Deep learning-enhanced shoreline dynamics and vulnerability assessment in Niger Delta area of Nigeria

Abstract

The Niger Delta region of Nigeria faces escalating coastal erosion and shoreline retreat, threatening ecosystems, infrastructure, and livelihoods. This study presents an integrated geospatial methodology combining deep learning-based shoreline detection, multi-temporal satellite imagery, and statistical modeling to assess coastal vulnerability and shoreline evolution. Landsat imagery from 2002, 2015, and 2023 was processed using the CoastSat toolkit, which employs a U-Net convolutional neural network for semantic segmentation of land-water boundaries. Shoreline positions were extracted with sub-pixel accuracy using the Modified Normalized Difference Water Index (MNDWI). The Digital Shoreline Analysis System (DSAS) quantified shoreline dynamics, revealing that 75.3% of transects experienced erosion, with maximum retreat exceeding 8,000 m. High Shoreline Change Envelope values reflect barrier island migration and tidal channel dynamics characteristic of deltaic systems. A Coastal Vulnerability Index (CVI) was computed by integrating elevation, slope, distance to shoreline, tidal height, and tidal range through an AHP-weighted multi-criteria framework. Statistical validation using multiple regression, following Principal Component Analysis to address multicollinearity between slope and elevation (VIF > 14), confirmed that topographic factors are the dominant predictors of coastal vulnerability (Adjusted R² = 0.716, p < 0.001). The Topographic Component emerged as the primary predictor (β = -0.795), followed by distance to shoreline (β = -0.273), while tidal variables contributed significantly but with smaller effect sizes. Despite temporal gaps in Landsat coverage and moderate spatial resolution (30 m), the methodology demonstrates robust performance for regional-scale assessments. This framework provides a scalable, data-driven approach for coastal monitoring and highlights the urgent need for adaptive management in the Niger Delta.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-39405-7.

Introduction

The Niger Delta, located in southern Nigeria and bordered by the Atlantic Ocean, is one of Africa’s largest and most environmentally complex deltaic systems. Its low-lying terrain, dense population, and socio-economic dependence on fragile coastal ecosystems make it highly vulnerable to coastal hazards. In recent decades, the region has experienced escalating shoreline retreat, resulting in the loss of arable land, infrastructure damage, and displacement of communities^1–5. These trends are in line with global patterns, where sea-level rise, coastal erosion, and climate variability increasingly threaten deltaic environments^{6, 7}. In the Niger Delta, natural drivers such as tidal dynamics and sediment transport, alongside human pressures including oil exploration, sand mining, and mangrove degradation, further exacerbate shoreline instability^8–11

The Coastal Vulnerability Index (CVI) provides a structured framework for quantifying susceptibility to hazards by integrating geomorphological, hydrodynamic, and topographic variables^{12, 13}. Multi-criteria decision analysis (MCDA) approaches have proven effective for spatial vulnerability assessment in coastal and riverine environments, as demonstrated in recent flood risk evaluations in the Niger Delta region¹⁴. However, in West Africa, many CVI-based assessments still rely on static or manually derived inputs, limiting temporal resolution and replicability^{15, 16}. Traditional shoreline monitoring methods, such as ground surveys and manual image digitization, are labor-intensive, subjective, and poorly suited to capturing rapid coastal changes^{17, 18}. These limitations justify the need for scalable, automated approaches that combine temporal coverage with statistical rigor.

Recent advances in remote sensing (RS) and machine learning (ML) have transformed shoreline monitoring and coastal hazard assessment. Automated toolkits such as CoastSat apply deep learning to multispectral imagery and allow for sub-pixel shoreline detection across multi-decadal time series^{19, 20}. When CoastSat is used with the Digital Shoreline Analysis System (DSAS), they generate robust metrics, including Shoreline Change Envelope (SCE) and Net Shoreline Movement (NSM), facilitating quantitative assessments across diverse coastal settings^21–23. In the Niger Delta and surrounding riverine systems, DSAS-based analyses have documented significant shoreline dynamics, including high-mobility zones influenced by tidal channel migration and sediment redistribution^{24, 25}. Similarly, recent studies have successfully integrated machine learning with DSAS to forecast shoreline changes along the River Niger, demonstrating the predictive potential of hybrid approaches²⁶. Machine learning techniques have also been applied to analyze shoreline erosion and accretion patterns using geospatial methods, and these outputs serve effectively as regional comparisons for the present study²⁷.

Beyond shoreline dynamics, ML and geospatial techniques have been increasingly applied to related coastal and environmental challenges in the Niger Delta. Studies have employed ML combined with morphometric analysis for runoff dynamics and flood management in Bayelsa State²⁸, while others have utilized satellite imagery and ML algorithms for flood extent mapping and damage assessment²⁹. Climate-driven flood pattern analysis using ML and GIS has revealed significant temporal shifts in flood behavior across downstream Nigeria³⁰. Geospatial monitoring of sandbars and sedimentary processes in deltaic rivers has highlighted the dynamic nature of these systems and their implications for sustainable resource management^{31, 32}. Furthermore, ML approaches have been extended to groundwater quality modeling and assessment of drainage density and infiltration characteristics in coastal catchments, demonstrating the broad applicability of these techniques for environmental monitoring in the region.

U-Net and convolutional neural networks (CNNs) have advanced shoreline segmentation and coastal classification¹⁷, while hybrid workflows that integrate optical, SAR, gravimetric, and DEM datasets allow monitoring of erosion, subsidence, and mangrove loss³³. Coupling these inputs with multi-criteria decision analysis (e.g., AHP) and statistical validation enhances the robustness and policy relevance of CVI frameworks³⁴.

Despite these advances, applications in West Africa remain limited. Few studies have integrated automated shoreline detection with CVI, and most lack rigorous validation of the relative contribution of individual vulnerability factors. Multiple regression analysis provides a means to quantify how drivers such as slope, elevation, and tidal range shape coastal vulnerability^{35, 36}. In Nigeria, recent efforts have advanced shoreline change detection^{21, 37}, but integration with CVI and regression analysis is still underdeveloped. Furthermore, no documented prior research in the West African region has integrated deep learning-based shoreline detection, DSAS metrics, CVI modeling, and regression validation into a unified vulnerability framework. Addressing this gap, the present study develops a reproducible framework that combines CoastSat-based shoreline detection, DSAS metrics, multi-criteria CVI, and regression validation to generate spatially explicit vulnerability maps for the Niger Delta.

Study area

The Niger Delta (Fig. 1), located in southern Nigeria and bordered by the Atlantic Ocean, is one of the world’s largest and most complex river deltas, spanning approximately 70,000 km² with an extensive network of rivers, creeks, estuaries, barrier islands, tidal flats, and mangrove swamps^{38, 39}. The adjacent continental shelf is relatively shallow (5–20 m nearshore, extending to ~ 200 m at the shelf break), strongly influencing tidal energy, wave dynamics, and sediment deposition⁴⁰. These features, combined with a tropical climate and high rainfall, create a dynamic and ecologically sensitive coastal environment.

Fig. 1.

The Niger Delta region.

Over recent decades, the region has experienced significant shoreline retreat driven by natural processes (wave action, tidal currents, sea level rise) and anthropogenic activities including oil exploration, sand mining, urbanization, and mangrove deforestation^{8, 9}. This geomorphic instability causes constant shoreline reconfiguration, with profound implications for biodiversity, infrastructure, and livelihoods dependent on fisheries and agriculture^{15, 16, 41}. Climate-induced sea level rise and increasing extreme weather events underscore the urgency of managing coastal vulnerability in this area⁴². The delta’s coastal states lie between 4°00ʹN–8°00ʹN and 4°30ʹE–9°30ʹE, encompassing rich natural and mineral resources that complicate the balance between conservation and development⁴³.

Methodology

Figure 2 illustrates the workflow diagram summarizing the methodological framework employed in this study. The process begins with data acquisition, which includes multi-temporal satellite imagery (Landsat series), Digital Elevation Models (DEM), and tidal measurements obtained from local tide gauge stations⁴⁴. These datasets were preprocessed to ensure consistency in spatial resolution and coordinate systems. Shoreline positions for the years 2002, 2015, and 2023 were extracted using the CoastSat Python toolkit, which leverages deep learning-based image segmentation algorithms for sub-pixel shoreline detection from multispectral imagery²⁰.

Fig. 2.

Flowchart summarizing the methodology for the evaluation of coastal vulnerability index for shoreline. MCDM multi-criteria decision-making, deep learning, AHP analytical hierarchy process, GIS geographic information systems.

Following shoreline extraction, the Digital Shoreline Analysis System (DSAS) was used to compute quantitative shoreline change metrics, including the Shoreline Change Envelope (SCE), Net Shoreline Movement (NSM), End Point Rate (EPR), and Linear Regression Rate (LRR). These metrics allowed for the temporal analysis of shoreline dynamics and erosion/accretion patterns^{45, 46}. Subsequently, a Coastal Vulnerability Index (CVI) was developed by integrating key geophysical and hydrodynamic variables, namely, elevation, slope, distance to shoreline, tidal range, and tidal height. Each variable was standardized, reclassified, and weighted based on both expert judgment and the observed shoreline change trends.

The resulting CVI values were used to delineate spatial zones of varying vulnerability levels across the study area. To statistically validate the influence of each input factor on coastal vulnerability, a multiple regression analysis was performed using CVI scores as the dependent variable. This final step identified the most significant predictors of vulnerability, thereby enhancing the explanatory power of the model and supporting targeted coastal risk management interventions.

Datasets

The spatial and environmental datasets used in this study were obtained from a combination of satellite remote sensing and ancillary geospatial sources (Table 1). Landsat imagery was acquired for three distinct time periods—2002 (Landsat 7), 2015, and 2023 (Landsat 8)—with spectral bands selected to optimize land–water boundary detection. Specifically, bands in the green (Band 2 for Landsat 7; Band 3 for Landsat 8) and shortwave infrared (SWIR) ranges (Band 5 for Landsat 7; Band 6 for Landsat 8) were utilized.

Table 1.

Summary of satellite datasets used.

Year(s)	Sensor	Bands used
2002	Landsat 7	Band 2 (green), band 5 (SWR)
2015–2023	Landsat 8	Band 3 (green), band 6 (SWR)

A Digital Elevation Model (DEM) derived from the Shuttle Radar Topography Mission (SRTM) with a 30-meter spatial resolution was used to generate both elevation and slope data; both being key indicators of topographic exposure to coastal hazards. Hydrodynamic variables, including tidal range and tidal height, were sourced from regional tide gauge stations to characterize marine forcing conditions. Additionally, distance-to-shoreline was calculated by measuring horizontal proximity from inland points to extracted shoreline positions, allowing the assessment of exposure gradients across the study area.

Satellite image processing

To extract historical shoreline positions with high spatial accuracy, Landsat imagery was processed using the CoastSat Python toolkit, which implements deep learning-based shoreline detection. CoastSat employs a modified U-Net architecture, a fully convolutional neural network originally developed for biomedical image segmentation⁴⁷. The model performs pixel-wise classification of satellite imagery into four classes: water, whitewater, sand, and other land covers. The network was pre-trained on over 8,000 manually labeled Landsat and Sentinel-2 images from diverse coastal environments²⁰.

The Modified Normalized Difference Water Index (MNDWI) was applied to each scene to enhance land–water contrast and accurately delineate coastlines. The threshold of 0 (being the standard practice) for MNDWI was adopted, as positive values indicate water presence due to higher green reflectance relative to SWIR⁴⁸.

1

Where;

G = Reflectance in the Green band.

SWIR = Reflectance in the Shortwave Inferred Band.

Consequently, shoreline positions for 2002, 2015, and 2023 were extracted with sub-pixel precision and exported as shapefiles for further geospatial and temporal analysis. This enabled consistent tracking of shoreline dynamics over the two-decade study period.

Shoreline statistical analysis using DSAS

Quantitative assessment of shoreline dynamics was conducted using the Digital Shoreline Analysis System (DSAS), a widely used extension within ArcGIS developed by the U.S. Geological Survey. DSAS enables the computation of standardized shoreline change metrics along systematically spaced transects projected perpendicular to a user-defined baseline. In this study, transects were generated at 500-meter intervals, intersecting shoreline positions extracted for the years 2002, 2015, and 2023. This shoreline change metrics are used to detect both erosional and accretional trends along the Niger-Delta coastline. They were computed as follows;

2

Where;

SCE = Shoreline change envelop being the maximum horizontal distance between the farthest landward and seaward shorelines across analysis period.

3

Where

NSM = Net Shoreline movement, being the horizontal displacement between the oldest and most recent shoreline positions.

4

Where

EPR = End Point Rate, being the average annual rate of change between successive shoreline dates.

5

Where

LRR = Linear regression rate, being an estimate of the long-term shoreline change by fitting a least squares regression line to all available shoreline positions over time.

 

 

Coastal vulnerability index (CVI) calculation

To assess spatial variations in coastal risk, a Coastal Vulnerability Index (CVI) was computed by integrating five key vulnerability indicators: elevation, slope, distance to shoreline, tidal range, and tidal height. Each factor was normalized and assigned a weight (a₁ to aₙ) based on its relative contribution to coastal vulnerability, using an AHP-based multi-criteria decision-making framework⁴⁹. Land use was not directly included as an AHP criterion but served as supplementary data.

The resulting CVI provided a composite vulnerability assessment to inform targeted coastal management strategies.

6

Where:

X1, X2​,…,Xn​ are the normalized values of the factors (elevation, slope, distance to shoreline, tidal range, tidal height),

a1, a2,………a_n are the weights assigned to each factor, based on their relative contribution to vulnerability.

The shoreline change metrics derived from DSAS, while not directly incorporated into the Coastal Vulnerability Index (CVI) calculation, play a critical interpretive role in validating and contextualizing vulnerability patterns. These metrics help inform the weighting of CVI input variables, such as slope, elevation, and proximity to the shoreline, by highlighting which physical features are most associated with erosion or accretion trends⁵⁰. Additionally, spatial congruence between high erosion zones and elevated CVI scores serves as a cross-validation measure, enhancing the model’s reliability for coastal risk assessment. This integrated approach strengthens both temporal shoreline analysis and spatial vulnerability mapping in dynamic deltaic environments^{20, 24}.

Statistical analysis and regression model

To statistically validate the influence of each input factor on coastal vulnerability, correlation analysis and multiple regression modeling were performed using CVI scores as the dependent variable. This analysis aimed to identify the most significant predictors of vulnerability, quantify their relative contributions, and ensure the methodological robustness of the CVI framework.

Correlation analysis

Pearson correlation coefficients were computed to examine bivariate relationships between the CVI and each predictor variable. This preliminary analysis provided insights into the strength and direction of associations between individual factors and coastal vulnerability. Strong correlations between slope and elevation (r > 0.95) were anticipated, given that slope is mathematically derived from elevation gradients.

Multicollinearity assessment

Prior to regression modeling, multicollinearity among predictor variables was assessed using the Variance Inflation Factor (VIF). Multicollinearity occurs when predictor variables are highly correlated with each other, and this can inflate standard errors, produce unreliable regression coefficients, and lead to misleading R² values⁵¹. VIF values exceeding 10 indicate severe multicollinearity; hence, they correction⁵².

Initial VIF analysis revealed severe multicollinearity between slope and elevation, with VIF values of 14.2 for both variables (Table 3). This finding was expected, given the high correlation between these terrain variables (r = 0.964), as slope is a direct derivative of elevation. Such collinearity, if uncorrected, would produce inflated R² values and unreliable coefficient estimates, potentially leading to erroneous conclusions about the relative importance of predictors.

Table 3.

Variance inflation factor (VIF) analysis.

Variable	VIF	Interpretation
Elevation	14.20	Severe (> 10)
Slope	14.21	Severe (> 10)
Topographic component	1,00	Acceptable
Distance to shoreline	1.00	Acceptable
Tidal height	1.00	Acceptable
Tidal range	1.00	Acceptable

VIF > 10 indicates severe multicollinearity; VIF > 5 indicates moderate concern; VIF < 5 is acceptable.

Principal component analysis (PCA) for multicollinearity correction

To address the multicollinearity between slope and elevation, Principal Component Analysis (PCA) was applied to combine these two correlated variables into a single composite variable termed the “Topographic Component”⁵³. PCA transforms correlated variables into orthogonal (uncorrelated) principal components while retaining maximum variance from the original data. This approach is methodologically justified because slope and elevation share substantial redundant information, with slope being derived from elevation gradients.

The PCA results indicated that the first principal component (PC1) explained 98.2% of the total variance in the slope-elevation data, confirming that these variables can be effectively combined without significant information loss (Table 4). This “Topographic Component” was retained for subsequent regression analysis, effectively capturing the combined influence of terrain characteristics on coastal vulnerability while eliminating multicollinearity.

Table 4.

Principal component analysis results.

Component	Eigenvalue	Variance (%)	Cumulative (%)
PC1 (topographic component)	1.965	98.2	98.2
PC2	0.036	1.8	100.0

Component loadings: Elevation = 0.707, Slope = 0.707 for PC1.

Multiple regression model

Following multicollinearity correction, the multiple regression model was developed using four predictor variables. The regression model is specified as:

7

Where:

TC = PCA-derived composite of slope and elevation.

DtS = Euclidean distance from each point to the nearest shoreline (meters).

TH = Mean tidal elevation at the coast (meters).

TR = Difference between mean high water and mean low water (meters).

β₀ = Intercept.

β₁ - β₄ = Regression coefficients for each factor.

ε = Error term.

Post-correction VIF analysis confirmed that all predictor variables were within acceptable limits (all VIF < 3.0), indicating that multicollinearity had been successfully addressed and that the regression model would produce reliable coefficient estimates.

Results and discussion

This section presents the findings of the integrated shoreline evolution and coastal vulnerability analysis for the Niger Delta, structured across five analytical stages: (i) MNDWI-based shoreline extraction; (ii) temporal shoreline change analysis; (iii) coastal vulnerability mapping; (iv) correlation analysis of influencing variables; and (v) statistical validation using regression models. Each component builds toward a comprehensive understanding of the coastal dynamics and associated risks in the region.

Shoreline extraction using MNDWI and CoastSat

The Modified Normalized Difference Water Index (MNDWI) was successfully applied to Landsat imagery from 2002, 2015, and 2023, enhancing the contrast between water and land surfaces. Resulting index values ranged from − 1 to + 1, with water features showing strongly positive values (Figs. 3a–c). The threshold of 0 for MNDWI classification is standard practice, as positive values indicate water presence due to higher green reflectance relative to SWIR⁴⁸. These values enabled precise shoreline delineation using the CoastSat Python toolkit, which utilizes a modified U-Net architecture—a fully convolutional neural network originally developed for biomedical image segmentation⁴⁷. The model performs pixel-wise classification into four classes: water, whitewater, sand, and other land covers, having been pre-trained on over 8,000 manually labeled Landsat and Sentinel-2 images from diverse coastal environments²⁰. The outputs formed the baseline for subsequent shoreline evolution analysis.

Fig. 3.

MNDWI for (a top) 2002, (b middle) 2015, (c bottom) 2023.

Automated shoreline maps derived from MNDWI layers revealed consistent changes in coastline position across the two decades. Visual enhancements further improved land–water separation, enabling a high-confidence spatial representation of erosion and accretion zones (Fig. 4). These shorelines were exported for statistical analysis using the Digital Shoreline Analysis System (DSAS), facilitating the quantification of historical shoreline movement.

Fig. 4.

Shoreline derived from MNDWI of three years.

Shoreline change dynamics

Temporal shoreline evolution was analyzed using the four standard DSAS metrics, which are Shoreline Change Envelope (SCE), Net Shoreline Movement (NSM), End Point Rate (EPR), and Linear Regression Rate (LRR). Positional uncertainty for Landsat-derived shorelines was estimated at ± 15–30 m, accounting for pixel resolution (30 m), georeferencing error (RMSE < 10 m), and water level variations²⁰. The baseline was generated using the 2002 shoreline position, with transects cast perpendicular at 100 m spacing. Validation was performed at 25 randomly selected transects using Google Earth Pro historical imagery (submeter resolution), yielding a mean absolute error of 22.4 m, consistent with expected Landsat accuracy.

The findings confirm dynamic and spatially variable shoreline behavior across the Niger Delta (Fig. 5);

Fig. 5.

Shoreline change showing areas with a high risk of coastal erosion.

• (i)SCE values indicated high spatial variability, with an average envelope of 1,199.53 m and maximum displacement of 9,463.95 m (Transect 991).
• (ii)NSM analysis revealed that 74.92% of shoreline transects experienced net erosion, averaging − 665 m, while 25.08% exhibited accretion, with a maximum of 2,037.17 m (Transect 718).
• (iii)EPR showed an average annual retreat rate of − 31.27 m/year, with extremes ranging from − 383.07 m/year (Transect 990) to + 193.99 m/year (Transect 1262).
• (iv)LRR confirmed long-term erosion trends in 75.3% of transects, with a maximum erosion rate of − 414.99 m/year and peak accretion of + 95.1 m/year.

The extreme SCE values (> 8,000 m) and high erosion rates (> 400 m/year) are concentrated in areas characterized by: (1) barrier island mobility, where sandy barriers migrate landward in response to storm events and sea-level rise; (2) tidal channel migration, where meandering tidal creeks continuously reshape shoreline positions; and (3) sediment depletion from upstream dam construction (particularly the Kainji and Jebba dams), which has reduced sediment flux to the delta by an estimated 60–80%⁵⁴. Similar high-mobility zones have been documented in comparable deltaic systems in the region^{24, 31}.

These results reveal widespread coastal erosion across the study area, punctuated by local accretion near river mouths and sediment deposition zones (Fig. 5). The prevalence of erosion confirms the region’s vulnerability to hydrodynamic pressures and land-use disturbances.

Coastal vulnerability mapping

A composite Coastal Vulnerability Index (CVI) was computed by integrating five key variables: elevation, slope, distance to shoreline (Fig. 6), tidal range, tidal height, and land use (Fig. 7). The resulting vulnerability map (Fig. 8) classifies the Niger Delta coastline into three risk zones: (i) high vulnerability zones (CVI 3.5) occurred in low-lying coastal belts with steep slopes and high tidal exposure. These areas, particularly in Rivers, Akwa Ibom, and southern Delta State, are adjacent to active river mouths and estuaries, making them prone to frequent shoreline retreat. (ii) Moderate vulnerability zones (0.5), observed in Bayelsa and parts of Delta, exhibited transitional elevations and slope gradients, implying moderate susceptibility to future sea-level rise and extreme tidal events. (iii) Low vulnerability zones (CVI 2.5), such as Ilaje in Ondo State, were characterized by higher elevation, flatter slopes, and greater inland distances from the active coastline, reducing their exposure to immediate coastal hazards.

Fig. 6.

Elevation, slope, and distance to shoreline.

Fig. 7.

Tidal height, tidal range, and land use.

Fig. 8.

Coastal vulnerability in the Niger Delta.

This spatial stratification allows for evidence-based prioritization of intervention and adaptation strategies, especially in high-risk communities.

Correlation analysis of vulnerability drivers

A Pearson correlation analysis was conducted to assess the relative influence of the input variables on CVI scores (Table 2). The results revealed strong correlations between CVI and topographic variables, with slope and elevation showing the highest associations. Notably, the correlation between slope and elevation was extremely high (r = 0.964), indicating substantial shared variance between these terrain variables. This is expected given that the slope is mathematically derived from elevation gradients.

Table 2.

AHP pairwise comparison matrix for CVI factors.

A: Pairwise comparison matrix
Criterion	Elevation	Slope	Distance to shoreline		Tidal height		Tidal range	Weight
Elevation			1	1	2	3	6	0.34
Slope			1	1	2	4	3	0.31
Distance to shoreline			1/2	1/2	1	2	3	0.18
Tidal height			1/3	1/4	1/2	1	2	0.10
Tidal range			1/6	1/3	1/2	1/2	1	0.07

B: Consistency analysis
Parameter	Value
Principal eigenvalue (λmax)	5.180
Number of criteria (n)	5
Consistency index (CI)	0.045
Random index (RI)	1.12
Consistency ratio (CR)	0.04

C: Priority weights summary
Rank	Factor		Weight	Interpretation
1	Elevation		0.34	Highest priority - low elevation increases flood exposure
2	Slope		0.31	High priority - affects drainage and mass wasting susceptibility
3	Distance to shoreline		0.18	Moderate priority - proximity amplifies coastal hazard exposure
4	Tidal height		0.10	Lower priority - influences inundation extent
5	Tidal range		0.07	Lowest priority - affects tidal flooding frequency
	Total		1.00

Distance to the shoreline exhibited a moderate correlation with CVI (r = −0.264), suggesting that exposure increases closer to the coast but is moderated by elevation and slope. Weak yet statistically significant correlations were observed for tidal height and tidal range (r ≈ 0.08–0.10), indicating that while hydrodynamic factors contribute to vulnerability, their influence is secondary to topographic characteristics.

These findings align with established research emphasizing the dominant role of terrain-related variables in determining coastal hazard susceptibility^{12, 15, 16, 24}.

Statistical analysis

A multiple linear regression analysis was performed to statistically validate the influence of each variable on the computed CVI scores using Eq. (7). The regression model, using four predictors (Topographic Component, distance to shoreline, tidal height, and tidal range), yielded an adjusted R² of 0.716 (p < 0.001), with all VIF values reduced to 1.0 (Tables 4 and 5). The standardized coefficients used for each factor are given in Table 6.

Table 5.

Correlation analysis.

Correlations
		CVI	Distance to shorelines	Slope	Tidal height	Tidal range	DEM	MNDWI
CVI	Pearson correlation	–
	N	2750
Distance to shorelines	Pearson correlation	0.192**	–
	Sig. (2-tailed)	0.000
	N	2750	2750
Slope	Pearson correlation	1.000**	0.194**	-
	Sig. (2-tailed)	0.000	0.000
	N	2750	2750	2750
Tidal height	Pearson correlation	0.081**	− 0.018	0.081**	–
	Sig. (2-tailed)	0.000	0.337	0.000
	N	2750	2750	2750	2750
Tidal range	Pearson correlation	0.081**	− 0.018	0.081**	1.000**	–
	Sig. (2-tailed)	0.000	0.337	0.000	0.000
	N	2750	2750	2750	2750	2750
DEM	Pearson correlation	0.970**	0.201**	0.971**	− 0.005	− 0.005	–
	Sig. (2-tailed)	0.000	0.000	0.000	0.809	0.802
	N	2750	2750	2750	2750	2750	2750
MNDWI	Pearson correlation	0.024	0.161**	0.024	− 0.004	− 0.004	0.025	–
	Sig. (2-tailed)	0.203	0.000	0.204	0.829	0.840	0.185
	N	2750	2750	2750	2750	2750	2750	2750

**Correlation is significant at the 0.01 level (2-tailed).

Table 6.

Regression model comparison.

Statistic	Corrected model
R-squared	0.716
Adjusted R-squared	0.716
F-statistic	1732.29
p-value	< 0.001
Maximum VIF	1.00
Multicollinearity	None
Number of predictors	4

The regression model explains 71.6% of CVI variance, with all predictors statistically significant (p < 0.001). The regression model confirms that the Topographic Component (combining elevation and slope effects) is the strongest predictor of CVI (β = −0.795), followed by Distance to Shoreline (β = −0.273). The negative coefficients indicate that higher topographic values (representing higher elevation and steeper slopes combined) are associated with lower vulnerability, which aligns with the conceptual framework where low-lying, gently sloping coastal areas are most susceptible to hazards. Tidal variables (Tidal Height: β = 0.091 and Tidal Range: β = 0.063), while statistically significant (p < 0.001), contributed comparatively less to the overall vulnerability prediction, suggesting that hydrodynamic forcing is secondary to topographic exposure in determining overall vulnerability patterns in the Niger Delta (Table 7).

Table 7.

Regression model coefficients.

Variable	B	SE	t	p-value	Significance
(Constant)	3.420	0.039	87.89	< 0.001	***
Topographic component	−0.434	0.006	−78.21	< 0.001	***
Distance to shoreline	−0.00002	0.000	−26.87	< 0.001	***
Tidal height	0.144	0.016	8.95	< 0.001	***
Tidal range	0.082	0.013	6.23	< 0.001	***

** p < 0.001, ** p < 0.01, * p < 0.05*.

These findings are consistent with previous research emphasizing the dominant role of terrain-related variables in coastal hazard susceptibility^{12, 15}. The spatial coincidence of high CVI zones with areas of active erosion, as documented by DSAS metrics, validates the model’s structure and affirms the importance of combining topographic, hydrodynamic, and geomorphic datasets for comprehensive coastal risk mapping (Table 8). Similar patterns have been observed in related Niger Delta studies, including shoreline change assessments²⁴ and ML-based coastal analyses^{26, 27}.

Table 8.

Standardized coefficients (beta).

Variable	Beta	Rank
Topographic component	−0.795	1
Distance to shoreline	−0.273	2
Tidal height	0.091	3
Tidal range	0.063	4

Discussion

The results of this study reveal a dynamic and spatially heterogeneous pattern of coastal change and vulnerability across the Niger Delta region, consistent with global findings in deltaic environments subject to natural and anthropogenic pressures^55–57. The widespread coastal retreat, captured through DSAS-derived metrics such as NSM, EPR, and LRR, shows the intensity of erosion processes, particularly in low-lying coastal stretches adjacent to estuarine and riverine zones.

Erosion pattern and physical drivers

The extremely high shoreline retreat rates exceeding 400 m/year and maximum SCE values approaching 9,500 m reflect the dynamic geomorphology characteristic of the Niger Delta. Four primary mechanisms explain these magnitudes: (1) sediment depletion from upstream dams, particularly the Kainji and Jebba dams, which have reduced sediment delivery by 60–80% since their construction⁵⁴; (2) barrier island mobility, where unconsolidated sandy barriers migrate landward or alongshore in response to wave energy and storm events; (3) tidal channel migration, as distributary channels shift position through bank erosion and point bar accretion; and (4) anthropogenic pressures, including mangrove deforestation and sand mining, which destabilize coastal sediments. These findings align with Oborie et al.²⁴, who documented similar high-mobility zones using DSAS, and Mogo et al.³¹, who characterized sediment mobility in deltaic sandbars. The integration of machine learning for shoreline detection, as demonstrated by Jonathan and Charles²⁷and Eteh et al.²⁶, provides methodological validation for capturing such extreme but physically plausible changes.

Vulnerability patterns and topographic dominance

The CVI reveals that topographic variables (elevation and slope), dominate vulnerability patterns across the Niger Delta, consistent with established theory and empirical findings^{12, 15, 16}. Also, the regression model explains 71.6% of CVI variance, with the Topographic Component emerging as the strongest predictor (β = −0.795), followed by distance to the shoreline (β = −0.273), while tidal variables showed smaller but significant effects (β = 0.063–0.091). These results make physical sense in that (i) low elevations increase flood exposure from marine and fluvial sources, (ii) slope affects drainage and mass-wasting susceptibility, and (iii) proximity to the coast amplifies risk regardless of terrain characteristics⁵⁸. For coastal management, these findings empirically justify prioritizing elevation and slope in vulnerability assessments, support buffer zone approaches and setback policies in high-risk areas, and suggest that interventions should target low-lying, gently sloping coastal zones where vulnerability concentrates. The relatively modest contribution of tidal factors may partly reflect the interpolation of tidal data from limited gauge stations across the broad delta extent, hence, potentially underestimating localized tidal influences.

Methodological considerations

Integrating CoastSat’s U-Net architecture for automated shoreline detection with empirically grounded CVI modeling presents a scalable, replicable, and objective approach to vulnerability assessment, reducing the subjectivity of manual digitization while leveraging freely available multi-decadal Landsat imagery. The statistical validation employed by using VIF analysis to detect multicollinearity and PCA to correct it, improves on studies that report inflated R² values without such diagnostics. The Adjusted R² of 0.716 aligns with comparable coastal vulnerability studies^{36, 59, 60}, indicating that while the model captures most CVI variance, approximately 28.4% remains attributable to unmeasured factors. These likely include sediment dynamics and littoral transport, anthropogenic modifications such as coastal structures and land reclamation, localized geomorphic features below Landsat resolution, spatially variable subsidence rates, and groundwater extraction effects^{29, 30, 32}.

Implications for coastal management

These findings call for urgent action in high-risk coastal areas, with identified vulnerability hotspots (CVI ≥ 3.5) serving as clear targets for shoreline stabilization, mangrove restoration to reduce wave energy and trap sediments, zoning restrictions on new development, and early warning systems for erosion-prone communities. Moderate-risk zones in Bayelsa and inland Delta warrant careful management of future development to limit long-term exposure^{61, 62}, while integrating flood risk approaches^{14, 28}with shoreline vulnerability mapping could yield more comprehensive hazard assessments for coastal populations.

Limitations and uncertainty

Inspite the reliability of the methodology used, this study has some limitations, which include that (i) the analysis relies on three time points (2002, 2015, 2023), hence potentially missing short-term changes or storm impacts; (ii) Landsat’s 30 m resolution may not capture fine-scale shoreline features, hence a positional uncertainty of ± 15–30 m²⁰; (iii) tidal data interpolated from limited gauge stations may not fully represent local dynamics; (iv) the regression model’s 71.6% explained variance indicates that unmeasured factors such as sediment dynamics, anthropogenic modifications, and subsidence also influence vulnerability; and (v) CVI outputs remain sensitive to AHP weight assignments, affecting cross-study comparability. Despite these constraints, the methodology performs robustly for regional-scale assessment and offers a reproducible framework transferable to other deltaic settings.

Conclusion

This study employed an integrated geospatial framework to assess shoreline evolution and coastal vulnerability in the Niger Delta, combining deep learning-based shoreline extraction using the CoastSat toolkit, quantitative change detection through DSAS, and spatial risk modeling using the Coastal Vulnerability Index (CVI). Results revealed widespread shoreline retreat between 2002 and 2023, with more than 75% of transects exhibiting significant erosion, maximum retreat exceeding 8,000 m (reflecting barrier island mobility and tidal channel migration characteristic of deltaic systems), and erosion rates reaching 414.99 m/year in high-risk zones. The CVI, incorporating elevation, slope, distance to shoreline, tidal height, and tidal range through an AHP-weighted multi-criteria framework, identified vulnerability hotspots concentrated in low-lying areas of Rivers, Akwa Ibom, and southern Delta States, providing targets for coastal management interventions.

Statistical validation through correlation and regression analysis confirmed that topographic variables are the dominant predictors of coastal vulnerability. Following multicollinearity correction using Principal Component Analysis, the regression model explained 71.6% of CVI variance (Adjusted R² = 0.716, F = 1732.29, p < 0.001). The Topographic Component emerged as the primary predictor (β = −0.795), followed by Distance to Shoreline (β = −0.273), while tidal variables contributed significantly but with smaller effect sizes. The remaining 28.4% of variance is attributable to unmeasured factors including sediment dynamics, anthropogenic activities, and localized geomorphic processes. Future research should expand this framework to incorporate socio-economic indicators, ecosystem services valuation, and climate projection scenarios. Also, replicating the study across other West African deltas would strengthen regional coastal adaptation strategies.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

NVC built the conceptual design, and methodological framework for this study. IU and SL carried out the analysis using ArcGIS and WEKA to conduct geospatial and ML analyses respectively; while OJO interpreted the results. Finally, NVC and IU wrote the draft manuscript; while OJO revised the draft and prepared the final manuscript.

Data availability

The datasets generated and/or analyzed during the current study are available on request from the authors.

Declarations

Competing interests

The authors declare no competing interests.