Comparative Analysis of Multi-Spectral Shoreline Delineation Using Landsat-8, Sentinel-2, and PlanetScope Imageries in Coastal Environments of Nigeria

Nigeria coastline extends approximately 800km alone the Gulf of Guinea. The shoreline is assumed to be one of the curved river deltas global with about 21 major river inlet and intersect the coast which breaks it up into series barrier islands. Identification of shoreline position is imperative to understand the coastal area dynamics and a vital parameter for coastal vulnerability assessment (Sexton W.J. and Murday, 1994). Change in the shoreline position is in response to climate change-induced sea level rise and morphological changes due to the response of coastal processes (Pardo-Pascual et al., 2012). Shoreline position has been posing a significant hazard to socioeconomics of coastal communities (Lane et al., 2013), coastal land cover (Hadley, 2009; Lo and Gunasiri, 2014), coastal ecosystems, and cultural heritage site (Mazurczyk et al., 2018).  In order to determine change in shoreline position, it is important to understand the underpinning indicators for definition shoreline. Because the nature of this position and definition chosen must be considered both spatial and temporal knowledge and the dependence of this position at a given time. (Boak and Turner, 2005) outline several indicators of ascertaining shoreline positions in terms of the vertical/horizontal sense to the physical water boundary. The challenge is the ability to develop a robust and repeatable technique to extract shoreline position is needed. Shoreline extraction technique in the low-lying region dependences on the data source and shoreline indictors used. The physical indictors for extraction of shoreline position considering the dynamic nature of coastline comprises (i) instant tidal level such as high water line (Pajak and Leatherman, 2002) Water line (Valderrama-Landeros et al., 2019), high tide wrack line (Thieler et al., 2013), Mean high water line (Chang et al., 2005)  (ii) wetting limit (Zarillo and Synder, 2001), (iii) tidal datums such as Mean sea level (Aagaard et al., 2004), Mean water level (Morton & Miller, 2005); (Miller and Dean, 2007), Mean higher high water line (Allan et al., 2015), Mean low water line (Reeve and Fleming, 1997) and Lowest astronomical images (iv) vegetation limits such as Permanent vegetation line (Priest, 1999), Seaware edge of dune vegetation (Pajak and Leatherman, 2002), Upper limit of algae and marine lichen (Morton and Paine, 1985) (v) beach contour, (vi) storm lines such as Storm-surge penetration boundary and crest of geowash-over terrace and (vii) geomorphological reference lines (Boak and Turner, 2005); (Toure et al., 2019). While Waug (1995) described other indicators such as the movement of earth crust such as tectonic, geology, weathering, deposition and biotic factors. However, the optimal utilisation of the indicator depends on the material quality, operator experience and working conditions. Nevertheless, changes in shoreline position mapping have been determined through a number of data acquisition methods which includes Argus video camera, traditional field survey, aerial photography, and remote sensing. Remote sensing techniques are deemed to be particularly useful given data availability, geographic coverage, low cost, appropriateness, technological advancement in the satellite imagery and wide range of image-processing technology (Duarte et al., 2018). Although aerial photography and some satellite imageries provide high-resolution images for analysing shoreline position, this only capture the instant position at a given time. There are several factors that can potentially influence shoreline position, this are divided into primary and secondary factors. The primary factors includes; tidal stage, beach slope and groundwater position. Meanwhile, at short temporal scale, runup maxima/minima have the abilty to influence the shoreline position of instantenaous water level by tens of meters in low-lying beach area. The secondary factors are sediment grain size, mineralogy, solar zenith angle and geometery of the sensor. The positional accuracy of delineating the shoreline position is a difficult task, especially in the low-lying delta regions. Several research have been adopted different technique to assess the accuracy of change in shoreline position. Although most research have adopted confusion matrix method for the assessment of shoreline position. For instant, (Kelly and Gontz, 2018) adopt compared two GPS-surveyed intertidal zone with seven generic water indices to evaluate the degree of shoreline extraction performance. In their research High water or wet/dry sediment was chosen as an indicator for extraction of shoreline position using Landsat-8 OLI image.   (Acharya et al., 2018), employ confusion matrix to evaluate the performance of four water indices (i.e. NDVI, NDWI, MNDWI and AWEI) for extraction of shoreline position using Landsat-8 OLI image in Nepal. Elsahabi et al. (2016), exploit similar method to detect better technique for the extraction of surface water using eight technique and Landsat ETM⁺ in the Aswan Dam Lake. Other assessment adopted includes,  (Liu et al., 2017), exploits downscaling the pansharpening of Landsat-8 OLI images improve the accuracy. (Sekovski et al., 2014), employed zone-based technique to statistically compare reference shoreline and three supervised and one unsupervised classification technique using high-resolution multispectral Worldview-2 imagery. (Zhang et al., 2016) used Modified Histogram Bimodal to assess the accuracy of automated dynamic threshold to compare different water indexes (NDWI, MNDWI and AWEI) using Landsat-8 OLI. (Liu & Jezek, 2004), integrated Levenberg and Marquardt and Canny edge detector to speed up the convergence of iterative Gaussian curve fitting process and improve the accuracy of the bimodal Gaussian parameter for extraction of shoreline position using SAR and Landsat image. Rishikeshan and Ramesh (2017) exploit mathematics morphology-based algorithm for extraction shoreline position. The accuracy of the analysis was evaluated using computed-based technique as such F-scoring equation, accuracy equation and Matthew’s Correlation coefficient equation. (Pardo-Pascual et al., 2018) evaluated the accuracy of shoreline position using multisource infrared (IR) data (Landsat-7 ETM+, Landsat-8 OLI and Sentinel-2 MSI) and Polynomial Radiometric Correction (PCR) technique on a natural beach. In their study, a comparative analysis of shoreline position using multisource data of same date, indicates that Landsat have a bias of 2.17±3.38m. It was observation that surface brightness variation is the factor affecting shoreline position. Although, the data have similarity in environmental disturbance factor, Land zone brightness and wavelength of the incident wave. In recent years, there has been a growing number of open-access optical satellite imageries. Historically, Landsat missions are the most used optical satellite images for delineation of shoreline position due to large data coverage, historic record, and methodology. Other open-access multispectral imageries are the Sentinel-2 MSI, which was launched in June 2015 by the European Space Agency, while the PlanetScope is a multiple launch of a group of individual multispectral satellites (DOVEs), launched on 22 June 2016. In order to determine the potential of combining this data for coastal management and assessment, shoreline positions from multi-sensor imagery acquired at different spatial resolutions on the same day with slightly higher tides are required to be compared to ascertain the positional accuracy. Therefore, this study will compare the shoreline positions derived using threshold and unsupervised classification methods from two indices (NDVI and NDWI).

MATERIALS AND METHODOLOGY

Study Area

The study area is located between Andoni and Imo River, and it is a typical coastal state in the southern part of the Eastern Niger Delta region of Nigeria (figure 1). In this study, a portion of the state was selected which covers approximately 27.90 kilometres of the coastline. The seasonal climate in the region is tropical, dry and wet. The coastal environment is characterised by wetlands, low-lying tidal flats, beaches, estuaries and sandy barrier islands with an elevation of 0-45metres (Dike et al., 2024; Ochege et al., 2017; Sexton and Murday, 1994). Also, the mean tidal range increases from 1.9 to 3.0 metres from Bonny River and Ibo River, respectively (Abam, 2016). The average temperature in the region is 27°C, the annual average rainfall is 1626 mm/yr. The annual temperature of the area varies from 20 ?C to 35 ?C with high cloud cover and relative humidity ranging from 70 to 80%

Figure 1: Location of the Eastern Niger Delta Region, NigeriaData collection

In order to evaluate the shoreline position multispectral datasets were obtained from Landsat – 8 OLI, Sentinel – 2 and Planetscope imageries.

Satellite Imageries Data

The study utilised on multispectral satellite dataset for the analysis of shoreline position and change. Details different sources are represented Tables 1 and 2 respectively.

Table 1: Multispectral satellite data obtained from different sources

The methodological flowchart for delineation and positional accuracy of shoreline position from multisource satellite image is illustrated in Figure 2.

Figure 2: Flowchart of Shoreline Position and Landcover Map

Table 2: Automated water indices for shoreline position and river feature delineation

Water Indices	Formula
NDVI	ρNIR-ρREDρNIR+ρRED
NDWI	ρGREEN-ρNIRρGREEN+ρNIR

(Ρ = surface reflectance)

Rouse et al., (1974) developed the Normalised Difference Vegetation Index (NDVI). This technique is based on the division of the difference of near-infrared (NIR) and Red (R) reflectance bands and the sum of the same reflectance bands (Table 2). NIR is a suitable spectral band for water delineation and the band values can be sliced to differentiate between vegetated area, bare ground and water bodies. McFeeters, (1996) developed the Normalised Difference Water Index (NDWI) technique which is a non-linear conversion of the green and NIR spectral bands of multispectral satellite imageries. This technique has been widely used for the extraction of shoreline position and river bodies because the green band reflects radiation while the NIR band is absorbed in the water body and used for detection of water turbidity (Xu, 2006). Also, there is higher reflectance in the NIR spectral band than the green spectral band in terrestrial vegetation and areas of bare ground area (Liu et al., 2017).

Threshold delineation

This study used a widely used threshold-based method on indices (NDVI and NDWI) to delineate the shoreline position. First, NDVI and NDWI indices were plotted using a histogram. In the context of the Normalised Difference Vegetation Index (NDVI), water pixels generally display negative values, which can be attributed to the low reflectance of vegetation. Conversely, the Normalised Difference Water Index (NDWI) utilises the significant absorption of water in the near-infrared (NIR) spectral bands  (McFeeters, 1996). The study empirically determines the index thresholds by analysing histograms and validating them against georeferenced Google Earth images. For NDVI the threshold for L8 (Landsat-8 OLI) =0.01, S2 (Sentienl-2 MSI) =0.01 and PS (Planetscope) = -0.03, while NDWI the threshold for L8=0.0, S2=0.0, and PS=0.01. This method takes into account the unique spectral and spatial features of each sensor, ensuring consistent results across different resolutions.

Unsupervised Classification

Iterative Self-Organizing Data (IsoData) Technique

In order to delineate the shoreline position, IsoData unsupervised classification algorithm was adopted to classify land use the land cover of the coastal area. The IsoData algorithm is an iterative method that uses Euclidean distance as a resemblance measure to cluster multispectral imageries data into different classes (Dhodhi et al., 1999). In general, the algorithm consists of a k-mean clustering which is entrenched within the loop that has heuristics for cluster split and merges (Ball and Hall, 1967). A total of one hundred (100) classes was generated from the water indices (NDVI and NDWI) which was reclassified into two classes namely water body and Land.

Accuracy Assessment

Land use classification, shoreline position and river channel feature delineation are incomplete without an assessment of the accuracy. In remote sensing research, kappa coefficient has been generally recommended in the early 1980s for standard field data validation and image training (Congalton and Green, 2009). Several studies have applied this method for a comprehensive measurement index for water indices delineation (Xu, 2006) and image classification accuracy (Tian et al., 2016). The study adopted confusion matrix function which is a Kappa coefficient indexing in QGIS software to compare the overall degree of agreement from the delineated water indices and unsupervised classification on all the multispectral imageries. 

Raster to Vector Analysis

In order to carry out positional accuracy of shoreline delineations derived from NDVI, NDWI, and ISO-Data unsupervised classification outputs, raster-to-vector conversion was performed using QGIS’s Raster-to-Vector and Features-to-Lines tools. This transformed segmented raster boundaries into vectorized shoreline polylines. Subsequently, systematic point sampling was achieved by generating perpendicular transects at 50-meter intervals along the shoreline vectors via the Transect tool, followed by intersection point extraction using the Intersect tool. These intersection points served as validation samples, enabling consistent spatial comparison between the classified shorelines and reference data. This method aligns with standardized approaches for positional accuracy assessment, which emphasize systematic sampling at fixed intervals to minimize spatial bias and enhance reproducibility (FGDA, 1998; QGIS, 2025). The 50-meter transect spacing balanced resolution and computational efficiency, ensuring robust quantification of shoreline positional discrepancies across all datasets.

Positional Accuracy Assessment

Positional Accuracy Assessment is considered as valuable method of qualifying positional different of two spatial data (FGDA, 1998). In order to compare the positional accuracy of delineated shoreline positions from different multi-source and spatial-resolution optical satellites acquired on the same day. The vectorised shoreline line was compared with Planetscope, Sentinel-2 MSI and LandSat-8 OLI using both QGIS and Python scripting which was used to measure the difference on Point-based standard methodology (PBSM) which has the ability as compare two spatial data. In this study, four metrics for comparing the positional accuracy were considered, such as mean error position (MEP), root mean square error (RMSE), standard deviation (SD), and circular error (CE) and linear error (LE) at confidence levels (CL) of 90%, 95%, and 99%.

RESULTS AND DISCUSSION

Bands Analysis (Top of Atmosphere (TOA) Reflectance)

The comparative analysis of Top of Atmosphere (TOA) reflectance relationships between PlanetScope (PS), Sentinel-2 MSI (S2), and Landsat-8 OLI (L8) across red, green, and near-infrared (NIR) bands (Figures 3 and 4) highlights critical insights into inter-sensor spectral consistency and divergence. These findings emphasise the influence of sensor design, spectral band features, spatial resolution (Planet Labs, 2023; SUHET, 2015; USGS, 2019) and tidal states (Tides4fishing, 2018) on cross-platform data harmonisation, with implications for multisource satellite applications. The result reveals near-perfect correlations in red (R=0.99) and green (R=1.0) bands between PS and S2 (Figures 3a, 3c) and NIR (R=0.99; Figure 4a) underscoring their high compatibility. The dense clustering of data points along the 1:1 line in these bands reveals minimal spectral distortion, likely attributable to their comparable spatial resolutions (PS:3m; S2:10-20m) and spectral band alignment. Sentinel-2’s red and green bands are optimised for vegetation and land-cover monitoring, with central wavelengths closely matching those of PlanetScope (Drusch et al., 2012; Planet Labs, 2023). This alignment facilitates reliable data fusion for applications such as vegetation mapping and land use classifications, where fine-scale spectral consistency is critical (Houborg and McCabe, 2018). The result reveals strong NIR agreement which further supports their combined use in deriving indices like NDVI, where NIR reflectance is pivotal (Zhang et al., 2018). In contrast, PS and L8 results reveal weaker correlations in visible bands (red: R=0.30; green: R=0.40; Figures 3b, 3d), characterised by significant scatter and deviation from the 1:1 line. These discrepancies are likely attributable to variations in spatial resolution (L8: 30m) and spectral response functions. The finer resolution of PlanetScope (PS) captures sub-pixel heterogeneity, which is averaged out by the coarser pixels of Landsat 8 (L8), resulting in mismatches in Top-of-Atmosphere (TOA) reflectance values. Additionally, variations in bandpass widths and central wavelengths particularly in the green band (e.g., PS:535-585nm vs. L8:530-590nm) may introduce spectral biases (Wulder et al., 2016). These discrepancies hinder direct comparisons in the visible spectrum, thereby necessitating careful consideration when integrating Landsat 8 (L8) with high-resolution sensors such as PlanetScope (PS) for applications that demand precise spectral reliability.

Figure 3. Top of Atmosphere (TOA) for red and green bands

Despite diminished harmony in visible spectral bands, the analysis revealed robust near-infrared (NIR) correlation between PlanetScope (PS) and Landsat 8 (L8) datasets (R = 0.98; Figure 4b), while L8 and Sentinel-2 (S2) exhibited near-perfect NIR alignment (R = 0.99; Figure 4c). The enhanced NIR agreement may be attributed to the broader spectral range of the NIR band and its reduced susceptibility to atmospheric scattering, which collectively attenuate sensor-specific discrepancies, fostering cross-platform interoperability (Pahlevan et al., 2017). This consistency is particularly advantageous for longitudinal environmental monitoring, wherein L8’s extensive temporal archive (2013–present) and S2’s operational continuity (2015–present) can be synergistically integrated with PS’s high temporal resolution to elucidate long-term trends. However, marginally elevated scatter in PS-L8 NIR comparisons (relative to PS-S2) suggests residual spatial resolution effects, as L8’s 30-m pixel resolution may homogenize fine-scale features resolved by PS’s 3-m resolution. These findings underscore the necessity of band-specific calibration in multisensor workflows. While PS and S2 exhibit interchangeability for visible and NIR applications, incorporating L8 necessitates spectral and spatial adjustments—particularly in visible bands—where L8’s red and green bands may require spectral bandpass calibration or spatial downscaling to align with PS/S2 specifications (Yang et al. 2015). Conversely, the pronounced NIR concordance across all platforms supports their amalgamation for large-scale vegetation dynamics or hydrological monitoring, leveraging NIR’s sensitivity to biomass fluctuations and moisture gradients (Pekel et al., 2016). When combining data from PS, S2, and L8 for applications using Near-Infrared (NIR), it's essential to follow sensor-specific guidelines to minimize errors in blending the data. While this study utilized top-of-atmosphere (TOA) reflectance data, surface reflectance retrievals would introduce variability due to atmospheric interference and canopy angular anisotropy. Methodological limitations include the exclusion of viewing geometry and overpass temporal offsets, which may introduce temporal inconsistencies in cross-sensor comparisons. Future research should focus on creating specific correction methods for different sensors and using machine learning to improve data consistency across various platforms. Additionally, field studies in different types of terrains are needed to identify the causes of data discrepancies, such as terrain reflectance and sensor angles. These improvements will help combine satellite data from multiple sources, enhancing their effectiveness for precise environmental monitoring and global change studies.

Figure 4. Top of Atmosphere (TOA) for Near Infare band

threshold method for Landsat-8, Sentinel-2 and Planetscope Using unsupervised classification (ISODATA) to compare the NDVI and NDWI from different optical satellite sources shows a slight difference in the overall classification of land and water bodies, especially in the complex creek water body and accuracy assessment (figure 6). The results show that the NDVI has a high kappa coefficient compared to other optical satellite imageries. For instance, the kappa coefficient includes L8: 0.95, S2: 0.97, and PS: 0.99, while for NDWI, the kappa coefficient includes L8: 0.96, S2: 0.98, and PS: 0.99. The inclusion of these indices is relevant for environmental monitoring especially in the complex coastal environment and the data is vital for managing natural resources and mitigating the impacts of climate change. The comparative analysis across different satellite data sources provides valuable insights, allowing for more informed decision-making in environmental management.

Imageries Segmentation and Accuracy Assessment

Figure 5 presents comparison of various methods for analysing different spatial resolution satellite images, using Normalised Difference Vegetation Index (NDVI) and the Normalised Difference Water Index (NDWI). NDVI measures the health of vegetation through its spectral reflectance, while NDWI is used to assess moisture levels in vegetation and to identify water bodies. This analysis is performed using three multispectral satellite platforms: Landsat 8 (L8), Sentinel-2 (S2), and PlanetScope (PS). A quantitative accuracy assessment, based on kappa coefficient values obtained from raster-based analysis, shows variability in NDVI performance across the platforms (L8: 0.94, S2: 0.96, PS: 0.98). Likewise, NDWI accuracy also varies by platform, with L8 at 0.96, S2 at 0.97, and PS at 0.99. These results highlight the effectiveness of optical satellite systems in assessing vegetation health and hydrological features, providing essential insights for precise environmental monitoring, especially in coastal areas. The findings indicate that Landsat 8 (L8), Sentinel-2 (S2), and PlanetScope (PS) each have distinct advantages and drawbacks. For example, Landsat 8 (L8) has a moderate spatial resolution, making it suitable for large-scale environmental monitoring. Sentinel-2 (S2) offers higher spatial resolution, which is ideal for in-depth vegetation studies. In contrast, PlanetScope (PS), despite its lower resolution, provides frequent monitoring, which is beneficial for observing rapid changes. A thorough understanding of these differences enables researchers to make informed choices regarding satellite-derived datasets that are best suited for specific applications, thus enhancing the accuracy of environmental monitoring efforts and assessment protocols (Gao, 1996).

Figure 5. Indices images showing the retrieved and segmented (land and water body) using threshold method for Landsat-8, Sentinel-2 and Planetscope.

Using unsupervised classification (ISODATA) to compare the NDVI and NDWI from different optical satellite sources shows a slight difference in the overall classification of land and water bodies, especially in the complex creek water body and accuracy assessment (figure 6). The results show that the NDVI has a high kappa coefficient compared to other optical satellite imageries. For instance, the kappa coefficient includes L8: 0.95, S2: 0.97, and PS: 0.99, while for NDWI, the kappa coefficient includes L8: 0.96, S2: 0.98, and PS: 0.99. The inclusion of these indices is relevant for environmental monitoring especially in the complex coastal environment and the data is vital for managing natural resources and mitigating the impacts of climate change. The comparative analysis across different satellite data sources provides valuable insights, allowing for more informed decision-making in environmental management.

Figure 6. Indices images showing the retrieved and segmented (land and water body) using unsupervised classification method for Landsat-8, Sentinel-2 and Planetscope.

The comparative analysis of NDVI (Normalized Difference Vegetation Index) and NDWI (Normalized Difference Water Index) values from Landsat 8 (L8), Sentinel-2 (S2), and PlanetScope (PS) reveals significant differences in vegetation and water body detection capabilities. The NDVI results show that L8 and S2 provide more detailed vegetation patterns, with higher values indicating denser vegetation. For instance, the NDVI high values are 0.425594 for L8 and 0.298274 for S2, compared to 0.180212 for PS. This suggests that L8 and S2 might have higher spatial resolution or better spectral sensitivity for vegetation detection, which is crucial for accurate environmental monitoring and agricultural assessments.

Vector-based Shoreline Positions     

Figure 7 presents a comparative shoreline analysis using three different datasets: PS (PlanetScope), S2 (Sentinel-2), and L8 (Landsat 8) NDVI-derived shorelines. The study assesses shoreline variability across different remote sensing platforms. The results indicate variations in shoreline positions depending on the dataset used. In figure 7(a), the distance differences between datasets range from -1.34 to 19.52 meters, suggesting that higher-resolution imagery (PS) captures more detailed shoreline features than coarser resolution data (L8). Figure 7(b) shows a big difference between PS and L8, with large difference of 138.31 meters. Figure 7(c) shows similar trends, with differences reaching up to 72.19 meters. These findings suggest that higher-resolution sensors like PlanetScope can provide more precise shoreline mapping, but the differences between datasets must be considered when assessing coastal dynamics. The study underscores the critical importance of dataset selection in coastal monitoring decision-making processes, as empirical findings demonstrate that lower-resolution data may result in distorted representations of shoreline changes compared to actual geomorphological dynamics. The observed shoreline positioning discrepancies align with findings from previous research that emphasize spatial resolution impacts on shoreline detection and coastal erosion assessments (Smith et al., 2021).

Figure 7. Overlaid vectorised shoreline position for PS, S2 and L8 using NDVI and Threshold method

Figure 8 shows the comparative analysis of shoreline positions derived from three satellite datasets: PlanetScope (PS), Sentinel-2 (S2), and Landsat 8 (L8). The PS and S2 shorelines are very much in line with each other, but the L8 shoreline is very far off. This is especially clear in figure 8(b), where the differences between the PS and L8 shorelines range from 4.71 meters to 125.49 meters. The PS data, with its high spatial resolution, provides the most precise shoreline delineation, capturing finer coastal details compared to the coarser resolution of L8. This precision is essential for accurate environmental monitoring and coastal management. The discrepancies between L8 and the other datasets highlight the limitations of using lower-resolution data for detailed shoreline mapping. These findings are consistent with previous research that emphasises the importance of high-resolution satellite imagery for accurately detecting shoreline changes and managing coastal erosion (Smith et al., 2023). The integration of multiple satellite datasets allows for a more comprehensive understanding of coastal dynamics, ultimately improving the accuracy of shoreline monitoring. The study highlights the imperative for developing sensor-specific calibration protocols and machine learning-based harmonisation frameworks to improve cross-platform data interoperability, along with systematic ground-truth initiatives to isolate and address sources of observational variability inherent in multi-sensor environmental monitoring systems.

Figure 8. Overlaid vectorised shoreline position for PS, S2 and L8 using NDWI and Threshold method

The figure 9 shows comparative shoreline analysis using the Isodata (ISO) classification of the Normalized Difference Vegetation Index (NDVI) from three different satellite sources: PlanetScope (PS), Sentinel-2 (S2), and Landsat 8 (L8). The results indicate that the PlanetScope and Sentinel-2 shorelines exhibit minimal differences in most areas, as shown by small deviation values (e.g., PS vs. S2: 1.61m, 12.02m in figure 9(a)). However, more significant discrepancies appear in dynamic coastal regions, such as figure 9(b), where differences between PS and S2 reach -40.48m, suggesting active erosion or accretion. The L8-derived shoreline consistently shows greater deviations from PS and S2, reinforcing the influence of Landsat 8’s coarser spatial resolution (~30m) on shoreline accuracy (Pardo-Pascual et al., 2018). Notably, the figure 9(b) also highlights irregularities in shoreline positioning, likely due to vegetation dynamics affecting NDVI-based shoreline classification. The results suggest that higher-resolution sensors such as PS (~3m) and S2 (~10m) provide more accurate shoreline delineation for fine-scale coastal monitoring, whereas L8-derived shorelines may be less precise in detecting subtle changes   (Luijendijk et al., 2018). These findings emphasize the importance of using high-resolution imagery for shoreline change analysis and coastal management in erosion-prone areas

Figure 9.Overlaid vectorised shoreline position for PS, S2 and L8 using NDVI and Unsupervised Classification (ISODATA)

Figure 10 shows comparative shoreline extraction analysis using the ISODATA (ISO) classification of the Normalised Difference Water Index (NDWI) from three satellite sources: PlanetScope (PS), Sentinel-2 (S2), and Landsat 8 (L8). Figure 10(a) shows relatively small deviations among the datasets (e.g., PS vs. S2: 4.23m, 9.96m), suggesting a high degree of consistency in stable coastal regions. However, in figure 10(b), significant deviations emerge, with PS vs. L8 reaching up to 118.49m, and S2 vs. L8 showing a difference of 115.28m. These large differences indicate the sensitivity of NDWI-derived shorelines to sensor spatial resolution, with PlanetScope (~3m) and Sentinel-2 (~10m) offering higher accuracy compared to Landsat 8 (~30m) (Pardo-Pascual et al., 2018). Figure 10(c) shows minimal deviations (e.g., PS vs. S2: 0.56m, 6.62m), reaffirming that high-resolution sensors provide more precise shoreline extraction in stable regions. The study suggests that while NDWI-based classification effectively captures the land-water boundary, its accuracy varies based on resolution and coastal dynamics (Luijendijk et al., 2018). Therefore, integrating high-resolution datasets enhances coastal monitoring, particularly in dynamic environments prone to erosion and accretion.

Figure 10. Overlaid vectorised shoreline position for PS, S2 and L8 using NDWI and Unsupervised Classification (ISODATA)

Positional Accuracy Assessment

The comparative analysis of positional accuracy among PlanetScope (PS), Sentinel-2 MSI (S2), and Landsat-8 OLI (L8) using NDVI, NDWI, and ISO Data unsupervised classification reveals significant insights into inter-sensor agreement and variability. These results reveal the importance of sensor characteristics, spatial resolution, and spectral band configurations in influencing positional accuracy assessments, particularly for vegetation and water-related indices. The strongest agreement was observed between PS and S2 across both NDVI and NDWI analyses. For NDVI, PS vs S2 revealed the lowest MEP (2.24), RMSE (4.36), and SD (3.63), with narrow confidence intervals (CE: 5.96-9.34m at 90-99%), indicating minimal variability. Similarly, NDWI comparisons revealed low MEP (2.66) and RMSE (4.46), supported by tight confidence intervals (CE: 5.89-9.2m at 90-99%). This alignment likely stems from the higher spatial resolutions of PS (3m) and S2 (10-20m), which better capture fine-scale landscape features compared to L8 (30 m). The spectral overlap in red and near-infrared bands, critical for NDVI and NDWI calculations, may further enhance consistency between these sensors (Drusch et al., 2012; Planet Labs, 2023). In ISO Data unsupervised classification, PS vs S2 maintained superior agreement (MEP: 2.88m; RMSE: 4.83m), reinforcing their interoperability for land-cover mapping. These results align with studies emphasizing the complementary use of PS and S2 for high-resolution monitoring (Houborg and McCabe, 2018).

Table 3. Shoreline Position Accuracy Assessment using Threshold

In contrast, comparisons involving L8 showed markedly higher errors. NDVI analysis for PS vs L8 yielded elevated MEP (43.17) and RMSE (46.24), with SD (16.58) and wide confidence intervals (CE: 27.27-42.70%; LE: 76.06-119.11%). Similarly, NDWI comparisons for PS vs L8 and S2 vs L8 revealed moderate disagreement (MEP: 17.76-16.71; RMSE: 19.35-18.09), reflecting L8’s coarser resolution and potential spectral mismatches in green and shortwave infrared bands, which are critical for NDWI (Roy et al., 2016). The ISO Data classification further highlighted L8’s limitations, with NDWI-ISO comparisons showing the highest variability (e.g., PS vs L8 MEP: 6.34m; CE up to 17.34%). These findings are consistent with research demonstrating L8’s reduced sensitivity to small water bodies and heterogeneous vegetation due to its 30-m resolution (Pekel et al., 2016). Table 4. Shoreline Position Accuracy Assessment using Unsupervised Classification