Bayesian Projections of Shoreline Retreat Under Climate Change Along the Arid Coast of Duba, Saudi Arabia

Abstract

Coastal zones in arid regions are particularly vulnerable to climate change because of their limited sediment supply and high sensitivity to marine and aeolian forces. This study provides probabilistic projections of coastal evolution for a 130 km segment of the Duba shoreline, Saudi Arabia, a rapidly developing region that includes the NEOM mega-project. An integrated modeling framework was developed by combining a four-decade (1985–2024) diachronic analysis of shoreline evolution from Landsat imagery with a cascade of numerical models. Specifically, climate projections from CMIP6 (under SSP1-2.6, SSP2-4.5, and SSP5-8.5 scenarios) were dynamically downscaled using the regional climate models COSMO-CLM and RegCM, which provided boundary conditions for the SWAN hydrodynamic model to simulate the wave dynamics. The SWAN outputs were then used to force the Delft3D morphodynamic model to project future shoreline evolution. A Bayesian framework was applied to systematically quantify and integrate the uncertainties across all modeling steps, enabling robust probabilistic forecasts. Results indicate an accelerated trend of shoreline retreat, with mean Net Shoreline Movement (NSM) by 2100 ranging from −8.1 m under the low-emission SSP1-2.6 scenario to a critical −25.6 m under the high-emission SSP5-8.5 scenario, with 95% confidence intervals reaching −47.9 m. This erosion is mainly driven by a projected relative sea-level rise of up to 48.3 cm (±15.8 cm) and an increase in significant wave height of up to 40% (mean of 1.95 m). By delivering probabilistic rather than deterministic results, this study provides a solid scientific basis to guide sustainable coastal management, inform the design of risk-sensitive infrastructure, and support the development of climate-resilient adaptation strategies in one of the world’s most rapidly transforming coastal regions.

1. Introduction

Coastal zones represent dynamic interfaces where geological, oceanographic, and atmospheric processes interact, supporting unique ecosystems and vital socioeconomic activities [1].

Hosting over 40% of the world’s population and concentrating critical infrastructure, as well as exceptionally rich ecosystems, these areas are intrinsically vulnerable to the increasing pressures of global climate change. These pressures include accelerated sea level rise, intensification of extreme weather events, and changes in wave regimes [2]. Such phenomena directly threaten their stability and sustainability, with coastal erosion leading to land loss, degradation of natural habitats, and increased risk to property and human safety [3,4,5]. Understanding and anticipating these transformations has become a scientific and societal imperative for developing effective adaptation strategies and integrated coastal management [6,7].

In this global context, the coastlines of arid and semi-arid regions, such as the Red Sea, present unique challenges [8]. Characterized by low fluvial sediment input, high evaporation, and sensitivity to aeolian and marine forces, these coastal systems respond specifically to climatic pressures [8]. The northwest coast of Saudi Arabia, and more specifically, the Duba coastline in the Tabuk region, perfectly illustrates this situation. This region is undergoing unprecedented economic and demographic transformation, embodied by large-scale development projects such as NEOM [9], which profoundly alter sedimentary balances and increase the vulnerability of coastlines to natural hazards. Paradoxically, despite their strategic importance and proven sensitivity, these coasts remain under-studied. Knowledge of their historical dynamics and even more so their future response to climate change remains fragmentary, limiting decision-makers’ ability to develop effective adaptation strategies.

This study focused on a 130 km coastal segment along the Duba coastline. This site was chosen because of its remarkable geomorphological diversity, including sandy beaches, rocky coasts, semi-enclosed lagoons, and sabkhas, making it a representative natural laboratory for the Red Sea coastline. Specifically, the unique interplay of these features, such as sediment trapping in lagoons influenced by aeolian transport from adjacent sabkhas, wave refraction along rocky headlands that modulate beach erosion, and episodic storm-driven exchanges between sandy beaches and offshore systems under arid conditions with minimal fluvial input, creates distinct morphodynamic processes that are highly sensitive to climatic variations [10,11]. Comparative studies have highlighted how such diverse geomorphological confluences in arid or semi-arid settings amplify vulnerability to climate change, positioning Duba as an ideal site for studying these impacts in the context of the Red Sea. Subject to a hot desert climate and an oceanic regime dominated by northwestern swells, this coastline is already experiencing active erosion. The morpho-sedimentary dynamics of these beaches are complex. Furthermore, increasing anthropogenic pressures and future climatic threats significantly heighten its vulnerability, making it a priority site for coastal risk assessment. For instance, recent localized land loss has affected beachfront properties and minor infrastructure, with erosion rates exceeding 1–2 m per year in some segments, threatening coastal habitats valued at more than USD 500 million in ecosystem services annually, similar to the challenges identified in coastal retreat studies for SW Spain [12] and ongoing erosion concerns in Great Britain [13].

Therefore, the central scientific question is how this complex coastal system will evolve in the 21st century under the combined effects of climatic forcing and local pressure. Answering this question requires moving beyond traditional approaches that are often deterministic and based on simple extrapolations of past trends [14]. These traditional approaches include methods that rely solely on historical shoreline change rates, such as those using the Digital Shoreline Analysis System (DSAS) without dynamic modeling [15,16], or simpler regression models that assume static conditions [17]. Such methods suffer from limitations such as a lack of probabilistic outputs, an inability to integrate multi-model climate data, and static assumptions that ignore nonlinear interactions, as discussed in studies on historical shoreline error analysis [18], quantitative trend analysis [19], and grey relation analysis [20]. By contrast, our integrated framework, which combines Bayesian probabilistic methods with CMIP6 downscaling, RCMs, SWAN, and Delft3D, offers a superior alternative by providing uncertainty-quantified projections and addressing these gaps.

Coastal evolution is the product of nonlinear interactions between multiple factors (sea-level rise, waves, currents, winds, and sediment inputs), and its prediction is fraught with considerable uncertainty. These uncertainties stem from the intrinsic variability of the climate system, diversity of future socioeconomic trajectories (SSP scenarios), limitations of numerical models, and inaccuracies in input data. Therefore, a robust assessment of coastal risks cannot forgo rigorous quantification of these uncertainties. The development of probabilistic projections, which provide not only an estimate of the most probable evolution, but also a range of possible outcomes with associated confidence levels, is thus necessary for informed decision-making.

In light of these challenges, the main objective of this study was to develop and apply an integrated modeling framework to produce probabilistic projections of the evolution of the Duba coastline by 2100. To achieve this objective, we adopted a multifaceted and innovative methodological approach, which is one of the main contributions of this study. This approach offers 2–3 unique technical advancements over single or sequential frameworks in prior studies (e.g., [14,21]), including the novel coupling of hydrodynamic (SWAN/Delft3D) and probabilistic (Bayesian) components for real-time uncertainty propagation, integration of downscaled CMIP6 data via RCMs for region-specific forcing, and performance-based weighting of models to generate probability distributions, addressing limitations such as isolated Bayesian networks [7] and the need for advanced models highlighted in [21]. This approach combines a diachronic analysis of shoreline evolution over the past four decades (1985–2024) with Landsat satellite imagery to establish a robust historical baseline. It also integrates the use of climate projections from the CMIP6 project, dynamically downscaled for the region via the regional climate model (RCMs) COSMO-CLM and RegCM under three contrasting socioeconomic scenarios (SSP1-2.6, SSP2-4.5, and SSP5-8.5) [22,23]. These climatic data were then used for force-coupled hydrodynamic and morphodynamic modeling, relying on the reference models SWAN (for waves) and Delft3D (for currents and sediment transport). Finally, the integration of all these sources of information and their respective uncertainties is carried out within a Bayesian probabilistic framework [24,25], an approach that allows weighting different models based on their historical performance and generating probability distributions for key coastal change indicators, such as retreat rates and sediment balances. This study extends beyond the Duba coast. Numerous arid and semi-arid coastal regions worldwide, from the North African coast to parts of Australia and South America, face similar challenges of sediment scarcity, intense wave regimes, and increasing anthropogenic pressures due to climate change. Thus, the methodologies developed and the results obtained provide a transferable framework for comparable assessments in these contexts, contributing to a broader global understanding of arid coastal dynamics.

2. Materials and Methods

2.1. Study Area

This study focused on a 130 km coastal segment along the Duba shoreline in the Tabuk region of northwestern Saudi Arabia (approximate coordinates between 28.1° N, 35.0° E and 26.7° N, 36.1° E) (Figure 1). This sector presents wide geomorphological diversity, characterized by sandy beaches, rocky shores, semi-enclosed lagoons, and typical arid sebkha environments. A hot desert climate strongly influences erosive processes through the combined effects of dominant winds and low rainfall input. The oceanic regime is dominated by a semi-diurnal tide of low-amplitude and seasonal cyclonic coastal currents associated with a prevailing northwesterly swell that significantly contributes to sediment transport [26,27].

Figure 1. Study area.

2.2. Methodological Approach

The methodological approach adopted in this study combines diachronic shoreline analysis based on satellite data, numerical and statistical modeling, and probabilistic integration within a Bayesian framework (Figure 2). To improve readability and sequencing, the workflow was organized into three successive stages: (i) data preprocessing and shoreline extraction, (ii) numerical and statistical modeling, and (iii) Bayesian probabilistic integration.

Figure 2. Flowchart of the methodological approach integrating satellite data processing, numerical and statistical modeling, and Bayesian probabilistic framework. Each step is color-coded according to data type: blue for remote sensing, green for numerical models, orange for statistical/AI models, and purple for probabilistic integration.

First, Landsat imagery and LiDAR surveys were processed to extract the shoreline by applying radiometric and atmospheric corrections, and semi-automatic methods based on spectral indices and supervised classification methods. Shoreline dynamics were quantified using the DSAS module, which calculates morphodynamic indicators such as Net Shoreline Movement (NSM), endpoint rate (EPR), Linear Regression Rate (LRR), and Weighted Linear Regression (WLR). In parallel, climate and socioeconomic projections (SSP scenarios) were integrated into numerical models (SWAN and Delft3D) to simulate hydrodynamic and morphodynamic processes and into statistical and artificial intelligence models (vector regression and LSTM networks) to capture the nonlinear relationships and effects of anthropogenic pressures. The vector regression model was employed to establish quantitative relationships between historical shoreline changes (derived from DSAS indicators such as NSM and LRR) and external drivers including wave height, sea-level variability, and coastal development intensity. The Long Short-Term Memory (LSTM) network, a recurrent neural network architecture, was implemented to reproduce the temporal evolution of shoreline displacement and to capture delayed nonlinear responses to climatic and anthropogenic forcings. The results from these AI-based models were used to refine the statistical trends of shoreline dynamics and to complement the process-based outputs of the SWAN and Delft3D simulations before integration into the Bayesian probabilistic framework. The outputs of these various models were then combined within a Bayesian framework and weighted according to historical performance and associated uncertainty to produce multiscale probabilistic projections of future coastal evolution. Finally, integrated uncertainty management allows for the estimation of confidence intervals for morphodynamic indicators and sediment budget.

Bayesian Integration Framework

A Bayesian hierarchical framework was implemented to integrate the model outputs and observational datasets. Priors for shoreline retreat rates and wave parameters were defined as weakly informative normal distributions centered on the observed historical means. The likelihood function was assumed to be Gaussian, with the variance representing both observational uncertainty and model error. Posterior inference was obtained using a Markov Chain Monte Carlo (MCMC) sampling scheme (10,000 iterations with 2000 burn-in), and convergence was verified using Gelman–Rubin diagnostics ($\widehat{R}<1.1$) and effective sample size (ESS). Model weights were derived from historical performance using normalized inverse RMSE values calculated over the calibration period (1985–2015). This weighting scheme ensured that models with better agreement with the historical shoreline and wave observations contributed more strongly to the posterior distribution.

2.3. Data Sources

The datasets used in this study were obtained from publicly accessible sources. Multitemporal satellite imagery (1985, 1995, 2005, 2015, and 2024), including Landsat 5 MSS/TM, Landsat 8 OLI, and Landsat 9 OLI-2 scenes, was downloaded from the United States Geological Survey (USGS) Earth Explorer portal (https://earthexplorer.usgs.gov/) (accessed on 1 June 2025). Climate projections were retrieved from the Coupled Model Intercomparison Project Phase 6 (CMIP6) through the Earth System Grid Federation (ESGF) portal (https://esgf-node.llnl.gov/projects/cmip6/) (accessed on 1 June 2025), considering three Shared Socioeconomic Pathways (SSP1-2.6, SSP2-4.5, and SSP5-8.5). Regional downscaling was carried out using data from the CORDEX-MENA initiative, which provides dynamically downscaled outputs from COSMO-CLM and RegCM regional climate models at a spatial resolution of approximately 10 km (https://cordex.org/domains/cordexregion-mena-cordex/ ) (accessed on 10 June 2025). For shoreline change analysis, the Digital Shoreline Analysis System (DSAS, v6), developed by the USGS, was employed to compute morphodynamic indicators such as Net Shoreline Movement (NSM), End Point Rate (EPR), Linear Regression Rate (LRR), and Weighted Linear Regression (WLR) (https://www.usgs.gov/centers/whcmsc/science/digital-shoreline-analysis-system-dsas) (accessed on 10 June 2025). All datasets were freely available and will be accessed between January and April 2024. To address variations in data quality and differences in radiometric calibration among the Landsat 5 MSS/TM, Landsat 8 OLI, and Landsat 9 OLI-2 sensors, sensor-specific radiometric corrections and normalization procedures were applied. The preprocessing steps included atmospheric correction using the Landsat Ecosystem Disturbance Adaptive Processing System (LEDAPS) for Landsat 5 and Landsat Surface Reflectance Code (LaSRC) for Landsat 8 and 9 data to obtain consistent surface reflectance values. Cross-calibration techniques and inter-sensor harmonization methods were employed to minimize spectral discrepancies and ensure temporal comparability across the multi-temporal dataset. These procedures followed the established guidelines in the remote sensing literature to reduce the impact of sensor differences on the analysis results.

2.4. Satellite Data

The assessment relied on a multitemporal analysis of five Landsat images (1985, 1995, 2005, 2015, and 2024) selected according to strict quality criteria: absence of clouds, consistent tidal levels (0.48–0.62 m), and winter season acquisitions (January) to ensure temporal comparability. Landsat 5 MSS, TM, as well as OLI sensors from Landsat 8 and 9, were used with spatial resolutions ranging from 60 m (1985) to 15 m through a pan-sharpening technique, enabling higher precision in shoreline extraction. The images were georeferenced using the WGS84 UTM Zone 37N system [28]. To ensure an accurate combination of data from different sensors, a rigorous co-registration validation was performed using ground control points distributed throughout the study area. The root mean square (RMS) error for each image pair was calculated, with values consistently below 1 pixel (<15 m), confirming the high spatial agreement necessary for reliable shoreline extraction.

2.5. Preprocessing and Shoreline Extraction

The images were geo-cropped to isolate the study area and processed to enhance the land–water discrimination using natural and infrared color composites. The shoreline was manually extracted, referenced to the mean tidal level, digitized by a single operator following a strict protocol, and validated through random double entry to minimize subjectivity. The inter-digitization variance was calculated by comparing repeated digitizations of 10% of the shoreline segments, yielding an average positional difference of 1.32 m, indicating acceptable consistency across the dataset. This method was preferred over standard automated approaches, such as NDWI or machine learning-based techniques, which are unreliable in this complex landscape context, particularly because of spectral confusion between wet sabkha surfaces, bright sandy substrates in shallow waters, and the actual water line, resulting in significant classification errors [29,30].

2.6. Coastal Change Analysis with DSAS

Quantitative shoreline modification analyses were conducted using a Digital Shoreline Analysis System (DSAS, v6, USGS). A baseline parallel to the coast was placed 1000 m offshore, and perpendicular transects spaced 100 m apart were generated (n = 2300) across the entire area. Standard indicators computed include the End Point Rate (EPR), Linear Regression Rate (LRR), Weighted Linear Regression (WLR) (where shorelines are weighted based on the inverse of their positional uncertainty), Net Shoreline Movement (NSM), and the Shoreline Change Envelope (SCE), with a positional uncertainty estimated at ±10 m, a value derived from the quadratic sum of errors, including image resolution, georeferencing accuracy, and digitization subjectivity. These indicators provide a robust characterization of the spatial and temporal trends of change along the coast [31]. Although a fixed 100 m transect spacing was applied across the study area, a thorough visual inspection and manual adjustments were performed in areas with high coastline curvature to ensure proper transect alignment. Adaptive transect spacing based on shoreline morphology could further improve local accuracy but is not currently available within the DSAS tool.

2.7. Climate Projections and Socioeconomic Scenarios

This study focuses on a coastal segment of Duba approximately 130 km long, the modeling of which requires adequate resolution of the climatic and hydrodynamic fields. To ensure an accurate representation of the spatial and temporal variability of climatic parameters, the projections used were based on multi-model simulations of the Coupled Model Intercomparison Project Phase 6 (CMIP6). Three Shared Socioeconomic Pathways (SSPs) reflecting contrasting greenhouse gas emission trajectories were selected: SSP1-2.6 (low emissions), SSP2-4.5 (intermediate), and SSP5-8.5 (pessimistic) [22]. These scenarios cover three time horizons: short-term (2025-2045), medium-term (2046–2070), and long-term (2071–2100).

The native spatial resolution of CMIP6 projections, on the order of 100–200 km, is insufficient to capture the fine-scale dynamics required at the coastal level. Therefore, two dynamic Regional Climate Models (RCMs), COSMO-CLM and RegCM, were used within the CORDEX-MENA project, providing a horizontal resolution of approximately 0.08°, that is, approximately 10 km, which is better suited to the complex geoclimatic and orographic conditions of the Duba region [32,33]. Along the Duba coast, the models were gridded with a spatial resolution ranging from 10 to 60 m, obtained by statistical downscaling from the 10 km resolution of the RCMs. This process involved bilinear interpolation for continuous fields, such as temperature and wind, combined with a nearest-neighbor approach for precipitation, to preserve storm cell characteristics, ensuring that local topographic and bathymetric effects were implicitly accounted for in the high-resolution model grid. This approach ensures the fidelity of climatic forcing while enabling a resolution appropriate to the scales of interest for coastal processes. The combined use of COSMO-CLM and RegCM allows for the comparison and validation of results given their recognized performance in the MENA region and their different physical methodologies, thereby strengthening the robustness of the projections. While acknowledging that a 10 km horizontal resolution for RCMs (COSMO-CLM and RegCM) may be considered coarse for directly resolving fine-scale coastal dynamics such as nearshore currents and headland effects, it is the highest resolution currently achievable through dynamic downscaling for multi-decadal climate projections across large regional domains like MENA. To bridge this gap, a statistical downscaling approach was applied to achieve resolutions ranging from 10 m to 60 m along the Duba coast. The climatic variables integrated into the SWAN (waves) and Delft3D (currents and sediments) hydrodynamic models were expressed as multidimensional spatiotemporal functional fields, corrected using a statistical bias-adjustment method (quantile mapping) to reduce systematic biases in the climate models. The principle of the quantile mapping correction is described in detail in Appendix A Equation (A1).

The observed distribution functions $F_{\mathrm{obs}}^{-1}$ were constructed from the ERA5 reanalysis dataset [34], which provides consistent hourly to monthly reference data for the historical period (1981–2010). Temporal matching was ensured by aligning the bias-correction procedure to the native resolution of each variable (hourly, daily, or monthly).

The respective temporal resolutions were as follows: daily precipitation $pr=pr(t_{daily},x,y)$, daily air temperature of 2 m $tas=tas(t_{daily},x,y)$, zonal wind corrected for bias $uas=uas(t_{hourly},x,y)$, meridional wind corrected for bias $vas=vas(t_{hourly},x,y)$, monthly sea surface temperature $tos=tos(t_{monthly},x,y)$ and relative sea level monthly to annual $zos=zos(t_{monthly/annual},x,y)$.

The SWAN model simulates the evolution of the wave spectral energy $E(t,x,y)$ (see Appendix A Equation (A1) for a detailed formulation). The energy transfer from wind to waves $S_{in}$, as well as the wind speed magnitude $U_{10}$ and direction ${\theta }_u$, are detailed in Appendix A Equation (A1).

Delft3D is based on the fundamental Navier–Stokes equations that describe the current dynamics and free surface evolution (Appendix A Equation (A1)). Hydrostatic pressure, wind-induced friction, dynamic bathymetry update, and sediment flux formulations are fully described in Appendix A Equation (A1), including the definitions of each variable and parameter.

Finally, a temporal interpolation procedure was implemented to ensure the temporal consistency of the corrected climatic variables used for forcing in SWAN and Delft3D. Daily variables such as pr and tas were linearly interpolated to produce continuous hourly values consistent with the time steps required by the models. This interpolation preserved the overall trends while avoiding temporal discontinuities. Synchronization of multi-frequency data was ensured by directly incorporating hourly wind files and aligning the interpolated variables. Physical consistency checks were performed to verify that the interpolations did not generate unrealistic values such as negative precipitation or sudden temperature fluctuations. Specifically, interpolated precipitation values were floored at zero, and the rate of change in air temperature was flagged for review if it exceeded a threshold of 3 °C per hour. Any flagged inconsistencies were corrected using linear interpolation from adjacent validated data points to ensure the physical integrity of model forcings. This approach guarantees homogeneous and reliable integration of corrected climatic data into hydrodynamic simulations, thereby optimizing the relevance of the results and the quality of climate impact analyses on the coastal dynamics of Duba.

2.8. Uncertainty Management Using a Bayesian Approach

Assessing coastal processes using climate projections and hydrodynamic models inherently involves substantial uncertainties arising from natural system variability, numerical model limitations, and input data inaccuracies. To address this, a Bayesian approach is adopted, which systematically combines a priori model information with available observations D, enabling the probabilistic updating of parameters $\theta$. The posterior distribution is given by

$$P(\theta \mid D)={\displaystyle \frac{P(D\mid \theta )·P\left(\theta \right)}{P\left(D\right)}},$$

(1)

where $P\left(\theta \right)$ denotes the prior distribution, $P(D\mid \theta )$ denotes the likelihood, and $P(\theta \mid D)$ denotes the posterior distribution.

In this study, the random variables $\theta$ correspond to the key morphodynamic indicators significant wave height ($H_s$), relative sea level (RSL), and net shoreline movement (NSM), each represented by a probability distribution that reflects both model-based mean estimates and associated uncertainties. Specifically, the prior distributions of the key parameters Net Shoreline Movement (NSM) and Significant Wave Height (SWH) were defined as normal distributions centered on their respective historical means, with variances reflecting the interannual variability observed over the 1985–2024 period. Similarly, the likelihood functions were modeled as Gaussian distributions. This configuration allows the posterior distribution to integrate both empirical observations and model performance, thereby ensuring a robust and transparent probabilistic inference.

For example, the significant wave height $H_s$ can be represented as a normal distribution, reflecting uncertainties from climate and numerical modeling:

$$H_s\sim \mathcal{N}({\mu }_{H_s},{\sigma }_{H_s}^2),$$

(2)

where ${\mu }_{H_s}$ is the simulation mean and ${\sigma }_{H_s}^2$ represents the combined uncertainties.

The Bayesian framework offers a key advantage in that it explicitly quantifies uncertainty and provides confidence intervals for morphodynamic indicators. Unlike deterministic approaches, it delivers probabilistic projections, thereby enhancing the robustness and credibility of the information available to coastal managers. It also allows the integration of multiple data sources and the continuous updating of knowledge as new observations are made.

Previous studies have highlighted the efficacy of Bayesian approaches in coastal applications. Kroon et al. (2020) applied a Bayesian network to quantify uncertainty in decadal-scale coastal morphology predictions, incorporating uncertainties from model inputs, calibration, and processes [35]. Similarly, Plant and Stockdon (2012) employed a Bayesian approach for uncertainty estimation in coastal barrier island response modeling, underscoring the importance of probabilistic assessment in coastal projections [36].

Although alternative methods, such as Monte Carlo simulations, multi-model ensembles, or error propagation, exist, the Bayesian approach stands out for its ability to integrate multiple sources of uncertainty and provide a comprehensive statistical interpretation of results, which is essential for informed coastal management under conditions of high uncertainty.

3. Results

3.1. Shoreline Change Assessment

The diachronic analysis of Net Shoreline Movement (NSM) (Figure 3) highlights a progressive intensification of shoreline retreat in Duba by 2100, with notable contrasts between the emission scenarios and regional climate models considered. During the historical reference period, the NSM medians were approximately $-10$ to $-15$ m, reflecting moderate shoreline retreat with relatively limited variance.

Figure 3. Observed and projected evolution of net shoreline movement (NSM) across multiple time horizons under SSP1–2.6, SSP2–4.5, and SSP5–8.5 scenarios using COSMO–CLM and RegCM models.

To quantitatively validate the agreement between the modeled and observed NSM during the historical period, the MAE was below 2.5 m, the RMSE did not exceed 3.0 m, and the Pearson correlation coefficient reached 0.98. The modeled interquartile ranges encompassed those of the observations, confirming that the projections accurately reproduced historical shoreline dynamics. These metrics provide direct evidence of strong quantitative consistency between the modeled and observed NSM, supporting the reliability of subsequent scenario-based projections and management recommendations.

Under the low-emission scenario (SSP1-2.6), projections suggest relative stability of historical conditions, with NSM medians remaining between $-10$ and $-20$ m and low dispersion, reflecting the potential resilience of the coastal system under effective mitigation of greenhouse gas emissions and climate change (Figure 4).

Figure 4. Projected shoreline retreat patterns under the low-emission scenario SSP1–2.6 in Duba.

In contrast, the intermediate (SSP2-4.5) and high-emission (SSP5-8.5) scenarios highlighted a marked intensification of coastal retreat. Under SSP2-4.5, medians progressively decrease to $-25$ to $-30$ m in the long term, accompanied by an increase in variability (Figure 5). SSP5-8.5 projects the most concerning values, with medians below $-30$ to $-40$ m and extremes sometimes below $-70$ m, indicating substantial uncertainty and a high risk of extreme shoreline retreat (Figure 6). This amplification reflects the combined effects of accelerated sea-level rise, altered wave regimes, and an increased frequency of extreme events [3,4,37].

Figure 5. Projected shoreline retreat patterns under the intermediate-emission scenario SSP2–4.5 in Duba.

Figure 6. Projected shoreline retreat patterns under the high-emission scenario SSP5–8.5 in Duba.

A comparison of the two regional climate models revealed increasing divergence with longer time horizons. COSMO-CLM tends to project wider dispersions and more pronounced extremes, reflecting a higher sensitivity to atmospheric and oceanographic dynamics. The RegCM produced more moderate trajectories but converged towards similar long-term trends. This divergence supports the use of multi-model approaches, which capture the full range of uncertainties and prevent excessive reliance on a single model prediction.

These findings align with studies in the Mediterranean and MENA regions [5,21,38], demonstrating that high-emission trajectories are systematically associated with intensified coastal erosion and a pronounced shoreline retreat. The literature emphasizes the combined factors of sea-level rise (0.5 m to >1 m by 2100 under SSP5-8.5), intensified wave regimes, and reduced sediment supply, which exacerbate the vulnerability of arid and semi-arid coasts [2,39].

The substantial increase in variability and extremes under high-emission scenarios necessitates the revision of coastal management strategies and incorporating the uncertainty ranges derived.

3.2. Assessment of Shoreline Change Rates

The analysis of shoreline change rates (EPR, in m/year) presented in Figure 7 highlights a dynamic that is strongly influenced by both the climate scenarios and the models applied. During the historical period, observations and simulations (COSMO-CLM and RegCM) showed predominantly negative or near-zero values, indicating relative shoreline stability or a minor retreat trend. This consistency between the measured data and modeling supports the robustness of future projections, as reported in similar studies using regional models to evaluate historical coastal erosion [40,41], which validated the use of RegCM for climate simulations in Mediterranean areas and showed good agreement with in situ observations.

Figure 7. Observed and projected evolution of shoreline change rates (EPR) across multiple time horizons under SSP1–2.6, SSP2–4.5, and SSP5–8.5 scenarios using COSMO–CLM and RegCM models.

Under the SSP1-2.6 (low-emission) scenario, EPR variability remains limited, with slightly negative medians and narrow dispersions, reflecting the potential resilience of the coastal system under effective mitigation of greenhouse gas emissions. These results are consistent with similar findings, such as those of [4], who projected moderate erosion under low-emission scenarios with a significant reduction in shoreline retreat due to the limitation of sea-level rise.

In the intermediate SSP2-4.5, shoreline retreat intensified, and the range of values widened over time, indicating more dynamic erosion processes. This trend is in line with the findings of [38], who, using coupled climate-wave models, demonstrated a global increase in coastal erosion under SSP2-4.5, with enhanced variability driven by intensified storm activity.

Under the SSP5-8.5 (high-emission) scenario, the figure reveals an accentuated retreat, with maximum dispersions ranging from $-2.5$ to $+2$ m/year, reflecting greater uncertainty and amplification of extreme events. Extreme projection echo studies, such as [3], modeled shoreline erosion rates reaching 2–3 m/year under high-emission scenarios, driven by the increase in extreme wave events and sea-level rise, with similar dispersions linked to model uncertainties.

3.3. Assessment of Significant Wave Height Changes

The analysis of significant wave heights (SWH) in Duba demonstrated a strong agreement between historical observations (ranging from 0.5 to 2.3 m, with a median of 1.1 m) and simulations generated by the regional climate models COSMO-CLM and RegCM. To quantitatively substantiate this agreement, validation metrics were calculated for the historical period. The mean absolute error (MAE) between the observed and modeled median wave heights from COSMO-CLM and RegCM did not exceed 0.2 m, whereas the root mean square error (RMSE) remained below 0.3 m, indicating close proximity between the simulations and observations. Moreover, the Pearson correlation coefficient calculated between observed and modeled medians is 0.97, evidencing a strong linear relationship.

Notably, RegCM exhibited a slight tendency to overestimate the wave heights, offering a conservative assessment of potential future risks. Projected under different emission scenarios, wave conditions revealed progressive intensification: the low-emission SSP1-2.6 scenario forecasted a moderate increase, with maximum wave heights reaching 3.0 to 3.2 m by the end of the century. In contrast, the intermediate SSP2-4.5 scenario anticipates more substantial intensification, with extreme wave heights ranging from 3.5 to 3.7 m.

The high-emission SSP5-8.5 scenario predicts a pronounced transformation of wave dynamics, with frequent exceedances of 3.7–3.9 m, amounting to increases exceeding 60% relative to current conditions (Figure 8). These findings are consistent with regional climatic patterns, including the pronounced warming of the Arabian Gulf, where recent extreme precipitation events have been linked to the exacerbation of climate change [42,43]. Global studies have similarly established a direct link between climate forcing and changes in global wave regimes [44], confirming the increasing vulnerability of coastal regions to ocean–climate hazards. Consequently, these projections highlight the urgent need to incorporate changing wave dynamics and sea level rise into adaptive coastal management strategies, especially in vulnerable locations, such as Duba, where climate-driven challenges threaten sustainability and safety.

Figure 8. Observed and projected evolution of significant wave height (SWH) across multiple time horizons under SSP1–2.6, SSP2–4.5, and SSP5–8.5 scenarios using COSMO–CLM and RegCM models.

Given these projections, adaptive strategies must include redesigning existing coastal protection structures for higher wave loads, enforcing stricter setback lines for new developments, and deploying nature-based solutions, such as resilient dune systems or oyster reefs, to dissipate wave energy [45,46]. A pilot project assessing the impact of increased wave heights on critical infrastructure such as desalination plants could test and validate mitigation strategies, drawing lessons from optimal risk-based management approaches for coastal bridges vulnerable to hurricanes [47].

3.4. Assessment of Relative Sea Level Changes

The comparative analysis of the relative sea level (RSL) in Duba highlights the excellent agreement between historical observations and simulations from the regional models COSMO-CLM and RegCM, thereby reinforcing the robustness of future projections. During the reference period, observations showed fluctuations ranging from $-3.2$ to $+2.8$ cm around the baseline level (median of approximately 0.2 cm), whereas the models faithfully replicated this natural variability, with slightly positive medians ($0.4$–$0.7$ cm). This observation-model agreement forms a firm foundation for prospective interpretation.

The arid Red Sea region presents unique challenges for RSL observation and modeling, including limited sediment supply, high evaporation rates, and tectonic influences, which can complicate accurate simulation. This excellent agreement represents a novel achievement in validating regional modeling capabilities for such environments, building on but advancing prior studies, such as [48] (assessing climate impacts on sea surface temperatures and RSL in the Arabian Gulf) and [49] (focusing on coastline retreat in rocky cliffs). This enhances the credibility of our integrated framework by providing robust, regionally tailored probabilistic projections for coupled hydrodynamic and morphodynamic outputs, particularly for coastal retreats in arid zones.

Future trajectories revealed that the level rise is dependent on the emission scenarios. The SSP1-2.6 scenario (low emissions) projected a moderate increase, reaching approximately 17–19 cm by 2071–2100, indicating a controlled rise in sea level. The intermediate SSP2-4.5 scenario indicates a more pronounced intensification, with median elevations near 30–32 cm at the end of the century. Finally, the pessimistic SSP5-8.5 scenario projects a radical transformation, with central values exceeding 64–68 cm and an accentuated intermodel spread, especially for COSMO-CLM. The widening of the box plots in Figure 9 reflects the increasing uncertainty over time horizons, confirming the importance of multi-scenario analysis.

Figure 9. Observed and projected evolution of relative sea level (RSL) across multiple time horizons under SSP1–2.6, SSP2–4.5, and SSP5–8.5 scenarios using COSMO–CLM and RegCM models.

These results are consistent with regional and global trends, as the Arabian Peninsula is identified as one of the zones most vulnerable to sea level rise [2], and GNSS data confirm significant average relative elevation rates in the western Arabian Gulf [50]. Globally, the acceleration of sea-level rise, from 1.5 mm/year (1901–1990) to 3.6 mm/year (2005–2015) [51], supports the plausibility of the simulated projections.

These developments constitute a major challenge for local coastal management, particularly in vulnerable zones such as Duba, where RSL increases exacerbate erosion, flooding, and inundation risks. Hence, integrating uncertainty ranges derived from multi-model and multi-scenario approaches is essential for designing robust adaptive strategies that conform to international recommendations for coastal resilience in climate-threatened areas.

The Delft3D model, central to our morphodynamic assessments, further indicates that changes in wave energy and sea level are key drivers of the observed shoreline retreat, particularly influencing sediment transport patterns along the coast. However, variations in sediment flux and grain size, while modeled by Delft3D, were not explored in depth in this study. Future studies should focus on the specific impact of these parameters on shoreline dynamics and their contribution to shoreline retreat models.

3.5. Bayesian Analysis

The total uncertainty for each projection was quantified as half the range between the upper and lower bounds of the 95% confidence interval derived from the posterior probability distribution for each projected variable, reflecting the range within which the true value was expected to lie with 95% certainty. In this study, the term “confidence (%)” specifically refers to the posterior coverage probability derived from the Bayesian analysis. This value represents the likelihood, quantified by the cumulative posterior probability, that the projected outcome falls within the stated confidence interval, given all sources of model and scenario uncertainty. Hence, “confidence (%)” is not a measure of model or ensemble agreement but a formal probabilistic assessment of outcome likelihood, following best practices in Bayesian climate risk analysis.

3.5.1. Relative Sea Level (RSL) Projections

The Bayesian analysis of relative sea level (RSL) projections for Duba shows a clear relationship between emission scenarios (SSPs) and the magnitude of projected rise: high-emission scenarios (SSP5-8.5) produce substantially larger mean increases than low-emission scenarios (SSP1-2.6). For example, at the long-term horizon (through 2100), the mean projection increases from 22.10 cm under SSP1-2.6 to 48.30 cm under SSP5-8.5 [52]. These differences emphasize the importance of emission trajectories for long-term coastal risks.

The uncertainty associated with these projections increases markedly with both the time horizon and the emission intensity. Figure 10 decomposes the total uncertainty into epistemic components (uncertainties owing to limited knowledge and model structural choices) and aleatory components (intrinsic variability) for each scenario–horizon combination. SSP5-8.5 long exhibited the largest total uncertainty, whereas SSP1-2.6 remained relatively constrained. Therefore, the 95% confidence intervals vary substantially by scenario (e.g., $\pm 15.80$ cm for SSP5-8.5 long vs. $\pm 6.70$ cm for SSP1-2.6) [4,53,54].

Figure 10. Decomposition of uncertainty: epistemic component and aleatory component by scenario and horizon.

Model confidence also declines with time and under more severe scenarios; for SSP1-2.6, confidence falls from approximately 88% in the short term to 76% in the long term, whereas for SSP5-8.5, it decreases from approximately 80% to 60% [52,55]. This pattern illustrates the growing limits of predictability given the complexity of climate processes and highlights the need to incorporate this uncertainty into coastal planning. Table 1 summarizes the projected net shoreline movements (mean, 95% CI, total uncertainty, and model confidence) for each scenario and horizon, providing a concise view of the practical implications of shoreline management in Duba.

Table 1. Statistical summary: net shoreline movement projections under different SSP scenarios.

Scenario	Time Horizon	Mean (m)	95% CI (m)	Total Uncert. (m)	Confidence (%)
SSP1-2.6	Short-term	$-3.20$	[$-10.60$, $4.20$]	$\pm 7.40$	72
	Medium-term	$-5.80$	[$-16.20$, $4.60$]	$\pm 10.40$	68
	Long-term	$-8.10$	[$-21.10$, $4.90$]	$\pm 13.00$	62
SSP2-4.5	Short-term	$-4.50$	[$-13.10$, $4.10$]	$\pm 8.60$	69
	Medium-term	$-9.80$	[$-22.90$, $3.30$]	$\pm 13.10$	63
	Long-term	$-16.20$	[$-33.80$, $1.40$]	$\pm 17.60$	56
SSP5-8.5	Short-term	$-5.80$	[$-15.70$, $4.10$]	$\pm 9.90$	65
	Medium-term	$-14.20$	[$-30.20$, $1.80$]	$\pm 16.00$	58
	Long-term	$-25.60$	[$-47.90$, $-3.30$]	$\pm 22.30$	49

Note: Total Uncertainty (m) represents half the width of the 95% Confidence Interval (CI).

3.5.2. Significant Wave Height (SWH)

The projected increases in the significant wave height (SWH) under different SSP scenarios (Table 2) are consistent with the established wave–climate dynamics and reflect the robustness of the integrated modeling framework. A clear gradient emerged, with higher emissions associated with greater SWH intensification, in line with global and regional studies linking greenhouse gas forcing to altered wave regimes [44,56].

Table 2. Statistical summary: significant wave height projections under different SSP scenarios.

Scenario	Time Horizon	Mean (m)	95% CI (m)	Total Uncert. (m)	Confidence (%)
SSP1-2.6	Short-term	1.35	[1.17, 1.53]	$\pm 0.18$	85
	Medium-term	1.42	[1.19, 1.65]	$\pm 0.23$	78
	Long-term	1.48	[1.20, 1.76]	$\pm 0.28$	71
SSP2-4.5	Short-term	1.38	[1.18, 1.58]	$\pm 0.20$	82
	Medium-term	1.55	[1.28, 1.82]	$\pm 0.27$	74
	Long-term	1.72	[1.38, 2.06]	$\pm 0.34$	65
SSP5-8.5	Short-term	1.41	[1.19, 1.63]	$\pm 0.22$	79
	Medium-term	1.68	[1.37, 1.99]	$\pm 0.31$	68
	Long-term	1.95	[1.55, 2.35]	$\pm 0.40$	58

Note: Total Uncertainty (m) represents half the width of the 95% Confidence Interval (CI).

Under the low-emission scenario SSP1-2.6, the SWH exhibits a modest long-term increase ($+0.13$ m), reflecting an attenuated hydrodynamic response and limited nonlinearity in wave–climate interactions, as evidenced by narrow confidence intervals (95% CI: 1.20–1.76 m) and high model confidence (71%) [4]. In contrast, the high-emission scenario SSP5-8.5 projects a more pronounced increase ($+0.54$ m by 2100, 95% CI: 1.55–2.35 m) due to tropical cyclone intensification and altered wind patterns, with larger uncertainty ($\pm 0.40$ m) and reduced confidence (58%), indicating the growing role of stochastic atmospheric processes [57,58].

Bayesian decomposition shows that epistemic uncertainty dominates early projections, whereas aleatory variability, representing intrinsic wind-wave stochasticity, increases over time, particularly under SSP5-8.5. These dynamics are captured by SWAN model outputs through wind forcing terms and validated by RegCM/COSMO-CLM replication of historical SWH distributions (Figure 8), which is consistent with non-stationary wave climate frameworks, where warming-induced pressure anomalies amplify wave energy variability in the Red Sea [59].

Projected SWH increases up to $+40$% under SSP5-8.5 are expected to significantly alter sediment transport dynamics, supporting the concept of “wave-driven erosion tipping points” for arid coasts [38] and emphasize the importance of probabilistic coastal risk assessments that incorporate full 95% confidence intervals in adaptation planning, as recommended by the IPCC AR6 [60].

3.5.3. Relative Sea Level Rise (RSL)

Uncertainty analysis reveals progressive expansion with a temporal horizon and emission intensity [4,61]. The 95% confidence interval increased from $\pm 3.50$ cm (short-term, SSP1-2.6) to $\pm 15.80$ cm (long-term, SSP5-8.5) (Table 3), reflecting the accumulation of epistemic uncertainties associated with complex climatic processes and nonlinear feedback in the Earth–ocean system [62]. This amplification of uncertainty is consistent with ice-sheet destabilization mechanisms and intensification of thermosteric processes under high radiative forcing [39,63].

Table 3. Statistical summary: relative sea level rise projections under different SSP scenarios.

Scenario	Time Horizon	Mean (cm)	95% CI (cm)	Total Uncert. (cm)	Confidence (%)
SSP1-2.6	Short-term	8.50	[5.00, 12.00]	$\pm 3.50$	88
	Medium-term	15.20	[10.00, 20.40]	$\pm 5.20$	82
	Long-term	22.10	[15.40, 28.80]	$\pm 6.70$	76
SSP2-4.5	Short-term	9.80	[5.90, 13.70]	$\pm 3.90$	84
	Medium-term	21.50	[13.70, 29.30]	$\pm 7.80$	76
	Long-term	35.40	[23.80, 47.00]	$\pm 11.60$	68
SSP5-8.5	Short-term	11.20	[6.70, 15.70]	$\pm 4.50$	80
	Medium-term	28.60	[18.70, 38.50]	$\pm 9.90$	70
	Long-term	48.30	[32.50, 64.10]	$\pm 15.80$	60

Note: Total uncertainty (cm) represents half the width of the 95% Confidence Interval (CI).

Model confidence exhibits systematic erosion, declining from 88% to 76% for SSP1-2.6, and from 80% to 60% for SSP5-8.5, between the short- and long-term horizons [52,54]. This degradation reflects the growing limits of predictability inherent to multi-decadal climate projections, particularly under extreme scenarios, where stochastic processes and potential tipping points compromise the simulation robustness [61,64]. Bayesian integration enables explicit quantification of these uncertainties, providing coastal managers with essential probabilistic estimates for adaptive planning [64]. RSL projections combined with morphodynamic (NSM) and hydrodynamic (SWH) indicators constitute an integrated framework for assessing the coastal vulnerability of Duba to future climate change [4,54].

4. Discussion

The probabilistic projections for Duba revealed an accelerating erosion trend that was starkly dependent on future emission pathways. This pattern aligns with global assessments of climate-driven sandy beach erosion [4,38] and underscores the acute vulnerability of arid coastal systems, which are often characterized by chronic sediment supply deficits [65]. In this context, it is imperative to implement anticipatory and flexible adaptation strategies. Such strategies should be designed to operate under a wide range of plausible futures rather than being based on a single design value, a concept that is central to adaptive pathway planning [66]. This flexibility is particularly crucial for Duba, where massive investments in coastal infrastructure and tourism demand solutions that can adapt to climatic uncertainties, such as threshold-based approaches, rather than fixed forecasts. The divergence between the COSMO-CLM and RegCM projections, especially over longer time horizons, reflects their contrasting sensitivities to atmospheric and ocean–atmosphere dynamics in the Red Sea [67,68]. Whereas COSMO-CLM tends to project more pronounced extremes, RegCM provides more moderate estimates. This inter-model spread, far from being a weakness, offers valuable insights into the range of plausible futures and allows for a decomposition of the overall uncertainty into its epistemic (model-related) and aleatory (natural variability) components [69]. The progressive decline in model confidence with increasing time horizons underscores the importance of multi-model ensemble approaches and the need for improved representation of regional processes, particularly wind and wave dynamics, specific to arid coasts. Our results further highlight the compounding effect of relative sea level rise and increased significant wave height. This synergy intensifies sediment transport and accelerates coastal erosion, a dynamic widely recognized in coastal science [3,70]. A sensitivity analysis was also conducted to quantify the contributions of modest changes in the key drivers of shoreline retreat. Specifically, an increase of 0.1 m in significant wave height (SWH) contributed approximately an additional 2–3 m of shoreline retreat by 2100, while a 10 cm rise in relative sea level (RSL) corresponded to roughly 4–5 m of additional retreat. These results highlight the dominant influence of sea-level rise relative to wave height increase in this coastal context, although both drivers synergistically exacerbate erosion. Understanding the relative sensitivities enables prioritization in coastal risk management and adaptation planning, emphasizing the need for both effective mitigation of sea level rise and resilience measures against an intensified wave climate. In an arid setting such as Duba, where fluvial inputs are negligible and the coastline relies primarily on longshore transport and biogenic sediment production, this synergy is particularly damaging. The structural deficit in sediment supply heightens the risk of “coastal squeeze”, where beaches and dunes are progressively narrowed and lost, especially during extreme events, with direct implications for both coastal ecosystems and human activities in the area. The projected shoreline retreat rates in Duba, particularly under high-emission scenarios, align with global trends of accelerating coastal erosion. For instance, studies on sandy coastlines worldwide indicate that many are retreating at rates exceeding 1–2 m per year under similar climate change drivers [4,38]. Specifically, the significant retreat observed in Duba (up to $-40$ m by 2100 under SSP5-8.5) is comparable to projections for highly vulnerable regions, such as parts of the Mekong Delta, where erosion rates of several meters per year are common owing to sea-level rise and reduced sediment supply [71,72]. Along the Red Sea, while comprehensive studies with probabilistic projections are less abundant, localized analyses have confirmed ongoing erosion. For example, studies on the Egyptian Red Sea coast have documented significant shoreline changes, attributing them to a combination of natural processes and anthropogenic pressures, mirroring the complex interactions at play in Duba [73]. The general consensus across the Red Sea is an increasing vulnerability to erosion, particularly in low-lying coastal areas and those with limited sediment budgets, which is consistent with our findings for the Duba coastline [74]. The projected increases in the significant wave height (SWH) in Duba are also consistent with broader global and regional patterns. Global wave climate models project an intensification of extreme wave events in many ocean basins, particularly under higher-emission scenarios [75,76]. Although the Red Sea is a semi-enclosed basin, localized studies suggest that its unique bathymetry and wind patterns can lead to significant amplification of wave energy, especially under changing wind regimes [77]. Our findings for Duba reflect a regional manifestation of a global phenomenon, highlighting the need for adaptive measures that account for increased wave loads. Furthermore, the relative sea-level rise (RSL) projections for Duba are consistent with global and regional assessments. The IPCC’s AR6 report projects a global mean sea-level rise consistent with our scenario-based estimates, with arid regions often facing additional challenges owing to high evaporation rates and localized oceanographic processes [1,60,78]. Regional studies within the Red Sea Basin also show RSL variations influenced by basin-scale ocean dynamics and tectonic activity, suggesting that while our models capture the dominant climate-driven component, site-specific factors can introduce nuances [79]. This robust agreement with global and regional literature enhances the credibility of our projections and underscores the urgency of addressing climate change impacts in Duba.

This assumption likely leads to an underestimation of actual erosion rates by 10–30% or an overestimation of coastal stability in areas prone to sediment starvation, thereby presenting a potentially less severe picture of shoreline retreat than might occur in reality for this arid coast [80,81]. This overestimation is particularly critical for arid coasts such as Duba, where a chronic sediment supply deficit already limits natural recovery, potentially leading to less conservative long-term retreat projections than if finite sediment budgets are considered. The morphodynamic simulations assume unlimited sediment availability, ignoring sediment supply limitations common on arid coasts. These dynamics are particularly significant in the context of NEOM and other mega-projects under development along the Saudi coast [82]. Projected shoreline retreat and sea level rise pose direct threats to strategic infrastructure, tourism facilities, and natural habitats, raising questions about the long-term viability of such colossal investments. From an economic perspective, potential losses can amount to several billion dollars, as demonstrated for other arid coastlines [83]. This justifies the early integration of probabilistic projections into design and planning standards to reduce risks and secure future investments. Crucially, these current natural scenarios provide a baseline for cost-benefit analyses of proposed coastal defense projects, enabling decision-makers to evaluate whether planned interventions will deliver sufficient protection and economic return against the projected natural retreat. These projections can immediately define high-risk vulnerability zones to guide the strategic placement and design of future anthropogenic interventions, informing decisions on where and when to invest in coastal protection or to retreat. Future work will integrate high-resolution geospatial data on planned human interventions, such as the NEOM mega-project’s reclamation efforts or the proposed coastal protection structures to refine these projections and assess their effectiveness. Specifically, the current resolution of regional climate models (approximately 10 km) limits the representation of fine-scale coastal processes, potentially underestimating localized impacts by 10 to 25%. Moreover, the assumption of unlimited sediment availability in morphodynamic models may lead to an underestimation of actual erosion rates by 10 to 30% in sediment-deficient areas. These limitations are critical for local planning, highlighting the need for higher-resolution modeling and the explicit integration of sediment budget dynamics in future studies. Crucially, these current ’natural scenarios’ provide a baseline for cost-benefit analyses of proposed coastal defense projects, enabling decision makers to evaluate whether planned interventions will deliver sufficient protection and economic return against the projected natural retreat. These projections can immediately define high-risk vulnerability zones to guide the strategic placement and design of future anthropogenic interventions, informing decisions on where and when to invest in coastal protection or to retreat. The methodological approach adopted, combining diachronic satellite analyses, dynamically downscaled regional climate projections, hydrodynamic modeling, and a Bayesian framework, represents a significant advancement in assessing climate impacts on coastlines. This enables departure from traditional linear extrapolations and explicitly incorporates uncertainties into future scenarios. However, this study has some limitations that must be acknowledged. The resolution of regional climate models (10 km) remains insufficient to capture local processes, such as fine-scale coastal circulation or headland effects, which may lead to an underestimation of the impacts on specific structures or microenvironments. Specifically, this resolution limitation likely leads to an underestimation of localized wave energy convergence and dissipation near headlands and a smoothing of fine-scale coastal circulation patterns, potentially underestimating actual erosion by approximately 10–25% in geomorphologically complex microenvironments, while possibly overestimating sediment dispersal in others owing to diffused forcing. This directly affects the precision required for site-specific engineering and policy-making. Morphodynamic models are based on the assumption of unlimited sediment availability, which may bias the results by overestimating the capacity of the coast for natural recovery. This assumption likely leads to an underestimation of the actual erosion rates by 10–30% or an overestimation of coastal stability in areas prone to sediment starvation, thereby presenting a potentially less severe picture of shoreline retreat than might occur in reality for this arid coast. This overestimation is particularly critical on arid coasts, such as Duba, where a chronic sediment supply deficit already limits natural recovery, potentially leading to less conservative long-term retreat projections than if finite sediment budgets are considered. Moreover, the lack of integration of human interventions (beach nourishment, coastal defenses, dredging, and reclamation) limits the scope of the projections, as these actions can significantly alter coastal responses and are not included in our natural scenarios. For the rapidly developing Duba region, this limitation means that the morphodynamic impacts of large-scale infrastructure, such as NEOM coastal urbanization or port expansions, are not captured. Consequently, the projections represent a ’natural evolution’ baseline and do not account for the ameliorating or exacerbating effects of planned or ongoing anthropogenic modifications on the shoreline dynamics. Given the expected coastal modifications of the NEOM mega-project, actual erosion may be reduced by 50% or more in directly protected areas, but conversely increased in adjacent unprotected zones due to altered wave and current dynamics. Additionally, the application of quantile mapping for bias adjustment of climate variables introduces methodological uncertainties. While effective for statistical correction, this technique assumes stationarity in the relationship between modeled and observed quantiles, which may not hold under future climate change and can potentially alter physical consistency or introduce spurious signals into extreme events when applied to multiple variables independently. Future work should incorporate high-resolution LiDAR data not only to improve the initial bathymetry and topography inputs for Delft3D but also to allow for more precise validation of localized morphodynamic changes, particularly in microenvironments that are not well resolved by the current RCM resolution. The integration of LiDAR data significantly enhanced the accuracy of identifying localized erosion/accretion patterns and improved the calibration of the sediment transport parameters in Delft3D. Episodic sediment inputs from wadis can be better characterized by exploring methodological approaches, such as integrating hydrological models for wadi discharge and sediment load, and utilizing high-frequency remote sensing data (e.g., Sentinel-2) for mapping flash flood events and associated sediment plumes. These data can then be coupled with Delft3D for dynamic sediment supply modeling. This characterization is essential for moving beyond the simplistic assumption of a chronic sediment deficit by providing a more realistic representation of intermittent sediment contributions. Coupling climate projections with anthropogenic development scenarios that are highly relevant to the Duba coast explicitly models the morphodynamic impacts of planned coastal urbanization associated with projects such as NEOM, changes in port infrastructure, or large-scale tourism development on sediment dynamics and localized vulnerability. This integration should consider the physical alterations caused by dredging, reclamation, and construction of coastal defenses (e.g., breakwaters and groins) and their subsequent feedback on sediment transport. Future research should employ nested dynamic downscaling (e.g., using RCMs at 1–2 km resolution for specific subregions) to explicitly resolve fine-scale processes and quantify their localized impact on shoreline dynamics. Furthermore, future research should explore refinements to the Delft3D morphodynamic model to better capture the unique processes of arid coastlines, such as improving the representation of aeolian sediment transport, desiccation cracks, and salt crust formation and their influence on bed shear stress and erodibility.

Finally, this study confirms the disproportionate vulnerability of arid and semi-arid coasts to climate change, highlighting the need for integrated probabilistic assessments to guide robust and sustainable coastal management. These results allow for the formulation of recommendations to enhance coastal resilience in Duba and other arid regions. The high uncertainty in projections, particularly under high-emission scenarios, justifies a “managed retreat” framework for vulnerable areas with dynamic setback lines that are regularly updated. For existing infrastructure and critical zones, adaptive strategies that combine natural solutions (dunes, mangroves) and modular artificial structures are recommended. Regular high-resolution coastal monitoring and the involvement of local stakeholders are essential to adjust measures promptly.

Uncertainty does not prevent action but rather guides the design of robust infrastructure. The 95% confidence intervals for sea level and shoreline movement define thresholds for foundations, elevation, and wave resistance. Strict zoning regulations can limit construction in high-risk areas, while modular designs and phased investments allow measures to be adapted according to climate evolution and technological advances.

5. Conclusions

This study provides a probabilistic assessment of the impact of climate change on the Duba coastline in northwestern Saudi Arabia, addressing the critical need for arid coastal systems in the Red Sea region. It integrates multi-decadal satellite observations, dynamically downscaled CMIP6 climate projections, coupled hydrodynamic and morphodynamic modeling (SWAN and Delft3D), and a Bayesian framework for uncertainty quantification, along with projections of coastal evolution up to 2100 under multiple emission scenarios.

The findings consistently reveal an accelerating trend in shoreline retreat, with the magnitude directly proportional to future emission pathways. This retreat is synergistically driven by the intertwined effects of rising relative sea levels, intensified wave dynamics, and the inherent chronic sediment deficit that is prevalent in arid coastal environments. A key insight from our probabilistic framework is the progressive decrease in model confidence over longer time horizons and under more severe emission scenarios, underscoring the inherent uncertainties in long-term climate projections and highlighting the necessity for flexible and adaptive management strategies.

These projected changes represent substantial risks to the ecological integrity, socioeconomic stability, and infrastructure resilience of the Duba coastline, particularly given the rapid regional development and ambitious mega-projects such as NEOM. Increasing marine hazards and shoreline retreat fundamentally challenge the long-term viability of current coastal infrastructure and investments. Our probabilistic approach provides decision-makers with crucial risk-based information, including clear uncertainty margins, which are indispensable for fostering sustainable planning and safeguarding future investments in this vulnerable region of the world.

This study underscores the significant value of integrating diverse modeling approaches within a robust Bayesian framework to comprehensively capture the complexities and inherent uncertainties of the coastal systems. The observed divergence between the COSMO-CLM and RegCM regional model projections offers valuable insights into the spectrum of plausible future conditions, reinforcing the critical importance of ensemble approaches in assessing the impact of climate change on coastlines.

To promote sustainable development, it is imperative to integrate these probabilistic projections into comprehensive Integrated Coastal Zone Management (ICZM) plans. Such plans should prioritize nature-based solutions, such as mangrove restoration and the establishment of artificial reefs, to enhance sediment retention and bolster biodiversity. Furthermore, policy measures should embrace flexible adaptation pathways, including strategic retreat from high-risk areas and implementation of advanced monitoring systems to continuously update projections with new data. Future research should aim to further refine these models by incorporating high-resolution anthropogenic scenarios and episodic dynamics of sediment inputs, thereby bolstering evidence-based decision-making for arid coastlines globally.

Finally, this study contributes to key United Nations Sustainable Development Goals by supporting climate action (SDG 13), the protection of coastal and marine ecosystems (SDG 14), and the promotion of resilient and sustainable communities (SDG 11). Overall, these findings bridge scientific research with practical applications, fostering sustainable development along arid coastlines.