torsri and Wuttichai Boonpook; formal analysis: Chakrit Chawalit and Wuttichai Boonpook; investigation: Chakrit Chawalit and Wuttichai Boonpook; resources: Chakrit Chawalit and Apised Suwansaard; data curation: Chakrit Chawalit and Attawut Nardkulpat; writing—original draft preparation: Chakrit Chawalit and Wuttichai Boonpook; writing—review and editing: Chakrit Chawalit, Wuttichai Boonpook, Asamaporn Sitthi, Daroonwan Kamthonkiat and Yumin Tan; visualization: Chakrit Chawalit and Wuttichai Boonpook; supervision: Wuttichai Boonpook and Asamaporn Sitthi; project administration: Chakrit Chawalit and Wuttichai Boonpook; funding acquisition: Wuttichai Boonpook and Apised Suwansaard. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the Faculty of Social Sciences, Srinakharinwirot University (No. 313/2566).

Data Availability Statement

The data presented in this study are available on request from the corresponding author. The data are not publicly available due to the permission of the data sharing policy.

Acknowledgments

All authors would like to thank the Faculty of Social Science and the Graduate School of Srinakharinwirot University for their research funding and facilities. We also extend our gratitude to the Hydrographic Department of the Royal Thai Navy, the Land Development Department, the Department of Marine and Coastal Resources, the Marine Department, the Meteorological Department, and the Department of Mineral Resources in Thailand, as well as the United States Geological Survey (USGS) and the National Aeronautics and Space Administration (NASA) in the USA for providing the spatial data used in the analysis process. We sincerely thank the reviewers and editors for their beneficial, careful, and detailed comments and suggestions that helped improve the paper.

Conflicts of Interest

The authors declare no conflicts of interest.

References

Li, K.; Zhang, L.; Chen, B.; Zuo, J.; Yang, F.; Li, L. Analysis of China’s coastline changes during 1990–2020. Remote Sens. 2023, 15, 981. [Google Scholar] [CrossRef]
WMO. State of the Global Climate 2022; UN: New York, NY, USA, 2023. [Google Scholar]
Department of Marine and Coastal Resources. Annual Report on the State of Marine and Coastal Resources and Coastal Erosion in Thailand, 2022, 1st ed.; Do Connection Company Limited: Bangkok, Thailand, 2024. [Google Scholar]
Ritphring, S.; Nidhinarangkoon, P. The comparative study of shoreline detection methods. In Proceedings of the 24th National Convention on Civil Engineering, Udonthani, Thailand, 10–12 July 2019. [Google Scholar]
Bird, C.O.; Bell, P.S.; Plater, A.J. Application of marine radar to monitoring seasonal and event-based changes in intertidal morphology. Geomorphology 2017, 285, 1–15. [Google Scholar] [CrossRef]
Sesli, F.A. Mapping and monitoring temporal changes for coastline and coastal area by using aerial data images and digital photogrammetry: A case study from Samsun, Turkey. Int. J. Phys. Sci. 2010, 5, 1567–1575. [Google Scholar]
Gonçalves, J.A.; Bastos, L.; Pinho, J.; Granja, H. Digital aerial photography to monitor changes in coastal areas based on direct georeferencing. In Proceedings of the 5th EARSeL Workshop on Remote Sensing of the Coastal Zone, Prague, Czech Republic, 1–3 June 2011; pp. 1–3. [Google Scholar]
Novais, J.; Vieira, A.; Bento-Gonçalves, A.; Silva, S.; Folharini, S.; Marques, T. The Use of UAVs for Morphological Coastal Change Monitoring—A Bibliometric Analysis. Drones 2023, 7, 629. [Google Scholar] [CrossRef]
Ahmed, T.; Creedon, L.; Gharbia, S.S. Low-Cost Sensors for Monitoring Coastal Climate Hazards: A Systematic Review and Meta-Analysis. Sensors 2023, 23, 1717. [Google Scholar] [CrossRef]
Geo-Informatics and Space Technology Development Agency Public Organization. Textbook on Space Technology and Geo-Informatics, 1st ed.; Amarin Printing and Publishing Public Company Limited: Bangkok, Thailand, 2009. [Google Scholar]
Zheng, H.; Li, X.; Wan, J.; Xu, M.; Liu, S.; Yasir, M. Automatic coastline extraction based on the improved instantaneous waterline extraction method and correction criteria using SAR imagery. Sustainability 2023, 15, 7199. [Google Scholar] [CrossRef]
Xu, N. Detecting coastline change with all available landsat data over 1986–2015: A case study for the state of Texas, USA. Atmosphere 2018, 9, 107. [Google Scholar] [CrossRef]
Kamthonkiat, D.; Koshimura, S.; Matsuoka, M. Time series analysis of shoreline positions in the pre-and post-2004 Indian Ocean Tsunami in Phang Nga Province, Thailand. J. Earth Sci. Eng. 2012, 2, 35–41. [Google Scholar]
Thiangtrong, A. Remote Sensing Applications for Change Detection in Coastal Region, Ban Laemsing Samut Parkarn Province. Master’s Thesis, Chulalongkorn University, Bangkok, Thailand, 2006. [Google Scholar]
Supawat Inkerd, N.O.; Kwanthong, N.; Klompong, N.; Khankajorn, T. Remote Sensing Applications for Change Detection in The Andaman Sea Coastal Region, Sikao District, Trang Province. Wichcha J. Nakhon Si Thammarat Rajabhat Univ. 2018, 37, 84–98. [Google Scholar]
Kim, M.K.; Sohn, H.G.; Kim, S.P.; Jang, H.S. Automatic coastline extraction and change detection monitoring using Landsat imagery. J. Korean Soc. Geospat. Inf. Sci. 2013, 21, 45–53. [Google Scholar]
Ghosh, M.K.; Kumar, L.; Roy, C. Monitoring the coastline change of Hatiya Island in Bangladesh using remote sensing techniques. ISPRS J. Photogramm. Remote Sens. 2015, 101, 137–144. [Google Scholar] [CrossRef]
Maglione, P.; Parente, C.; Vallario, A. Coastline extraction using high resolution WorldView-2 satellite imagery. Eur. J. Remote Sens. 2014, 47, 685–699. [Google Scholar] [CrossRef]
Vitousek, S.; Buscombe, D.; Vos, K.; Barnard, P.L.; Ritchie, A.C.; Warrick, J.A. The future of coastal monitoring through satellite remote sensing. Camb. Prism. Coast. Futures 2023, 1, e10. [Google Scholar] [CrossRef]
Fatah, K.K.; Mustafa, Y.T.; Hassan, I.O. Groundwater potential mapping in arid and semi-arid regions of kurdistan region of Iraq: A geoinformatics-based machine learning approach. Groundw. Sustain. Dev. 2024, 27, 101337. [Google Scholar] [CrossRef]
Rash, A.; Mustafa, Y.; Hamad, R. Quantitative assessment of Land use/land cover changes in a developing region using machine learning algorithms: A case study in the Kurdistan Region, Iraq. Heliyon 2023, 9, e21253. [Google Scholar] [CrossRef]
Sun, W.; Chen, C.; Liu, W.; Yang, G.; Meng, X.; Wang, L.; Ren, K. Coastline extraction using remote sensing: A review. GIScience Remote Sens. 2023, 60, 2243671. [Google Scholar] [CrossRef]
Kannan, R.; Kanungo, A.; Murthy, M. Detection of shoreline changes Visakhapatnam coast, Andhra Pradesh from multi-temporal satellite images. J. Remote Sens. GIS 2016, 5, 157. [Google Scholar]
Choung, Y.-J.; Jo, M.-H. Comparison between a Machine-Learning-Based Method and a Water-Index-Based Method for Shoreline Mapping Using a High-Resolution Satellite Image Acquired in Hwado Island, South Korea. J. Sens. 2017, 2017, 8245204. [Google Scholar] [CrossRef]
Minghelli, A.; Spagnoli, J.; Lei, M.; Chami, M.; Charmasson, S. Shoreline extraction from WorldView2 satellite data in the presence of foam pixels using multispectral classification method. Remote Sens. 2020, 12, 2664. [Google Scholar] [CrossRef]
Bamdadinejad, M.; Ketabdari, M.J.; Chavooshi, S.M.H. Shoreline extraction using image processing of satellite imageries. J. Indian Soc. Remote Sens. 2021, 49, 2365–2375. [Google Scholar] [CrossRef]
Zhou, X.; Wang, J.; Zheng, F.; Wang, H.; Yang, H. An Overview of Coastline Extraction from Remote Sensing Data. Remote Sens. 2023, 15, 4865. [Google Scholar] [CrossRef]
Vos, K.; Splinter, K.D.; Harley, M.D.; Simmons, J.A.; Turner, I.L. CoastSat: A Google Earth Engine-enabled Python toolkit to extract shorelines from publicly available satellite imagery. Environ. Model. Softw. 2019, 122, 104528. [Google Scholar] [CrossRef]
Curoy, J.; Ward, R.D.; Barlow, J.; Moses, C.; Nakhapakorn, K. Coastal dynamism in Southern Thailand: An application of the CoastSat toolkit. PLoS ONE 2022, 17, e0272977. [Google Scholar] [CrossRef] [PubMed]
Hossen, M.F.; Sultana, N. Shoreline change detection using DSAS technique: Case of Saint Martin Island, Bangladesh. Remote Sens. Appl. Soc. Environ. 2023, 30, 100943. [Google Scholar] [CrossRef]
Nassar, K.; Mahmod, W.E.; Fath, H.; Masria, A.; Nadaoka, K.; Negm, A. Shoreline change detection using DSAS technique: Case of North Sinai coast, Egypt. Mar. Georesour. Geotechnol. 2019, 37, 81–95. [Google Scholar] [CrossRef]
Goksel, C.; Senel, G.; Dogru, A.O. Determination of shoreline change along the Black Sea coast of Istanbul using remote sensing and GIS technology. Desalination Water Treat. 2020, 177, 242–247. [Google Scholar] [CrossRef]
Wang, X.; Liu, Y.; Ling, F.; Liu, Y.; Fang, F. Spatio-temporal change detection of Ningbo coastline using Landsat time-series images during 1976–2015. ISPRS Int. J. Geo-Inf. 2017, 6, 68. [Google Scholar] [CrossRef]
Darwish, K.; Smith, S. A comparison of landsat-8 oli, sentinel-2 msi and planetscope satellite imagery for assessing coastline change in el-alamein, egypt. Eng. Proc. 2021, 10, 23. [Google Scholar] [CrossRef]
Rumahorbo, B.T.; Warpur, M.; Hamuna, B. Detection of Shoreline Changes due to Abrasion and Accretion Using Landsat Imagery–A Case Study in the Coastal Areas of Supiori Regency, Indonesia. Ecol. Eng. Environ. Technol. 2023, 24, 161–171. [Google Scholar] [CrossRef]
Chettiyam Thodi, M.F.; Gopinath, G.; Surendran, U.P.; Prem, P.; Al-Ansari, N.; Mattar, M.A. Using RS and GIS Techniques to Assess and Monitor Coastal Changes of Coastal Islands in the Marine Environment of a Humid Tropical Region. Water 2023, 15, 3819. [Google Scholar] [CrossRef]
Quang, D.N.; Ngan, V.H.; Tam, H.S.; Viet, N.T.; Tinh, N.X.; Tanaka, H. Long-term shoreline evolution using dsas technique: A case study of Quang Nam province, Vietnam. J. Mar. Sci. Eng. 2021, 9, 1124. [Google Scholar] [CrossRef]
Oraon Sarajit, K.N. Geo-information Application for Coastal Erosion Situation, Phetchaburi Province. Thai Sci. Technol. J. 2014, 6, 789–800. [Google Scholar]
Wuttipong Sangmanee, C.T. Integration remote sensing data for monitoring shoreline changesin Pattani province. J. Humanit. Soc. Sci. Prince Songkla Univ. 2017, 13, 69–102. [Google Scholar]
Aekkarak Faiboon, W.S. Geospatial Monitoring and Forecasts of Coastal Engineering Structures’ Shoreline Transformational Impact at Songkhla Lake Mouth. Princess Naradhiwas Univ. J. 2019, 11, 165–179. Available online: https://li01.tci-thaijo.org/index.php/pnujr/article/view/160033 (accessed on 14 February 2025).
Khongkhan, A. Incorporating the Use of Time Series Remote Sensing Data to Assess the Vulnerability of the Thai Coast to Climate Change Induced Coastal Hazards. J. Sci. Technol. 2021, 4, 1–18. [Google Scholar]
Pawornprach Imlamai, S.W.; Phaksopa, J. Monitoring Study of Coastal Morphological Change from Coastal Engineering Structures at Mrigadayavan Palace, Phetchaburi Province Using Geo-Informatics. Burapha Sci. J. 2023, 28, 730–751. [Google Scholar]
Buranapratheprat, A. Circulation in the Upper Gulf of Thailand: A Review. Burapha Sci. J. 2024, 13, 75–83. [Google Scholar]
Lorpittayakorn, P. Circulation Patterns in the Gulf of Thailand from Model. Thai Sci. Technol. J. 2015, 23, 446–465. [Google Scholar]
Kim, Y.I.; Park, W.H.; Shin, Y.; Park, J.-W.; Engel, B.; Yun, Y.J.; Jang, W.S. Applications of Machine Learning and Remote Sensing in Soil and Water Conservation. Hydrology 2024, 11, 183. [Google Scholar] [CrossRef]
Agarwal, V.; Akyilmaz, O.; Shum, C.; Feng, W.; Yang, T.-Y.; Forootan, E.; Syed, T.H.; Haritashya, U.K.; Uz, M. Machine learning based downscaling of GRACE-estimated groundwater in Central Valley, California. Sci. Total Environ. 2023, 865, 161138. [Google Scholar] [CrossRef] [PubMed]

Figure 1. Map of the study area in the Upper Gulf of Thailand which is divided into six regions based on physical characteristics.

Figure 2. Workflow of the research methodology used for shoreline change analysis in the Upper Gulf of Thailand.

Figure 3. Compare the performance of classification algorithms including Minimum Distance, Maximum Likelihood Classifier, Support Vector Machine, and Random Forest in overall accuracy and Cohen’s Kappa Coefficient.

Figure 4. Classification results using four ML methods—Random Forest (a), Support Vector Machine (b), Maximum Likelihood Classifier (c), and Minimum Distance (d), for the Upper Gulf of Thailand. Each classification result illustrates the boundary between land and water in sample areas, including beach (aA,bA,cA,dA), mangrove forest (aB,bB,cB,dB), coastal fishing areas (aC,bC,cC,dC), shoreline protection structures (aD,bD,cD,dD), and steep cliffs (aF,bF,cF,dF).

Figure 5. Overall accuracy and Cohen’s Kappa Coefficient for the Random Forest classification method applied to 65 satellite images from 1988 to 2023.

Figure 6. Overlay of the extracted shorelines from seven time periods (1988, 1994, 2000, 2006, 2011, 2018, and 2023) in the Upper Gulf of Thailand. (A) represents shoreline changes at the Klong Yi San Kao estuary, (B) represents shoreline changes at Pak Thalenai, (C) represents shoreline changes at the mangrove area in Bang Krachao, (E) represents shoreline changes at Khun Samut Chin, and (D) represents shoreline changes at Khlong Nang Hong.

Figure 7. Assessment of annual shoreline extraction compared to the reference shorelines in 2018 (a) and 2023 (b) in the Upper Gulf of Thailand. Shoreline locations in 2018: (aA) Hua Hin Beach, (aB) Chaosamran Beach, (aC) Pak Thale Nok, (aD) Bang Khun Thian, (aE) Bang Pu, (aF) Udom Bay, (aG) Na Chom Thian Beach, and (aH) Bang Sare. Shoreline locations in 2023: (bA) Hua Hin Beach, (bB) Chaosamran Beach, (bC) Bang Tabun estuary, (bD) Bang Khun Thian, (bE) Udom Bay, (bF) Jomtien Beach, (bG) Na Chom Thian Beach, and (bH) Bang Sare.

Figure 8. Results of shoreline change analysis using the Digital Shoreline Analysis System (DSAS) for the Upper Gulf of Thailand.

Figure 9. Trends in global mean sea level and average temperature, along with mean sea level, average temperature, and accumulated shoreline erosion in the Upper Gulf of Thailand.

Figure 10. Correlation analyses between sea level, temperature, and coastal erosion: (a) Global mean sea level vs. global average temperature (b). Coastal erosion in the Upper Gulf of Thailand vs. global mean temperature (c). Coastal erosion in the Upper Gulf of Thailand vs. global mean sea level (d). Mean sea level vs. mean temperature in the Upper Gulf of Thailand (e). Coastal erosion vs. mean temperature in the Upper Gulf of Thailand (e), (f) Coastal erosion vs. mean sea level in the Upper Gulf of Thailand.

Figure 11. Shoreline changes over six time periods from Hua Hin District to Laem Phak Bia region. (A) represents shoreline changes in the northern part of Cha-Am Beach, and (B) represents shoreline changes in Bang Kao Beach.

Figure 12. Shoreline change analysis: sample of shoreline changes (a) over six time periods in Saphan Pla Cha-am (c), and sample of shoreline changes (b) over six time periods in the coastal area Bang Kao Subdistrict, Cha-am District, Phetchaburi (d).

Figure 13. Shoreline changes over six time periods from Laem Phak Bia–Mae Klong River. (A) represents shoreline changes at the Klong Yi San Kao estuary, and (B) represents shoreline changes at Pak Thalenai.

Figure 14. Shoreline change analysis: sample of shoreline changes (a) over six time periods in the coastal area between the Mae Klong estuary and the Khlong Bang Tabun estuary (c), and sample of shoreline changes (b) over six time periods in the coastal area Pak Thale Conservation Area, Pak Thale Subdistrict, Ban Laem District, Phetchaburi Province (d).

Figure 15. Shoreline changes over six time periods from Mae Klong River to Tha Chin River. (A) represents shoreline changes at the mangrove area in Bang Krachao, and (B) represents shoreline changes at Ao Mahachai Mangrove Forest Study Centre.

Figure 16. Shoreline change analysis: sample of shoreline changes (a) over six time periods in the coastal area Bang Phraek Subdistrict, Mueang District, Samut Sakhon Province (c), and a sample of shoreline changes (b) over six time periods in the coastal area Ao Mahachai Mangrove Forest Natural Education Center, Bang Phraek Subdistrict, Mueang District, Samut Sakhon Province (d).

Figure 17. The shoreline changes over six time periods from Tha Chin River to Chao Phraya River. (A) represents shoreline changes at Khun Samut Chin, and (B) represents shoreline changes at the Tha Chin estuary.

Figure 18. Shoreline change analysis: sample of shoreline changes (a) over six time periods in the coastal area Ban Khun Samut Chin, Laem Fa Pha Subdistrict, Phra Samut Chedi District, Samut Prakan Province (c), and sample of shoreline changes (b) over six time periods in the coastal area Marine and Coastal Resources Office, Samut Sakhon Mueang District, Samut Sakhon Province (d).

Figure 19. Shoreline changes over six time periods from Chao Phraya River to Bang Pakong River. (A) represents shoreline changes at Khlong Nang Hong, and (B) represents shoreline changes at Bang Pu Mai.

Figure 20. Shoreline change analysis: sample of shoreline changes (a) over six time periods in the coastal area Khlong Dan Subdistrict, Bang Bo District, Samut Prakan Province (c), and sample of shoreline changes (b) over six time periods in the coastal area Bang Pu Subdistrict, Mueang District, Samut Prakan Province (d).

Figure 21. Shoreline changes over six time periods from Bang Pakong River to Sattahip District. (A) represents shoreline changes at the Bang Pakong estuary, and (B) represents shoreline changes at Laem Chabang Port.

Figure 22. Shoreline change analysis: sample of shoreline changes (a) over six time periods in the coastal area Bang Pakong estuary, Khlong Tamhru Subdistrict, Mueang District, Chonburi Province (c), and sample of shoreline changes (b) over six time periods in Laem Chabang coastal area, Thung Sukhla Subdistrict, Sri Racha District, Chonburi Province (d).

Table 1. Summary of datasets used for shoreline analysis in the Upper Gulf of Thailand.

Data Type	Period	Details	Source
Landsat 5, 7, 8, 9	1988–2023	A 30 m spatial resolution collected from Path 129/Row 051 which underwent atmospheric correction, representing the Upper Gulf of Thailand for land–water classification and shoreline extraction.	United States Geological Survey (USGS)
Land Use/Land Cover Types	1988, 1994, 2000, 2006, 2011, 2018, 2023	ESRI shapefiles covering coastal provinces in the Upper Gulf of Thailand, used for analyzing land use and coastal changes.	Land Development Department, Thailand
Shoreline Data	2019, 2022	ESRI shapefiles derived from digitization combined with GNSS surveys, used to validate extracted shorelines.	Department of Marine and Coastal Resources, Thailand
Digital Elevation Model	2000	A 90 m spatial resolution covering the Upper Gulf of Thailand, used to represent onshore topography	United States Geological Survey (USGS)
Annual Global Mean Sea Level Data	1987–2023	Represents global annual sea level averages in centimeters.	The National Aeronautics and Space Administration (NASA), USA
Annual Global Average Temperature Data	1987–2023	Represents global annual temperature averages in degrees Celsius.	The National Aeronautics and Space Administration (NASA), USA
Annual Mean Sea Level (Upper Gulf of Thailand)	1987–2023	Collected from four stations around the Gulf of Thailand (Bang Pakong, Ban Laem, Samut Sakhon, Ao Udom), representing local annual sea level trends in centimeters.	Marine Department, Thailand
Annual Average Temperature (Upper Gulf of Thailand)	1987–2023	Collected from weather stations (Phetchaburi, Bangkok, Samut Prakan, Chonburi) to monitor regional temperature changes.	Meteorological Department, Thailand
Sea Depth Data	-	Sixteen nautical chart sections of the Upper Gulf of Thailand presented in raster format representing underwater topography.	Hydrographic Department, Royal Thai Navy
Geological Maps	-	Scale of 1:250,000 covering four sections of the Upper Gulf of Thailand, used to analyze geological characteristics.	Department of Mineral Resources, Thailand.

Table 2. Number of data samples used for training and testing in the land and water classification process.

Sample Sets	Training	Testing
Land cover types (Land/Water)	8400	3600

Table 3. The average rate of shoreline erosion in each region of the Upper Gulf of Thailand from 1988 to 2023.

Period	The Average of Shoreline Erosion (−) and Accretion Rate (Meters/Year)
Period	Region 1	Region 2	Region 3	Region 4	Region 5	Region 6
1988–1994	2.01	10.09	0.01	−13.55	−6.02	−0.64
1994–2000	−0.40	8.89	−5.22	−12.50	−6.68	−0.48
2000–2006	−0.04	2.94	−1.21	−10.35	−6.77	−0.10
2006–2011	2.81	2.41	−1.80	−4.55	−5.18	0.19
2011–2018	−0.71	−0.18	−1.31	−3.88	0.10	−2.05
2018–2023	−1.55	−0.54	1.31	−0.40	3.30	−0.40

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

© 2025 by the authors. Published by MDPI on behalf of the International Society for Photogrammetry and Remote Sensing. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).

Share and Cite

MDPI and ACS Style

Chawalit, C.; Boonpook, W.; Sitthi, A.; Torsri, K.; Kamthonkiat, D.; Tan, Y.; Suwansaard, A.; Nardkulpat, A. Geoinformatics and Machine Learning for Shoreline Change Monitoring: A 35-Year Analysis of Coastal Erosion in the Upper Gulf of Thailand. ISPRS Int. J. Geo-Inf. 2025, 14, 94. https://doi.org/10.3390/ijgi14020094

AMA Style

Chawalit C, Boonpook W, Sitthi A, Torsri K, Kamthonkiat D, Tan Y, Suwansaard A, Nardkulpat A. Geoinformatics and Machine Learning for Shoreline Change Monitoring: A 35-Year Analysis of Coastal Erosion in the Upper Gulf of Thailand. ISPRS International Journal of Geo-Information. 2025; 14(2):94. https://doi.org/10.3390/ijgi14020094

Chicago/Turabian Style

Chawalit, Chakrit, Wuttichai Boonpook, Asamaporn Sitthi, Kritanai Torsri, Daroonwan Kamthonkiat, Yumin Tan, Apised Suwansaard, and Attawut Nardkulpat. 2025. "Geoinformatics and Machine Learning for Shoreline Change Monitoring: A 35-Year Analysis of Coastal Erosion in the Upper Gulf of Thailand" ISPRS International Journal of Geo-Information 14, no. 2: 94. https://doi.org/10.3390/ijgi14020094

APA Style

Chawalit, C., Boonpook, W., Sitthi, A., Torsri, K., Kamthonkiat, D., Tan, Y., Suwansaard, A., & Nardkulpat, A. (2025). Geoinformatics and Machine Learning for Shoreline Change Monitoring: A 35-Year Analysis of Coastal Erosion in the Upper Gulf of Thailand. ISPRS International Journal of Geo-Information, 14(2), 94. https://doi.org/10.3390/ijgi14020094

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Menu