Next Article in Journal
Assessing Pacific Madrone Blight with UAS Remote Sensing Under Different Skylight Conditions
Previous Article in Journal
Estimation of Downlink Signal Transmitting Antenna PCO and Equipment Delays for LEO Navigation Constellations with Limited Regional Stations
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Remote Sensing
Volume 17
Issue 18
10.3390/rs17183140
remotesensing-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Assessing the Impacts of Marine Ranching Construction on Water Quality and Fishery Resources in Adjacent Coastal Waters

by
Jianqu Chen
1,2,3,
Xue Feng
1,2,3,
Chunya Guo
4,
Yuxiang Chen
1,2,3,
Fei Tong
1,2,3,
Lei Zhang
1,5,
Zhangbin Liu
6,
Jian Zhang
7,
Huanrong Yuan
1,2,3,* and
Pimao Chen
1,2,3
1
South China Sea Fisheries Research Institute, Chinese Academy of Fishery Sciences, Guangzhou 510000, China
2
Key Laboratory of Marine Ranching, Ministry of Agriculture and Rural Affairs, Guangzhou 510000, China
3
National Digital Fisheries (Marine Ranching) Innovation Sub-Center, Guangzhou 510000, China
4
Marine Weather Forecast Center of South China Sea, Guangdong Meteorological Observatory, Guangzhou 510080, China
5
College of Marine Living Resource Sciences and Management, Shanghai Ocean University, Shanghai 201306, China
6
Graduate School of Bioresource and Bioenvironmental Sciences, Kyushu University, Fukuoka 819-0395, Japan
7
Graduate School of Environmental Science, Hokkaido University, Hokkaido 060-0811, Japan
*
Author to whom correspondence should be addressed.
Remote Sens. 2025, 17(18), 3140; https://doi.org/10.3390/rs17183140
Submission received: 23 July 2025 / Revised: 1 September 2025 / Accepted: 6 September 2025 / Published: 10 September 2025
Browse Figures
Versions Notes

Abstract

Highlights

What are the main findings?
  • By integrating satellite remote sensing data with in-situ water quality sampling data, this study conducted water quality parameter retrieval before and after the construction of the marine ranch. The SVR model exhibited relatively superior inversion performance. SHAP analysis revealed that the contribution pathways of optical and non-optical sensitive parameters differed significantly.
  • This study quantified the impact of artificial reef deployment on surrounding fishery resources and water quality parameters. Following the placement of 38,048 m3 of artificial reefs near Wailingding Island in Zhuhai, fishery resources increased by 318 kg/km2 in spring and 660 kg/km2 in autumn. Moreover, the deployment may have influenced concentrations of Chla, DO, COD, and PO4-P in surface waters within a radius of approximately 10 km.
What is the implication of the main finding?
  • This study identified the optimal model and key spectral bands for water quality parameter inversion in the waters adjacent to Wailingding Island’s marine ranch. These findings provide a scientific reference for the future optimization and upgrading of remote sensing models for water quality assessment, as well as for ecological environment monitoring and fishery resource prediction in related regions.
  • The results offer essential data support for calculating the input-output ratio of marine ranches. This contributes to a scientific evaluation of the economic and ecological benefits of marine ranch construction, helping managers avoid blind planning caused by uncertainties in the extent and degree of environmental impacts, and laying a foundation for environmentally sustainable marine ranch management.

Abstract

This paper aims to explore the impact of marine ranching construction on water quality and fishery resources in the surrounding marine areas. Utilizing in situ water quality and fishery resource data collected before and after the establishment of marine ranching, the study analyzes changes in water quality parameters from both temporal and spatial perspectives. A quantitative evaluation of the water quality data is conducted using several models to assess the accuracy of different evaluation methods. By integrating the SHAP algorithm with physical significance, the study examines the differences between optically sensitive and non-optically sensitive water quality parameters during the machine learning evaluation process. Finally, based on the inverted water quality data, the potential impact range and resource output following the deployment of artificial reefs are investigated. The results indicate that in the marine area near Wailingding Island, Zhuhai, the deployment of artificial reefs with a volume of 38,048 cubic meters led to an increase in fishery resources by 318 kg/km2 in spring and 660 kg/km2 in autumn. Additionally, deployment had varying degrees of impact on the concentrations of chlorophyll a (Chla), dissolved oxygen (DO), chemical oxygen demand (COD), and phosphate (PO4-P) in the surface water within an approximate range of 10 km. This study provides a valuable reference for calculating input–output ratios, as well as for the management and evaluation of marine ranching.

1. Introduction

Marine ranching is a sustainable fishery model [1,2,3] designed to enhance and conserve inshore fishery resources while promoting the health and stability of ecosystems [4,5,6]. This is achieved through the construction of suitable artificial habitats [7] and the implementation of fishery enhancement practices [8]. Additionally, marine ranching has emerged as a significant strategy for fishery carbon sequestration [9,10,11]. The creation of artificial habitats in marine ranching includes the construction of benthic habitats [12] and pelagic habitats [13]. The deployment of bottom-placed artificial reefs changes the local seabed environment, while the construction of floating and surface facilities alters the spatial environment of the sea area, causing disturbances to the original hydrodynamics, biological communities, and habitat patterns of the sea area [14,15,16]. The alterations in flow fields induced by underwater structures, such as bottom-placed artificial reefs, facilitate the transport of seabed nutrients to the pelagic zone. This process enhances the primary productivity of the marine area, increases the biomass of bait for plankton and small fish, and promotes the circulation and transfer of substances and energy throughout the food chain. Consequently, these changes lead to an increase in the density of fishery resources within the reef area [16,17,18,19].
Against the dual backdrop of the rapid development of the global marine economy and the increasing demand for ecological protection, marine ranching has emerged as a vital model for promoting sustainable fishery development and marine ecological restoration. It serves as a significant vehicle for achieving “blue growth” [20,21,22]. The United Nations Sustainable Development Goals (SDGs) have identified “conserving and sustainably using the oceans, seas, and marine resources” as a critical issue [23]. Within this framework, the construction and development of marine ranching are clearly important areas of practice. China has also integrated marine ranching development into the 14th Five-Year Plan for Marine Economic Development [24]. As of 2024, China has established nearly 200 national marine ranching demonstration zones, which have developed distinctive models tailored to different marine environments [25].
However, as marine ranching continues to expand, the uncertainty surrounding the potential impacts of engineering activities on adjacent marine biological resources and ecological environments has become increasingly pronounced [17,26,27]. Calculating the input–output ratio of marine ranching construction is a crucial aspect of marine and coastal fishery management. However, due to the ambiguous effects of artificial reef deployment, the calculation of relevant input–output ratios lacks a reliable foundation, which hinders the planning and management of environmentally sustainable marine ranches. Furthermore, current research on the effects of marine ranching construction primarily focuses on changes in fishery catches [18], while their indirect benefits, such as environmental restoration and carbon sequestration, are challenging to quantify. Additionally, there is a significant lack of accurate quantitative data regarding the scope and extent of impacts on the surrounding ecological environment following the deployment of artificial reefs.
As critical indicators of the health of marine ecosystems, water quality parameters can reflect the impacts of artificial reef deployment through their dynamic changes [16,28]. The introduction of artificial reefs alters local hydrodynamic conditions, facilitating the transport of nutrients from the seabed to the surface, which directly affects the concentrations of nutrients, such as phosphates [29]. When nutrient concentrations fluctuate, the phytoplankton in the area (i.e., chlorophyll a concentration) will also respond accordingly [30]. Upwelling can also directly or indirectly influence dissolved oxygen concentrations, either through changes in phytoplankton levels or other mechanisms [31]. Additionally, artificial reefs provide attachment surfaces and shelters from predators for various aquatic organisms, leading to the aggregation of biological communities [32]. The metabolic activities of these organisms can cause fluctuations in chemical oxygen demand [33]. Furthermore, water quality parameters can also serve as indicators of the status of fishery resources to some extent [34,35,36]. Therefore, by monitoring changes in seawater quality parameters, one can indirectly assess the extent and scope of the impacts of artificial reef deployment on the environment and biological resources.
In this study, in situ survey data on water quality and fishery resources were collected before and after the deployment of artificial reefs in marine ranching. These data were combined with satellite remote sensing information to construct a quantitative inversion model of water quality parameters tailored to the local marine environment. The optimized model was then utilized to investigate the temporal and spatial variations in the concentrations of chlorophyll a (Chla), dissolved oxygen (DO), chemical oxygen demand (COD), and phosphate (PO4-P) before and after the deployment of artificial reefs, thereby clarifying the scope and extent of their impacts on water quality. Additionally, a correlation analysis was performed by integrating the SHAP algorithm with physical significance to examine the differences in machine learning approaches for evaluating optically sensitive versus non-optically sensitive water quality parameters. Finally, based on the evaluation results of the inverted water quality parameters, the increase in fishery resources was quantified, and the yield corresponding to the scale of the deployed reefs was determined. This analysis provides valuable support for relevant managers in calculating the input–output ratio and in the rational utilization of marine resources.

2. Materials and Methods

2.1. Data Collection and Processing

Water quality parameters and fishery resource data were collected from in situ surveys conducted in the national marine ranching demonstration zone and the adjacent sea areas near Wailingding Island, Zhuhai, during April, September, and November from 2020 to 2024 (Figure 1). These data include sampling conducted both before and after the deployment of artificial reefs. The deployment of these artificial reefs was completed in August 2021, with a total volume of 38,048 cubic meters.
Seawater samples were collected on-site during the seawater quality survey to assess Chla, DO, COD, and PO4-P. On-site sampling and analysis were conducted in accordance with the methods specified in relevant standards [37,38]. Specifically, Chla was measured using the spectrophotometric method [39]; seawater DO was determined through the iodometric titration method [40]; COD was analyzed using the alkaline potassium permanganate method [41]; and reactive phosphate was measured using the phosphomolybdate blue method [42]. For the fishery resource trawl survey, one trawl was conducted at each station, with a single net deployed per trawl. The average towing speed was approximately 3.0 knots. The survey vessel deployed the net approximately 1–2 nautical miles before reaching the station and towed it toward the predetermined location for 60 min. The calculation of trawl time begins when the trawl warp stops being deployed, the trawl touches the seabed, and the warp is tightened under tension (marking the start of trawling). It ends when the vessel stops and the winch begins to retrieve the warp (indicating the end of net retrieval). Based on the catch data of various species at each station, parameters like catch composition, catch rate, and fishery resource density for each station and species were calculated. The study evaluated fishery resources by investigating species composition, biomass composition, quantity distribution, dominance, the distribution of economically significant species, and the biological characteristics of nekton, including fish, shrimp, crabs, mantis shrimp, and cephalopods [43]. The methods for processing and determining water quality data and fishery resources were primarily based on relevant survey standards and the literature.
The satellite data utilized in this study consisted of PlanetScope images corresponding to the survey sampling dates, ensuring that cloud cover conditions and time differences were within three days. PlanetScope satellite images are employed to acquire data with high spatial and spectral resolution. The PlanetScope data, provided by PLANET’s Dove satellite constellation (SuperDove, PSB.SD), feature high spatial resolution and rapid revisit capabilities, and the downloaded data encompass eight spectral bands. We adopted the Level 3B analytical surface reflectance products [44,45,46], which have undergone orthorectification and radiometric correction. The 2020 dataset includes only four bands (blue, green, red, and near-infrared), whereas the 2021 and 2024 datasets contain eight bands (coastal blue, blue, green I, green II, yellow, red, red edge, and near-infrared). For 2020, analyses were restricted to the four available bands due to data limitations. For 2021 and 2024, all eight bands were used, as the additional spectral information improved the accuracy of water quality assessment.

2.2. Data Analysis

With the advancement of artificial intelligence, numerous researchers have employed various machine learning techniques to invert water quality parameters [47,48,49]. Machine learning models developed based on the nonlinear relationships between information from different spectral bands and water quality parameters generally achieve higher inversion accuracy [50]. K-fold cross-validation is a widely used model evaluation method in machine learning particularly applicable in scenarios with limited data volume [51]. By repeatedly partitioning and validating the dataset, this method can evaluate model performance more robustly and mitigate the randomness introduced by a single data division. In this study, several machine learning models were combined with 5-fold cross-validation, and model parameters were optimized through a grid search to identify a suitable evaluation model for water quality parameters (Supplementary Materials Table S10).
In addition, indicators like Chla, DO, COD, and PO4-P are considered to have significant impacts on the growth of marine organisms [52,53,54]. However, for the satellite remote sensing evaluation of optically sensitive water quality parameters like Chla, colored dissolved organic matter, and suspended solids, there are clear radiative transfer models [55]. In contrast, for the satellite remote sensing evaluation of non-optically sensitive water quality parameters, such as DO, COD, and PO4-P, the radiative transfer models are unclear, and indirect or empirical methods are usually used for evaluation [55]. These two types of water quality parameters may differ in the evaluation process of machine learning algorithms. This paper attempts to conduct a correlation analysis between the SHAP algorithm and physical significance to explore the differences between optically sensitive and non-optically sensitive water quality parameters in the calculation of machine learning algorithms.
In addition, indicators like Chla, DO, COD, and PO4-P are recognized for their significant impacts on the growth of marine organisms [52,53,54]. However, for the satellite remote sensing evaluation of optically sensitive water quality parameters, such as Chla, colored dissolved organic matter, and suspended solids, established radiative transfer models exist [55]. In contrast, the evaluation of non-optically sensitive water quality parameters, including DO, COD, and PO4-P, lacks clear radiative transfer models; instead, indirect or empirical methods are typically employed for assessment [55]. These two categories of water quality parameters may exhibit differences in the evaluation processes utilized by machine learning algorithms. To explore the correlation degree and statistical reliability between water quality parameters, satellite bands, and fishery resource indicators in the Zhuhai sea area, this study adopted correlation analysis and significance test methods, with all statistical analysis processes implemented via the PyCharm platform. This paper aims to conduct a correlation analysis between the SHAP algorithm and physical significance to investigate the distinctions between optically sensitive and non-optically sensitive water quality parameters in the context of machine learning algorithm calculations.
To analyze the changes in water quality data between the reef deployment area and the control area, point data with consistent distance intervals were generated along the direction of both areas. The evaluation results of water quality parameters were then extracted for these points. By examining the variations in water quality parameters at different spatial locations, the impact of artificial reef deployment on these parameters was assessed. Furthermore, using the evaluation results of water quality parameters in conjunction with fishery resource data, the water quality data served as the feature set while the fishery resource data functioned as the label set. This approach facilitated the prediction and evaluation of fishery resources based on seawater quality parameters, allowing for an analysis of the distribution of fishery resources.

3. Results

3.1. In-Site Survey Results

In accordance with the survey specifications outlined in the methodology, as well as the relevant processing and analysis methods for water quality and fishery resources, water quality parameters and fishery resources in the marine ranching area and the control area were measured during April 2020, September 2020, September 2021, April 2024, and November 2024. The original data were collated to obtain in situ data for the constructed artificial reef area, the planned artificial reef area, the marine ranching area, and the control area, as presented in Figure 2 and Table 1.
Before the deployment of artificial reefs, the water quality parameter data in each area exhibited seasonal variations; the concentrations of DO, PO4-P, and Chla were all higher in autumn than in spring, while the concentration of COD was lower in autumn than in spring. After reef deployment, the concentration of DO in spring was generally higher than that in autumn, and the concentration of COD in autumn was higher than that in spring. The changes in PO4-P and Chla concentrations were relatively complex.
As illustrated in Figure 2a, prior to the deployment of the reef, both the planned reef area and the control area exhibited similar hydrodynamic conditions and biological activities, characterized by a uniform oxidizing environment. DO values in each area were relatively comparable, with a variation of ±0.3. In the spring of 2024, the DO levels in each area increased, revealing differences among the areas (±0.8), with the most significant change observed in the marine ranching area, which experienced an increase of +0.94. In autumn of 2021, the DO value in the constructed reef area surpassed that of the other areas; however, during other time periods, the DO in the constructed reef area was lower than in the other areas. Notably, before reef deployment, the DO levels in autumn were higher than those in spring.
As illustrated in Figure 2b, prior to reef deployment, the maximum difference in COD among various areas was ±0.1; following deployment, this maximum difference increased to ±0.5. In spring before reef deployment and spring after, COD levels decreased, with the most significant reduction occurring in the planned reef area, which saw a decrease of −0.6. In September, with the exception of the planned reef area where COD decreased, levels in other areas increased. Notably, before reef deployment, COD levels in spring were higher than those in autumn.
As illustrated in Figure 2c, prior to reef deployment, the maximum variation in PO4-P levels among different areas was ±0.004. Following deployment, the maximum variation among areas increased to ±0.008. In spring after reef deployment, all areas, except for the marine ranching area, experienced an increase in PO4-P concentrations. However, in autumn, the concentrations in all areas decreased.
As illustrated in Figure 2d, prior to reef deployment, the maximum variation in Chla among different areas was ±0.4. Following deployment, the maximum variation among areas decreased to ±0.2. In spring after reef deployment, the Chla concentration in each area increased, with the most significant rise observed in the constructed reef area. In September (autumn), the Chla concentration decreased but rose again in November.
After assessing the fishery resources and categorizing them by area, a table illustrating the differences before and after construction was generated (Table 1).
According to the results of the trawl survey conducted in April (spring), both before and after reef deployment, the number of nekton species in the construction area increased from 12 to 29, representing an increase of 17 species. In the control area, the number of nekton species rose from 19 to 54, indicating an increase of 35 species. The resource density in the construction area increased from 181 kg/km2 to 499 kg/km2, reflecting an increase of 318 kg/km2. Conversely, in the control area, it decreased from 212 kg/km2 to 169 kg/km2, resulting in a decline of 43 kg/km2. The density of individual organisms in the construction area changed from 12,994 ind/km2 to 11,408 ind/km2, marking a decrease of 1586 ind/km2. In the control area, this density decreased from 12,816 ind/km2 to 5596 ind/km2, representing a decline of 7220 ind/km2.
In autumn, following the deployment of the reef, the number of nekton species in the construction area increased from 18 to 34, representing an increase of 16 species. In contrast, the control area saw an increase from 16 to 26 nekton species, with a gain of 10 species. The resource density in the construction area rose from 275 kg/km2 to 935 kg/km2, reflecting an increase of 660 kg/km2. Meanwhile, in the control area, resource density increased from 255 kg/km2 to 456 kg/km2, resulting in a gain of 201 kg/km2. Additionally, the density of individual organisms in the construction area increased from 24,169 ind/km2 to 38,615 ind/km2, marking an increase of 14,446 ind/km2. Conversely, in the control area, the density of individuals decreased from 16,391 ind/km2 to 13,742 ind/km2, indicating a decline of 2649 ind/km2.

3.2. Statistical Analysis of Survey Results and Satellite Bands

Correlation analysis between the 8 Planetscope bands (B1–B8) and water quality parameters revealed distinct and significant correlation patterns for different indicators across the spectral bands (Figure 3).
The correlation of chlorophyll a (Chl-a) was relatively weak, with significant positive correlations observed only in bands B2, B3, and B4 (R2 = 0.37–0.51, p < 0.05 or p < 0.01), while correlations with the remaining bands were low or not significant. Dissolved oxygen (DO) exhibited significant positive correlations with all bands (R2 = 0.45–0.49, p < 0.01). Chemical oxygen demand (COD) showed significant negative correlations across all bands (R2 = −0.43 to −0.59, p < 0.05 or p < 0.01). For PO4-P, strong negative correlations were identified with all bands (R2 = −0.42 to −0.84, p < 0.01), particularly in bands B3 and B4, where the correlations were strongest (R2 < −0.80).
Overall, the statistical analysis demonstrated that the sensitivity of Planetscope bands to water quality parameters varied across the spectrum. Specifically, bands B2–B4 exhibited stronger responses to COD, PO4-P, and Chl-a, whereas bands B6–B8 were more strongly correlated with COD and DO. Correlation and significance heatmaps for individual months are provided in Supplementary Materials Figure S1.

3.3. Establishment of Water Quality Parameter Inversion Model

Using PyCharm 2020.1 x64 software, several models were developed and trained utilizing five-fold cross-validation. The reflectance values obtained from satellite data (comprising four bands) served as the feature set, while the water quality parameter data were employed as the label set. Model parameters were optimized through a grid search. The models generated using the four-band data are illustrated in Figure 4.
In the inversion of chlorophyll a (Chla Models), support vector regression (SVR) demonstrated exceptional performance, achieving a correlation coefficient close to one, a high degree of alignment between the standard deviation and the true values, and data points that were closely clustered around the true values, resulting in the best inversion accuracy. In contrast, other models performed relatively poorly. The evaluation accuracy for non-optically sensitive water quality parameters, such as DO, COD, and PO4-P, was notably low (R2 < 0.7). For the inversion of dissolved oxygen (DO Models), SVR exhibited the highest correlation coefficient with the actual data points and optimal accuracy. In the cases of chemical oxygen demand (COD Models) and phosphate (PO4-P Models), random forest (RF) achieved the highest correlation coefficient, with the standard deviation and data points closest to the true values, thereby attaining the highest accuracy.
Similarly, the reflectance values from satellite data (comprising eight bands) were utilized as the feature set, while water quality parameters served as the label set. Model parameters were optimized through a grid search. The models generated using the eight-band data are illustrated in Figure 5.
Compared to the four-band data, the eight-band data provide richer features, resulting in improved inversion accuracy for most models. Notably, the R2 value for Chla increased by approximately 0.3, while the R2 values for other non-optically sensitive water quality parameters rose from below 0.7 to above 0.7. SVR performed optimally in the inversion of all parameters, benefiting from the multidimensional information contained in the eight-band data, which enabled SVR to better capture complex relationships within the data. In contrast, linear regression achieved the lowest accuracy due to its inability to account for nonlinear responses. This suggests that the accuracy of water quality evaluation results in 2020 was lower than that in 2021 and 2024. The specific parameter ranges, step sizes, and optimal parameters are provided in Supplementary Materials Tables S1–S4. The evaluation results, including RMSE, bias, and MAE, are presented in Supplementary Materials Tables S5–S8.

3.4. Water Quality Evaluation Analysis Based on SHAP Decision Plots

To further analyze the inputs and outputs of the models, SHAP (SHapley Additive exPlanations) was utilized to evaluate each optimized model. The visualization of SHAP values provides an intuitive understanding of the influence of various spectral bands on the outputs of water quality assessment models. This approach facilitates the examination of model decision making processes, the identification of key spectral bands, and the subsequent optimization of marine water quality inversion models. For the established inversion models, the SHAP algorithm was employed to assess the contributions of four and eight input parameters (reflectance values from four and eight bands of the PlanetScope satellite) in calculating the concentrations of different water quality parameters, resulting in SHAP decision plots (Figure 6).
Figure 6 presents the SHAP decision plots based on four satellite bands. The horizontal axis represents the output values of the water quality parameter models, while the vertical axis denotes the different satellite bands. The trends and distributions of the SHAP contribution path lines reflect the contributions of each band to the prediction results of the water quality parameters. Here, b1–b8 correspond to the coastal blue, blue, green I, green, yellow, red, red edge, and NIR bands of the PlanetScope satellite, respectively. As shown in Figure 6, B4 is a key contributing feature for Chla and COD, while b8 and b2 are important features for DO and PO4-P. As indicated in Figure 4 and Figure 5, utilizing eight bands for water quality assessment yields higher accuracy. Therefore, the models based on eight bands are analyzed in detail in Figure 7.
As illustrated in Figure 7, the SHAP decision plot for Chla reveals significant variations in the influence of different spectral bands on the prediction of water quality parameters. Bands like Band 3, Band 4, and Band 1 exhibit lines with relatively extensive coverage on the horizontal axis, indicating that these bands have a more substantial impact on the prediction outcomes for water quality parameters and serve as key reference signals for the model when assessing water quality. Notably, certain bands (e.g., Band 3) consistently appear in high-contribution regions within the plots for Chla, DO, and COD. The importance of Band 3 for predicting water quality parameters remains stable, suggesting that its spectral information is essential for the model’s assessments, regardless of variations in the sample data.
In addition, Chla is a typical optically active parameter, and its concentration directly influences the spectral characteristics of aquatic environments. The correlation between satellite bands and Chla is governed by clear optical mechanisms. This strong physical correlation ensures that the contribution rules learned by the model are highly consistent; specifically, the direction in which bands promote (or inhibit) the predicted values remains stable, with few intersections in the SHAP contribution path lines.
For non-optically active parameters, such as DO, COD, and PO4-P, the correlation with spectral data is indirect and weakly coupled. In various water bodies, the proportional relationship between their concentrations and optical components is inconsistent, resulting in a lack of stable correlations between the spectral signals captured by satellite bands and these parameters. The model must adapt to complex interferences, as the direction and intensity of band contributions fluctuate depending on the sample scenarios, leading to increased intersections in the SHAP contribution path lines. Overall, due to strong optical mechanisms, Chla exhibits clear contribution rules and is suitable for highly interpretable models enhanced by physical mechanisms. In contrast, DO, COD, and PO4-P are constrained by non-optical properties and necessitate the integration of multi-source features and mechanistic constraints. For the density scatter plots of SHAP values, please refer to the attached document.

3.5. Application of the Optimized Water Quality Parameter Evaluation Model

Using the optimized models for each water quality parameter, we conducted quantitative evaluations of Chla, DO, COD, and PO4-P concentrations in satellite images. This analysis aimed to determine the temporal and spatial variations in water quality in the marine area near Wailingding Island, both before and after the deployment of artificial reefs (Figure 8).
Similarly, the water quality parameters in the sea area near Wailingding Island were evaluated after the deployment of artificial reefs using the optimized models. The results are illustrated in Figure 9.
To analyze the spatial changes in water quality parameters following reef deployment, equidistant points (0.15 km apart) were established along the trajectory from the reef area to the control area. Data from the water quality evaluation results were extracted at these designated points to assess the spatial distribution characteristics of water quality parameters before and after reef deployment (Figure 10).
Figure 10 presents the analysis results of Chla distribution and the associated profile changes in the marine ranching area, with the aim of exploring the dynamic variations of Chla at different spatial locations. The two curves—representing the constructed area and the planned area—illustrate the trends in chlorophyll a concentration along the profile in the construction area and its surrounding regions at various time points. Prior to reef deployment, there were no significant changes in Chla concentration across the different areas (R2 = 0.01 and 0.23). However, following reef deployment, the Chla concentration in the reef area increased significantly, while it continued to decrease as the distance from the reef area increased (R2 = 0.78 and 0.81). Additionally, as shown in Figure 10a, within the first 3 km, there was no noticeable area where Chla concentration decreased. On average, the concentration decreased by 0.008 mg/L for every 1 km distance from the reef area.
Similarly, the concentration data of DO, COD, and PO4-P were extracted using the points generated in Figure 10. The same analysis was conducted, resulting in Figure 11.
As illustrated in Figure 11a,b, prior to the deployment of the reef, DO concentration exhibited an increasing trend (R2 = 0.50 and 0.57). Following reef deployment, the DO concentration in the reef area increased significantly by 0.8 mg/L, but it continued to decline as the distance from the reef area increased (R2 = 0.70 and 0.79). Furthermore, Figure 11a indicates that within the first 3 km, there was no significant area where the DO concentration decreased (±0.3 mg/L). On average, the concentration decreased by 0.12 mg/L for every kilometer away from the reef area until reaching a distance of approximately 10 km, where the concentration was comparable to that of the control area.
As illustrated in Figure 11c,d, the concentration of COD in the northern region exhibited minimal cross-regional variation before and after the deployment of the reef (R2 = 0.18 and 0.30). Following reef deployment, the COD concentration decreased by approximately 0.6 mg/L. In contrast, the southern region displayed an increasing trend in COD concentration from the reef area to the control area prior to reef deployment (R2 = 0.73). However, this increasing trend diminished after reef deployment (R2 = 0.47). Similarly, the COD concentration in the southern region also decreased by about 0.6 mg/L post-deployment.
As illustrated in Figure 11e,f, prior to the deployment of the reef, the concentration of PO4-P exhibited a decreasing trend (R2 = 0.22 and 0.69). Following reef deployment, the concentration of PO4-P in the reef area decreased, and this decline persisted as the distance from the reef area increased (R2 = 0.52 and 0.63). Overall, there was a discernible trend indicating that the farther one is from the reef area, the lower the concentration of PO4-P.

3.6. Fishery Resources Assessment Based on Water Quality Parameters

Variance inflation factor (VIF) analysis was conducted on the selected water quality parameters (Chl-a, DO, COD, and PO4-P) (Table 2). The results showed that Chl-a (VIF = 15.93) and DO (VIF = 19.07) exhibited high multicollinearity, while COD (VIF = 6.18) and PO4-P (VIF = 7.88) showed relatively lower levels of collinearity.
Using the water quality assessment results calculated in Section 3.5 as inputs and fishery resources data as outputs, several assessment models were established (Figure 12).
In the analysis of fishery resources, the linear regression predictions for April 2020 demonstrated a high degree of correlation, standard deviation, and alignment with actual values (represented by black dots). In contrast, the results from other models exhibited varying degrees of deviation from the actual values. For April 2024, the Lasso regression predictions performed well in relation to the actual values, achieving a high correlation coefficient and appropriate standard deviation. Overall, based on the indicators from the Taylor diagram, Lasso regression exhibited more stable and superior predictive performance in fishery modeling at these two time points. Consequently, Lasso regression was selected for the assessment of fishery resources. By utilizing the four water quality data points collected before and after reef deployment as feature sets and the corresponding fishery resources data as label sets, the Lasso model was employed to generate the fishery resources distribution map (Figure 13). Chla, COD, and PO4-P were used to assess fishery resources (refer to Supplementary Materials Figure S2).
As illustrated in Figure 13, prior to the deployment of artificial reefs, fishery resources were predominantly concentrated in the control area (212 kg/km2), while the distribution in the coastal area was comparatively low (181 kg/km2). This observation is consistent with the on-site trawling results presented in Table 1. Following the deployment of artificial reefs, the quantities of resources in both the planned area and the control area experienced an increase, which also corresponds with the on-site trawling results (Table 1). According to the findings from on-site sampling and assessment, the fishery resources in the planned area increased by 516 to 1162 kg after the deployment of artificial reefs.

4. Discussion

Marine ranching plays a crucial role in marine ecological restoration and the sustainable utilization of fishery resources. Consequently, the evaluation and management of its construction effects have become significant research areas. The deployment of artificial reefs can induce multi-scale changes in water quality and fishery resources, necessitating systematic analysis. In this study, in situ data on water quality and fishery resources were collected before and after the establishment of marine ranching to analyze the temporal and spatial changes associated with this development. Additionally, several models were employed to quantitatively assess water quality data, examining the accuracy of different models in water quality evaluation. The SHAP algorithm was further utilized to analyze the differences in calculation methods among machine learning algorithms when evaluating optically sensitive and non-optically sensitive water quality parameters. Finally, based on the inverted water quality data, the potential impact range and resource output following the deployment of artificial reefs were investigated.

4.1. Evaluation of Water Quality Parameter Assessment Models

Overall, SVR demonstrated optimal performance in the inversion of all water quality parameters. This effectiveness is attributed to the fact that marine water quality parameters are significantly influenced by nonlinear processes, which are intertwined with physical, biological, and chemical factors. SVR is capable of capturing these complex relationships through multi-factor interactions and decision tree structures. In contrast, linear regression exhibited the lowest accuracy in the inversion of all parameters, as it fails to account for the nonlinear responses of water quality parameters. In terms of input parameters, the eight-band feature set shows greater diversity in band contributions compared to the four-band feature set. The additional bands enhance the interpretive dimensions through interactive effects, allowing for the capture of more intricate coupling relationships while providing more informational references for model optimization.
By correlating SHAP decision plots with their physical interpretations, the contribution of spectral bands to Chla is based on direct spectral responses, meaning there is a one-to-one relationship between satellite bands and Chla concentration parameters. This correlation results in clear trends in SHAP contribution paths with minimal intersections. In contrast, for non-optically sensitive parameters, such as DO, COD, and PO4-P, the relationship follows an indirect pathway: satellite bands → intermediate optical components → DO, COD, or PO4-P concentration parameters [55]. For instance, when considering COD, the direction of band contributions to its prediction can vary significantly depending on the pollution source of the samples. This variability can lead to frequent reversals in the direction of SHAP contribution paths; for example, the same band may initially increase COD predictions due to colored dissolved organic matter (CDOM) and subsequently decrease them due to suspended solids in different samples [56]. Such multi-link and multi-interference pathways disrupt the stability of band contributions, resulting in disordered and highly intersecting SHAP contribution paths. Furthermore, among the non-optically sensitive water quality parameters (DO, COD, and PO4-P), B3 is a significant contributor to the assessment of both DO and COD. Specifically, for DO, the most influential bands are B3 and B4, which align with the optically sensitive parameter Chla. This suggests that variations in Chla concentration have a substantial impact on local DO levels and a certain influence on COD [57].
Therefore, the difference in the regularity of SHAP contribution paths may also define the boundary of remote sensing interpretability for parameters; SHAP plots of optically sensitive parameters (e.g., Chla) have clear physical meanings, while non-optically sensitive parameters (e.g., DO, COD, PO4-P) require more complex feature engineering and mechanism analysis. This provides a direction for the development and validation of water quality remote sensing models—for parameters like COD, focus should be placed on feature interaction and interference factor separation to improve model reliability.
Because water quality parameters, such as Chla, DO, COD, and PO4-P, may exhibit skewed distribution, logarithmic transformation can effectively reduce the impact of extreme values, make data dispersion more uniform, and prevent the model’s generalization ability from decreasing due to excessive fluctuations in some samples, thereby improving model performance. Therefore, this study also performed logarithmic transformation on the four water quality parameters and compared the results with those of the raw data (Supplementary Materials Table S9). The results show that the use of logarithmic transformation can improve the accuracy of the assessment model.

4.2. Changes in Fishery Resources and Water Quality Parameters Before and After Reef Deployment

Fishery resources in the Zhuhai sea area exhibit clear seasonal dynamics, with autumn serving as the traditional peak fishing season due to superior individual counts, biomass, and catchable sizes (Table 1). These seasonal differences are shaped not only by life history traits (e.g., reproduction, maturation, and migration) but also environmental stability. Autumn conditions, characterized by reduced Pearl River runoff, stable salinity, and calm seas, are generally more favorable for fish aggregation, whereas spring conditions are often influenced by monsoon transitions and runoff fluctuations.
Reef deployment was associated with enhanced phytoplankton growth (Chla), forage organisms, and fishery resources, with stronger effects observed in autumn (Table 1, Figure 2). DO exhibited contrasting seasonal dynamics; increases in spring reflected low biomass and limited oxygen consumption, while autumn declines were likely driven by intensive respiration and organic decomposition. Seasonal differences in COD and PO4-P concentrations were also evident, with spring decreases linked to rapid plankton uptake and autumn increases associated with excreta input and sediment nutrient release [58,59].
Overall, the deployment of artificial reefs was observed to amplify seasonal differences in both resources and biogeochemical processes, particularly during autumn. Nonetheless, it is important to acknowledge that these dynamics are also shaped by external drivers, such as climate variability and fishing pressure. The inherent complexity and unpredictability of these factors make it difficult to disentangle the specific proportion of changes directly attributable to reef deployment versus broader environmental influences [60,61,62].

4.3. Decision Making for Artificial Reef Deployment

In marine ranching management, understanding species-specific responses to reef-related environmental changes is fundamental. Reef deployment alters physical, chemical, and biological conditions, and species vary in their responses depending on their ecological and physiological traits [63,64,65].
Species distribution models, integrated with simulations of post-deployment environmental changes, offer a promising approach for predicting resource distribution and supporting management decisions [66,67]. For example, changes in water temperature or salinity following reef placement can be incorporated into such models to estimate shifts in resource abundance and distribution, thereby guiding fishing and conservation strategies.
Moreover, incorporating variables like water quality parameters, reef area and volume, and dominant species into resource increment models enables quantification of factor-specific contributions to resource growth. To enhance robustness, case studies across different regions should be accumulated, reducing the influence of site-specific conditions and improving generalizability.
Attention must also be given to multicollinearity among environmental variables. As noted in Section 4.1, interactions between Chla and DO highlight the need for diagnostic tools, such as variance inflation factor (VIF) analysis, to prevent overfitting and improve model stability [68,69].
Finally, management decisions must account for the carrying capacity of local ecosystems. Excessive reef deployment or resource exploitation beyond ecological limits can undermine sustainability. Integrating environmental carrying capacity models will help determine appropriate deployment scales and exploitation intensities, ensuring a balance between ecological protection and resource utilization.

5. Conclusions

In this study, in situ data on water quality and fishery resources before and after the establishment of marine ranching were used to analyze temporal and spatial changes in the marine ranching area. Additionally, several models were applied for quantitative assessment of water quality data to explore the accuracy of different models in water quality evaluation. The SHAP algorithm was correlated with physical significance to investigate differences in machine learning calculation methods for evaluating optically sensitive and non-optically sensitive water quality parameters. Finally, based on the inverted water quality data, the potential impact range and resource output after artificial reef deployment were explored. The results show that in the sea area near Wailingding Island, Zhuhai, the deployment of artificial reefs with a volume of 38,048 cubic meters led to an increase in fishery resources by 318 kg/km2 in spring and 660 kg/km2 in autumn. Moreover, the deployment affected the concentrations of Chla, DO, COD, and PO4-P in surface waters within a range of approximately 10 km around the reef area.

Supplementary Materials

The following supporting information can be downloaded at https://www.mdpi.com/article/10.3390/rs17183140/s1, Figure S1: Heatmap of correlation and significance between satellite bands and water quality parameters in various months. (a–e) respectively represent April 2024, November 2024, September 2021, April 2020, and September 2020, and (f) shows the mutual correlation and significance between water quality parameters and fishery resources. Figure S2: Model application for assessing fishery resources based on water quality parameters (kg/km2). (a) 202004, (b) 202404. Table S1: The parameter table of the optimal model (Ridge regression). Table S2: The parameter table of the optimal model (Lasso regression). Table S3: The parameter table of the optimal model (Random Forest regression). Table S4: The parameter table of the optimal model (Support Vector Regression). Table S5: Stability of evaluation results (Chla). Table S6: Stability of evaluation results (DO). Table S7: Stability of evaluation results (COD). Table S8: Stability of evaluation results (PO4-P). Table S9: Comparison of Assessment Results Based on Raw Data and Log-Transformed Data. Table S10: Performance Comparison of Different Models.

Author Contributions

H.Y., Y.C., X.F., F.T. and C.G. acquired the in situ data and sampling data; J.C. processed/analyzed the data and interpreted the results, together with P.C., H.Y., Z.L., L.Z., J.Z., Y.C., X.F., F.T. and C.G.; J.C., H.Y. and P.C. prepared the manuscript. All authors contributed to the review of the manuscript. Project administration, J.C., H.Y., C.G. and P.C. All authors conceived the initial design of the research. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the Fund of the Guangdong Basic and Applied Basic Research Foundation (2024A1515510037) and the Central Public-interest Scientific Institution Basal Research Fund, CAFS (2023TD06), and it was supported by the Foundation of Guangdong Provincial Field Observation and Research Station for Marine Ecosystem in Hanjiang River Estuary-Nanao Island Area (HNS202404).

Data Availability Statement

The relevant data in this paper may be obtained by contacting the corresponding author or first author, who should be informed of specific data requirements and intended use.

Acknowledgments

We appreciate the support of our funding agency. We also thank the editor and the anonymous reviewers and editors, whose comments significantly improved the manuscript. We also thank PlanetScope for the data.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Jiao, M.; Yue, W.; Suo, A.; Zhang, L.; Li, H.; Xu, P.; Ding, D. Construction and influencing factors of an early warning system for marine ranching ecological security: Experience from China’s coastal areas. J. Environ. Manag. 2023, 335, 117515. [Google Scholar] [CrossRef]
  2. Zhou, X.; Zhao, X.; Zhang, S.; Lin, J. Marine Ranching Construction and Management in East China Sea: Programs for Sustainable Fishery and Aquaculture. Water 2019, 11, 1237. [Google Scholar] [CrossRef]
  3. Yu, J.; Zhang, L. Evolution of marine ranching policies in China: Review, performance and prospects. Sci. Total Environ. 2020, 737, 139782. [Google Scholar] [CrossRef] [PubMed]
  4. Liu, S.; Zhou, X.; Zeng, C.; Frankstone, T.; Cao, L. Characterizing the development of Sea ranching in China. Rev. Fish Biol. Fish. 2022, 32, 783–803. [Google Scholar] [CrossRef]
  5. Wang, Y.; Zhang, W. Maximum sustainable yield estimation of enhancement species with the characteristics of movement inside and outside marine ranching. J. Ocean Limnol. 2021, 39, 2380–2387. [Google Scholar] [CrossRef]
  6. Yan, J.; Chen, Y.; Cao, Y.; Sun, J.; Wen, B.; Gao, X.; Wang, G.; Gong, L.; Liu, H.; Li, Q.; et al. Marine ranching enhances ecosystem stability and biological carbon sequestration potential: Insights from Ecopath with Ecosim model simulation of 30-year ecological path of a national marine ranching in China. Front. Mar. Sci. 2025, 12, 1583896. [Google Scholar] [CrossRef]
  7. Wang, X.; Ren, J.S.; Zhong, Y.; Wu, W.; Liu, Y.; Jiang, Y.; Zhang, J. A dynamic energy budget model for black rockfish Sebastes schlegelii: Parameterization and application in marine ranching areas, Yellow Sea, China. Ocean Coast. Manag. 2023, 245, 106852. [Google Scholar] [CrossRef]
  8. Qin, M.; Wang, X.; Du, Y. Factors affecting marine ranching risk in China and their hierarchical relationships based on DEMATEL, ISM, and BN. Aquaculture 2022, 549, 737802. [Google Scholar] [CrossRef]
  9. Liang, J.; Zhang, M.; Yin, Z.; Niu, K.; Li, Y.; Zhi, K.; Huang, S.; Yang, J.; Xu, M. Tripartite evolutionary game analysis and simulation research on zero-carbon production supervision of marine ranching against a carbon-neutral background. Front. Ecol. Evol. 2023, 11, 1119048. [Google Scholar] [CrossRef]
  10. Le, C.Y.; Feng, J.C.; Sun, L.; Yuan, W.; Wu, G.; Zhang, S.; Yang, Z. Co-benefits of carbon sink and low carbon food supply via shellfish and algae farming in China from 2003 to 2020. J. Clean. Prod. 2023, 414, 137436. [Google Scholar] [CrossRef]
  11. Yu, L.; Zhu, J.; Guan, X.; Zhai, L.; Zhao, Z.; Mu, Y.; Liu, T. Carbon sink efficiency of Chinese mollusk mariculture. Aquaculture 2025, 596, 741781. [Google Scholar] [CrossRef]
  12. Zhang, Y.; Ou, Z.; Tweedley, J.R.; Loneragan, N.R.; Zhang, X.; Tian, T.; Wu, Z. Evaluating the effectiveness of baited video and traps for quantifying the mobile fauna on artificial reefs in northern China. J. Exp. Mar. Biol. Ecol. 2024, 573, 152001. [Google Scholar] [CrossRef]
  13. Yu, J.; Wang, Z.; Zhang, S.; Cai, P.; Mao, J.; Li, J. Nursing Effects of Large-Scale Floating Raft Aquaculture Habitats on Conger myriaster: A Perspective from Marine Ranching. Water 2025, 17, 431. [Google Scholar] [CrossRef]
  14. Tang, S.; Xu, P.; Zhang, S.; Yin, Y.; Liang, J.; Jiang, Y.; Xu, K.; Li, J.; Feng, J.; Gao, J. An idea for marine ranching planning based on ocean currents and its practice in Zhongjieshan Archipelago National Marine Ranching, China. Aquaculture 2025, 596, 741780. [Google Scholar] [CrossRef]
  15. Fang, G.; Yu, H.; Sheng, H.; Tang, Y.; Liang, Z. Comparative analysis of microbial communities between water and sediment in Laoshan Bay marine ranching with varied aquaculture activities. Mar. Pollut. Bull. 2021, 173 Pt A, 112990. [Google Scholar] [CrossRef]
  16. Qin, C.; Chen, P.; Zhang, A.; Feng, X.; Yuan, H.; Li, X.; Yu, J.; Leber, K.M. Impacts of marine ranching construction on sediment pore water characteristic and nutrient flux across the sediment-water interface in a subtropical marine ranching (Zhelin Bay, China). Appl. Ecol. Environ. Res. 2018, 16, 163–179. [Google Scholar] [CrossRef]
  17. Lee, S.I.; Zhang, C.I. Evaluation of the Effect of Marine Ranching Activities on the Tongyeong Marine Ecosystem. Ocean Sci. J. 2018, 53, 557–582. [Google Scholar] [CrossRef]
  18. Cheng, X.; Wang, Z.; Zhang, S.; Zhao, X.; Lin, J.; Huang, H.; Chen, Y.; Zou, Q. Analysis of short-term and local scale variations in fish community structure in Dachen Island waters. Front. Mar. Sci. 2023, 10, 1199524. [Google Scholar] [CrossRef]
  19. Lee, J.; Nam, J. Lessons and Outcomes from the Marine Ranching Program in Korea. J. Korean Isl. 2020, 32, 179–199. [Google Scholar] [CrossRef]
  20. Zhang, S.; Zhou, X.; Wang, K.; Li, J.; Zhao, J.; Zhao, X.; Guo, Y.; Liu, S.; Cheng, X. Review of marine livestock ecological urbanization hypothesis and marine ranching construction key-technology against blue growth background. J. Fish. China 2019, 43, 81–96. [Google Scholar]
  21. Du, Y.-W.; Jiang, J.; Li, C.-H. Ecological efficiency evaluation of marine ranching based on the Super-SBM model: A case study of Shandong. Ecol. Indic. 2021, 131, 108174. [Google Scholar] [CrossRef]
  22. Seaman, W. Artificial habitats and the restoration of degraded marine ecosystems and fisheries. Hydrobiologia 2007, 580, 143–155. [Google Scholar] [CrossRef]
  23. United Nations General Assembly. Transforming Our World: The 2030 Agenda for Sustainable Development; United Nations: New York, NY, USA, 2015. [Google Scholar]
  24. State Council of the People’s Republic of China. Outline of the Fourteenth Five—Year Plan for National Economic and Social Development of the People’s Republic of China and the Vision for 2035; The State Council of the People’s Republic of China: Beijing, China, 2021.
  25. Du, Y.-W.; Huang, Y.-N. Evaluation of marine ranching socio-ecological system resilience in China. Reg. Stud. Mar. Sci. 2024, 77, 103658. [Google Scholar] [CrossRef]
  26. Kitada, S. Economic, ecological and genetic impacts of marine stock enhancement and sea ranching: A systematic review. Fish Fish. 2018, 19, 511–532. [Google Scholar] [CrossRef]
  27. Zeng, L.; Tang, Z.; Chen, P.; Yu, J.; Chen, G. Optimization of fishery resources assessment methods and ecological effects evaluation of artificial reefs. Mar. Biol. Res. 2021, 17, 72–85. [Google Scholar] [CrossRef]
  28. Rizzi, J.; Torresan, S.; Critto, A.; Zabeo, A.; Brigolin, D.; Carniel, S.; Pastres, R.; Marcomini, A. Climate change impacts on marine water quality: The case study of the Northern Adriatic sea. Mar. Pollut. Bull. 2016, 102, 271–282. [Google Scholar] [CrossRef]
  29. Leftheriotis, G.A.; Sibetheros, I.A.; Dimas, A.A. Large-eddy simulation of the turbulent oscillatory flow around an artificial reef. Ocean Eng. 2025, 318, 120142. [Google Scholar] [CrossRef]
  30. Gao, S.; Zhang, S.; Feng, Z.; Lu, J.; Fu, G.; Yu, W. The ecological risk and fate of microplastics in the environmental matrices of marine ranching area in coastal water. J. Hazard. Mater. 2024, 473, 134570. [Google Scholar] [CrossRef]
  31. Zhang, J.; Zhu, L.; Liang, Z.; Sun, L.; Nie, Z.; Wang, J.; Xie, W.; Jiang, Z. Numerical Study of Efficiency Indices to Evaluate the Effect of Layout Mode of Artificial Reef Unit on Flow Field. J. Mar. Sci. Eng. 2021, 9, 770. [Google Scholar] [CrossRef]
  32. Inoue, N.; Nambu, R.; Kuwahara, H.; Kuwamoto, J.; Yokoyama, J.; Kanaiwa, M. The present situation of utilization and the effects of artificial reefs on the resource density of Japanese butterfish Hyperoglyphe japonica and red seabream Pagrus major in the waters of the Iki Islands, Nagasaki, Japan. Nippon Suisan Gakkaishi 2018, 84, 1010–1016. [Google Scholar] [CrossRef]
  33. Zhou, W.; Xu, P.; Ding, D.; Suo, A.; Xie, M.; Zhang, L.; Yue, W.; He, W. Trophic structure of fishes and macroinvertebrates in relation to environmental indicators in artificial reef ecosystems of Pearl River Estuary. Ecol. Indic. 2022, 138, 108823. [Google Scholar] [CrossRef]
  34. Meshesha, T.W.; Wang, J.; Melaku, N.D. Modelling spatiotemporal patterns of water quality and its impacts on aquatic ecosystem in the cold climate region of Alberta, Canada. J. Hydrol. 2020, 587, 124952. [Google Scholar] [CrossRef]
  35. Scapin, L.; Zucchetta, M.; Pranovi, F.; Franzoi, P. Influence of Seascape on Coastal Lagoon Fisheries: The Role of Habitat Mosaic in the Venice Lagoon. Estuaries Coasts 2022, 45, 793–811. [Google Scholar]
  36. Harris, J.M.; Vinobaba, P.; Kularatne, R.K.A.; Khan, S.A. Fish diversity and assemblage in the Batticaloa lagoon, Sri Lanka. J. Fish. Biol. 2023, 102, 773–793. [Google Scholar] [CrossRef]
  37. GB17378-2007; The Specification for Marine Monitoring. Standards Press of China: Beijing, China, 2007.
  38. GB/T 12763.6-2007; General Administration of Quality Supervision, Inspection and Quarantine of the People’s Republic of China, & Standardization Administration of the People’s Republic of China. Specifications for Oceanographic Survey Part 6: Marine Biological Survey (S). Standards Press of China: Beijing, China, 2007.
  39. Chen, J.; Wang, K.; Li, X.; Zhao, X.; Cheng, X.; Liu, Z.; Zhang, J.; Zhang, S. Remote sensing inversion of water quality parameters (TSM, Chl-a, and CDOM) in subtidal seaweed beds and surrounding waters. Ecol. Indic. 2024, 167, 112716. [Google Scholar] [CrossRef]
  40. Banse, K.; Naqvi, S.W.A.; Narvekar, P.V.; Postel, J.R.; Jayakumar, D.A. Oxygen minimum zone of the open Arabian Sea: Variability of oxygen and nitrite from daily to decadal timescales. Biogeosciences 2014, 11, 2237–2261. [Google Scholar] [CrossRef]
  41. Fujimori, K.; Ma, W.; Moriuchi-Kawakami, T.; Shibutani, Y.; Takenaka, N.; Bandow, H.; Maeda, Y. Chemiluminescence Method with Potassium Permanganate for the Determination of Organic Pollutants in Seawater. Anal. Sci. 2001, 17, 975–978. [Google Scholar] [CrossRef] [PubMed]
  42. Altahan, M.F.; Esposito, M.; Achterberg, E.P. Improvement of On-Site Sensor for Simultaneous Determination of Phosphate, Silicic Acid, Nitrate plus Nitrite in Seawater. Sensors 2022, 22, 3479. [Google Scholar] [CrossRef] [PubMed]
  43. Kronbak, L.G.; Nielsen, J.R.; Jørgensen, O.A.; Vestergaard, N. Bio-economic evaluation of implementing trawl fishing gear with different selectivity. J. Environ. Manag. 2009, 90, 3665–3674. [Google Scholar] [CrossRef]
  44. Frazier, A.E.; Hemingway, B.L. A Technical Review of Planet Smallsat Data: Practical Considerations for Processing and Using PlanetScope Imagery. Remote Sens. 2001, 13, 3930. [Google Scholar] [CrossRef]
  45. Zhao, Y.; Lee, C.K.; Wang, Z.; Wang, J.; Gu, Y.; Xie, J.; Law, Y.K.; Song, G.; Bonebrake, T.C.; Yang, X.; et al. Evaluating fine-scale phenology from PlanetScope satellites with ground observations across temperate forests in eastern North America. Remote Sens. Environ. 2022, 283, 113310. [Google Scholar] [CrossRef]
  46. Roy, D.P.; Huang, H.; Houborg, R.; Martins, V.S. A global analysis of the temporal availability of PlanetScope high spatial resolution multi-spectral imagery. Remote Sens. Environ. 2021, 264, 112586. [Google Scholar] [CrossRef]
  47. Zhu, L.; Cui, T.; Runa, A.; Pan, X.; Zhao, W.; Xiang, J.; Cao, M. Robust remote sensing retrieval of key eutrophication indicators in coastal waters based on explainable machine learning. ISPRS J. Photogramm. Remote Sens. 2024, 211, 262–280. [Google Scholar] [CrossRef]
  48. Wang, Y.; Wang, K.; Wu, D.; Xu, D.; Meng, Q.; Zhou, F.; Zheng, M.; Zheng, H. Subsurface hypoxia observation in the Changjiang estuary based on a wave-driven profiler, satellite data, and machine learning. J. Geophys. Res. Ocean. 2025, 130, e2024JC022142. [Google Scholar] [CrossRef]
  49. Qin, T.; Liang, T.; Fan, D.; He, H.; Lan, G.; Fu, B. A novel hybrid machine learning approach for accurate retrieval of ocean surface chlorophyll-a across oligotrophic to eutrophic waters. Environ. Res. 2025, 279 Pt 2, 121864. [Google Scholar] [CrossRef]
  50. Kalu, I.; Ndehedehe, C.E.; Okwuashi, O.; Eyoh, A.E.; Ferreira, V.G. Reconstructing terrestrial water storage anomalies using convolution-based support vector machine. J. Hydrol. Reg. Stud. 2023, 46, 101326. [Google Scholar] [CrossRef]
  51. Poursaeid, M.; Poursaeed, A.H.; Shabanlou, S. Water Resources Quality Indicators Monitoring by Nonlinear Programming and Simulated Annealing Optimization with Ensemble Learning Approaches. Water Resour. Manag. 2025, 39, 1073–1087. [Google Scholar] [CrossRef]
  52. Zhou, F.; Lu, X.; Chen, F.; Zhu, Q.; Meng, Y.; Chen, C.; Lao, Q.; Zhang, S. Spatial-Monthly Variations and Influencing Factors of Dissolved Oxygen in Surface Water of Zhanjiang Bay, China. J. Mar. Sci. Eng. 2020, 8, 403. [Google Scholar] [CrossRef]
  53. Omoni, V.T.; Bankole, P.O.; Omoche, O.; Obida, C.; Igben, C.; Stephen, O.E.; Ogwo, E.I.; Torjir, D.N. Evaluation of the effects of abattoir effluent on the physicochemical and bacteriological quality of River Benue, Nigeria. Environ. Monit. Assess. 2023, 195, 146. [Google Scholar] [CrossRef]
  54. Azizpour, J.; Manbohi, A.; Rahnama, R.; Hamzepour, A.; Bastami, K.D.; Bagheri, H.; Taheri, M.; Farjami, H.; Mehdinia, A. Environmental impacts of fish cage cultures in the southern Caspian Sea. Environ. Res. 2025, 266, 120574. [Google Scholar] [CrossRef]
  55. Yang, H.; Kong, J.; Hu, H.; Du, Y.; Gao, M.; Chen, F. A Review of Remote Sensing for Water Quality Retrieval: Progress and Challenges. Remote Sens. 2022, 14, 1770. [Google Scholar] [CrossRef]
  56. Feng, Y.; He, H.; Wang, L.; Ji, Y.; Wang, B.; Chen, B.; He, S.; Feng, Y.; Xue, L.; Xing, B. Hydrothermal carbonization aqueous phase applied to the rice paddy: Interaction between soil DOM and bacterial community on runoff water quality. Chem. Eng. J. 2025, 512, 162495. [Google Scholar] [CrossRef]
  57. Majnooni, S.; Fooladi, M.; Nikoo, M.R.; Al-Rawas, G.; Haghighi, A.T.; Nazari, R.; Al-Wardy, M.; Gandomi, A.H. Smarter water quality monitoring in reservoirs using interpretable deep learning models and feature importance analysis. J. Water Process Eng. 2024, 60, 105187. [Google Scholar] [CrossRef]
  58. Zhang, S.; Li, M.; Zhu, J.; Xu, S.; Chen, Z. Diversity in habitat use by the East Asian fourfinger threadfin Eleutheronema rhadinum revealed by otolith Sr:Ca and Ba:Ca profiles. Aquat. Biol. 2022, 31, 89–97. [Google Scholar] [CrossRef]
  59. Liu, Z.; Zhang, J.; Chen, J.; Zeng, J.; Hu, Y.; Wang, K.; Zhang, S. The impact of dissolved oxygen and sediments on the decomposition of Sargassum thunbergii. Mar. Biol. 2025, 172, 28. [Google Scholar] [CrossRef]
  60. Zhao, X.; Zhang, Y.; Tang, H.; Zhang, Q.; Cui, C.; Qin, B. Water quality improvements and ecological inertia: A basin-wide assessment of ecosystem recovery in the Yangtze River system (2005–2022). Ecol. Indic. 2025, 176, 113700. [Google Scholar] [CrossRef]
  61. Chen, Z.; Liang, C.; Xian, W.; Pauly, D. Drivers of spatiotemporal community variations in estuarine ecosystems: A case study of the waters adjacent to the Yangtze Estuary. Mar. Pollut. Bull. 2025, 217, 118078. [Google Scholar] [CrossRef] [PubMed]
  62. Martin, P.; Haniffa, M.A.; Arunachalam, M. Abundance and diversity of macroinvertebrates and fish in the Tamiraparani river, South India. Hydrobiologia 2000, 430, 59–75. [Google Scholar] [CrossRef]
  63. Santos, M.N.; Monteiro, C.C. A fourteen-year overview of the fish assemblages and yield of the two oldest Algarve artificial reefs (southern Portugal). Hydrobiologia 2007, 580, 225–231. [Google Scholar] [CrossRef]
  64. Wang, Z.; Yao, L.; Yu, J.; Chen, P.; Li, Z.; Yang, W. Evaluation of the ecological carrying capacity of Wailingding marine ranching in Zhuhai, China by high-resolution remote sensing. Front. Mar. Sci. 2024, 11, 1354407. [Google Scholar] [CrossRef]
  65. Li, C.; Liu, Y.; Yin, Z.; Si, Z.; Li, Q.; Saitoh, S.-I. Evaluation of the Pacific oyster marine aquaculture suitability in Shandong, China based on GIS and remote sensing. Front. Mar. Sci. 2024, 11, 1402528. [Google Scholar] [CrossRef]
  66. Layman, C.A.; Allgeier, J.E.; Yeager, L.A.; Stoner, E.W. Thresholds of ecosystem response to nutrient enrichment from fish aggregations. Ecology 2013, 94, 530–536. [Google Scholar] [CrossRef]
  67. Hylkema, A.; Debrot, A.O.; Osinga, R.; Bron, P.S.; Heesink, D.B.; Izioka, A.K.; Reid, C.B.; Rippen, J.C.; Treibitz, T.; Yuval, M.; et al. Fish assemblages of three common artificial reef designs during early colonization. Ecol. Eng. 2020, 157, 105994. [Google Scholar] [CrossRef]
  68. Zhang, J.; Fu, P.; Meng, F.; Yang, X.; Xu, J.; Cui, Y. Estimation algorithm for chlorophyll-a concentrations in water from hyperspectral images based on feature derivation and ensemble learning. Ecol. Inform. 2022, 71, 101783. [Google Scholar] [CrossRef]
  69. Yang, J.; Deng, R.; Ma, Y.; Li, J.; Guo, Y.; Lei, C. Satellite Retrieval and Spatiotemporal Variability in Chlorophyll-a for Marine Ranching: An Example from Daya Bay, Guangdong Province, China. Water 2025, 17, 780. [Google Scholar] [CrossRef]
Remotesensing 17 03140 g001
Figure 1. Sampling stations. (a) Approximate locations of the sampling points. (b) Specific distribution of the sampling points.
Figure 1. Sampling stations. (a) Approximate locations of the sampling points. (b) Specific distribution of the sampling points.
Remotesensing 17 03140 g001
Remotesensing 17 03140 g002
Figure 2. Changes in water quality parameters in various areas before and after artificial reef deployment. (a–d) represent the concentration data of water quality parameters at different sampling areas, in order of DO, COD, PO4-P, and Chla.
Figure 2. Changes in water quality parameters in various areas before and after artificial reef deployment. (a–d) represent the concentration data of water quality parameters at different sampling areas, in order of DO, COD, PO4-P, and Chla.
Remotesensing 17 03140 g002
Remotesensing 17 03140 g003
Figure 3. Correlation and significance analysis between water quality parameters and satellite spectral bands.
Figure 3. Correlation and significance analysis between water quality parameters and satellite spectral bands.
Remotesensing 17 03140 g003
Remotesensing 17 03140 g004
Figure 4. Model accuracy of water quality parameter assessed through four-band satellite data. (a–d) respectively represent the Taylor diagrams of chla, DO, COD, and PO4-P.
Figure 4. Model accuracy of water quality parameter assessed through four-band satellite data. (a–d) respectively represent the Taylor diagrams of chla, DO, COD, and PO4-P.
Remotesensing 17 03140 g004
Remotesensing 17 03140 g005
Figure 5. Model accuracy of water quality parameter assessed through eight-band satellite data. (a–d) respectively represent the Taylor diagrams of chla, DO, COD, and PO4-P.
Figure 5. Model accuracy of water quality parameter assessed through eight-band satellite data. (a–d) respectively represent the Taylor diagrams of chla, DO, COD, and PO4-P.
Remotesensing 17 03140 g005
Remotesensing 17 03140 g006
Figure 6. SHAP decision plots based on 4-band data. (a–d) correspond to Chla, DO, COD, and PO4-P, respectively.
Figure 6. SHAP decision plots based on 4-band data. (a–d) correspond to Chla, DO, COD, and PO4-P, respectively.
Remotesensing 17 03140 g006
Remotesensing 17 03140 g007
Figure 7. SHAP decision plots based on 8-band data. (a–d) correspond to Chla, DO, COD, and PO4-P, respectively.
Figure 7. SHAP decision plots based on 8-band data. (a–d) correspond to Chla, DO, COD, and PO4-P, respectively.
Remotesensing 17 03140 g007
Remotesensing 17 03140 g008
Figure 8. Water quality parameters before artificial reef deployment (mg/L).
Figure 8. Water quality parameters before artificial reef deployment (mg/L).
Remotesensing 17 03140 g008
Remotesensing 17 03140 g009
Figure 9. Water quality parameters after artificial reef deployment (mg/L).
Figure 9. Water quality parameters after artificial reef deployment (mg/L).
Remotesensing 17 03140 g009
Remotesensing 17 03140 g010
Figure 10. Spatial changes in Chla concentration before and after artificial reef deployment. (a) and (b) respectively represent the variation of Chla concentration along the offshore direction in different regions.
Figure 10. Spatial changes in Chla concentration before and after artificial reef deployment. (a) and (b) respectively represent the variation of Chla concentration along the offshore direction in different regions.
Remotesensing 17 03140 g010
Remotesensing 17 03140 g011
Figure 11. Spatial changes in DO, COD, and PO4-P concentration before and after artificial reef deployment. (a–f) respectively represent the changes in concentrations of DO, COD, and PO4-P along the offshore direction in different regions.
Figure 11. Spatial changes in DO, COD, and PO4-P concentration before and after artificial reef deployment. (a–f) respectively represent the changes in concentrations of DO, COD, and PO4-P along the offshore direction in different regions.
Remotesensing 17 03140 g011
Remotesensing 17 03140 g012
Figure 12. Model accuracy of fisheries’ resources assessed based on water quality parameters. (a) and (b) respectively represent the Taylor diagrams of fishery resource assessment in April 2020 and April 2024.
Figure 12. Model accuracy of fisheries’ resources assessed based on water quality parameters. (a) and (b) respectively represent the Taylor diagrams of fishery resource assessment in April 2020 and April 2024.
Remotesensing 17 03140 g012
Remotesensing 17 03140 g013
Figure 13. Model application for assessing fishery resources based on water quality parameters (kg/km2). (a) and (b) respectively represent the Taylor diagrams of fishery resource assessment in April 2020 and April 2024.
Figure 13. Model application for assessing fishery resources based on water quality parameters (kg/km2). (a) and (b) respectively represent the Taylor diagrams of fishery resource assessment in April 2020 and April 2024.
Remotesensing 17 03140 g013
Table 1. Statistics on catches of nekton in each trawl survey.
Table 1. Statistics on catches of nekton in each trawl survey.
SpeciesResource Density (kg/km2)Resource Density of Bycatch (ind/km2)
Constructed
Area		Control AreaConstructed
Area		Control AreaConstructed
Area		Control Area
2020041219181.310212.26512,994.612,816.6
2024042954499.252169.51111,408.15596.4
2020091816275.185255.41224,169.516,391.5
2024113426935.522456.38538,615.613,742.8
Table 2. VIF results of water quality parameters.
Table 2. VIF results of water quality parameters.
VariableVIF
Chla15.93
DO19.07
COD6.18
PO4-P7.88
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Chen, J.; Feng, X.; Guo, C.; Chen, Y.; Tong, F.; Zhang, L.; Liu, Z.; Zhang, J.; Yuan, H.; Chen, P. Assessing the Impacts of Marine Ranching Construction on Water Quality and Fishery Resources in Adjacent Coastal Waters. Remote Sens. 2025, 17, 3140. https://doi.org/10.3390/rs17183140

AMA Style

Chen J, Feng X, Guo C, Chen Y, Tong F, Zhang L, Liu Z, Zhang J, Yuan H, Chen P. Assessing the Impacts of Marine Ranching Construction on Water Quality and Fishery Resources in Adjacent Coastal Waters. Remote Sensing. 2025; 17(18):3140. https://doi.org/10.3390/rs17183140

Chicago/Turabian Style

Chen, Jianqu, Xue Feng, Chunya Guo, Yuxiang Chen, Fei Tong, Lei Zhang, Zhangbin Liu, Jian Zhang, Huanrong Yuan, and Pimao Chen. 2025. "Assessing the Impacts of Marine Ranching Construction on Water Quality and Fishery Resources in Adjacent Coastal Waters" Remote Sensing 17, no. 18: 3140. https://doi.org/10.3390/rs17183140

APA Style

Chen, J., Feng, X., Guo, C., Chen, Y., Tong, F., Zhang, L., Liu, Z., Zhang, J., Yuan, H., & Chen, P. (2025). Assessing the Impacts of Marine Ranching Construction on Water Quality and Fishery Resources in Adjacent Coastal Waters. Remote Sensing, 17(18), 3140. https://doi.org/10.3390/rs17183140

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop