Assessing the Long-Term Changes in the Suspended Particulate Matter in Hangzhou Bay Using MODIS/Aqua Data

Highlights

What are the main findings?

• Long-term MODIS/Aqua data (2003–2024) revealed a significant decline in suspended particulate matter (TSM) in Hangzhou Bay (HZB), linked to reduced sediment discharge from the Yangtze River.
• Interannual TSM variations in winter were associated with Yangtze River sediment discharge and regional wind forcing, with bridge construction causing spatial contrasts across the Hangzhou Bay Bridge and reclamation altering TSM around Yushan Island.

What is the implication of the main finding?

• The results highlight the combined influence of natural factors and human activities on sediment dynamics, revealing significant spatiotemporal changes and regulatory mechanisms of TSM in HZB.
• The findings provide valuable insights into the long-term changes in suspended sediment and water quality in HZB, supporting sustainable water management and effective water strategies.

Abstract

Hangzhou Bay (HZB) has become a hot spot in hydro-morphodynamic research due to human impacts and natural influences, as well as the substantial quantities of water discharge and sediment load of the Yangtze River and Qiantang River. Although many previous studies have analyzed the spatial–temporal variations in suspended particulate matter (TSM) from in situ and satellite observations, the long-term changes in suspended sediment dynamics remain unclear. In this study, we quantified the long-term variation in TSM load using MODIS/Aqua data during 2003–2024. The TSM products in the HZB displayed a decreasing trend from 2003 to 2024 (k = −1.90 mg/L/year, p < 0.05), which may be attributed to decreased sediment discharge from the Yangtze River. The spatial variation in TSM provided quantitative results for HZB, with a substantially increasing trend in the southern shallow areas and a decreasing trend in the northern deep troughs and central bay. The interannual variations in TSM in winter displayed a positive correlation with the sediment load from the Yangtze River (R = 0.640 for the data during 2014–2022) and with wind speed (R = 0.676 for the data during 2009–2021). The TSM of HZB was partly affected by the combined impacts of human activities and climate change. A distinct difference in TSM concentrations on both sides of the Hangzhou Bay Bridge was observed, with higher TSM on the western side than on the eastern side for most of the year during 2003–2024. A decline in TSM was observed near Yushan Island from 2003 to 2024, attributed to large-scale land reclamation and associated alterations in tide-dominated areas. This study provides valuable insights into the long-term changes in suspended sediment and water quality in HZB, which is crucial for managing water resources, creating effective water strategies, predicting future needs, and ensuring sustainable water management.

1. Introduction

Hangzhou Bay (HZB) is the largest bay located on the west coast of the East China Sea, where suspended particulate matter (TSM) concentrations are significantly influenced by tidal oscillations and amplified by human activities and climate variability. Large-scale reclamation in this region in the past few decades has altered the hydrodynamics of the bay and TSM distribution patterns. For example, since 2000, over 40,000 dams and reservoirs [1] have been constructed along the Yangtze River, which significantly reduced the sediment load and impacted the TSM concentrations in HZB water [2]. These structural changes and urban land expansion adjacent to the economically developed Yangtze River Delta have intensified in recent years. Large-scale coastal reclamation projects have been implemented [3,4,5], along with the construction of deep-water ports such as Zhoushan and Ningbo, and major infrastructure projects, like the Hangzhou Bay Bridge. These reclamation and construction activities have altered the distribution of suspended sediments and resulted in complex hydrodynamic and TSM distribution within HZB. In 2022, the Chinese Ministry of Ecology and Environment, along with six other governmental departments, issued a plan for significantly improving the ecological environment of adjacent seas, including the Bohai Sea, the Yangtze River Estuary–Hangzhou Bay, and the Pearl River Estuary [6]. The goal was to improve surface water quality by approximately 2% by 2025. It is thus crucial to identify the driving forces and understand the historical and present trends of TSM in HZB water for effective environmental and water resource management.

Compared to traditional water quality monitoring methods that involve costly and time-exhaustive sample collection and analysis for fixed point locations, remote sensing technology offers the advantages of providing rapid and periodic data products over large areas. This not only improves the temporal frequency but also enables dynamic monitoring and long-term trend analysis. Recent studies have demonstrated the practical utility of remote sensing technology for studying the spatial and temporal variations in TSM in coastal and estuarine systems [7,8,9,10,11,12,13,14,15].

According to recent investigations of spatiotemporal variations and regulatory mechanisms of TSM in HZB using satellite data and in situ measurement data, the sediment load from the Yangtze River and human activities in the region caused dramatic changes in TSM transport [2,7,16,17]. Since the year 2000, a sharp decline in the annual sediment load of the Yangtze River occurred in association with the sediment interception caused by the upstream dam construction and the relative contributions of climate change [16,18]. Shen et al. (2013) [19] reported a delayed response of TSM in HZB to a reduction in sediment load from the Yangtze River during 2003–2010. Although short-term sediment load changes have a minimal effect on TSM in HZB, a long-term decrease could reduce TSM over time. Additionally, human interventions such as offshore construction [20,21,22] and land reclamation [3,4,5,23] caused significant changes in the suspended sediment load in HZB, resulting in complex impacts on TSM distributions.

The substantial seasonal variations in TSM in HZB are also largely caused by natural factors, such as the Yangtze River discharge and wind speed. Gao et al. (2008) [24] concluded that the TSM of the Yangtze River is relatively much lower than that of HZB, suggesting that flood discharges through the river had a diluting effect on the hydro-sediment dynamics of the bay. Dan et al. (2014) [25] and Liu et al. (2013) [26] noted that higher wind speeds in the autumn and winter months could easily cause sediment resuspension and transport in the shallow regions of HZB, where initiation of sediment resuspension is modulated by prevailing winds.

Despite many studies focusing on hydro-sediment dynamics in HZB, the spatiotemporal variations and regulatory mechanisms of sediment load in the bay remain unclear. Firstly, previous studies have only focused on short-term or seasonal variations in TSM without further analysis of its long-term trends. Secondly, studies on the spatial and temporal distribution characteristics and evolution of TSM in HZB are limited. Although many studies qualitatively analyzed the influencing factors of TSM, only a few studies employed quantitative methods to explore the correlations between the above factors and TSM concentrations. To address this gap, our study produced long-term TSM products, providing quantitative, long-term analysis of the spatiotemporal variations in TSM and its primary driving factors.

In this study, we utilized the time-series satellite data from the Moderate Resolution Imaging Spectroradiometer (MODIS) to investigate the long-term changes in TSM in HZB. To deliver high-accuracy TSM products, we processed the MODIS data using the shortwave infrared atmospheric correction algorithm (SWIR-AC) and a regional TSM retrieval algorithm. Finally, we analyzed the spatiotemporal variations and regulatory mechanisms of TSM in HZB for the period from 2003 to 2024.

2. Data and Methods

2.1. Study Area

HZB is the largest macro-tidal bay located south of the Yangtze River Estuary and bounded by 29.8°N to 31°N latitude and 120.15°E to 122.5°E longitude (Figure 1). It features a funnel-shaped geometry, measuring 86 km in length and 100 km in width, and covers an area of about 8500 square kilometers. It is characterized by high TSM concentrations (50–5300 mg/L) [27], as a result of the sediment load of the Yangtze River and Qiantang River, as well as high wind speeds and human activities. For this study, five regions were selected to extract satellite-derived TSM data for spatial analysis (Figure 1). These regions (box0, box1, box2, box3, and box4) are located at the top of Hangzhou Bay, the north shore, the middle of the bay, the south shore, and the northern entrance of Hangzhou Bay, respectively.

2.2. Data

2.2.1. Satellite Data

The MODIS/Aqua L1B data were obtained from the Atmosphere Archive and Distribution System Distributed Active Archive Center (LAADS DAAC, https://ladsweb.modaps.eosdis.nasa.gov, last accessed on 9 August 2025). The MODIS/Aqua L1B products include data over a number of spectral bands at 412 nm, 443 nm, 488 nm, 531 nm, 551 nm, 667 nm, 678 nm, 748 nm, and 869 nm, with a spatial resolution of 1 km, and 469 nm, 555 nm, 645 nm, 859 nm, 1240 nm, 1640 nm, and 2130 nm, with a spatial resolution of 500 m.

Using the MODIS Cloud Mask product (MOD35; NASA, Washington, D.C., USA) with a 30% cloud coverage threshold for HZB, MODIS/Aqua L1B data from January 2003 to December 2024 were processed, and 2817 high-quality cloud-screened scenes were obtained for this study. These products ensure data continuity at up to daily time steps over 22 years, beginning from the first full calendar year of MODIS/Aqua observations (2003) to the last complete year (2024). The data density analysis for the L1B data after atmospheric correction shows index values ranging from 0 to 1 (Figure 2a), where the values approaching 1 indicate more complete data coverage and the values near 0 reflect significant data deficiencies. The overall number of valid data from 2003 to 2024 is fairly good for the analysis (Figure 2a), which represents a density above 50% for most areas of HZB and 80% for the bay entrance. A stacked monthly time series dataset from 2003 to 2024 shows colored segments representing the monthly counts and the contribution of each month to the annual total (Figure 2b). The black line on the top panel represents the annual total every year, highlighting the interannual variability and long-term trends. The analysis revealed a pronounced seasonal pattern, with data availability reaching its highest level in summer and dropping to its lowest level in winter. Notably, data availability in December is often limited due to frequent cloud cover and bad weather conditions. Despite this limitation, the annual image count remains consistently above 100 every year throughout the study period, highlighting the dataset’s reliability for long-term analysis.

2.2.2. In Situ TSM Data

The TSM data were collected from two cruises in the East China Sea, which spanned from 29°N to 33°N latitude and 120°E to 128°E longitude (including HZB, the Yangtze River Estuary, and the East China Sea shelf) [17,27]. The summer cruise was conducted from 18 July to 23 August 2006, while the winter cruise was conducted from 23 December 2006 to 4 February 2007. During both cruises, TSM and remote sensing reflectance ($R_{rs}$) data were concurrently obtained at the sampling stations. TSM was measured gravimetrically using pre-weighed cellulose acetate membrane filters (47 mm diameter, 0.45 μm pore size). $R_{rs}$ was measured aboard the ship with an ASD FieldSpec HandHeld Spectroradiometer (FSHH). The $R_{rs}$ data covered wavelengths from 413 nm to 865 nm, and the corresponding TSM concentrations ranged from 8 to 5300 mg/L. This range fully encompasses the TSM variations in HZB [27].

2.2.3. In Situ $R_{rs}$ Data

The in situ $R_{rs}$ data for the validation of the atmospheric correction method were obtained from a high-frequency water spectral observation system established at the Hai-Tian-Yi-Zhou (HTYZ) site (121.12528°E, 30.46278°N), located approximately 16 km and 19 km from the southern and northern coasts of the Hangzhou Bay Bridge, respectively. This system was equipped with two hyperspectral radiance sensors and one hyperspectral irradiance sensor to measure sea surface upwelling radiance ($L_w$), sky radiance ($L_{sky}$), and downwelling irradiance ($E_s$). The observation system operated daily from 07:00 to 17:00 local time, recording measurements at 15 min intervals. The radiometric data were collected over the spectral range of 320–950 nm, with a spectral resolution of 3.3 nm [28,29]. $R_{rs}$ was derived using the following equation:

$$R_{rs}=\left(L_w-{\rho L}_{sky}\right)/E_s$$

(1)

where $\rho$ represents the sea surface reflectance coefficient, set to 0.028 [30,31].

2.2.4. Environmental Data

To explore spatiotemporal variations in TSM and its driving mechanisms in HZB, surface environmental data, such as monthly discharge, sediment load, and wind speed, were collected for the period from 2003 to 2024. Due to the unavailability of 2024 records, the monthly discharge and sediment load of the Yangtze River and Qiantang River were obtained (up to 2023) from the China River Sediment Bulletin (http://mwr.gov.cn/zzsc/tjgb/zghlnsgb/, last accessed on 9 August 2025). To ensure temporal consistency with the discharge and sediment load data, the time series of TSM used in the subsequent correlation analyses were also limited to the period of 2003–2023. Monthly average wind speeds at a height of 10 m were obtained from the CCMP dataset, with a spatial resolution of 0.25° $\times$ 0.25° (http://www.remss.com/measurements/ccmp/, last accessed on 9 August 2025).

2.3. Atmospheric Correction Method

Atmospheric correction (AC) is the process of retrieving the water-leaving reflectance from the top-of-atmosphere (TOA) reflectance, ${\rho }_t\left(\lambda \right),$ measured by the satellite sensor. The water-leaving reflectance (${\rho }_w\left(\lambda \right)$) is then converted to $R_{rs}$ to derive the water color products, such as TSM, according to:

$$R_{rs}\left(\lambda \right)={\displaystyle \frac{{\rho }_w\left(\lambda \right)}{\pi ·t_s\left(\lambda \right)}}$$

(2)

where $t_s\left(\lambda \right)$ is the diffuse transmission coefficient from the sun to the sea surface.

In this process, when avoiding the sun glint area and empirically removing white cap reflectance on the sea surface [32,33,34,35,36], the total reflectance ${\rho }_t\left(\lambda \right)$ measured at a wavelength $\lambda$ can be decomposed into different contributions simply as follows [37,38]:

$${\rho }_t\left(\lambda \right)={\rho }_r\left(\lambda \right)+{\rho }_a\left(\lambda \right)+{t\left(\lambda \right)\rho }_w\left(\lambda \right)$$

(3)

where ${\rho }_r\left(\lambda \right)$ is the Rayleigh reflectance due to air molecules, ${\rho }_a\left(\lambda \right)$ is the aerosol reflectance (including Rayleigh–aerosol interactions) due to aerosol particles, ${\rho }_w\left(\lambda \right)$ is the water-leaving reflectance [37,39,40,41,42], and $t\left(\lambda \right)$ is the diffuse transmittance at the sensor viewing angle [43]. To derive ${\rho }_w\left(\lambda \right)$, the atmospheric effects, i.e., ${\rho }_r\left(\lambda \right)$ and ${\rho }_a\left(\lambda \right)$, are removed. The standard atmospheric correction algorithm computes ${\rho }_r\left(\lambda \right)$ from Rayleigh lookup tables with the inputs of the solar-sensor geometry, atmospheric pressure, and wind speed [39,40,41,44,45].

For open ocean water, the AC algorithm estimates ${\rho }_a\left(\lambda \right)$ with the assumption of negligible water-leaving reflectance (black ocean approximation) at two NIR wavelengths [46]. For turbid coastal waters, the black ocean approximation becomes quickly invalid because ${\rho }_a\left(\lambda \right)$ at NIR wavelengths can be significant [47,48,49]. This makes the standard atmospheric correction algorithm invalid in HZB water. An alternative AC algorithm was proposed by He et al. (2012) [50] using the ultraviolet (UV) wavelength for turbid waters, but it is not suitable for MODIS data due to the absence of UV bands. Another AC algorithm, developed by Wang and Shi (2007) [51], estimates ${\rho }_a\left(\lambda \right)$ using two SWIR bands (1240 nm and 2130 nm for MODIS), where the black ocean assumption may still hold, and then obtains ${\rho }_w\left(\lambda \right)$ in the visible bands by extrapolation, from which $R_{rs}\left(\lambda \right)$ is subsequently derived using Equation (2). In addition, Wang et al. (2009) [52] developed an NIR-SWIR-based AC algorithm based on the turbidity level, which effectively combines the strengths of both SWIR-based and NIR-based AC methods. Before atmospheric correction, turbid water detection is performed using the turbid water index criterion. The SWIR-AC algorithm is applied to identify turbid water pixels, whereas the standard NIR-AC algorithm is employed for non-turbid ocean waters. Recently, AC approaches have explored the application of machine learning under complex environmental conditions. For instance, the Ocean Color–Simultaneous Marine and Aerosol Retrieval Tool (OC-SMART) platform proposed a multilayer fully connected neural network (FCNN), combined with extensive radiative transfer simulations, to directly retrieve $R_{rs}$ from TOA radiances over coastal waters. In this study, three well-established atmosphere correction approaches, namely, SWIR-AC, L2gen from SeaDAS with the NIR-SWIR AC option, and OC-SMART, were employed on the MODIS/Aqua data. The performance of the three atmosphere correction algorithms was validated by comparing the satellite-derived $R_{rs}\left(\lambda \right)$ at 555 nm (Figure 3a–c).

Both L2gen and OC-SMART yielded numerous invalid retrievals, which necessitated masking out the affected pixels during the atmospheric correction process (Figure 3b,c). Thus, the SWIR-AC method was applied to the MODIS data to correct the atmospheric effects. The satellite-derived $R_{rs}$ values were matched with in situ $R_{rs}$ data of HTYZ during 2021, and the fitting results are shown in Figure 3e. To further assess the performance of the SWIR-AC algorithm, the MODIS-retrieved and in situ $R_{rs}$ data were compared across ten spectral bands in the visible and near-infrared range at HTYZ on three dates: 14 July, 6 August, and 18 September 2021 (Figure 3g–i). In general, the MODIS-retrieved $R_{rs}$ matched the in situ values well in both quantity and spectral shape.

To evaluate the specific correction effect of the SWIR-AC algorithm, the root mean squared errors (RMSEs) and mean absolute percentage errors (MAPEs) of the $R_{rs}$ products were calculated through Equations (4) and (5). Although there is a slight underestimation at 645 nm, the RMSE values across all bands remain below 0.005 and the MAPE values below 24%, suggesting that the SWIR-AC algorithm performs well overall (

Table 1. RMSE and MAPE of HTYZ in situ $R_{rs}$ validation of the SWIR-AC algorithm.

Wavelength (nm)	Number	RMSE (${\mathit{s}\mathit{r}}^{-1}$)	MAPE (%)
412	10	0.00316	19.09
443	10	0.00274	12.56
469	10	0.00172	8.32
488	10	0.00173	7.22
531	10	0.00135	4.73
555	10	0.00126	3.24
645	10	0.00290	6.56
859	10	0.00402	23.60

).

$$RMSE=\sqrt{{\displaystyle \frac{{\sum }_{i=1}^N{\left({Rrs}_{corrected,i}-{Rrs}_{in-situ,i}\right)}^2}{N}}}$$

(4)

$$MAPE={\displaystyle \frac{100\%}{n}}\sum _{i=1}^n{\displaystyle \frac{\left|{Rrs}_{in-situ,i}-{Rrs}_{corrected,i}\right|}{{Rrs}_{in-situ,i}}}$$

(5)

2.4. Description of the TSM Retrieval Algorithm

Over the past decades, several empirical algorithms for TSM concentration were reported using single or combined visible band reflectance data, but their applicability is limited in extremely turbid waters, such as the Changjiang River Estuary and HZB [53,54]. To overcome this limitation, Bai et al. (2010b) [27] established good linear relationships between TSM and the $R_{rs}$ band ratio of the near-infrared (779 nm) and visible (560nm) bands for HZB water from MERIS (Medium Resolution Imaging Spectrometer) data. Similarly, He proposed an empirical model between TSM and the $R_{rs}$ band ratio of 745 nm and 490 nm bands for moderate to extreme turbid waters of the Changjiang River Estuary and HZB (TSM from 8 mg/L to 5275 mg/L) from GOCI (Geostationary Ocean Color Imager) data. Considering 1 km resolution MODIS NIR saturation (745 and 869 nm) in highly turbid waters, the 500 m resolution 859 nm band was selected as a substitute in this study. Using the in situ data of $R_{rs}$ and TSM reported by Bai et al. (2010a) [27] and He et al. (2013) [17], we developed an empirical algorithm for a wide range of TSM concentrations in HZB (8~5300 mg/L) (Figure 4). This algorithm relies on MODIS $R_{rs}\left(859\right)$ and $R_{rs}\left(555\right)$ data to estimate TSM concentrations using the following equation:

$$TSM={10}^{\left(1.991\times Ratio+1.258\right)}$$

(6)

$$Ratio={\displaystyle \frac{R_{rs}\left(859\right)}{R_{rs}\left(555\right)}}$$

(7)

While SWIR-AC-retrieved $R_{rs}\left(859\right)$ showed a relatively high MAPE (23.60%) compared to the in situ measurements (Table 1), the impact on TSM retrieval was minimal. This is because the proposed TSM algorithm uses the spectral ratio $R_{rs}\left(859\right)$/$R_{rs}\left(555\right)$, which demonstrated significantly lower error (MAPE = 6.85%, Figure 3f). The ratio-based approach effectively reduces atmospheric correction error propagation, ensuring the robustness and reliability of derived TSM products.

3. Results

3.1. TSM Retrieval Algorithm Validation

For comparative validation, nine other empirical TSM algorithms constructed by Miller and McKee [55], Cheng et al. [56], Doxaran et al. [57], Han et al. [58], D’Sa et al. [59], Chen et al. [60], Ma et al. [61], and Tang et al. [62] were also applied to the same dataset. The formulas of these algorithms are summarized in

Table 2. The empirical TSM algorithms.

Author	Equation	$\mathit{X}$
Miller and McKee [55]	$TSM=1140.25X-1.91$	$X=R_{rs}(620~670)$
Cheng1 [56]	$TSM=217266X^2+8608.5X+3.4644$	$X=R_{rs}(750~900)$
Cheng2 [56]	$TSM=255.84X^2+1141X+429.86$	$X=R_{rs}(750~900)/R_{rs}(620~700)$
Doxaran [57]	$TSM=exp(3.132X+3.01)$	$X=R_{rs}\left(862\right)/R_{rs}\left(551\right)$
Han [58]	$TSM={10}^{6.2244X+0.892}$	$X={[R}_{rs}\left(550\right)+R_{rs}\left(670\right)]/[R_{rs}\left(550\right)/R_{rs}\left(670\right)$]
D’Sa [59]	$TSM=17.783X^{1.11}$	$X=R_{rs}\left(670\right)/R_{rs}\left(555\right)$
Chen [60]	$lgTSM=-4.782\left[R_{rs}\left(555\right)+R_{rs}\left(670\right)\right]+1.490\left[{\displaystyle \frac{R_{rs}\left(555\right)}{R_{rs}\left(670\right)}}\right]+0.384$
Ma [61]	$lgTSM=3.76862\left[R_{rs}\left(555\right)+R_{rs}\left(670\right)\right]-3.87735\left[{\displaystyle \frac{R_{rs}\left(555\right)}{R_{rs}\left(670\right)}}\right]+3.71737$
Tang [62]	$lgTSM=23.84071\left[R_{rs}\left(555\right)+R_{rs}\left(670\right)\right]-0.48532\left[{\displaystyle \frac{R_{rs}\left(555\right)}{R_{rs}\left(670\right)}}\right]+0.58213$

. Their performances were evaluated by comparing the algorithm-derived TSM with in situ measured TSM concentrations (Figure 5). The algorithms proposed by Miller and McKee [55], Cheng et al. [56], Han et al. [58], D’Sa et al. [59], Chen et al. [60], and Ma et al. [61] exhibit poor performance in HZB, which is characterized by high TSM concentrations, with RMSE exceeding 0.7 and MAPE over 83%. Doxaran’s algorithm overestimates TSM at low concentrations, while Tang’s algorithm shows larger deviations (RMSE = 1.245, MAPE = 259.3%) than our model (RMSE = 0.140, MAPE = 41.0%). Overall, the model proposed in this study has better accuracy and is more suitable for retrieving TSM in HZB.

Using MODIS/Aqua derived $R_{rs}$ data from 2003 to 2024, a long-term TSM dataset for HZB waters was obtained. Note that that TSM concentrations exceeding 5300 mg/L were considered anomalies and excluded from the analysis according to previous studies [17,27,50]. Based on this dataset, the spatiotemporal distribution patterns and long-term trends were analyzed.

3.2. Spatial Distribution of TSM in HZB

3.2.1. Long-Term Average Spatial Distribution

The SWIR-AC method and TSM retrieval algorithm were applied to MODIS/Aqua L1B data to derive the long-term average spatial distribution of TSM in HZB from 2003 to 2024 (Figure 6). The TSM concentrations are significantly higher in the upper reaches of the bay (>450 mg/L) and in shallow areas of Andong Shoal on the southern shore (>415 mg/L), while the TSM range is mostly at a lower level (<350 mg/L) in the region between Luchaogang and Jinshan on the northern shore. These variations are primarily influenced by underwater topography, coastal geometry, and tidal dynamics (Figure 1). The deep coastal trough near the northern shore reduces sediment resuspension, whereas Andong Shoal on the southern shore, with its flat topography and shallow water depth, experiences strong wave-driven sediment resuspension [63]. Additionally, high TSM at the northern part of the bay mouth near the Yangtze River Estuary and around the islands at the bay mouth may be influenced by the sediment load from the Yangtze River [2,64].

3.2.2. Long-Term Monthly Average Spatial Distribution

Monthly average TSM variations from 2003 to 2024 were analyzed (Figure 7,

Table 3. Long-term (2003–2024) monthly mean, maximum, and minimum TSM in HZB (units: mg/L).

Month	Mean	Max	Min
January	589.80	5871.32	36.16
February	566.34	5799.39	39.53
March	499.38	5785.83	29.25
April	485.98	5943.95	55.62
May	469.87	5676.12	53.56
June	337.82	5853.43	22.44
July	327.10	5962.10	26.26
August	353.05	5793.73	29.49
September	393.80	5929.99	25.75
October	434.61	5990.35	30.47
November	477.77	5975.67	30.42
December	773.80	5994.22	38.60

). Generally, monthly mean TSM in HZB ranges from 327.10 mg/L in July, when values reach their annual minimum, to 773.80 mg/L in December. In winter (December–February), the strong northeast monsoon enhances sediment resuspension, resulting in the highest TSM concentrations, particularly along the southern shore, where wave-driven turbulent conditions in shallow waters likely caused more resuspended sediments. Additionally, the high-concentration zone also extends further during winter. In spring (March–May), the monsoon weakens, leading to a general decrease in TSM concentration. While the high-concentration zone remains near the southern coast and islands at the bay mouth, its extent is significantly reduced compared to winter. In summer (June–August), despite the maximum discharge over the year, sediment loads from the Yangtze River are transported to the northeast direction instead of entering HZB [65]. Moreover, due to the weak southwest monsoon, sediment resuspension weakens, causing low TSM concentrations in HZB water, except for the south shore and islands at the mouth of the bay. In autumn (September–November), the northeast monsoon intensifies, causing strong sediment resuspension and a gradual rise in TSM. The high-TSM zone expands, and the distribution pattern transitions towards that of winter [2,19]. Due to high cloud coverage in winter, data in December are available for only a few years, but the available data reliably capture the spatial patterns of TSM.

3.3. Long-Term Changes in TSM in HZB

3.3.1. Long-Term Changes in Annual Mean TSM in HZB

To detect TSM dynamics over the past two decades, we employed the Python (v3.9.16) package ruptures (v1.1.10) [66], which provides algorithms for the detection of change points in time series data. The algorithms combine a cost function, a search method, and a constraint on the number of changes. Based on this package, we selected the Pruned Exact Linear Time (PELT) algorithm, using a radial basis function (rbf) cost model and a penalty value of 0.01, to detect change points in the annual mean TSM data from 2003 to 2024. PELT is a dynamic programming-based algorithm known for its high efficiency and ability to identify change points. The penalty value controls the sensitivity of the algorithm to the number of change points, with a lower penalty leading to more change points. Three change points were detected in the years 2008, 2013, and 2018 (Figure 8a). Based on this result, the MODIS/Aqua-derived TSM data were grouped into four sub-periods (2003–2007, 2008–2012, 2013–2017, and 2018–2024), and the distribution patterns were analyzed using box plots (showing median, quartiles, and outliers). As shown in Figure 8b, the median TSM in HZB notably decreased from the period 2008–2012 to the period 2013–2017. Additionally, the minimum TSM consistently decreased across the time intervals (2003–2007, 2008–2012, and 2013–2017). For quantitative analysis, we extracted the annual mean TSM values in HZB and collected the annual sediment discharge from the Yangtze River during 2003–2024 (Figure 8c). Both these variables exhibit significant decreasing trends, with a rate of TSM decline of 1.90 mg/L·year⁻¹ (p = 0.029) and a decrease in sediment discharge of 370.36 × 10⁴ t/year (p = 0.021). To better highlight long-term trends and reduce interannual fluctuations, a centered five-year moving average was applied to both the annual mean TSM and sediment discharge. The smoothed curves further displayed their consistent decreasing trends over the past two decades. These findings indicate that the reduction in TSM in HZB is likely associated with a declining Yangtze River sediment load.

To further analyze the long-term trend in TSM changes across specific areas of HZB, we selected five representative locations from the upper bay to the bay mouth and calculated the monthly mean TSM values for each year (Figure 1). These locations were chosen based on the spatial distribution patterns of TSM shown in Figure 6, capturing different topographic and TSM conditions across the bay. As shown in Figure 9c–f, the TSM concentrations at the north shore display a significant decreasing trend (p < 0.05) and little variation in the middle and south shores and northern entrance of HZB, which are consistent with the decreasing trend of sediment load from the Yangtze River (Figure 8b). This decline in sediment load inputs is directly related to the construction of the Three Gorges Dam, which reduced the sediment load movements entering HZB and thereby decreased the TSM concentrations. However, TSM at the top of HZB (Figure 9b) displays an observed increasing trend, though not statistically significant over the study period, possibly influenced by the increased sediment load of the Qiantang River (Figure 9a). The long-term regression analysis for TSM concentrations at each location detected these variations (

Table 4. Statistical regression analysis of TSM at different locations in HZB.

Location	k	p-Value
box0	2.23	0.122
box1	−3.27	0.009 **
box2	−3.90	0.021 *
box3	−0.19	0.885
box4	−2.00	0.250

Notes: k is the slope, and p is the p-value. * indicates p < 0.05, while ** indicates p < 0.01.

).

3.3.2. Long-Term Changes in Seasonal Mean TSM in HZB

Further analysis of the TSM concentrations in HZB waters showed significant seasonal differences (Figure 10). The TSM concentration was the lowest in summer, with an average value of 200.70 mg/L, and the highest in winter, with an average value of 431.89 mg/. In spring and autumn, the TSM concentrations averaged 334.27 mg/L and 282.75 mg/L, respectively.

From 2003 to 2024, the seasonal TSM products display a long-term trend (Figure 11). For example, TSM shows a decreasing trend (p < 0.05) in both spring and summer, a slightly increasing trend in autumn, and a slightly decreasing trend in winter (

Table 5. Seasonal statistical regression analysis of TSM in HZB.

Season	k	p-Value
Spring	−2.99	0.014 *
Summer	−2.38	0.017 *
Autumn	0.20	0.879
Winter	−0.88	0.685

Notes: k is the slope, and p is the p-value. * indicates p < 0.05.

). Although data in December are available for only a limited number of years due to frequent winter cloud coverage, the available data are reliable for interpreting long-term seasonal variations.

3.3.3. The Rate of TSM Variation in HZB

To estimate the long-term variation rate of TSM at each pixel in HZB, a linear regression was applied to the annual TSM data from 2003 to 2024. The slope of the regression line represents the rate of TSM change over time. The calculation is based on the equation:

$${TSM}_t=a\times t+b$$

(8)

where a is the annual rate of change.

Trend analysis of TSM shows an increasing trend in the southern bay, particularly at Andong Shoal and the bay mouth islands (Figure 12). In contrast, there is a decreasing trend in TSM in the nearshore waters off Jinshan on the north shore and the central area of the bay. These spatial differences are likely due to the combined effects of the underwater topography of HZB and human activities. The relatively flat and shallow Andong Shoal in the south promotes strong sediment resuspension, making it more responsive to human impacts. Sediment resuspension is weak in the northern bay due to deep coastal troughs, which leads to the decreased TSM concentration, attributable to sediment settling and declining sediment load inputs from the Yangtze River.

4. Discussion

4.1. Uncertainties in the Atmospheric Correction Method and theTSM Retrieval Algorithm

The atmospheric correction method and the TSM retrieval algorithm are crucial for obtaining reliable TSM products. Therefore, we assess the performance of the atmospheric correction method and the TSM retrieval algorithm applied in this study.

While SWIR-AC-retrieved $R_{rs}\left(859\right)$ showed a relatively high error (Table 3), the use of the spectral ratio $R_{rs}\left(859\right)$/$R_{rs}\left(555\right)$ in the TSM retrieval algorithm significantly reduces error propagation (Figure 3f). Furthermore, SWIR-AC provides better data validity compared to L2gen and OC-SMART, ensuring more reliable results. Therefore, SWIR-AC remains a robust and suitable atmospheric correction method for our study.

Although the proposed algorithm outperforms the other tested models in terms of accuracy, the resulting MAPE of 41.0% still indicates uncertainty. This is attributed to the inherent challenges of TSM retrieval in highly turbid coastal waters, where extremely high TSM concentrations significantly enhance surface reflectance and lead to the saturation of $R_{rs}$. As a result, the sensitivity of $R_{rs}$ to changes in TSM concentrations decreases, increasing the retrieval errors. In addition, dynamic processes, such as wind-induced waves, tidal forcing, and sediment resuspension, take place regularly in highly turbid coastal waters, leading to pronounced spatiotemporal variability in TSM concentrations. Consequently, while the algorithm provides relatively reliable estimates, it is advisable to apply the model cautiously to regions or periods with rapidly changing turbidity.

4.2. Factors Influencing TSM Interannual Variations

The sediments of HZB are transported by the freshwater flow from the Yangtze River [67]. The Yangtze River flow displays a clear seasonal pattern, with the sediment load accounting for about 78% of the annual discharge during the flood season and 22% during the dry season [68]. However, due to the high river discharge, the surface TSM is dispersed in HZB during summer, which implies that the sediment discharge carried by the river flow does not directly affect the TSM concentration in HZB waters. Previous studies on sediment transport in HZB found that the turbulence of wind–wave actions is the primary driving force for the resuspension of sediments in shallow HZB waters [26,69]. As a consequence, the wind-induced vertical mixing and stirring mechanism caused by wave kinetic energy induces surface TSM fluctuations in previously settled sediments in HZB waters during summer, leading to high TSM in HZB waters during winter [26,70,71,72,73,74].

To investigate the driving factors of the interannual variations in TSM of HZB, the Yangtze River sediment load and regional mean wind speed were selected for correlation analysis with annual mean TSM concentrations (Figure 13). However, both correlation coefficients were statistically insignificant, suggesting no clear linear relationship between TSM and either sediment load or wind speed (Figure 13b). Therefore, further analysis using Granger causality tests was performed to investigate potential lagged effects of the Yangtze River sediment load and wind speed on TSM, which is a statistical method used in time series analysis to determine whether there is a causal relationship between variables. Granger causality tests compare two time series of data by calculating a Pearson correlation coefficient (R) between the corresponding values at each time step within a time window. Additionally, one of the time series is time-lagged relative to the other, and new correlations are calculated for various time lags. This allows for the estimation of delayed effects between the two time series. At a one-year lag, the test reveals a significant influence of the Yangtze River sediment load on the TSM in HZB (p < 0.01, Figure 13c), suggesting that the sediment load from the Yangtze River is an important driving factor of TSM in HZB on interannual timescales.

To further support this finding, the Pearson correlation coefficient between TSM and the Yangtze River sediment load delayed by one year was calculated (R = 0.58), indicating that changes in the TSM in HZB respond to variations in the Yangtze River sediment load with a one-year delay, which is consistent with the findings reported in previous studies [18,19]. This may be attributed to the fact that sediment transported into HZB during the flood season is initially diluted and settles to the bottom; it is then resuspended in winter due to wind-induced vertical mixing, resulting in increased surface TSM concentrations in January and February of the next year (Figure 10).

Due to the strong resuspension of sediments in winter, the TSM concentration peaks during winter, making winter TSM a good indicator for analyzing its long-term driving mechanisms. Since winter TSM is influenced by both the Yangtze River sediment load and wind speed, Pearson correlation coefficients between TSM and each factor were calculated using an 8-year moving window (Figure 14b). Overall, during 2014–2022, the sediment load had a statistically significant influence on TSM, with the Pearson correlation coefficient consistently exceeding 0.5, which was confirmed by a recalculated positive correlation coefficient (R = 0.640) for the same period. Similarly, in 2009–2021, the correlation coefficient between wind speed and winter TSM remained above 0.5, indicating a significant impact of wind speed on TSM during the period, as confirmed by a recalculated positive correlation coefficient (R = 0.676). Additionally, except for a few years, the anomalies of TSM and sediments in 2014–2022 displayed consistent positive and negative trends. During 2009–2021, TSM and wind anomalies also showed consistent trends, indicating that wind-induced hydrodynamic changes impacted the sediment resuspension processes.

Due to the significant reduction in the sediment load following the construction of the Three Gorges Dam, which significantly reduced the annual sediment load by approximately 60% [19], the long-term variability in TSM in HZB may correspondingly be influenced by the sediment load from the Yangtze River. Wind-driven resuspension during winter has also played a crucial role in influencing TSM concentrations in recent years. Moreover, the spatial distribution and temporal variations in TSM concentrations in HZB could be explained by other complex factors, such as human activities (e.g., offshore construction and land reclamation) and tidal dynamics.

The TSM concentration in HZB is influenced not only by sediment load and wind speed, but also by many other dynamic processes, such as water stratification, wind-induced mixing, tidal dynamics, and sediment resuspension [26,75,76,77,78,79,80,81]. For example, the blocking effect of the Zhoushan Islands restricts current velocity on the southern side of HZB, causing it to be slower than on the northern side. The weaker currents lead to reduced turbulence and vertical water motions that normally cause sediment resuspension, resulting in lower surface TSM on the southern side. However, the higher velocity between and along the Zhoushan Islands triggers sediment resuspension from the bottom, causing elevated TSM in this region [75]. In addition, HZB is a tide-dominated estuary with one of the world’s strongest tides, where tidal amplitudes range from 3–4 m at the mouth to 4–6 m at the head, and can reach 9 m [82,83]. High-velocity tidal currents trigger resuspension of previously deposited sediments, resulting in high TSM, whereas the reduction in current velocity towards the end of the ebb tide allows for the deposition of suspended sediments [75,78,79,80,81]. However, our study does not aim to comprehensively address all potential driving factors of seasonal and interannual variability in TSM. We focused on important environmental factors with pronounced seasonal and interannual variability, including river discharge, sediment load, and wind speed, based on a thorough review of previous studies [2,7,16,17,18,19,24,25,26]. Hydrodynamic processes, such as tidal dynamics, can cause short-term variations in TSM on hourly or daily scales, while others, such as sediment resuspension influenced by current velocity, take place on longer time scales but are not major drivers of seasonal and interannual variability in TSM. Considering space limitations, we do not discuss these factors in detail here.

4.3. Impacts of Human Activities

Bridge engineering facilities in nearshore areas are known to cause changes in the hydrodynamic environment through the interaction between currents and bridge piers, leading to increased sediment concentrations in downstream areas of bridges [20,21,22,56,84,85,86]. Bridge piers can block water currents from upstream [20,21,22], thereby promoting sediment deposition in downstream areas while simultaneously inducing local scouring near the bridge and resulting in sediment resuspension [84,85]. A pronounced increase in downstream suspended sediment concentration is observed during westward flows, whereas eastward flows result in only a modest increase [87,88]. The Hangzhou Bay Bridge spans approximately 36 km across the main channel of HZB, close to the Zhoushan Archipelago. It began operation in 2003, providing enhanced transportation connectivity for cities in northeastern Zhejiang Province and connecting Jiaxing, Ningbo, and Cixi cities. This study analyzed the impacts of the Hangzhou Bay Bridge on the distribution of TSM in the surrounding sea areas. For this analysis, eight equally spaced points were chosen along the Hangzhou Bay Bridge. At each point, cross-sections extending 3 km on either side perpendicular to the bridge were created (Figure 15b). The TSM variation data along these transects were then extracted for seasonal analysis (Figure 15c,d). Since the construction of the bridge began in 2003, TSM distributions in 2003 (at the beginning of the bridge construction) and 2024 (after the bridge construction) were also considered for comparison to investigate the influence of the bridge (Figure 15e).

A distinct difference in TSM concentrations on both sides of the Hangzhou Bay Bridge is shown in Figure 15c. The details are highlighted in Figure 15d, showing that the western side of the bridge consistently displays significantly higher TSM concentrations than the eastern side in spring, summer, and autumn. This difference in TSM concentrations between the western and eastern sides of the bridge can be attributed to two factors. First, the hydrodynamic influence of the bridge plays a key role. When the current flows westward, the TSM concentrations in downstream areas of the bridge increased significantly, whereas the increase in downstream TSM concentrations was less pronounced during the eastward flow [87,88]. Second, the western side of the bridge is affected by sediment input from the Qiantang River, which contributes to higher TSM concentrations on the western side for most of the year. However, this spatial pattern weakens in winter, during which high TSM concentrations are observed on both sides of the bridge. As discussed previously in Section 3.3, strong winds in winter lead to resuspension of bottom sediments, resulting in overall high surface TSM concentrations, which weakens the influence of the bridge on the TSM distribution during winter. A comparison between 2003 and 2024 indicates that the distinct concentration difference between the western and eastern sides observed in 2024 was not evident in 2003, when the TSM distribution was more continuous and the west–east concentration difference was less pronounced. In general, the clear boundary shown along the bridge in Figure 15c–e supports the conclusion that the TSM distribution on both sides is indeed influenced by the bridge.

Yushan Island, a part of the Zhoushan Archipelago, was once small, covering 8.44 square kilometers with a population of 3000. Since 2017, the island has been developed into an oil refining base, with the aim of becoming a world-class chemical industrial hub. Initially, the island comprised a limited flat terrain. However, by 2020, the area of the islands expanded to 21 square kilometers due to the extensive land excavation and reclamation activities. From 2003 to 2024, TSM in the waters near Yushan Island showed an overall declining trend (Figure 15f). This decrease may be attributed to large-scale land reclamation and tidal dynamic alterations [89,90].

5. Conclusions

The time series of MODIS/Aqua L1B data from 2003 to 2024 revealed the spatial and temporal characteristics of TSM and driving mechanisms in the HZB. The results generally indicated a decreasing trend in TSM concentration in HZB waters, which can be attributed to declines in sediment discharge from the Yangtze River. Spatially, TSM exhibited an increasing trend in the southern shallow regions and a decreasing trend in the northern deep troughs and the central bay. Interannual variations in the wintertime TSM products were closely associated with sediment discharge from the Yangtze River and wind-driven sediment resuspension. Furthermore, anthropogenic activities have increasingly altered the sediment dynamics of HZB. TSM on the western sides of the Hangzhou Bay Bridge remained higher than on the eastern side for most of the year from 2003 to 2024, which is likely attributable to the influence of the bridge on regional hydrodynamic conditions. Additionally, a notable and continuous decline in TSM near Yushan Island was observed, likely due to the extensive land reclamation projects and local tidal dynamics. These findings highlight the combined impacts of human activities and natural factors on the sediment dynamics of HZB over the past decades. The long-term TSM products and observed trends provide important insights into the regulatory mechanisms governing the suspended sediment concentrations and their temporal trends, supporting informed decision-making for water resource management and environmental protection in HZB and other similar urbanized estuarine systems.