Satellite Retrieval and Spatiotemporal Variability in Chlorophyll-a for Marine Ranching: An Example from Daya Bay, Guangdong Province, China

Abstract

With the planning and construction of marine ranching in China, water quality has become one of the critical limiting factors for the development of marine ranching. Due to geographical differences, marine ranches exhibit varying water quality conditions under the influence of the continental shelf. To the best of our knowledge, there is limited research on satellite-based water quality monitoring for marine ranching and the spatiotemporal variations in marine ranches in different geographical locations. Chlorophyll-a (Chl-a) is a key indicator of the ecological health and disaster prevention capacity of marine ranching, as it reflects the conditions of eutrophication and is crucial for the high-quality, sustainable operation of marine ranching. Using a physically based model, this study focuses on the retrieval of Chl-a concentration in Daya Bay. The coefficient of determination (R²) between the model retrieval values and the in situ Chl-a data is 0.69, with a root mean square error (RMSE) of 1.52 μg/L and a mean absolute percentage error (MAPE) of 44.25%. Seasonal variations in Chl-a concentration are observed in Daya Bay and are higher in spring–summer and lower in autumn–winter. In the YangMeikeng waters, Chl-a concentration shows a declining trend with the development of marine ranching. A comparison between the YangMeikeng (nearshore) and XiaoXingshan (offshore) marine ranches suggests that offshore ranching may be less impacted by terrestrial pollutants. The primary sources of Chl-a input in Daya Bay are the Dan’ao River and the aquaculture areas in the northeastern part of the bay. This study can provide valuable information for the protection and management of marine ranching.

1. Introduction

As the global population grows and economies develop, the demand for marine fishery products continues to rise. However, unsustainable fishing practices, such as overfishing and illegal fishing, have led to a significant decline in coastal marine fishery resources. Additionally, the continuous generation and discharge of terrestrial pollutants into the ocean have caused eutrophication in coastal waters, further degrading the marine ecological environment [1,2]. In this context, to promote the sustainable development of marine biological resources and the protection of the marine ecosystem, the Food and Agriculture Organization of the United Nations (FAO) introduced the concept of the “Blue Growth Initiative” [3,4]. Guided by the concept, scientists have proposed the idea of marine ranching, referring to ecosystems constructed in suitable sea areas based on ecological principles, utilizing natural productivity, modern engineering technology, and management models through habitat restoration and artificial breeding [5]. They serve the dual purpose of environmental protection and resource conservation and the sustainable production of fisheries. Marine ranching facilitates coastal ecological restoration, restores fishery resources, and ensures continuous production, attracting global attention. In recent years, the Chinese government has also released the “National Marine Ranching Demonstration Zone Construction Plan (2017–2025)”, aiming to accelerate the development of marine ranching [6,7].

However, the dual functions of ecological restoration and fishery production in marine ranching are closely related to the water quality of the coastal region [8,9]. According to the “Specifications for the Management of National Marine ranching Demonstration Zones” promulgated by the Chinese government, seawater quality in a marine ranching site should comply with Class II seawater quality standards. In practice, water quality monitoring plays a crucial role in the planning stage and throughout the entire life-cycle, including the construction and operation stages. Specifically, during the construction stage, water quality information can be used to monitor the potential environmental impacts caused by construction activities [10,11] and serves as a critical indicator of the health of the marine ranch ecosystem, helping to predict and warn of ecological disasters such as red tides during the operation stage [12,13]. Therefore, monitoring water quality in marine ranch regions is of significant practical importance. However, based on our understanding, there is currently limited focus and research on this aspect.

Currently, water quality monitoring for marine ranching primarily relies on manual water sample collection and buoy monitoring stations [14]. Manual sampling requires significant labor and time costs, especially in extensive coastal regions, making it difficult to effectively capture changes in water quality [15]. Buoy monitoring requires substantial financial investment in equipment, and issues such as sensor fouling and sampler blockages make maintenance and repairs challenging, resulting in poor long-term operational reliability. As a result, traditional water quality monitoring methods are time-consuming and labor-intensive, unable to meet the needs for large-scale, stable, and continuous monitoring. Additionally, they are insufficient for reflecting the complete condition of large-scale marine ranching [16,17,18,19]. With its continuous development, satellite remote sensing technology has become an effective method for large-scale, continuous, real-time water quality inversion and monitoring [20,21,22,23]. Remote sensing offers advantages such as wide spatial coverage, broad spectral range, high timeliness, and high spatial and spectral resolution [24,25,26,27]. Compared to traditional monitoring technologies, remote sensing overcomes the limitations of high costs and low timeliness, providing large-scale, periodic water quality monitoring results for target coastal regions [28,29,30,31].

In water quality remote sensing monitoring, Chl-a is a crucial indicator of the ecological health and disaster resilience of marine ranching [32,33,34]. Influenced by terrestrial nutrients, seawater movement, and the metabolic activity of phytoplankton, changes in Chl-a reflect the eutrophication condition in marine ranching areas [35,36,37]. High Chl-a concentrations are often associated with harmful algal bloom events, which threaten the marine ranching ecosystem and cause significant economic losses. Therefore, Chl-a monitoring is essential for the high-quality and sustainable operation of marine ranching. Current research on Chl-a retrieval models primarily focuses on data-driven empirical and semi-empirical models. Zhang et al. established a statistical model based on a univariate quadratic equation using band-Ultra Blue and band-Blue of Landsat-8 to estimate Chl-a in the coral reef areas of the South China Sea [34]. In fact, machine learning models are also a type of statistical model. Cen et al. established a Long Short-Term Memory (LSTM) neural network to estimate the Chl-a in the East China Sea [38]. While data-driven models demonstrate high accuracy in inversion regions, they often lack clear physical significance, which has been a point of contention. Additionally, in practical studies, the availability of water quality data in coastal areas is often limited, or there are information barriers, leading to the failure of data-driven models. Therefore, this study aims to establish a Chl-a retrieval model that has clear physical meaning and does not rely on large amounts of water quality data.

This study focuses on the marine ranching areas in Daya Bay, with the objectives of: (1) retrieving Chl-a in Daya Bay using a physically based model; (2) analyzing the spatiotemporal variability and seasonal patterns of Chl-a in Daya Bay; and (3) comparing Chl-a concentrations between YangMeikeng (nearshore marine ranching) and XiaoXingshan (offshore marine ranching) to better understand the spatial and temporal dynamics of marine ranching water quality in these distinct geographical environments.

2. Methodology

2.1. Study Area

Daya Bay is located between 114°30′ E and 114°50′ E, and 22°30′ N and 22°50′ N, in the northern part of the South China Sea. It is bordered by the Dapeng Peninsula in Longgang, Shenzhen (Dapeng New District, west coast), Tieluzhang in Huiyang, Huizhou (Aotou and Xiayong Streets, Daya Bay Economic and Technological Development Zone, north coast), and the Renping Peninsula in Huidong (east coast). To the east, it adjoins Honghai Bay, while to the west, it is adjacent to Dapeng Bay (as shown in Figure 1). Daya Bay is a typical subtropical, semi-enclosed bay and the northernmost bay in the South China Sea, extending furthest into the mainland. Covering an area of approximately 600 km², it has a maximum depth of 21 m and an average depth of 11 m. The tidal currents are predominantly reciprocating, with stronger currents at the bay mouth compared to the bay head. The bay’s coastline is varied and rugged, consisting mainly of rocky, sandy, gravelly, muddy, and mangrove shores. This area is home to a rich diversity of large benthic organisms, contributing to its high biodiversity. The XiaoXingshan marine ranch has been designated as a national-level marine ranching demonstration zone, while the YangMeikeng sea areas have also been included in the national marine ranching demonstration zone plan. Geographically, the YangMeikeng marine ranch is closer to the continental shelf, while the XiaoXingshan marine ranch is closer to the open ocean. Thus, monitoring water quality in this region is essential for the successful development and management of marine ranching.

2.2. Data

Sentinel-2 is a high-resolution multispectral imaging satellite constellation launched by the European Space Agency (ESA) in 2015, comprising two satellites: Sentinel-2A and Sentinel-2B. These satellites are primarily used for land management, agriculture and forestry, disaster control, and risk mapping. They inherit the legacy of the SPOT and Landsat programs, providing high-quality data support for multispectral observations. Sentinel-2 is equipped with a multispectral imager (MSI) that covers 13 spectral bands, with an imaging swath width of 290 km and a revisit period of 5 days. Due to their high resolution, short revisit period, and multispectral imaging capabilities, Sentinel-2 data are widely used in various environmental and Geographic Information System (GIS) applications, such as land cover classification, agricultural monitoring, forest health assessment, and disaster response [39,40,41]. Its multispectral data not only enhance the ability to identify surface features but also improve the accuracy and timeliness of environmental change monitoring [42,43,44]. In terms of water quality monitoring, its high-resolution images aid in detecting suspended matter and algal distributions in water system, thereby supporting environmental protection and water resource management.

The validation data were obtained from nearshore seawater quality monitoring information publicly available from the Guangdong Provincial Department of Ecology and Environment (GPDEE), and the observational data were supplied by the South China Sea Institute of Oceanology (SCSIO), Chinese Academy of Sciences. Chl-a concentrations are continuously monitored in real time using fluorescence-based sensors mounted on buoys [45], and the measurements are made at a depth of 0.5 m. These water quality validation data are further used to perform accuracy assessments for validating the Chl-a concentration retrieval model introduced in Section 2.4. The Sentinel-2 images are matched with the buoy station’s geographical coordinates and date information, with a time window of ±3 days. A previous study suggested that a time window of ±3 days between satellite data acquisition and in situ measurements can effectively improve Chl-a retrieval accuracy [46].

2.3. Preprocessing of Satellite Data

In this study, atmospheric correction was performed using the ACOLITE algorithm. ACOLITE is a program developed by the Royal Belgian Institute of Natural Sciences (RBINS) specifically designed for processing satellite water imagery. It defaults to using dark spectrum fitting for its algorithm [47]. The main advantage of ACOLITE is its ability to quickly and directly process high-resolution satellite imagery for aquatic applications (coastal/oceanic and inland waters). The algorithm can be downloaded from http://github.com/acolite (accessed on 20 July 2024) and used with the ACOLITE GUI. Sun glint masks used Cox–Munk model and Cloud masks was conducted via a threshold-based extraction method. Water segmentation was performed using the Normalized Difference Water Index (NDWI). The specific image and processing flow are shown in Figure 2.

2.4. Physically Based Retrieval Model

In our previous study, a physical-based retrieval model based on unit inherent optical parameters was established [48], as shown in the Formula (1). By measuring unit inherent optical parameters and substituting the band values, the three unknown variable “D_c”, “D_u”, and “D_s” can be solved. In this study, we utilizes four bands (RGB and NIR) from Sentinel-2 with a spatial resolution of 10 m to estimate Chl-a concentration (“D_c”). The unit inherent optical parameters were measured based on typical polluted water bodies within the study area, following the measurement methods outlined in our previous research [49,50]. All the unit inherent optical parameters of Daya Bay are shown in Table 1.

$$\begin{array}{c}{r}_{rs}={\displaystyle \frac{{(b}_w+b_c \ast D_c+b_u \ast D_u+b_s \ast D_s)\beta \left(\mathsf{\Theta }\right)}{4(\mathit{sec}{\theta }_{WIOZ}+\mathit{sec}{\theta }_{WOOZ})\left(a_w+b_w+{(a}_c+b_c)D_c+{(a}_u+b_u)D_u+{(a}_s+b_s)D_s\right)}}\end{array}$$

(1)

In Formula (1), “$r_{rs}$” is the water-leaving reflectance, “$\beta \left(\mathsf{\Theta }\right)$” represents the scattering phase function, and “${\theta }_{WIOZ}$” and “${\theta }_{WOOZ}$” are the zenith angle of the incident light and the upwelling light beneath the water surface, respectively. “$a_w$” and “$b_w$”, “$a_c$” and “$b_c$”, “$a_u$” and “$b_u$”, and “$a_s$ “ and “$b_s$” represent the absorption and scattering coefficients of water, Chl-a, suspended sediment, and aerobic organic matter, respectively. “$D_c$”, “$D_u$”, and “$D_s$” denote the concentration of Chl-a, suspended sediment, and aerobic organic matter, respectively.

Table 1. The unit inherent optical parameters of each component in seawater.

		Band 2	Band 3	Band 4	Band 8
central wavelength (nm)		490	560	665	842
a (m-1)	water	0.0196	0.0844	0.4221	3.7938
	Chl-a	0.9310	0.6308	0.9394	0.5348
	suspended sediment	0.1202	0.0463	0.0321	0.1144
	aerobic organic matter	0.1961	0.1404	0.1293	0.1001
b (m-1)	water	0.0031	0.0017	0.0008	0.0003
	Chl-a	0.0906	0.1721	0.0905	0.1418
	suspended sediment	0.0523	0.0589	0.0529
	aerobic organic matter	0.0078	0.0098	0.0047

Table 1. The unit inherent optical parameters of each component in seawater.

		Band 2	Band 3	Band 4	Band 8
central wavelength (nm)		490	560	665	842
a (m-1)	water	0.0196	0.0844	0.4221	3.7938
	Chl-a	0.9310	0.6308	0.9394	0.5348
	suspended sediment	0.1202	0.0463	0.0321	0.1144
	aerobic organic matter	0.1961	0.1404	0.1293	0.1001
b (m-1)	water	0.0031	0.0017	0.0008	0.0003
	Chl-a	0.0906	0.1721	0.0905	0.1418
	suspended sediment	0.0523	0.0589	0.0529
	aerobic organic matter	0.0078	0.0098	0.0047

2.5. Accuracy Assessment

To evaluate the accuracy of the Chl-a inversion model for Daya Bay, three indicators were used: coefficient of determination (R²), root mean square error (RMSE), and the mean absolute percentage error (MAPE).

The formula for calculating R² is

$$R^2=1-{\displaystyle \frac{\sum {\left(y_i-f_i\right)}^2}{\sum {\left(y_i-Y\right)}^2}}$$

(2)

where “$R^2$” is the coefficient of determination, “$y_i$” is Chl-a at the validation point, “$f_i$” is the value retrieved from inversion model, and “$Y$” is the mean Chl-a at the validation points.

The formula for calculating RMSE is

$$RMSE=\sqrt{{\displaystyle \frac{\sum {\left(y_i-f_i\right)}^2}{n}}}$$

(3)

where “$n$” is the sample size, “$y_i$” is Chl-a at the validation point, and “$f_i$“ is the value retrieved from inversion model.

The formula for calculating MAPE is

$$MAPE={\displaystyle \frac{1}{n}}\sum \left|y_i-f_i\right|\ast 100\%$$

(4)

2.6. Spatiotemporal Evolution Analysis

We conducted a spatiotemporal evolution analysis of Daya Bay from 2020 to 2024. For the intra-annual calculation of Chl-a, at least three years of monthly data are required to compute the monthly mean Chl-a concentration. For the interannual calculation of Chl-a, one image per quarter is selected, resulting in a total of four images, to calculate the annual mean.

Kernel Density Estimation (KDE) is a non-parametric statistical method that treats each raster cell as a point and utilizes a kernel function (such as the Gaussian kernel) to weight these points, generating a smooth probability density function. In this study, KDE is employed to effectively assess the spatial distribution characteristics of Chl-a in the Daya Bay region, identifying areas of high Chl-a concentration. To enhance computational efficiency, this study adopts bilinear interpolation to resample the original raster data to a resolution of 50 m.

To investigate interannual variation trends, the trend (also referred to as the rate of change) at each location within the study area was calculated and represented as the linear slope of the annual average Chl-a map, following the methodology used in many previous studies [51,52].

3. Results

3.1. Model Performance

The validation conducted using satellite-to-ground matching points shows that the Chl-a concentration estimated based on our model is in good agreement with the in situ data. The matching points generally lie along the 1:1 line (Figure 3), with a R² of 0.69, a RMSE of 1.52 μg/L and a MAPE of 44.25%. All validation points are listed in

Table 2. Statistical comparison between the in situ value and model retrieval value of Chl-a.

No.	Date	Measured Value (μg/L)	Inversion Value (μg/L)	MAPE
1	20 October 2020	4.5	5.68	26.22%
2	21 October 2020	1.9	2.31	21.58%
3	21 October 2020	2	1.32	34.00%
4	22 October 2020	1.4	2.60	85.71%
5	22 October 2020	1.35	1.31	2.96%
6	22 October 2020	1.3	0.81	37.69%
7	31 October 2020	4.49	5.31	18.35%
8	1 November 2020	0.33	0.83	151.52%
9	30 October 2020	1.05	1.78	69.52%
10	22 October 2020	3.5	3.12	10.86%
11	1 November 2020	0.6	1.44	140.00%
12	16 August 2019	6.04	4.35	27.98%
13	16 August 2019	6.82	5.77	15.40%
14	16 August 2019	1.53	1.82	18.95%
15	16 August 2019	1.75	3.15	80.00%
16	17 August 2019	9.48	5.33	43.78%
17	17 August 2019	5.68	7.62	34.15%
18	16 August 2019	3.51	4.22	20.23%
19	17 August 2019	4.85	4.77	1.63%
20	17 August 2019	4.94	6.21	25.71%
21	17 August 2019	9.83	12.54	27.57%
22	17 August 2019	9.98	13.61	36.37%
23	18 August 2019	9.9	8.66	12.48%
24	16 August 2019	2.06	4.51	118.93%
25	17 August 2019	1.61	0.48	70.19%
26	17 August 2019	2.61	3.27	25.29%
27	18 August 2019	4.5	2.78	38.22%
28	18 August 2019	2.46	6.48	163.41%
29	18 August 2019	2.55	3.66	43.53%
30	17 August 2019	1.6	1.53	4.38%
31	18 August 2019	1.57	2.34	49.04%
32	18 August 2019	1.53	2.19	43.14%
33	15 August 2019	1.46	1.28	12.33%
34	18 August 2019	4.47	4.69	4.92%
35	15 August 2019	1.36	0.72	47.06%
36	15 August 2019	0.18	0.33	83.33%
37	15 August 2019	1.66	1.54	7.23%
38	16 August 2019	3.54	4.52	27.68%
Min	-	0.18	0.33	1.63%
Max	-	9.98	13.61	163.41%
Average	-	3.42	3.81	44.25%

, revealing that the model tends to exhibit relatively larger errors at lower Chl-a concentrations. This may be due to the influence of suspended sediments in the shortwave blue and green regions, which significantly affect the absorption and scattering properties of Chl-a in waters with lower Chl-a concentrations [53]. Moreover, aerobic organic matter absorbs light in the visible spectrum, particularly at shorter wavelengths, increasing water absorbance and reducing the intensity of Chl-a reflection signals. This interference becomes more pronounced in waters with lower Chl-a concentrations. Therefore, further research and refinement of the model are needed in the future.

3.2. Intra-Annual Spatiotemporal Distribution Pattern of Chl-a

Chl-a exhibits spatial variation across different regions of Daya Bay (Figure 4). Generally, higher Chl-a concentrations are observed at the bay head, while lower concentrations are found at the bay mouth. In the middle of the bay, nearshore areas show elevated Chl-a levels, particularly around river estuaries and land-adjacent regions. The results indicate that Chl-a exhibit an increasing trend followed by a subsequent decline over the course of the year. Moreover, the spatial distribution patterns of Chl-a in Daya Bay show significant differences between the spring–summer and autumn–winter seasons. In winter (January and February), Chl-a concentrations in Daya Bay are relatively low, with higher values observed only in nearshore areas. By March, Chl-a concentrations within the bay remain low, but a significant increase is observed at the interface between the bay and the coast. In May, Chl-a in coastal regions increases substantially. Despite the impact of cloud cover on image quality between June and August, it is still evident that the Chl-a concentration reaches its annual peak (Figure 4 Jun–Aug). From September to November, Chl-a gradually declines, reaching lower levels by December.

3.3. Interannual Variation Trend of Chl-a

Figure 5 illustrates the rate of change in Chl-a concentration over a five-year period (2020–2024). During this observation period, the southwestern coastal region of Daya Bay exhibited a significant decreasing trend in Chl-a concentration, which may be linked to ongoing environmental remediation efforts associated with the establishment and development of the YangMeikeng marine ranch. These efforts likely contributed to improvements in water quality and the reduction in nutrient loads in the region. In contrast, certain areas within the central bay experienced a marked increase in Chl-a concentrations, which may be attributed to the expansion of aquaculture zones. The growth of aquaculture activities in this region has the potential to increase nutrient input into the water, stimulating phytoplankton growth and consequently raising Chl-a levels. This increase in Chl-a in the central bay warrants further investigation, as it could indicate shifts in the ecological balance due to anthropogenic influences.

3.4. Comparison Between Nearshore and Offshore Marine Ranching

The YangMeikeng marine ranch, a representative example of nearshore marine ranching, exhibits distinct seasonal variations in Chl-a concentration throughout the year, as shown in Figure 6. Chl-a concentrations are higher from May to October, while they remain relatively low from November to March. Figure 7 illustrates the seasonal changes in the XiaoXingshan marine ranch, an example of offshore marine ranching. Compared to YangMeikeng, the duration of elevated Chl-a concentrations at XiaoXingshan is shorter, with the highest peak occurring primarily in May and June, followed by lower concentrations during the remaining months. Nearshore marine ranching is more susceptible to terrestrial pollution due to its proximity to the coast [54]. In contrast, offshore marine ranching benefits from cleaner seawater with less exposure to terrestrial contaminants, stronger seawater circulation, and richer marine biodiversity. However, nearshore marine ranching has lower management and maintenance costs and is more accessible for routine monitoring and management. Therefore, it is essential to consider various factors when planning and developing marine ranching to ensure site-specific and sustainable practices.

4. Discussion

4.1. Spatiotemporal Analysis of Chl-a

The spatial distribution of Chl-a exhibits a general decreasing trend from land towards the ocean. Coastal waters are more susceptible to the input of land-based pollutants and human activities, resulting in higher Chl-a concentrations. However, certain areas within the bay also exhibit high Chl-a concentrations, potentially due to disturbances caused by suspended sediments. Suspended sediments act as key carriers for pollutant migration and recirculation. With their high specific surface area and adsorption capacity, these sediments can adsorb a variety of pollutants, including heavy metals, organic contaminants, and nutrients [55,56]. This adsorption process facilitates pollutant migration and diffusion alongside suspended sediments or their subsequent sedimentation and fixation [57]. At the bay mouth, Chl-a concentrations are relatively low, as offshore areas experience a dilution effect due to seawater influx. Offshore waters generally contain fewer pollutants and nutrients, and the dilution process further reduces the concentrations of pollutants, including Chl-a [58,59]. This dilution effect is particularly pronounced in the bay mouth and areas with active tidal exchange [60], contributing to a reduction in pollutant loads and an improvement in water quality.

In intra-annual variation, the seasonal fluctuations in Chl-a concentration are closely tied to phytoplankton growth cycles, oceanic dynamics, and environmental conditions, which together account for the pronounced seasonal changes observed in Daya Bay. During winter, most phytoplankton enter a dormant state, leading to lower Chl-a concentrations in this season. As early summer approaches, rising sea temperatures in Daya Bay, combined with upwelling along the eastern Guangdong coast, bring nutrient-rich deep waters to the surface, further stimulating phytoplankton growth. The rapid proliferation of algae during the summer results in a sharp increase in Chl-a concentrations over a short period [61]. Moreover, large river runoffs in summer deliver additional nutrients into Daya Bay, enhancing vertical mixing in the water column and providing more resources for phytoplankton growth [62,63].

4.2. Pollution Source Analysis

Figure 8a shows the kernel density estimation of Chl-a in Daya Bay, highlighting two hotspot regions with elevated Chl-a concentrations. Most of the Chl-a in nearshore areas is derived from terrestrial inputs [64]. According to marine and fishery statistical bulletins, Region I (Figure 8b) is designated as an aquaculture zone [65]. In aquaculture, large amounts of feed and bait are introduced, containing abundant nutrients, particularly nitrogen and phosphorus, which are essential for algal growth [66,67,68]. As the concentrations of nitrogen and phosphorus increase, algae proliferate rapidly, leading to a rise in Chl-a concentration in the water [69,70]. This phenomenon is especially evident in aquaculture areas, where excessive nutrient inputs not only promote algal growth but can also trigger eutrophication and other ecological issues. This explains the elevated Chl-a concentrations in Region I, consistent with the patterns observed in this study. A similar pattern of higher Chl-a concentrations is observed in the coastal areas of Region II (Figure 8c). Region II encompasses the estuary of the Dan’ao River, which may be influenced by terrestrial pollutants. As the largest tributary flowing into Daya Bay, the Dan’ao River is heavily impacted by domestic wastewater, contributing the majority of dissolved inorganic nitrogen and dissolved reactive phosphorus to the bay [71]. In addition to nitrogen and phosphorus from urban wastewater, agricultural chemical pollution is also a significant contributor to nutrient enrichment in Daya Bay [72]. Furthermore, discharges from the Dan’ao River and nearby industrial areas are also considered major sources of pollution [73]. These findings align with the patterns observed in this study.

4.3. Uncertainty Analysis

In remote sensing-based Chl-a retrieval, there exist several uncertainties that require further investigation. Firstly, over 90% of the radiation signal received by sensors originates from atmospheric effects, such as Rayleigh scattering, with water-related signals accounting for less than 10% [74]. Therefore, the accuracy of atmospheric correction directly influences the final accuracy of water quality remote sensing inversion. Additionally, the concentration and type of aerosols vary significantly across different regions and seasons [75,76]. The presence of residual transparent cirrus clouds, cloud cover over the water surface, and shadows from ships and riverside structures are difficult to fully eliminate, further increasing uncertainty in atmospheric correction.

The components in water exhibit both temporal and spatial variability, and they have different impacts on the absorption and scattering characteristics of light. This leads to heterogeneity in inherent optical properties (IOPs) over time and space. The radiative transfer model used in this study relies on IOPs, and their spatiotemporal heterogeneity adds uncertainty to the inversion results. Due to data availability constraints, this study’s model was unable to further account for the spatiotemporal heterogeneity of IOP. Moreover, in shallow water areas, bottom reflection may affect the water’s reflectance signal, potentially leading to biased estimates of Chl-a concentration. Future research will focus on further analyzing the temporal and spatial variations in IOPs and the bottom substrate in shallow water areas to reduce uncertainty in the model.

5. Conclusions

Regular Chl-a monitoring is essential for maintaining the ecological balance of coastal ecosystems and ensuring the high-quality operation of marine ranching. The spatial distribution of Chl-a concentrations derived from Sentinel-2 satellite inversion products reveals higher Chl-a concentrations at the top of the Daya Bay and lower ones at the bay mouth. Seasonal variation in Chl-a concentrations is evident throughout the year, with higher concentrations in summer and lower in winter. The main drivers behind this seasonal variation are water temperature, phytoplankton growth conditions, and ocean dynamics. In Yangmeikeng waters, Chl-a concentrations show a declining trend with the development of marine ranching. Additionally, compared to the YangMeikeng (nearshore) marine ranch, XiaoXingshan (offshore) is less impacted by terrestrial pollutants. These findings can help local governments take effective measures to improve marine ranching planning, construction, and water quality management.