Quantifying Tidal Asymmetry of Suspended Sediment Concentration in Macro-Tidal Embayments: A Sentinel-2 Based Framework

Highlights

What are the main findings?

• Outstanding Application Value: Fine-scale SSC dynamic reconstruction in macrotidal embayments.
• Mechanism Analysis Innovation: Conjugate regulation mechanism of sediment asymmetry in different geomorphic units.

What are the implications of the main findings?

• The core regulatory role of the “tidal pumping” mechanism is quantitatively verified. This provides spatially continuous high-precision basic data for coastal engineering planning, channel maintenance and ecological management.
• A high-resolution isomorphic spatial analysis framework is established. This study systematically reveals, for the first time, the differentiated regulatory laws between intertidal mudflats/shallow shoals and deep tidal channels. It clarifies the physical process by which tidal range and water depth modulate sediment transport through conjugate effects.

Abstract

Suspended sediment concentration (SSC) is a critical proxy for coastal water quality, geomorphological evolution, and biogeochemical cycles. In shallow macro-tidal embayments like Sanmen Bay (SMB), China, surface SSC exhibits highly dynamic spatiotemporal variations driven by intense, multi-scale tidal forcing. Using Sentinel-2 MSI imagery processed with the ACOLITE Dark Spectrum Fitting (DSF) algorithm, this study reconstructs the spatial distribution of surface SSC across the embayment. We then introduce the normalized Suspended Sediment Concentration Asymmetry Index ($A_{ssc}$) to quantitatively diagnose asymmetrical sediment responses across spring–neap and flood–ebb cycles. The results reveal a spatially divergent, dual-control mechanism governing sediment transport across the embayment’s hydro-geomorphic gradients. Quantitative trend-surface fittings and stratified regressions demonstrate that net sediment transport in deep bedrock channels is primarily governed by tidal pumping. Conversely, sediment dynamics on intertidal mudflats and shallow subtidal shoals are modulated by geomorphic resistance, exhibiting high morphodynamic sensitivity to minute water depth variations. By bridging process-based estuarine tidal theory with discrete satellite observations, this reproducible framework transforms multi-temporal remote sensing snapshots into spatially continuous diagnostics, providing a practical decision-support paradigm for coastal engineering and ecosystem management in dynamically analogous macro-tidal environments.

1. Introduction

Suspended sediment concentration (SSC) is a critical proxy for assessing estuarine and coastal water quality, regional biogeochemical cycles, and long-term geomorphological evolution [1,2]. In shallow macro-tidal embayments, the spatiotemporal distribution of surface SSC is exceptionally dynamic, regulated by the complex interplay between sediment supply and localized hydrodynamic forcing [3]. Sanmen Bay (SMB), situated on the central coast of Zhejiang Province, China, represents a quintessential semi-enclosed, tide-dominated coastal system characterized by high turbidity, expansive mudflats, and intricate subaquatic channels [4]. In such macrotidal environments, intense tidal currents control localized sediment resuspension and net transport fluxes, a process primarily governed by tidal hydrodynamics and morphodynamic feedbacks.

In macrotidal estuarine systems worldwide, tidal asymmetry—characterized by inequalities between flood and ebb durations or peak current velocities—acts as the primary physical driver of residual sediment transport [5,6]. Extensive process-based research has demonstrated that when peak flood velocities exceed ebb velocities (flood dominance), net-landward sediment transport is initiated through the tidal pumping mechanism [7]. Crucially, existing studies emphasize that this asymmetric transport is heavily modulated by non-linear sediment physics, including scour and settling lags, which systematically delay deposition and prolong particulate suspension [8]. Furthermore, in geomorphologically complex basins, lateral and vertical bathymetric gradients frequently segregate sediment transport pathways, causing highly localized and process-compensated dynamics between deep tidal channels and adjacent shallow shoals [9,10]. While advanced numerical modeling and continuous in situ high-frequency measurements (e.g., ADCP, OBS) have successfully decoded these time-dependent sediment flux variations at specific locations [3,11], their limited spatial representativeness makes it challenging to capture the continuous, whole-basin expression of sediment–response asymmetry.

To resolve this spatial limitation, satellite remote sensing has emerged as an indispensable paradigm for macro-scale, continuous aquatic observations [1,12]. Over the past few decades, low-to-medium-resolution sensors, such as MODIS or Landsat, have been extensively deployed to resolve large-scale estuarine turbidity features [1,13]. The recent advent of high-resolution land observation systems, including the Chinese GaoFen-1 (GF-1) and HaiYang-1C (HY-1C) satellites, has substantially enhanced coastal monitoring capabilities [14,15]. For instance, prior investigations demonstrated that the elevated radiometric sensitivity of the GF-1 WFV sensor effectively minimizes mixed-pixel uncertainty in narrow water corridors [16], while empirical band ratios have successfully mapped high-turbidity sediment plumes across dynamic river boundaries [17,18].

Despite this progress, a fundamental theoretical and methodological gap persists between remote sensing observations and estuarine morphodynamic theory. Existing coastal water quality remote sensing pipelines focus heavily on empirical or semi-analytical algorithm calibrations [19,20] and statistical machine-learning estimation workflows [21]. These approaches frequently treat multi-temporal satellite data as discrete, static snapshots or aggregate them into broad seasonal averages, which overlooks the short-term intra-tidal forcing. Conversely, classical tidal asymmetry diagnostics rely strictly on continuous hydrodynamic records to perform harmonic decomposition [6,22]. Consequently, a quantitative framework capable of extracting spatially continuous tidal asymmetry signatures directly from discrete satellite observations remains fundamentally lacking. Furthermore, most existing coastal sediment retrieval pipelines rely on highly localized, parameter-dependent empirical calibrations or site-specific machine-learning training datasets, which can limit their direct application to other dynamically analogous domains. Crucially, while the European Space Agency’s (ESA) Sentinel-2 Multi-Spectral Instrument (MSI) provides an optimized combination of fine spatial resolution (10 m) and a short 5-day revisit interval [20,23], its integration with advanced modeling to explicitly decouple localized sediment asymmetry from broader environmental backgrounds remains largely underexplored.

To address this specific limitation, this study presents a structured remote sensing diagnostic framework to analyze surface SSC asymmetry in SMB using an extensive Sentinel-2 MSI data archive [14,24]. The methodological repeatability of the proposed framework is rigorously secured by an automated, parameter-free preprocessing workflow utilizing standard open-access products. Meanwhile, the normalized mathematical structure of the Suspended Sediment Concentration Asymmetry Index ($A_{ssc}$) effectively mitigates regional baseline concentration shifts, thereby offering a conceptual basis that may facilitate its potential application in other macro-tidal environments. To ensure high physical fidelity, this study adopts a physically interpretable diagnostic paradigm specifically optimized for the unique hydro-geomorphic complexities of macro-tidal embayments. The specific objectives are as follows: (1) to construct and validate an optimized, structurally repeatable SSC retrieval model based on physically sensitive NIR bands; (2) to map the continuous, fine-scale spatial distribution of surface SSC under distinct tidal phases (spring–neap and flood–ebb cycles), highlighting the performance of 10 m imagery in resolving subaquatic channels and intertidal mudflats; and (3) to apply $A_{ssc}$ to quantitatively diagnose the spatially divergent sediment response regimes driven by localized tidal forcing and topographic constraints. Ultimately, this framework provides a practical approach to help connect process-based estuarine concepts with large-scale Earth observations, utilizing discrete satellite snapshots to support spatially continuous morphodynamic analysis.

2. Materials and Methods

2.1. Study Area

Sanmen Bay (SMB) is located on the central coast of Zhejiang Province, China (Figure 1). It is a typical semi-enclosed, tide-dominated embayment characterized by a barrier island at its mouth, expansive macro-tidal waters, and an intricate network of tidal inlets. The bay features extensive muddy tidal flats that exhibit distinct geomorphological zonation and clear sedimentary facies differentiation. Prominent subaquatic topographic features, such as the Maotou Deep Pool, are strictly governed by localized tidal dynamics and exhibit pronounced negative topographic relief.

Since the 1960s, continuous, large-scale reclamation projects have significantly altered the shoreline configuration and compressed the intertidal flat area [25,26]. The hydrodynamic environment within SMB is predominantly driven by tides, which play a critical role in maintaining channel stability and modulating suspended sediment transport. Hydrodynamic studies indicate that the tidal prism of SMB decreased by approximately 16.6% between 2003 and 2020 as a direct consequence of ongoing coastal reclamation activities [4]. Sediments in SMB consist primarily of fine-grained suspended particles originating from tidal flat erosion and open-sea inputs, exhibiting substantial spatiotemporal variability [27]. Under normal meteorological conditions, the SSC in the bay remains relatively high and demonstrates clear tidal-cycle fluctuations [28]. To comprehensively investigate these dynamics, multi-source datasets including satellite imagery, hydrological station measurements, and in situ SSC data were integrated in this study (

Table 1. Data sources.

Type	Source
Sentinel-2 MSI L1C	European Space Agency (ESA)
Sentinel-3 OLCI Level-2	European Space Agency (ESA)
SSC measured data	Second Institute of Oceanography, MNR
Hydrological measurement site data	Second Institute of Oceanography, MNR

).

2.2. ACOLITE Image Preprocessing

The overall technical flowchart of this study, encompassing atmospheric correction, cross-sensor validation, and empirical SSC modeling, is illustrated in Figure 2. The atmospheric correction of Sentinel-2 MSI L1C imagery was executed using ACOLITE (version 20260421) [29]. Developed by the Royal Belgian Institute of Natural Sciences (RBINS), ACOLITE incorporates the Dark Spectrum Fitting (DSF) algorithm, which is highly optimized for water color applications over highly turbid inland and coastal waters [30,31].

To streamline the processing framework, default climatological values for atmospheric profiles (i.e., ozone: 0.30 cm·atm, water vapor: 1.50 g·cm⁻², and surface pressure: 1013.25 hPa) were applied. The image was segmented into 19 × 19 tiles (each 600 × 600 pixels). Within each tile, the DSF independently identified dark pixels and estimated the aerosol optical thickness (AOT) based on the “minimum root mean square deviation” criterion. For specific bands exhibiting strong gas absorption (e.g., Band 9 and Band 10), masking criteria were applied (invalidated if transmittance < 0.70). Furthermore, residual direct sun glint was estimated and subtracted by analyzing residual reflectance in the shortwave infrared (SWIR) or NIR bands using the built-in wave slope model. Detailed mathematical formulations and algorithmic rationales for these correction steps can be found in the foundational studies by Vanhellemont and Ruddick [30,31,32]. Finally, all relevant L2W-level remote sensing reflectance ($R_{rs}$) bands were resampled to a uniform spatial resolution of 10 m.

2.3. Cross-Sensor Consistency Assessment of Atmospheric Correction

To evaluate the operational applicability of the ACOLITE Dark Spectrum Fitting (DSF) algorithm for Sentinel-2 MSI imagery over SMB, nearly synchronous Sentinel-3 OLCI Level-2 full-resolution water color products (OL2-WFR) were utilized as an independent reference for a band-by-band cross-sensor comparison [24,33]. To mitigate uncertainties arising from the spatial resolution gap between Sentinel-2 (10 m) and Sentinel-3 OLCI (300 m), a rigorous spatial up-scaling procedure was applied. Specifically, the 10 m Sentinel-2 imagery was spatially resampled t300 m00-m grid using ENVI 5.6 prior to comparison. This pixel aggregation ensured footprint alignment and effectively eliminated representation errors caused by sub-pixel spatial heterogeneity, allowing the radiometric consistency evaluation to be executed on a strictly unified spatial scale. Band pairing was performed based on a nearest-central-wavelength matching scheme. To establish robust spectral relationships while minimizing the influence of residual cloud shadows, adjacency effects, and other outliers, the Random Sample Consensus (RANSAC) robust regression algorithm was applied (Figure 3, Figure 4 and Figure 5).

It must be emphasized that this assessment represents a cross-sensor consistency analysis rather than a strict ground-truth validation. A critical source of uncertainty in this comparison arises from the different atmospheric correction methodologies employed. While the Sentinel-2 data were processed using the ACOLITE DSF algorithm, the Sentinel-3 official Level-2 products rely on a distinct standard atmospheric correction baseline. This algorithmic divergence introduces an additional source of systematic error when directly comparing the reflectance outputs. OLCI-derived reflectance products are inherently subject to atmospheric correction uncertainties, particularly in optically complex and highly turbid coastal waters where aerosol characterization and adjacency effects remain challenging. Nevertheless, Sentinel-3 OLCI Level-2 products have been widely validated and adopted in coastal water-quality studies. They demonstrate reliable performance in highly turbid environments, particularly within the red to near-infrared spectral regions [34,35]. Therefore, in the absence of concurrent in situ radiometric measurements, cross-sensor comparisons between Sentinel-2 MSI and Sentinel-3 OLCI remain a scientifically acceptable and practical approach for evaluating relative radiometric consistency [36].

Despite the inherent uncertainties associated with satellite-derived reference data and differing atmospheric correction algorithms, the comprehensive cross-sensor comparison confirms that the ACOLITE-derived reflectance achieved satisfactory consistency with the official OLCI ocean color products across the red to near-infrared bands (Figure 3, Figure 4 and Figure 5). This robust radiometric consistency provides a highly reliable foundation for subsequent regional SSC inversion.

2.4. Modeling Strategy and Match-Up Dataset

The remote sensing reflectance ($R_{rs}$, sr⁻¹) derived from Sentinel-2 MSI imagery via ACOLITE atmospheric correction was utilized as the primary input to establish a SSC (mg/L) retrieval model using an empirical regression approach. The modeling dataset comprised three images from distinct temporal phases (15 January 2021; 25 July 2017; 2 September 2021) coupled with synchronous in situ SSC measurements. Following quality control and removal of outliers based on plausibility screening, 30 valid match-up sample pairs were retained, with SSC values ranging from approximately 300 to 1000 mg/L, representative of typical highly turbid waters. The in situ data were collected during three hydrological surveys within a one-hour temporal window relative to the satellite overpasses. To mitigate the impacts of geometric positioning errors and adjacent pixel heterogeneity, the satellite reflectance at each sampling coordinate was extracted as the median value within a 3 × 3 pixel window centered on the target point. The dataset was randomly partitioned into a training set (n = 19) and an independent validation set (n = 11) at a ratio of approximately 65:35. Furthermore, a LOOCV approach was implemented to robustly assess the final model’s generalization capability under limited sample conditions.

2.5. Physical Justification for Band Selection

Band selection for SSC retrieval was determined a priori based on established bio-optical principles of highly turbid waters rather than solely through empirical statistical fitting. In macro-tidal embayments such as SMB, where SSC frequently exceeds 300 mg/L, the optical response of the water column is dominated by suspended particulate matter and strong multiple scattering effects. Under such conditions, remote sensing reflectance in the blue and green spectral regions becomes progressively less sensitive to SSC variations, as these wavelengths are heavily absorbed by colored dissolved organic matter (CDOM) and tend to reach signal saturation at high turbidity levels. Previous studies in highly turbid estuaries have explicitly shown that visible bands, particularly in the blue-green range, lose retrieval sensitivity rapidly as suspended sediment concentration increases [37].

In contrast, the red and near-infrared (NIR) wavelengths retain strong sensitivity to sediment concentration over a substantially wider SSC range. As suspended particulate matter increases, particulate backscattering in the red-edge and NIR spectral regions rises significantly. Meanwhile, the extremely high absorption coefficient of pure water in the NIR region effectively suppresses background water-column reflectance. This optical mechanism dramatically enhances the sensitivity to particle-induced backscattering and delays spectral saturation. Numerous studies have demonstrated that the red to NIR bands provide superior robustness for suspended sediment retrieval in highly turbid and sediment-rich coastal waters, especially under extreme SSC conditions [38].

Based on these physical and optical constraints, Sentinel-2 MSI bands B4 (665 nm), B5 (704 nm), B6 (740 nm), and B8A (865 nm) were pre-selected as the core candidate bands for SSC inversion. Notably, our prior cross-sensor comparison (Section 2.3) also confirmed that these specific physically relevant bands possess robust radiometric fidelity following ACOLITE atmospheric correction.

2.6. Hydrodynamic Simulation

To simulate the hydrodynamic conditions in SMB, the DHI MIKE 21 model (MIKE Zero 2024; DHI, Hørsholm, Denmark) [39], which solves the depth-integrated Navier–Stokes equations using an unstructured triangular mesh framework, was employed. The computational domain covered the entire bay and adjacent regions of the East China Sea (Figure 6). Open ocean boundaries were prescribed using tidal forcing composed of eight major harmonic constituents, including $M_2$, $S_2$, $K_1$, $O_1$, $N_2$, $K_2$, $P_1$, and $Q_1$, derived from TOPEX altimetry datasets. A wetting–drying scheme was implemented to accurately resolve the extensive intertidal flats within SMB.

A continuous 15-day simulation was performed for January 2021 to ensure reliable harmonic decomposition of the dominant semidiurnal tidal signal. The simulation duration was selected based on the Rayleigh resolution criterion, which is commonly utilized in tidal harmonic analysis. Because the synodic period between the principal lunar semidiurnal constituent ($M_2$; 12.42 h) and the principal solar semidiurnal constituent ($S_2$; 12.00 h) is approximately 14.77 days, a continuous record exceeding this duration is strictly required to resolve these two constituents independently and avoid spectral leakage associated with spring–neap modulation. This criterion is widely adopted in classical tidal analysis and coastal hydrodynamic studies [40,41].

Model performance was evaluated using contemporaneous field observations collected in SMB (Figure 7). Statistical comparisons, as detailed in

Table 2. Model validation indices.

Type	R2	RMSE	MAE	Skill	Bias
Water level	0.952	0.245 m	0.195 m	0.949	0.022 m
Current direction	0.816	40.431°	21.277°	0.807	6.958°
Current velocity	0.749	0.184 m/s	0.133 m/s	0.650	0.008 m/s

, indicate high explanatory power for water level ($R^2=0.9516$, $\mathrm{Skill}=0.9488$, and $\mathrm{Bias}=0.0219\mathrm{m}$). Simulated tidal currents also demonstrated satisfactory agreement with observations, with current velocity and direction yielding $R^2$ values of 0.7488 and 0.8161, and Skill scores of 0.6497 and 0.8072, respectively. Systematic bias remained marginal across all variables, indicating that the calibrated model successfully reproduced the principal hydrodynamic characteristics of SMB, thereby providing a reliable basis for subsequent tidal asymmetry analysis.

2.7. Image Selection and Tidal Phase Classification

To systematically decode the spatial distribution characteristics of SSC under various tidal conditions, eight high-quality, cloud-free Sentinel-2 images were meticulously selected based on temporal coverage and tidal phases (

Table 3. Sentinel-2 imagery used in this study and the corresponding tidal information.

Image Date	Sensor	Tidal Phase	Tide Type	Analysis Purpose
15 February 2017	S2A	Flood	Middle	Flood–ebb cycle
18 September 2017	S2A	Ebb	Middle	Flood–ebb cycle
30 April 2021	S2A	Flood	Spring	Flood–ebb cycle/spring–neap cycle
1 December 2021	S2B	Ebb	Middle	Flood–ebb cycle
6 December 2021	S2B	Ebb	Neap	Spring–neap cycle
4 January 2025	S2B	Flood	Middle	Flood–ebb cycle
30 November 2025	S2C	Flood	Neap	Spring–neap cycle
15 December 2025	S2C	Ebb	Spring	Flood–ebb cycle/spring–neap cycle

). The selected imagery archive encompasses discrete acquisition dates from 2017, 2021, and 2025. To ensure the rigorous classification of the hydrodynamic background for each scene, local astronomical tide tables and concurrent water level observations were utilized. The classification criteria are defined as follows:

Flood/Ebb Classification: An image was classified as a “flood” phase if the acquisition time occurred within the interval from the lower low water to the subsequent higher high water, characterized by landward net flow. Conversely, it was classified as an “ebb” phase if acquired between the high water and the subsequent low water.

Spring/Neap Classification: The spring–neap cycle was defined based on the lunar phase and the corresponding tidal amplitude. “Spring” tides correspond to periods immediately following the new or full moon, characterized by maximum tidal ranges. “Neap” tides correspond to the first and third quarter moons (quadrature), characterized by minimum tidal ranges. “Middle” tides represent the transitional phases between spring and neap cycles.

Although Figure 8 displays continuous multi-year tidal level process curves with numerous potential satellite overpass times, the analyses in this study focus strictly on the eight representative scenes detailed in Table 3. These specific scenes were selected because they are entirely free of cloud cover over the bay area and closely coincide with the peak velocities of their respective tidal phases, thereby maximizing the signal-to-noise ratio for identifying tide-induced sediment resuspension.

2.8. Definition of the SSC Asymmetry Index and Spatial Coupling Framework

To quantitatively evaluate the tidal asymmetry of sediment dynamics, the Suspended Sediment Concentration Asymmetry Index ($A_{ssc}$, dimensionless) was formulated as follows:

$$A_{ssc}={\displaystyle \frac{{\overline{SSC}}_{flood}-{\overline{SSC}}_{ebb}}{{\overline{SSC}}_{flood}+{\overline{SSC}}_{ebb}}}$$

(1)

where ${\overline{SSC}}_{flood}$ and ${\overline{SSC}}_{ebb}$ denote the mean surface SSC ($\mathrm{mg}\cdot {\mathrm{L}}^{-1}$) derived during the flood and ebb tidal phases, respectively. The value of $A_{ssc}$ is strictly bounded within the interval [−1,1]. Fundamentally, an $A_{ssc}>0$ indicates that the mean surface sediment concentration during the flood phase exceeds that during the ebb phase. Hydrodynamically, this denotes that landward tidal pumping or localized flood-dominant resuspension governs the system, driving net sediment trapping within the inner bay and prompting morphological accretion. Conversely, an $A_{ssc}<0$ signifies elevated mean sediment concentrations during the ebb phase, reflecting a robust, ebb-dominant scouring mechanism that effectively flushes suspended particulates seaward. When $A_{ssc}$ approaches zero, the sediment transport capacities induced by flood and ebb tidal currents reach a dynamic equilibrium, designating a transient stability zone of regional sediment flux.

The unstructured computational mesh of the MIKE 21 hydrodynamic model inherently mismatches the regular grid array of the Sentinel-2 MSI observations. To establish an isomorphic spatial database, a robust geo-spatial downscaling framework was executed. The mean tidal range ($MTR$) was defined as a spatially varying climatological hydrodynamic parameter rather than a single image-wide value. It was quantified by capturing and averaging the water level differences across the continuous 15-day complete spring–neap simulation record at each individual mesh node.

Subsequently, a Delaunay triangulation-based linear N-dimensional interpolation algorithm was employed to map the node-based MTR and bathymetric depth (h) fields onto the 10 m coordinate grid of the Sentinel-2 imagery. After applying valid water masking and physical range filters ($-0.75\le A_{ssc}\le 0.75$; $2.5\mathrm{m}\le MTR\le 3.5\mathrm{m}$), a synchronous dataset comprising over $1.51\times {10}^7$ valid sub-tidal and intertidal pixels was generated. This pixel-wise association framework directly coupled the multi-year composite sediment response with the spatial hydrodynamic background across the entire embayment conduit.

To visualize and quantify the joint influence of hydrodynamic forcing (Mean Tidal Range, $MTR$) and topographic constraints (Water Depth, h) on SSC asymmetry ($A_{ssc}$), a three-dimensional spatial analysis framework was employed. Visually, the 3D trend surfaces were generated using a 2D spatial binning technique, where the continuous MTR and h datasets were discretized into 25 × 25 regular grids, and the mean $A_{ssc}$ was calculated for each grid cell to construct the smoothed response surface. Statistically, the multi-variable coupling strength within each geomorphic zone was quantified using a multiple linear regression framework, yielding the specific coefficient of determination ($R^2$) and partial correlation coefficients ($r_{partial}$) to diagnose the spatial variance explained by these physical controls.

3. Results

3.1. Performance and Applicability of the SSC Retrieval Model

3.1.1. Spectral Feature Engineering and Optimal Model Determination

Building upon the physical justification of the red and NIR bands, the next step involved determining the optimal mathematical architecture for the retrieval model. To fully exploit the spectral response of highly turbid waters, 53 candidate spectral features (

Table 4. Candidate spectral feature combinations.

Number of Bands	Combination Form
Single-band	$B_i$
Dual-band	$B_i/B_j,$$B_i-B_j,$$B_i\cdot B_j,$$\left(B_i-B_j)/(B_i+B_j\right)$
Three-band	$B_i/(B_j+B_k),$$(B_i-B_j)/B_k,$$B_i\cdot B_j/B_k$
Four-band	$\left(B_i+B_j)/(B_k+B_l\right),$$\left(B_i-B_k)/(B_j+B_l\right)$

) were constructed based on four core bands, and four regression variants (

Table 5. Candidate regression models.

Fitting Model	Mathematical Form
Linear	$SSC=aX+b$
Logarithmic	$SSC=a\mathrm{l}\mathrm{n}X+b$
Exponential	$SSC=\mathrm{e}\mathrm{x}\mathrm{p}(aX+b)$
Power-law	$SSC=\mathrm{e}\mathrm{x}\mathrm{p}\left(b\right)\cdot X^a$

) were evaluated for each feature. To mitigate multicollinearity-induced model redundancy, a rigorous feature selection procedure was executed: features exhibiting a pairwise Pearson correlation coefficient of $\left|r\right|>0.8$ were classified as collinear, and only the feature demonstrating the highest correlation with in situ SSC was retained from each collinear group.

The evaluation revealed that the optimal retrieval model was based on the Near-Infrared (NIR) band difference feature ($R_{rs,\mathrm{B}6}-R_{rs,\mathrm{B}8\mathrm{A}}$), utilizing a linear fitting approach. This model achieved remarkable accuracy, yielding an $R^2$ of 0.923 and an RMSE of 88.36 mg/L on the training set, and an $R_{\mathrm{val}}^2$ of 0.918 and an RMSE of 90.26 mg/L on the independent validation set. The LOOCV results ($R_{\mathrm{LOOCV}}^2=0.911$, ${RMSE}_{\mathit{LOOCV}}=94.68$ mg/L) further confirmed the absence of significant overfitting. The governing equation for the SSC retrieval is given as:

$$SSC=90805.29\cdot (R_{rs,\mathrm{B}6}-R_{rs,\mathrm{B}8\mathrm{A}})-15.48$$

(2)

To further demonstrate that the random partitioning did not introduce significant bias and to verify the overall feasibility of the established model, the derived retrieval equation (Equation (2)) was applied to the entire dataset (N = 30). As illustrated in the comprehensive evaluation matrix (Figure 9), the full-sample evaluation yielded highly consistent results, achieving an overall $R^2$ of 0.921 and an RMSE of 89.06 mg/L. Furthermore, the model demonstrated excellent predictive capability and low systematic error, characterized by a Mean Absolute Error (MAE) of 69.30 mg/L, a high Willmott’s Skill score of 0.980, and a marginal positive bias of 5.04 mg/L, the Mean Absolute Percentage Error (MAPE) was evaluated, yielding 37.6%, 32.0%, and 35.5% for the training, validation, and full-sample datasets, respectively. These supplementary multi-metric evaluations comprehensively prove that the selected spectral feature and the derived empirical formula are structurally stable and free from random partitioning biases, ensuring highly reliable regional SSC retrieval.

3.1.2. Physical Basis and Applicability

The robust sensitivity of the NIR band difference to SSC is rooted in the distinct optical properties of highly turbid waters: particulate backscattering at 740 nm increases markedly with elevated SSC, whereas the strong absorption of pure water at 865 nm (>5 m⁻¹) suppresses excessive background reflectance growth. This contrasting spectral behavior establishes a stable, monotonic relationship with sediment concentration [37,38].

Sentinel-2A, Sentinel-2B and Sentinel-2C employ nearly identical MSIs with harmonized radiometric calibration, and previous validation studies have demonstrated excellent inter-platform radiometric consistency suitable for quantitative time-series applications, with residual inter-satellite differences generally remaining within a few percent [42,43].

Furthermore, cross-sensor validation between Sentinel-2A and Sentinel-2B images acquired five days apart yielded an acceptable $R^2$ of 0.60 (Figure 10). Considering the intense tidal modulation of SSC in macro-tidal estuaries, this level of agreement is highly reasonable. Short-term hydrodynamic variability—such as intra-tidal resuspension and advection—can introduce substantial natural variance that operates independently of model performance [11,44]. This observed correlation suggests that the proposed model is capable of consistently reproducing the macroscopic spatial distribution patterns of SSC (i.e., distinguishing persistent high-turbidity zones from low-turbidity areas) under varying hydrodynamic backgrounds.

3.2. Spatial and Tidal Distribution Characteristics of SSC

By matching Sentinel-2 imagery with reconstructed hourly tidal level data, the spatial heterogeneity of SSC during varying tidal phases was systematically mapped. To rigorously eliminate artifacts induced by varying spatial backgrounds, a common spatial footprint masking strategy was employed, ensuring that statistical comparisons were conducted strictly on overlapping geographic pixels with valid physical SSC values.

Based on this isomorphic spatial evaluation (Figure 11a,b and Figure 12), the average SSC within SMB during the flood tide was $608.19\pm 152.60$ mg/L. Through a rigorous paired spatial analysis, this concentration was proven to be systematically and statistically significantly higher than the $598.24\pm 144.39$ mg/L observed during the ebb tide (paired two-sample t-test, $t=5.19$, $p<0.001$). Interestingly, this asymmetric SSC accumulation pattern contrasts directly with the “inflow-north, outflow-south” residual hydrodynamic environment. This discrepancy arises because the intense flow velocities in the northern channels facilitate rapid water exchange, allowing sediment to cycle continuously, whereas topographical constraints in the south promote prolonged sediment retention.

Furthermore, extreme concentration variations were documented between the spring and neap cycles. The mean SSC during spring tides was significantly elevated, reaching approximately 2.07 times the concentration observed during neap tides (spring mean: 698.72 mg/L vs. neap mean: 337.71 mg/L). This robust statistical contrast, visually corroborated by the density distributions (Figure 11d–i and Figure 12), underscores the profound regulatory impact of intensified hydrodynamic energy inputs on localized sediment resuspension.

3.3. Analysis of Hydrodynamic Field Characteristics

To decipher the tide-induced flow field characteristics across the entire embayment (Figure 13), the MIKE 21 hydrodynamic model was deployed for a continuous 365-day two-dimensional simulation. Harmonic analysis of the principal M₂ tidal constituent (

Table 6. Statistical parameters of tidal ellipses at the bay mouth cross-section.

Parameter	Mean	Standard Deviation	Minimum	Maximum
Semi-major axis $L_{\mathrm{m}\mathrm{a}\mathrm{j}}$ (m/s)	0.678	0.124	0.390	0.898
Semi-minor axis $L_{\mathrm{m}\mathrm{i}\mathrm{n}}$ (m/s)	0.053	0.025	0.001	0.103
Ellipticity $e$	0.075	0.031	0.001	0.154

) revealed that the tidal ellipse morphology at the bay mouth is predominantly elongated (Figure 13a). The cross-sectional average ellipticity was remarkably low (0.075), indicating that the tidal regime is dominated by strong rectilinear (reversing) flows rather than rotational currents (Figure 13d,e). The semi-major axis exhibited a distinct lateral gradient, peaking near Nantian Bay (0.898 m/s), identifying the deep channel as the core conduit for tidal exchange (Figure 13b).

The Eulerian residual current, which dictates the net transport of fine suspended particulate matter, unveiled a highly stable spatial structure. The annual simulation demonstrated negative values (down to −0.109 m/s, directed offshore) on the southern side of the transect and positive values (up to 0.242 m/s, directed inshore) on the central-northern side (Figure 13c). This distinct and temporally persistent “tidal pumping” residual circulation is likely governed jointly by topography-induced nonlinear tidal effects and the Coriolis force, establishing a persistent dynamic mechanism for net sediment transport within the bay. The observed higher surface SSC during the flood phase is explicitly substantiated by two synchronized factors: (1) intensified flood tidal currents within the northern channels effectively lower the resuspension threshold, triggering massive sediment erosion from the seabed; and (2) suspended sediment retained in the southern shallow areas during the slack water phase is insulated from seaward flushing due to the significantly weakened velocity profiles of the subsequent ebb tide, leading to lower ebb SSCs.

3.4. Spatial Pattern of SSC Asymmetry Index ($A_{ssc}$) and Multi-Variable Coupling

High-spatial-resolution spatial characterization revealed an overarching depositional trend across SMB, with the index gradually decreasing from the inner bay toward the outer bay and from south to north (Figure 14a). To decouple the interactions among hydrodynamics, bathymetry, and sediment transport, a high-resolution isomorphic spatial analysis framework was established (Figure 14b,c). To further quantitatively diagnose the variation patterns of $A_{ssc}$ under different hydrodynamic and topographic conditions, a stratified binned trend fitting was performed across four distinct geomorphic zones (Figure 15 and Figure 16). The linear regression results indicated distinct morphodynamic sensitivities across different geomorphic zones:

1. Intertidal Mudflats ($h<1$ m) and Shallow Shoals ($1\le h<3$ m): Based on the full-year simulation of the regional hydrodynamic environment, the 2D linear regression between MTR and $A_{ssc}$ in the intertidal mudflats ($h<1$ m) shows a slope of 0.051 and an $R^2$ of 0.007 ($N>2.87\times {10}^4$), with a $p$-value of 0.688 (Figure 15a). The lack of a significant linear relationship suggests that in these shallow environments (Figure 15a,b), the influence of tidal amplitude is modulated by factors such as geomorphic resistance, bottom friction, and wetting-drying cycles.

2. Transitional Slopes ($3\le h<10$ m) and Deep Tidal Channels ($h\ge 10$ m): In the deep tidal channels ($h\ge 10$ m), the 2D linear regression yields a slope of 0.428 and an $R^2$ of 0.483 ($p<0.001,N>6.10\times {10}^6$). This demonstrates a statistically significant positive linear correlation between MTR and $A_{ssc}$. The data indicates that in deeper water where topographic constraints are reduced (Figure 15c,d), tidal pumping is the primary factor governing the spatial distribution of sediment asymmetry. The 3D surface trend metrics remain consistent across the analyzed annual period (Figure 16).

4. Discussion

4.1. Hydrodynamic Mechanisms Driving SSC Asymmetry: Tidal Pumping vs. Geomorphic Resistance

Although the absolute difference in mean SSC between flood and ebb phases appears relatively small, its morphodynamic implications are profoundly significant. In macro-tidal embayments, the tidal prism—the volume of water exchanged during each tidal cycle—is immense. When this seemingly minor concentration asymmetry is multiplied by the massive tidal volume and integrated over hundreds of tidal cycles annually, it translates into a substantial net sediment flux. Consequently, this persistent, albeit small, tidal pumping asymmetry acts as the primary driver for long-term morphological evolution, dictating the continuous net import of suspended sediment and the subsequent accretion of intertidal mudflats within the bay.

The weak statistical relationships observed between suspended sediment asymmetry and hydrodynamic/topographic controls do not indicate the absence of physical coupling, but rather reflect the intrinsically nonlinear, spatially heterogeneous, and process-compensated nature of sediment transport in macrotidal embayments. Similar weak but physically meaningful correlations have been widely reported in estuarine systems where multiple flood–ebb asymmetry mechanisms coexist and partially offset each other within a single tidal cycle [7].

In macro-tidal environments such as SMB, suspended sediment dynamics are not governed by a single hydrodynamic control, but by the competition between tidal pumping processes in channels and geomorphic resistance on intertidal shoals. In deep tidal channels, net sediment transport is primarily controlled by tidal pumping mechanisms arising from asymmetries in turbulence, stratification, and sediment availability (Figure 15c,d and Figure 16c,d). Enhanced flood-phase turbulence and reduced stratification can promote sediment resuspension and landward sediment flux even under weak mean advective circulation, as demonstrated in partially mixed estuaries where flood-dominant mixing leads to net up-estuary sediment pumping [7]. Similar channel-dominated flood–ebb partitioning has been widely observed in tide-dominated estuaries and is often associated with secondary estuarine turbidity maxima formation [9,10].

In contrast, shallow tidal flats and shoal environments exhibit fundamentally different dynamics. Here, sediment transport is strongly constrained by water depth, bed exposure duration, and wave–current interaction, leading to what can be interpreted as geomorphic resistance dominance. Small variations in depth substantially modify bed shear stress and resuspension thresholds, causing nonlinear sediment responses that are not linearly proportional to tidal forcing (Figure 15a,b and Figure 16a,b). This behavior is consistent with previous studies showing that shallow systems exhibit strong sensitivity to depth-controlled stress modulation and storage-limited sediment availability, resulting in highly intermittent erosion–deposition cycles [5,6,8].

The quantitative fitting based on a full-year hydrodynamic simulation demonstrates a fundamental shift in statistical relationships along the depth gradient. In the shallow intertidal mudflats ($h<1$ m), the linear correlation between MTR and SSC asymmetry is not statistically significant ($R^2=0.007,p=0.688,N>2.87\times {10}^4$). This indicates that in extremely shallow zones, sediment dynamics are not linearly proportional to tidal amplitude variations, as geomorphic resistance and bottom friction dominate the physical processes. Conversely, in the deep tidal channels ($h\ge 10$ m), the regression slope increases to 0.428 with an $R^2$ of 0.483 ($p<0.001,N>6.10\times {10}^6$). This mathematically indicates that in deep channels, water depth ceases to act as a limiting frictional constraint, and sediment dynamics exhibit a direct linear response to the energy of tidal pumping.

When aggregated at the whole-bay scale, the coexistence of these competing asymmetry mechanisms (nonlinear geomorphic resistance in shallows versus linear tidal pumping in deep channels) exhibits partial spatial cancelation effects. This scale-dependent behavior produces variations in global statistical correlations, consistent with findings from other tide-dominated estuaries where lateral and vertical bathymetric gradients segregate sediment flux pathways [9]. The data indicates that MTR and water depth act as boundary constraints, while the local geomorphic setting determines the dominant sediment response regime. This dual-control structure provides the physical basis for the spatial heterogeneity in SSC asymmetry across SMB (Figure 17).

4.2. Theoretical Basis and Diagnostic Advantages of the $A_{ssc}$ Framework

Estuarine morphodynamics heavily relies on evaluating tidal asymmetry. Traditional hydrodynamic indices—such as velocity asymmetry ($\left(V_{flood}-V_{ebb}\right)/\left(V_{flood}+V_{ebb}\right)$) or harmonic overtide ratios (e.g., $M_4/M_2$)—are well established for interpreting nonlinear tidal propagation [5,6]. However, these metrics fundamentally depend on continuous time-series data and only infer sediment transport indirectly. Because sediment dynamics are modulated by nonlinear processes like resuspension thresholds, settling lag, and turbulence–sediment feedbacks, hydrodynamic asymmetry does not always translate linearly into suspended sediment asymmetry [8].

Similarly, conventional sediment asymmetry assessments often utilize point-based net sediment flux integrations ($\int U\cdot SSCdt$) [7] or simple absolute concentration differences ($SS{C}_{flood}-SS{C}_{ebb}$). While net flux integrations provide precise localized mechanisms, they are constrained by limited spatial coverage and cannot be easily scaled for large-scale coastal mapping. Furthermore, absolute concentration differences are heavily biased by the background turbidity magnitude, making cross-regional comparisons unreliable.

To address these limitations, the $A_{ssc}$ offers a complementary perspective by directly quantifying sediment-response asymmetry from multi-temporal snapshot satellite observations. By adopting a normalized difference formulation, the $A_{ssc}$ framework completely isolates the structural asymmetry of the sediment response from the magnitude of absolute background turbidity. Confining the index to a strictly bounded range ([−1,1]) guarantees robust statistical comparability across complex geomorphic gradients—from highly turbid deep channels to lower-turbidity shoals.

More importantly, unlike traditional station-based indicators, $A_{ssc}$ enables the generation of spatially explicit, high-resolution (10 m) asymmetry maps. This facilitates the direct identification of geomorphologically controlled heterogeneity, aligning perfectly with recent advances that emphasize the critical need for spatially continuous sediment transport diagnostics in active estuaries [9,10].

It should be emphasized that $A_{ssc}$ is not intended to replace physically based tidal harmonic or net flux decomposition frameworks. Rather, it serves as a specialized diagnostic indicator tailored for remote sensing applications where continuous hydrodynamic observations are unavailable. Ultimately, its strength lies in bridging process-based estuarine theory with large-scale Earth observation capabilities, transforming discrete satellite snapshots into spatially continuous morphodynamic diagnostics.

4.3. Compounding Impacts of Seasonal Runoff and Interannual Variability

In macro-tidal and river-influenced coastal systems, SSC variability often arises from the superposition of tidal, seasonal, and interannual processes [11]. However, the physical characteristics of Sanmen Bay (SMB) inherently decouple its sediment dynamics from terrestrial and seasonal interferences. Morphologically, SMB lacks major river inputs, and upstream cascade reservoirs (e.g., Baixi Reservoir) effectively trap episodic terrestrial sediment [45,46]. Additionally, the bay’s semi-enclosed geometry buffers the energetic anomalies of winter monsoons [47]. Together, these features ensure that localized mean tidal range ($MTR$) and topographic depth (h) function as the primary boundary conditions governing sediment transport.

The minimal impact of these compounding temporal factors is directly corroborated by our empirical results. If seasonal runoff or interannual meteorological noise dominated the system, multi-year satellite compositing would yield highly scattered and random data points. Instead, our analysis reveals highly structured morphodynamic feedback. The robust linear coupling observed in deep channels ($R^2\approx 0.51$) and the significant 3D topographic control in shallow shoals ($R^2\approx 0.55$) strongly demonstrate that tide-topography interactions overwhelmingly dictate the SSC distribution, overriding transient seasonal variations (Figure 17).

Furthermore, the stability of this spatial framework is reinforced by the normalized mathematical formulation of $A_{ssc}$. By calculating a ratio, the index inherently cancels out absolute concentration fluctuations induced by interannual or seasonal baseline shifts, thereby isolating the pure tidal asymmetry signal. Thus, rather than being distorted by temporal noise, the multi-year statistical compositing framework successfully suppresses anomalous extremes, providing a robust, time-averaged expression of sediment-response regimes under stable boundary constraints.

4.4. Methodological Limitations and Future Outlook

While the multi-year compositing of Sentinel-2 imagery successfully extracts persistent tidal asymmetry patterns (as discussed in Section 4.3), this approach inevitably entails temporal limitations. First, the fixed local-time acquisition (~10:30 a.m.) of sun-synchronous orbits causes diurnal aliasing, restricting the observation of continuous intra-tidal hydrodynamics [48]. Second, aggregating images across varying seasons and years (2017–2025) smooths out transient environmental signals. Consequently, the derived $A_{ssc}$ spatial patterns represent a stable, climatological morphodynamic baseline, but they intrinsically obscure the high-frequency dynamic impacts of seasonal monsoon shifts and interannual hydrological fluctuations [11]. It is also important to contextualize the statistical significance of these differences. Given the high spatial resolution of Sentinel-2 imagery, the statistical analyses encompass an overwhelmingly large number of pixels. In such scenarios, traditional statistical tests (e.g., t-tests) become hyper-sensitive, frequently yielding $p$-values $<0.001$ even for marginal absolute differences. Therefore, rather than relying solely on global $p$-values, this study emphasizes the spatial coherence and geomorphic zonation of the SSC asymmetry, which offer a more physically meaningful interpretation of the estuarine dynamics.

To capture the full spectrum of sediment dynamics, future research should integrate high-frequency geostationary ocean color sensors (e.g., GOCI-II) to resolve intra-day variability and reduce temporal sampling bias [49]. Furthermore, coupling satellite observations with fully three-dimensional hydrodynamic models would enhance physical interpretability by explicitly resolving vertical stratification and turbulence-driven near-bed resuspension [8,39], bridging the gap between two-dimensional surface snapshots and complex morphodynamic mechanisms.

5. Conclusions

Drawing upon high-resolution satellite remote sensing observations and a spatially isomorphic analysis framework, this study elucidates the spatiotemporal evolution patterns and underlying driving mechanisms of SSC in SMB, a semi-enclosed macro-tidal embayment. The core conclusions and methodological contributions of this research are summarized as follows:

1. Innovation and Application of the $A_{ssc}$ Framework: While the mathematical formulation of the “Suspended Sediment Concentration Asymmetry Index” ($A_{ssc}$) relies on a classic normalized difference structure, its unique contribution lies in extending tidal asymmetry analysis from continuous hydrodynamic time-series to discrete, satellite-observable sediment fields. By normalizing the flood–ebb concentration differences, this framework effectively eliminates the interference of absolute background turbidity magnitudes. This advancement provides a standardized, spatially continuous diagnostic metric to quantitatively compare sediment-response asymmetry across diverse morphodynamic units, overcoming the spatial sparsity of traditional in situ observations.

2. Systemic Elucidation of Dual-Control Driving Mechanisms: Guided by the $A_{ssc}$ distribution and multi-variable regressions, our results systematically refine the understanding of sediment transport mechanisms in macro-tidal embayments. Rather than demonstrating an absolute dominance of a single forcing, the findings validate a spatially divergent, dual-control mechanism. Specifically, net sediment transport in deep tidal channels is primarily driven by the “tidal pumping” mechanism. In contrast, sediment dynamics on intertidal mudflats and shallow shoals are highly sensitive to localized depth variations, where “geomorphic resistance” acts as the dominant regulatory control.

3. Spatially Continuous Mapping and Coastal Management Implications: By pairing 10 m resolution Sentinel-2 imagery with the ACOLITE atmospheric correction scheme, this study achieves a robust reconstruction of fine-scale sediment dynamics across complex estuarine geometries. The resulting spatiotemporal diagnostic paradigm provides explicit observational evidence of how tidal range and bathymetry modulate sediment resuspension and localized transport fluxes. These insights establish critical baselines for coastal engineering planning, channel navigation maintenance, and nearshore ecological conservation.

In summary, the remote sensing quantitative diagnostic paradigm and the $A_{ssc}$ application approach demonstrated in SMB provide a highly instructive exemplar for analyzing spatially continuous morphodynamics. The underlying physical logic of this framework offers a promising conceptual basis for analyzing sediment asymmetry. Future research involving cross-regional empirical validations, coupled with site-specific atmospheric and bio-optical parameterizations, will be essential to ascertain its broader applicability across diverse coastal environments.