Coastal Vulnerability Assessment in Svalbard region of the Arctic exploiting geoinformation technologies and a cloud-based platform

Coastal Vulnerability Assessment in Svalbard region of the Arctic exploiting geoinformation technologies and a cloud-based platform Christos Kyrveis, George P. Petropoulos * , Danai Varvatsouli Department of Geography, Harokopio University of Athens, El. Venizelou 70, Kallithea, Athens, 17676, Greece ARTICLE INFO Keywords: climate change Coastal Vulnerability Index (CVI) remote sensing Google Earth Engine DSAS Svalbard Arctic ABSTRACT Monitoring coastal vulnerability is becoming increasingly important, due to climate-driven phenomena, such as sea level rise, glacier melt, and coastal changes, accelerate erosion and land loss, threatening infrastructure and ecosystems. In this study, a comprehensive methodology was developed and applied in part of Svalbard, Norway, using the Coastal Vulnerability Index (CVI). The approach is based on an automated workflow developed in the Google Earth Engine (GEE) platform, enabling automated and accurate extraction of coastlines from Landsat satellite data for the period 2014–2023. Additionally, geospatial analysis tools including ArcGIS and DSAS employed to calculate parameters such as shoreline change rates. The results revealed spatial variations in vulnerability and dynamic changes in the coastline, providing valuable information for coastal zone management in one of the most vulnerable environments on the planet. Specifically, approximately 4 % of the coastline was classified as high vulnerability, 29 % as moderate, 53 % as low and 14 % as very low. The proposed herein approach is expected to serve as an effective tool for researchers and decision-makers, enhancing adaptation planning to the impacts of climate change in Arctic coastal areas. 1. Introduction Arctic coastal zones undergo significant challenges due to climate change with warming at a rate up to three times greater than the global average (IPCC et al., 2019). They are shaped by high-energy dynamics and environmental variability and serve as a reference point for understanding the ongoing climatic changes (Esteves, 2014). Arctic coastlines are formed by geophysical processes such as erosion and flooding, and are increasingly threatened by permafrost thaw, glacier retreat and sea-level rise (McAllister et al., 2022; Hossain et al., 2022a, Mondal et al., 2025a. These processes have significant implications for the physical characteristics, local economy, biodiversity, and ecosystem stability, and therefore require careful monitoring (Nazeer et al., 2020; Hzami et al., 2021; Mondal et al., 2025b). Moreover, 40 % of the world’s population inhabits coastal areas, highlighting the importance of coastal changes (United Nations, 2017). Coastlines are the natural boundary separating land from sea, and therefore monitoring changes in them depends largely on their accurate mapping (Luijendijk et al., 2018; Almeida et al., 2021). Effective monitoring of Arctic coastal dynamics is crucial for understanding climate changes, and developing effective management and mitigation strategies in vulnerable areas as Svalbard (Sun et al., 2023; Petropoulos et al., 2024; Das et al., 2025). * Corresponding author. E-mail address: [email protected] (G.P. Petropoulos). Contents lists available at ScienceDirect Remote Sensing Applications: Society and Environment journal homepage: www.elsevier.com/locate/rsase https://doi.org/10.1016/j.rsase.2025.101752 Received 13 August 2025; Received in revised form 23 September 2025; Accepted 9 October 2025 Remote Sensing Applications: Society and Environment 40 (2025) 101752 Available online 15 October 2025 2352-9385/© 2025 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license ( http://creativecommons.org/licenses/by/4.0/ ). Depending on the coastline scale and accuracy, different techniques are employed to effectively monitor coastal dynamics and human impacts. Smaller areas typically rely on the use of GPS and photogrammetric techniques for detailed topographic mapping (Palomar-V´ azquez et al., 2018; Pardo-Pascual et al., 2018; Zhou et al., 2023), while larger areas require satellite remote sensing data, as they provide high-resolution and cost-effective multispectral images for long term and cost-effective monitoring (Palomar-V´ azquez et al., 2018; Zhou et al., 2023). In this framework, geoinformation technologies, such as remote sensing and Geographic Information Systems (GIS), are essential tools for monitoring and analysing coastal dynamics as they offer cost-effective, scalable and consistent methods for monitoring across large and remote areas such as the Arctic (Luijendijk et al., 2018; Almeida et al., 2021; Thakur et al., 2021; Mondal et al., 2024). Satellite imagery and data from optical sensors such as Landsat and Sentinel-2 offer long-term monitoring of coastline changes, while GIS integrate data and visualise coastal vulnerability (Gorelick et al., 2017; Nicu et al., 2020; Mondal et al., 2021; Petropoulos et al., 2024). The repeatability of satellite orbits allows time-series monitoring the assess storms or erosion effects (Almonacid-Caballer et al., 2016; Mondal et al., 2017; Pardo-Pascual et al., 2018; Melet et al., 2021). In addition, satellite images allow automated coastline detection though computational methods based on water indices, such as the Normalized Difference Water Index (NDWI), the Modified Normalized Difference Water Index (MNDWI), and the Automated Water Extraction Index (AWEI) (Нu and Wang, 2022; McAllister et al., 2022; Sun et al., 2023; Zhou et al., 2023), enhancing detection accuracy, and consistent coastline extraction in remote Arctic areas (Sun et al., 2023; Zhou et al., 2023). The integration of different geoinformation technologies enables a comprehensive approach and the development of mitigation strategies (De Serio et al., 2018; Thakur et al., 2021; Cruz-Ramírez et al., 2024; Petropoulos et al., 2024). Although coastal vulnerability has been analysed by different methods, the most commonly utilised are those which focus on indexes which are used in erosion and flooding assessment (Chakraborty, 2021; Mondal et al., 2025c). One of the most widely applied approaches involves the computation of the Coastal Vulnerability Index (CVI), combining dynamic coastal features such as geomorphology, sea level change, wave height, tidal range, and displacement of the coast (Toumasi et al., 2024). CVI has also been effectively integrated into GIS platforms to support decision-making in coastal zones (Petropoulos et al., 2024; Hossain et al., 2022). Specialised tools such as the Digital Shoreline Analysis System (DSAS) and CoastSat enable automated analysis of shoreline change and CVI parameters. DSAS allow consistent shoreline change analysis, while CoastSat analyses a time series from Landsat and Sentinel satellites with deep learning neural network algorithms (Thieler et al., 2009; Vos et al., 2019; Mondal et al., 2020; Himmelstoss et al., 2021). Despite the increasing interest on coastal vulnerability in the Arctic, previous studies have focused on static or geomorphological data without utilizing automated, time-series, and large-scale analyses (McAllister et al., 2022; Petropoulos et al., 2024). While CVI has been widely used to measure coastal erosion in the Arctic and Svalbard archipelago, it exhibits limitations in multi-temporal remote sensing data integration with geospatial processing (Gorelick et al., 2017; McAllister et al., 2022). Furthermore, there is a gap in the implementation of automated coastline extraction and monitoring with cloud-based platforms, particularly in remote and data-limited areas such as Svalbard (Zhou et al., 2023; Toumasi et al., 2024). In purview of the above, the present study aims to develop and apply a comprehensive methodology for mapping and assessing coastal vulnerability in the Arctic, focusing on the Svalbard region. The methodology integrates the Google Earth Engine (GEE) platform with multi-year Landsat 8 OLI satellite data (30m resolution), Landsat 8 satellite data (15m resolution) and Sentinel-2 composite images (10m resolution), for automated high accuracy coastline extraction for the period 2014–2023. A custom code was developed with the GEE environment to process Landsat 8 OLI panchromatic channel and Sentinel-2 composites to ensure high accuracy and interpretation. This code was originally based on the application “Automated Coastline Extraction from Satellite Images using GEE” by Ujaval Gandhi, but was subsequently modified to meet the specific needs of this study. Additionally, ArcGIS and DSAS tools were employed to efficiently calculate CVI parameters as a result of climate-related environmental dynamics over time. This integrated approach is expected to serve as a reliable decision-making tool that enables both sea-level rise research and supports policymakers to design and implement prevention policies in coastal areas at increased risk (Gorelick et al., 2017; Toumasi et al., 2024; Petropoulos et al., 2024; Mondal et al., 2025d). Consequently, the study is expected to contribute significantly to the sustainable management of coastal areas and to the development of mitigation strategies for prospective climatic changes in the Artic regions and globally. 2. Experimental setup 2.1. Study area Longyearbyen in Svalbard archipelago was chosen as the study area, which was selected based on scientific and environmental criteria, as it is a part of the Arctic that has been characterised by intense climate variability and rapid temperature rise in recent decades (Serreze and Meier, 2019). Processes such as the retreat of permafrost and the acceleration of coastal erosion are being recorded (Serreze and Meier, 2019). It is one of the northernmost inhabited areas on the planet, with scientific significance for studying the effects of climate change on the Arctic coastal zone (Nicu et al., 2020; Esau and Miles, 2024). It is therefore an ideal choice for testing and evaluating the CVI methodology and improving the application of Google Earth Engine in dynamic and vulnerable environments (Toumasi et al., 2024; Petropoulos et al., 2024). Geographically, it is in Vest Spitsbergen, the largest island of Svalbard in the Arctic Europe, with 2.556 inhabitants (Hanssen-Bauer et al., 2019; SSB, n.d.). The area has significant urban infrastructure and is serve as reference point for monitoring environmental and climate change in the Arctic (Martens and Krangnes, 2022) (Fig. 1). Geomorphologically, it is characterized by a mountainous relief C. Kyrveis et al. Remote Sensing Applications: Society and Environment 40 (2025) 101752 2 with flat plateaus up to 500 m, with the settlement located in the lower part of the Longyearelva valley near the mouth of the river of the same name (Jask´ olski et al., 2018). Svalbard is covered mostly by glaciers (60 %) with the remaining landscapes consisting of barren rocky (30 %) and sparsely vegetated areas (10 %) (Gu´ egan and Christiansen, 2017; Martens and Krangnes, 2022). The geomorphology of Spitsbergen has been shaped by successive glacial periods, which formed fjords, valleys, and steep slopes, while frequent avalanches have contributed to the reshaping of the local morphology. In Adventdalen, glaciers cover 18 % of the surface, while the rest of the soil consists mainly of river sediments, weathered materials, and deposits associated with permafrost (Gu´ egan and Christiansen, 2017). The entrance to the fjord is dominated by low rocky cliffs and pebble beaches, while important infrastructure such as the airport, ports, road network, and anti-erosion structures, are located in close proximity to this zone, highlighting the need for systematic monitoring of coastal change (Jask´ olski et al., 2018). Climatically, the region is categorized as polar tundra under the K¨ oppen system, although the West Spitsbergen Current (WSC) brings some warming (Hanssen-Bauer et al., 2019). The region has been warming by 3–5 ◦C since the 1920s (Nicu et al., 2020; Lousada et al., 2018). Longyearbyen’s position near the mouth of Adventfjorden makes it an important site for studying the impacts of climate change on coastal erosion, while also highlighting wave and tide driven processes on low cliffs within tidal zones (Jask´ olski et al., 2018). 2.2. Datasets The datasets for this study include a combination of spatial and thematic data sources, which were selected to represent the physical parameters contributing to coastal vulnerability in Longyearbyen area, covering the period 2014–2023. Spatial data include the Landsat 8 OLU images, which were used for the extraction of the coastline, the 20m resolution DEM for the coastal slope measurement, and the geomorphology maps from the Norwegian Polar Institute. Additionally, thematic data included the environmental variables used for the computation of CVI. The integration and detailed processing of these datasets provide the basis for calculating the CVI. More details about each dataset and its role in the analysis is provided below (Table 1). Fig. 1. Archipelago Svalbard (up left), Svalbard location on Europe (up right), Topographic map of Longyearbyen study area, https://toposvalbard. npolar.no/(down). C. Kyrveis et al. Remote Sensing Applications: Society and Environment 40 (2025) 101752 3 2.2.1. Coastal Vulnerability Index (CVI): datasets Coastal vulnerability in the Longyearbyen area was based on the analysis of six critical parameters, which were selected to consistently reflect the complexity and vulnerability of the Arctic coastal environment. The selection of variables was based on international literature (Weslawski, 2011; Jask´ olski et al., 2018) and the specific characteristics of the study area. The six key variables used to calculate the CVI are: geomorphology, coastal slope, relative sea level change, coastline displacement rate, mean tidal range, and mean significant wave height. Geomorphology variable, records the type of coastal formations and their resistance to erosion (Karymbalis, 2010). The data were mainly obtained from the Norwegian Polar Institute, with 1:250.000 cartographic spatial scale, which provides detailed geospatial datasets (https://geodata.npolar.no/#viewers). In addition, for areas with incomplete coverage (e.g., ports), photo interpretation methods were used with high-resolution images from Google Earth (spatial resolution ~0.5m). The types of landforms mapped include marine and glacial deposits, rocky coasts, and man-made structures. Coastal slope was assessed as a critical factor in flood exposure and erosion. A 20m resolution Digital Elevation Model (DEM) from the Norwegian Polar Institute (https://data.npolar.no/dataset/dce53a47-c7264845-85c3-a65b46fe2fea) was used. The slope was calculated in ArcGIS, from the baseline to the 4-m contour, providing a quantitative assessment of vulnerability (Table 1). The coastal slope describes the incline of the coastal area contributing to vulnerability in CVI assessments. Relative Sea Level Change was estimated with long-term sea level trends observations from the Barentsburg tide gauge (PSMSL, 1949–2018) available from the Permanent Service for Mean Sea Level (https://psmsl.org). Supplementary data were obtained from the MOSJ program (https://mosj.no), providing updated measurements for Svalbard. The area is characterized by isostatic uplift, with an average trend of −2.77 mm/year (Table 1). This variable, supports critical coastal flooding risk evaluation, derived from estimation of tide gauge records in the Arctic. For the Coastline Displacement Rate, the Linear Regression Rate (LRR) method for the period 2014–2023 was utilised for monitoring coastline changing positions over time. The coastline data were derived from composite Landsat 8 satellite images (30 m resolution) using Google Earth Engine (https://earthengine.google.com) and GIS tools (ArcGIS, QGIS, DSAS v6) (Table 1). The Mean Tidal Range was calculated from measurements by the Norwegian Mapping Authority at the Longyearbyen tide gauge for the period 1996–2014 (https://www.kartverket.no) (Table 1). Also, it was obtained from global tide models (e.g., FES2014), impacting flood risk and sediment transport processes. While the mean significant Wave Height (Hs) was estimated from observations by the Longyearbyen Port Office for the period 2014–2023 (Table 1), indicating potential wave energy affecting the coastline. 2.2.2. Satellite images The coastline extraction in this study was based on the Landsat 8 Operational Land Imager (OLI) mission for the period 2014–2023 and its Level 2 Surface Reflectance products. The period of 2014–2023 (May–September) was selected to mitigate the cloud cover effects, cloud shadows, snow and ice, improving the accuracy of coastline detection. Collection 2 with its radiometric and geometric correction, atmospheric correction, and WGS84 UTM projection using two systems were applied. In addition, the panchromatic channel (Band 8) of 15 m spatial resolution was used for pan-sharpening to enhance the resolution of the land-water composite images. The annual composite images were created in Google Earth Engine (GEE) using Otsu thresholding and water indices NDWI, MNDWI, AWEIsh, AWEInsh to mitigate cloud, ice, and seasonal influences on the precision of coastline detection. The final coastline detection products were exported as shapefile (.shp) and GeoTIFF for further analysis and visualisation. 3. Methodology This section describes in detail the overall methodology followed to assess coastal vulnerability (CVI) in a selected area of Svalbard. The approach consists of two main stages: data pre-processing, analysis and calculation of the CVI index, through automated coastline extraction, and analysis of the physical parameters that influence its change for the period 2014–2023. The methodology integrates remote sensing with satellite imagery and GIS tools for spatial data analysis using GEE, ArcGIS, and DSAS to assess risk areas (Gorelick et al., 2017; Himmelstoss et al., 2021). Fig. 2 includes a workflow illustration adopted herein to address the study objectives. Table 1 Detailed dataset along with their data sources. Data Type Spatial Resolution Data Source Year/Period Geomorphology 1:250.000 Norwegian Polar Institute 2024 Coastal Slope 20m (DEM) Norwegian Polar Institute 2024 Relative Sea-Level Change – PSMSL, MOSJ 1949–2018 (for PSMSL, MOSJ Records) Shoreline Change Rate 30m (Landsat) Landsat 8, USGS via GEE 2014–2023 Mean Tidal Range –Norwegian Mapping Authority 1996–2014 Significant Wave Height –Longyearbyen Port Office 2014–2023 Satellite Imagery 30m Landsat 8 Operational Land Imager (OLI) 2014–2023 (May–September) C. Kyrveis et al. Remote Sensing Applications: Society and Environment 40 (2025) 101752 4 Fig. 2. Workflow of the methodology adopted herein to satisfy the study objectives. C. Kyrveis et al. Remote Sensing Applications: Society and Environment 40 (2025) 101752 5 3.1. Pre-processing Acquiring the necessary geospatial data, delineating the coastline alongside computing the CVI parameters was the focus of this stage. Relevant datasets, such as satellite remote sensing data and elevation models, were complemented with bibliographic sources for deriving the variables of lithology, slope, and coastal dynamics. Specifically, the study utilised tidal gauge measurements and historical shoreline records from the Norwegian Polar Institute in addition to satellite imagery and DEM data for CVI computation. Initially, Landsat 8 OLI images were pre-processed from 2014 to 2023. Collection 2 (Tier 1) Surface Reflectance data, were utilised, which are geometrically and radiometrically corrected and georeferenced to the WGS 84 UTM Zone 33N system (Gorelick et al., 2017). To limit the effects of cloud cover and seasonal phenomena, the period May–September of each year was selected. A cloud mask was applied using Quality Assessment (QA) bands, and composite images were created using the median algorithm. The median method was chosen over alternatives (mean, percentile) due to its resistance to noise and its ability to limit extreme values (White et al., 2014; Almonacid-Caballer et al., 2016; Luijendijk et al., 2018). To enhance the spatial resolution of the multispectral images, a pan-sharpening technique was applied. This involves merging the higher-resolution panchromatic Band 8 (15 m) with the lower-resolution multispectral bands (30 m) using HSV (hue–saturation–value) transformation. Pan-sharpening improves the sharpness and detail of land–sea boundaries, which allows for more accurate coastline delineation and facilitates precise manual verification and correction (Tu et al., 2001; Zhou et al., 2023). This approach allows a robust extraction of coastlines seamlessly, ensuring the accuracy of subsequent CVI parameters, particularly in Arctic regions with complex seasonal and environmental conditions. Coastlines were then automatically extracted using the GEE platform. Annual composites were created on the GEE platform using the median algorithm (Wulder et al., 2019). For each annual composite, four water indices (NDWI, MNDWI, AWEIsh, AWEInsh) were calculated to improve coastline detection. First, the Normalized Difference Water Index (NDWI) proposed by McFeeters (1996) was calculated, which is used for mapping water in inland and coastal environments, ranging from −1 (dry) to 1 (water). It is calculated from the difference in reflectance in the green and NIR spectrum, exploiting the high reflectance of water in the green and low reflectance in the NIR, based on the following Equation (1): NDWI =Green +NIR Green −NIR (1) Then Modified NDWI (MNDWI) proposed by Xu (2006) was calculated, which uses the NIR band instead of SNIR, providing greater accuracy in cloudy or wavy waters. Although it often outperforms NDWI, it can incorrectly classify dry tidal areas as water. It was calculated based on the following Equation (2): MNDWI =(Green −SNIR 1) (Green +SNIR 1)(2) The Automatic Water Extraction Index (AWEI) proposed by Feyisa et al. (2014) improves water classification in areas with shadows or dark surfaces, maximizing spectral contrast with land. It is divided into two variants: AWEIsh for shaded/mountainous areas and AWEInsh for flat or urban areas. They were calculated based on the following Equations (3) and (4): AWEIsh=4× (Green −SWIR1) − (0.25 ×NIR +SWIR2)(3) AWEIsh=4× (Green −SWIR1) − (0.25 ×NIR +2.75 ×SWIR2)(4) Subsequently, Otsu thresholding and Canny edge detection filter methods were applied to separate land from water. The use of multiple water indices simultaneously and classification methods allow the methodology to effectively address challenges caused by seasonal presence of snow/ice and topographical diversity, and consequently enhancing the reliability of automated shoreline extraction (Hu and Wang, 2022). Specifically, the integration of NDWI, MNDWI and the two AWEI indices focused enhancement of land–water discrimination and reduced misclassification. According to literature, including Otsu thresholding and Canny edge detection, the multi-index approach has been reported effective for coastal extraction in challenging environmental conditions (Hu and Wang, 2022). Coastal slope data has been derived from the 20 m resolution Terrengmodell – Svalbard DEM (https://data.npolar.no/ dataset/dce53a47-c7264845-85c3-a65b46fe2fea) published by the Norwegian Polar Institute. Slope values calculated from the baseline to 4 m contour in ArcGIS and then reclassified as CVI vulnerability classes based on the thresholds of Hamid et al. (2019) and Jask´ olski et al. (2018) & Toumasi et al. (2024). This DEM based slope layer was essential to CVI for incorporating elevation information. Internal water bodies were removed and converted from raster to vector to produce a coastline as a polygon. For a more realistic representation of the coastline, geometry smoothing was performed with a tolerance parameter ranging from 35 to 40. The improved version of the code in GEE also allowed the application of the pan-sharpening technique (HSV method) using the panchromatic channel (Band 8, 15m resolution) to enhance spatial resolution (Amro et al., 2011). The developed code can be accessed here: https://code.earthengine.google.com/d6c7e4e4c64aa69b0f5e3ee217a66a4e. The final extraction of coastlines for the period 2014–2023 was performed in shapefile (.shp) format, while the pan-sharpened composite images were stored in GeoTIFF for potiential visual verification. Despite the automated nature of the export in GEE, manual improvement and correction of coastlines is also necessary, in cases where residual cloud cover, and ice/snow can lead to segmentation inaccuracies. Thus, visual correction based on high resolution imagery, ensures accurate land-water boundaries as recommended in similar studies (Pardo-Pascual et al., 2018; Almonacid-Caballer et al., 2016). Specifically, this correction is performed in ArcGIS to improve accuracy at the local level. The presence of ice/snow, complex topography, and seasonal variations can make automatic methods susceptible to false positives. The final processing and C. Kyrveis et al. Remote Sensing Applications: Society and Environment 40 (2025) 101752 6 improvement of coastlines was performed in ArcGIS using line and vertex processing techniques. Line simplification was performed with custom tolerance, removal or correction of false branches (spikes, artifacts), as well as alignment and joining of segments resulting from mosaic errors. Semi-automatic improvement was deemed necessary so that the final set of coastlines would be geometrically consistent and highly accurate in terms of mapping (Almeida et al., 2021). Manual correction and line simplification were employed to ensure geometric consistency and eliminate errors caused by residual clouds and snow/ice, critical for an accurate CVI computation. Finally, coastline change analysis was performed using DSAS. The coastlines retrieved from the GEE platform later used in ArcGIS with the DSAS v6 tool (Himmelstoss et al., 2021). DSAS allows the calculation of linear shoreline change over time, producing values such as: Shoreline Change Envelope (SCE), Net Shoreline Movement (NSM), End Point Rate (EPR) and Linear Regression Rate (LRR). For this analysis, a baseline was established parallel to the coastline from which perpendicular lines (transects) are drawn every 100m. Along these lines, the coastline change values for the 2014–2023 timeframe were computed, which enabled the assessment of temporal trends (Himmelstoss et al., 2021). Spatial analysis focuses on determination of the most prominent areas of erosion or accretion, which would be valuable for integrated coastal zone management in the future (McAllister et al., 2022). 3.2. CVI computation On the main processing, CVI was computed for sections of the coastline. CVI is a composite index used to assess vulnerability to climate change which includes physical and geomorphological features of the coastline concerning climate change impacts and sea level rise (Gornitz, 1990; Thieler and Hammar-Klose, 1999; Thieler et al., 2009). For each coastline segment, six physical parameters were calculated: geomorphology, coastal slope, relative sea level change (RSLC), shoreline change rate, mean tidal range (MTR), and mean wave height (MWH). To calculate the coastal slope, a 20m resolution DEM ("Terrengmodell – Svalbard") was used, with the following Equation (5): Coastal Slope (%) = (Elevation Dis tan ce)x100 (5) The slopes were categorized so that small slopes were characterized as having higher vulnerability due to increased erosion potential. Each parameter was then classified on a scale of 1–5 (very low to very high vulnerability) according to thresholds derived from literature and geospatial data (Hamid et al., 2019; Toumasi et al., 2024). The thresholds for each CVI parameter were selected based on established methodologies and adapted to the Arctic context. This normalization ensures comparability between parameters with different units and variations, while also aligning the analysis with previous CVI assessments. The classification table used here represents a modified version of the approach by Thieler and Hammar-Klose (1999, 2001), with specific adjustments to the coastal slope parameter to better capture the conditions of the study area. This approach has also been adopted in related vulnerability assessments, such as Jask´ olski et al. (2017) and Toumasi et al. (2024), which focus on regions with similar geological, geomorphological, and climatic settings. Table 2 shows the calibration of CVI variables on a five-point scale. Weighting factors were then assigned to each parameter. Equal weighting was applied to each normalized parameter, i.e., all factors were considered as equally important for the final calculation, as this allows for a homogeneous comparison between regions (Palomar-V´ azquez et al., 2018). The final CVI index for each coastline segment was computed by the geometric mean of all parameters as in the following equation (6): CVI = (a×b×c×d×e×f) 6 √(6) Where: a =geomorphology, b =coastal slope, c =relative sea level change (RSLC), d =shoreline change rate, e =mean tidal range (MTR), f =mean significant wave height (MWH). The geometric mean mitigates the impact of extreme values and allows for homogeneous integration of the parameters. Using ArcGIS Desktop, the computed CVI was represented in the form of a raster map. A cell size of 100 m was selected, to achieve a suitable resolution for spatial analysis in the study area. The CVI values were further classified within the ArcGIS environment and Table 2 Vulnerability classification for CVI parameters (as per Jask´ olski et al., 2017; Toumasi et al., 2024). Vulnerability Parameter Very Low (1) Low (2) Moderate (3) High (4) Very High (5) Geomorphology Rocky, cliffed coasts, fjords, fiards Medium cliffs, indented coasts Low cliffs, glacial drift, alluvial plains Cobble beaches, estuary Barrier beaches, sand beaches, deltas, sand spits Coastal slope (◦)>10.0 6.0–10.0 3.1–6.0 1.0–3.0 <1.0 Relative sea-level change (mm/yr) <1.8 1.8–2.5 2.5–2.95 2.95–3.16 >3.16 Shoreline erosion/ accretion (m/yr) >2.0 Accretion 1.0–2.0 Accretion (–1.0) – (+1.0) Stable (–1.1) – (−2.0) Erosion <−2.0 Erosion Mean tide range (m) >6.0 4.1–6.0 2.0–4.0 1.0–1.9 <1.0 Mean wave height (m) <0.55 0.55–0.85 0.85–1.05 1.05–1.25 >1.25 C. Kyrveis et al. Remote Sensing Applications: Society and Environment 40 (2025) 101752 7 divided into four categorical ranges (very low to high) using the Natural Breaks method for its accurate classification. Four categories were chosen for comparability with Jask´ olski et al. (2018) study while ensuring balanced distribution. The most vulnerable sections of the coastline were identified using the CVI ranking map generated, supporting better protective and management strategies (Petropoulos et al., 2024). Thematic mapping in ArcGIS, combined with DSAS, facilitated the computation of CVI segments along the coastline. It also enabled the temporal shift (LRR) and the stratification of CVI values into high, medium, and low vulnerability zones to support coastal risk management. This study achieved significant advancements in automated shoreline extraction in GEE by employing diverse water indices such as NDWI, MNDWI, and AWEI to enhance accuracy in Arctic settings (Hu and Wang, 2022) and utilizing pan-sharpened Landsat data for improved spatial resolution (Amro et al., 2011). These processes were implemented through a consistent batch processing method for the years 2014–2023. At the same time, the use of DSAS for multi-year shoreline change analysis and the use of slope/relief data in a GIS environment established a coherent framework for CVI assessment in areas with severe climate change impacts (Sun et al., 2023). This integrated approach utilizes the rapid automation of GEE, the accuracy of ArcGIS, and the quantitative assessment of DSAS, providing customized tools for monitoring vulnerable coastal areas even in environments with data limitations (Petropoulos et al., 2024). 4. Results & discussion The results of the study are presented in Figs. 3–7, illustrating the coastline extraction, retreat and accretion patterns, and the classification of coastal vulnerability (CVI) across the Svalbard region. At first, the coastlines were extracted for the period 2014–2023, using Landsat 8 OLI images within the GEE platform. A total of 10 annual coastlines (one per year) was extracted for a consistent temporal analysis, using automated water indices (MNDWI and AWEIsh). The statistical analysis showed changes on in their linear arrangement, with zones of retreat and advance in different parts of the coastline, highlighting the heterogeneity of the area. Parts of coastline characterised by rocky and steep slopes remained mostly stable, while parts with low slopes and sedimentary formations (e.g., alluvial fans, sand deltas), align with the findings of previous studies. Specifically, Almeida et al. (2021) observed erosion hotspots on gentle, low-angle slopes of sedimentary bedrock, while the adjacent rocky coasts appeared to be stable. Similarly, Palomar-V´ azquez et al. (2018) reported that coastal slope and sediment composition significantly influence the rate of shoreline retreat, further supporting the spatial structures we observed in the study. Additionally, the geospatial analysis with a 100m grid resolution, demonstrated parts of the coastline with stability and parts of retreat, highlighting the importance of targeted management (Sun et al., 2023; Zhou et al., 2023). Those variations highlight the influence of coastal slope, wave exposure and the base type, offering a scientific framework for future coastal vulnerability assessment (Sun et al., 2023). Also, to ensure a reliable interpretation of the coastline changes, a differential analysis between AWEIsh and MNDWI indices for 2014, 2018, and 2023 was implemented to further verify the precision of the automated coastline extraction (Figs. 3–5). The differential analysis indicated mean variations below 11m, with AWEIsh exhibiting lower variations than MNDWI. Given the absence of very high-resolution imagery for 2014, 2018, and 2023, a threshold of 11 m was adopted as the maximum acceptable error, following Apostolopoulos and Nikolakopoulos (2020), who identified a comparable maximum deviation between coastlines extracted from Landsat and WorldView-2 data using the SCE method. This threshold was therefore applied as a benchmark for evaluating the reliability of the automatically extracted coastlines from medium-resolution satellite imagery. This demonstrates enhanced effectiveness of AWEIsh in high latitudes and water conditions, such as those occurring in Svalbard (Sun et al., 2023; Zhou et al., 2023). These findings align with current scientific practicing, that incorporate satellite-based methods with coastal monitoring practices in limited-data area, such as the Arctic (Hu and Wang, 2022; Sun et al., 2023). Subsequently, the statistical analysis was performed employing the DSAS tool to measure LRR indices allowing the quantitative analysis of coastline changes over time (Vos et al., 2019; Petropoulos et al., 2024). There were measures over a 100m transects spacing Fig. 3. Accuracy verification of MNDWI and AWEIsh coastlines for the year 2014. C. Kyrveis et al. Remote Sensing Applications: Society and Environment 40 (2025) 101752 8 revealing a spatial variation along Svalbard coastline for the period 2014–2023. LRR values revealed significant erosion in the central and southeastern part of the coastline. Specifically, it demonstrated long-term retreat rates up to up to −4.05 m/year (LRR <−1.5 m/year), in parts characterized by sand deposits with low coastal slopes. On the contrary, parts characterized by rocky and slightly lower slopes demonstrated almost zero or positive rated, indicating stability or accretion. The thematic maps of LRR disclosed spatial patters, erosion hotspots on the east and southeast of the study area, linking those spatial variations with geomorphology, sediment composition and wave exposure (Petropoulos et al., 2024). Also, studies in tropical Fig. 4. Accuracy verification of MNDWI and AWEIsh coastlines for the year 2018. Fig. 5. Accuracy verification of MNDWI and AWEIsh coastlines for the year 2023. Fig. 6. Left: Pie CVI - Percentage ratio (%) of each vulnerability category along the total length of the coastline on a four-point scale (very low to high), Right: CVI bar chart - Depiction of vulnerability distribution in terms of coastline length (in meters) on a four-point scale (very low to high). C. Kyrveis et al. Remote Sensing Applications: Society and Environment 40 (2025) 101752 9