Spatiotemporal Trends and Zoning Geospatial Assessment in China’s Offshore Mariculture (2018–2022)

Abstract

Offshore mariculture is a critical component of China’s aquaculture sector, but its rapid expansion presents significant challenges to sustainable marine resource management. This study utilizes high-resolution remote sensing data (2017–2023) and advanced ConvNeXt V2 algorithms to quantitatively analyze the spatiotemporal dynamics of offshore mariculture and explore its spatial distribution in relation to marine functional zoning policies. Through a detailed classification of six mariculture types, this study reveals significant spatial shifts, with China’s offshore mariculture transitioning from a model characterized by a “coastal, concentrated layout” to a new paradigm of “deep-sea and far-sea expansion, multi-point distribution”. Notably, the area of deep-sea and far-sea mariculture increased by 41.8% in regions with water depths of 50 m or more from 2018 to 2022. However, in 2022, the actual mariculture area accounted for only 0.608% of the designated functional zones, while 61.79% of mariculture activities occurred outside these planned zones, indicating a considerable spatial mismatch between mariculture practices and zoning plans. This study underscores the urgent need to optimize spatial planning and regulatory frameworks to balance economic growth with environmental sustainability, offering novel insights and actionable recommendations for the coordinated development of China’s marine economy.

1. Introduction

Offshore mariculture plays a pivotal role in meeting the growing global demand for seafood, contributing significantly to food security, economic development, and environmental sustainability. According to the Food and Agriculture Organization (FAO), aquaculture biomass surpassed capture fisheries for the first time in 2022, with mariculture accounting for 54.79% of total global aquaculture output. China leads the world in mariculture production, contributing 66.46% to global output [1,2,3]. However, this rapid expansion has intensified pressures on coastal ecosystems, exacerbating issues such as water pollution, habitat degradation, and resource inefficiency, particularly in high-density mariculture zones like those in Shandong Province [4]. To address these challenges, China’s “National Marine Functional Zoning (2011–2020)” policy was implemented to regulate marine spatial usage by delineating zones for mariculture suitability, limitation, and prohibition. While the policy was designed to optimize marine resource utilization and mitigate environmental pressures, spatial analysis reveals that mariculture activities have become increasingly concentrated in certain designated zones, which may pose new ecological risks that warrant further investigation [5].

In this context, precise monitoring and assessment of offshore mariculture zones have become an urgent priority. On one hand, comprehensive and systematic monitoring can reveal the spatial correspondence between mariculture practices and marine functional zoning plans, providing scientific support for the optimization of mariculture spatial layout. On the other hand, as the scale of mariculture continues to expand, traditional monitoring methods are increasingly revealing their limitations. For instance, while field surveys can provide high-precision localized data, they are costly to implement, time-consuming, and it is challenging to cover large areas [6]. Additionally, methods based on statistical data, such as agricultural censuses, are often constrained by the timeliness and spatial resolution of the data, making it difficult to capture dynamic changes [7]. In some regions, researchers have attempted to evaluate mariculture density using a combination of water quality sampling and model predictions. However, the results are highly dependent on the quantity and quality of field data, and the models tend to be complex [8].

To overcome the limitations of traditional methods, there is a need for a monitoring technology that combines efficiency, precision, and broad applicability. With the rapid advancement of remote sensing technology, monitoring methods based on remote sensing images have gradually become essential tools in both academia and industry. Remote sensing technology not only enables low-cost coverage of extensive areas but also captures dynamic changes by integrating multi-temporal data, thereby facilitating comprehensive and long-term monitoring. Numerous scholars have conducted in-depth studies on offshore mariculture areas using various methods. For instance, in certain bays or provincial regions, Lu et al. [9] utilized Sentinel-2 satellite imagery in conjunction with an improved U-Net model to successfully extract raft and cage culture areas from the northeastern coastal region of Fujian Province, achieving an F1 score of 83.75%, which demonstrates the model’s excellent classification performance. On the other hand, Wang et al. [10] employed high-resolution GF-2 satellite images to accurately extract cage culture areas in the Zhoushan Archipelago, achieving a detection accuracy of 93%, providing critical data support for assessing environmental pollution loads. Marín Del Valle et al. utilized high-resolution satellite imagery and object-based classification methods to analyze the spatiotemporal dynamics of mariculture in Southeastern China, revealing significant expansion and diversification trends in the sector from 2003 to 2016 [11]. However, these studies are limited to localized regions, making it challenging to comprehensively reflect the overall situation of mariculture in China.

In a study conducted across the entire coastal region of China, ZDSYS. [12] successfully depicted the spatial distribution of China’s mariculture in 2020 using multi-temporal Sentinel-2 and Sentinel-1 images at a resolution of 10 m. The overall classification accuracy achieved was 94%, with a Kappa coefficient of 0.91, demonstrating a high level of classification precision. Additionally, Liu et al. [13] utilized Landsat and multi-temporal Sentinel-1 images to extract the distribution of mariculture along China’s coast for the years 2000, 2010, and 2020, obtaining F1 scores of 96.2%, 94.3%, and 92.7%, respectively, which revealed a rapid expansion trend in mariculture in northern regions. Fu et al. [14] employed GF-1 imagery with a resolution of 16 m and developed a distribution map of coastal mariculture in China based on the HCHNet deep convolutional neural network model, achieving an overall classification accuracy of 95.83%. However, their research did not delve into a detailed classification of mariculture types.

In the mariculture industry, a detailed classification of mariculture areas is crucial for a comprehensive understanding of their spatial distribution and impact on ecosystems [15]. Different types of mariculture methods, such as cage culture, raft culture, and longline culture, exhibit significant variations in their environmental effects. For instance, cage culture may lead to localized water quality degradation, whereas raft culture tends to have a relatively minor impact on water flow and sediment dynamics [16]. A study has thoroughly examined the effects of mariculture on marine microbial communities, including bacteria, fungi, viruses, and antibiotic-resistance genes present in seawater and sediments [17]. The absence of meticulous classification can result in biased assessments of the overall environmental impact of mariculture, and limit the applicability of research findings in the formulation of management policies [18]. If classifications are overly broad, it becomes challenging to effectively identify and manage mariculture activities in specific areas, potentially increasing the risks of environmental pollution and ecological degradation. Conversely, detailed classifications enable more targeted management measures, allowing for the development of specific and effective strategies tailored to the characteristics of different mariculture zones, such as implementing stricter discharge standards and regulatory measures in high-risk pollution areas [19]. Furthermore, such classifications can facilitate the scientific planning of mariculture zones, optimize mariculture layouts, and ensure the rational use of aquatic resources, thereby achieving a balance between economic development and environmental protection. This is of paramount importance for realizing the goals of structural adjustment in aquaculture and enhancing income, while reducing output [20]. Therefore, to achieve sustainable development in mariculture, it is imperative that we conduct detailed classifications of mariculture areas to ensure that our assessments and management measures are both scientifically sound and effective [21].

To address the gaps in existing research regarding insufficient classification detail, limitations in research scope, and a lack of dynamic monitoring, this study conducts a detailed spatiotemporal analysis of China’s offshore mariculture from 2017 to 2023, using high-resolution Google Earth imagery and the state-of-the-art ConvNeXt V2 [22] algorithm. This research uniquely classifies mariculture zones into six types: traditional cage culture, deep-water cage culture, floating raft culture, longline culture, raft culture, and bouchot culture to provide a granular understanding of the spatial distribution and environmental impacts. By integrating mariculture data with Marine Functional Zoning (MFZ) plans, we analyze the spatial alignment between planned zones and actual mariculture activities, identifying critical mismatches that may inform future spatial planning efforts. Furthermore, we analyze the expansion of deep-sea and far-sea mariculture, uncovering trends that highlight opportunities and challenges for sustainable development.

This study contributes to the field by addressing three key scientific questions: (1) What are the spatiotemporal trends in China’s offshore mariculture from 2017 to 2023? (2) To what extent do current Marine Functional Zoning (MFZ) plans align with the spatial distribution of mariculture activities? (3) What implications do deep-sea and far-sea mariculture trends have for the sustainable management of marine resources? By answering these questions, this research provides actionable insights to optimize spatial planning, enhance policy implementation, and promote the sustainable development of China’s marine economy.

2. Materials and Methods

2.1. Study Area

China’s coastline extends from the Yalu River estuary in the north to the Beilun River estuary in the south, encompassing approximately 18,000 km of continental and over 14,000 km of island coastline [23,24]. The coastal geomorphology varies significantly, with sandy beaches dominating the north and rocky shores in the south. Climate conditions range from tropical and subtropical to temperate zones, leading to notable regional differences in temperature and precipitation. This environmental diversity supports China’s leading role in global mariculture, which accounted for 66.46% of global production in 2022 [1]. More than 6500 islands further facilitate mariculture expansion [12]. However, intensive mariculture has led to ecological concerns, including water quality degradation and biodiversity loss [13,25].

This study focuses on offshore areas within 50 km of the mainland coastline, where over 90% of China’s mariculture activities are concentrated, predominantly within 20 km of the shore. As shown in Figure 1, the study region includes the coastal provinces of Liaoning, Hebei, Tianjin, Shandong, Jiangsu, Zhejiang, Fujian, Taiwan, Guangdong, Guangxi, and Hainan. This 50 km buffer represents the core mariculture area and a critical interface for technological, ecological, and economic interactions [12,13,25]. Additionally, to account for mariculture around islands further offshore [26], a 10 km zone around selected islands was incorporated into the study.

According to the OMAD-6 dataset proposed by Mo et al. [21], offshore mariculture in China is classified into six types: traditional cage culture, deep-water cage culture, floating raft culture, longline culture, raft culture, and bouchot culture. These categories differ in structure, operation, and spatial distribution. High-resolution Google Earth images and ground-truthing data (Figure 2) were used to clarify the spatial characteristics of these mariculture types.

2.2. Data Preparation

The datasets used in this study are summarized in

Table 1. Data Sources Used in This Study.

Data Name	Type	Spatial Resolution	Source
Google Earth Satellite Images	Optical Imagery	0.6 m	Google Earth
OMAD-6 Dataset	Optical Imagery and Vector	0.6 m	Mo et al. (2024) [21]
Coastline Data	Vector	1 m	2015 Global Shoreline Dataset [23,24]
Marine Functional Zoning Data	Vector	Province-level	National MFZ Plans (2011–2020) [27]
Seawater Depth Dataset	Raster	~400 m (15 arc-seconds)	GEBCO_2024 Grid [28]

, including satellite imagery, coastline data, marine functional zoning data, and in situ ground truth data.

2.2.1. Google Earth Satellite Images

This study utilizes high-resolution satellite images from Google Earth with a resolution of 0.6 m to conduct an in-depth analysis of the spatial distribution of mariculture along China’s coastal regions for the years 2018, 2020, and 2022. Given the relatively small annual variations in mariculture areas [13], the research selected the clearest, cloud-free satellite images closest to each specified year within the designated time frames. Specifically, the data for 2018 were collected from satellite images taken between 1 July 2017 and 30 June 2019; the 2020 data were sourced from images captured between 1 July 2019 and 30 June 2021; and the 2022 data were derived from images taken between 1 July 2021 and 30 June 2023. This approach ensures the accuracy and stability of the data, allowing for the meticulous identification and localization of mariculture areas, thereby establishing a solid data foundation for the research.

2.2.2. Coastline Data

This study utilizes the 2015 global meter-resolution coastline and island dataset for China’s region, as illustrated in Figure 1 [23,24]. This dataset is derived from high-resolution remote sensing images from Google Earth, collected over a period of 1–2 years before and after 2015, using a human–computer interaction method for extraction. It offers detailed coastline information for global continents and islands, achieving a spatial resolution at the meter level, with a minimum patch area of 6 square meters. This high-precision dataset not only provides a reliable foundation for the accurate delineation and analysis of offshore aquaculture boundaries but also effectively minimizes interference from terrestrial areas, enhancing the efficiency and accuracy of offshore mariculture zone detection, thereby offering more precise data support for research.

2.2.3. Marine Functional Zoning Data

Under the guidance of the “National Marine Functional Zoning (2011–2020)” [27], each coastal province has successively developed its marine functional zoning plans. The marine functional zoning data utilized in this study are derived from these documents. Specifically, this includes the “Liaoning Coastal Zone Protection and Utilization Plan”, “Jiangsu Province Marine Principal Functional Zone Plan”, “Fujian Province Nearshore Environmental Functional Zoning (Revised)”, “Hebei Province Marine Functional Zoning (2011–2020)”, “Guangxi Zhuang Autonomous Region Marine Functional Zoning (2011–2020)”, “Zhejiang Province Marine Functional Zoning (2011–2020)”, “Tianjin Marine Functional Zoning (2011–2020)”, and the “Shandong Province Marine Ecological Environment Protection Plan (2018–2020)”. It is noteworthy that while the marine functional zoning for Shandong Province is applicable from 2018 to 2020, the other provinces’ zoning plans are valid from 2011 to 2020.

Due to the inability to obtain marine functional zoning data from some coastal provinces nationwide, this study lacks relevant zoning information for three provinces: Guangdong, Hainan, and Taiwan. Consequently, the quantitative analysis of marine functional zoning presented in this paper only encompasses the coastal provinces for which functional zoning data have been acquired, excluding the three missing provinces from the scope of indicator calculations.

2.2.4. Seawater Depth Dataset

The GEBCO_2024 Grid dataset [28], jointly managed by the International Hydrographic Organization and the Intergovernmental Oceanographic Commission of the United Nations Educational, Scientific and Cultural Organization, provides high-precision bathymetric data for the world’s oceans. This study utilizes the GEBCO_2024 Grid dataset to obtain seawater depth data for the coastal regions of China. Given the dataset’s high accuracy and global coverage, it is widely employed in the field of marine science. For this research, the dataset is indispensable for analyzing the relationship between offshore mariculture areas and seawater depth.

2.3. Methods

2.3.1. ConvNeXt-V2 Mask R-CNN Algorithm

The instance segmentation model Mask R-CNN [29] is a widely utilized two-stage segmentation framework, whose overall architecture is presented in Figure 3. Initially, Mask R-CNN employs a Region Proposal Network (RPN) to generate candidate regions, subsequently conducting parallel tasks including object classification, bounding box regression, and mask prediction to achieve precise instance-level segmentation. Traditionally, the Mask R-CNN utilizes ResNet-50 as its backbone for feature extraction, leveraging residual connections to alleviate gradient vanishing issues and demonstrating strong capability in feature extraction and representation.

However, in offshore mariculture scenarios characterized by complex backgrounds, dense distribution, and intricate boundary details, the standard ResNet-50 shows limitations in accurately extracting fine-grained structural features. According to Mo et al., the QueryInst [30] algorithm applied to the OMAD-6 dataset demonstrated commendable segmentation performance, achieving a mean Average Precision (mAP) of 80.9% [21]. In comparison, our experimental analysis revealed that replacing ResNet-50 with ConvNeXt-V2 [22] resulted in an even higher mAP performance for the segmentation task on the same dataset, reaching 82.5%. Notably, under an Intersection over Union (IoU) threshold of 50%, the mAP value further increased to an impressive 96.2%.

Table 2. Structural Characteristics of ResNet-50 and ConvNeXt V2.

Structure	ResNet-50	ConvNeXt V2
Input Layer	RGB image input (224 × 224)	RGB image input (224 × 224)
Convolutional Layer	7 × 7 convolution, stride = 2	7 × 7 depthwise separable convolution, stride = 4
Normalization Layer	Batch Normalization (BN)	Layer Normalization (LN)
Activation Function	ReLU	GELU
Bottleneck Structure	1 × 1 (dimensionality reduction) → 3 × 3 (feature extraction) → 1 × 1 (dimensionality restoration)	7 × 7 → 1 × 1 (expansion) → GELU + GRN → 1 × 1 (dimensionality reduction)
Skip Connection	ResNet Block	ConvNeXt V2 block
Channel Dimension Adjustment	1 × 1 convolution for channel adjustment	1 × 1 convolution for channel adjustment
Downsampling Strategy	Stride = 2 convolution inside residual blocks	Independent 2 × 2 stride convolution for downsampling
Pooling Operation	3 × 3 MaxPool	No max pooling
Residual Connection	Skip connection in every block	Residual connection, but differs from ResNet
Kernel Size	3 × 3	7 × 7
Convolution Type	Standard convolution	Depthwise separable convolution

summarizes the structural differences between ConvNeXt-V2 and ResNet-50. Specifically, ConvNeXt-V2 introduces depth-wise separable convolutions with a larger receptive field (7 × 7), coupled with Layer Normalization (LN) and Global Response Normalization (GRN), to enhance network adaptability in densely structured and complex background environments. Additionally, ConvNeXt-V2 replaces the conventional ReLU activation function with GELU and utilizes independent 2 × 2 stride convolutions for downsampling, discarding the residual bottleneck approach characteristic of ResNet-50. These structural improvements significantly bolster the model’s ability to accurately delineate detailed boundaries, particularly vital in extracting offshore mariculture facilities characterized by complex backgrounds and intricate edge features.

In light of these results, we selected the ConvNeXt-V2 Mask R-CNN algorithm for the precise extraction and analysis of offshore mariculture areas across six categories from the years 2018, 2020, and 2022.

2.3.2. Extraction and Processing of Mariculture Areas

As shown in Figure 4, the technical roadmap of extraction and processing of mariculture areas is divided into 3 steps, as detailed below.

(a)

Data preprocessing

To extract offshore mariculture areas, we preprocessed remote sensing images and coastline data to enhance extraction efficiency. Google satellite imagery (2018, 2020, 2022) and coastline data of China’s coastal regions were collected. Coastline data were rasterized to generate mask images, where land areas were assigned 0 and marine areas 1, ensuring segmentation was restricted to marine regions. Marine functional zoning maps from seven Chinese provinces were collected, georeferenced, and manually digitized to generate aquaculture functional zone vector maps. During inference, each annual image was overlaid with its corresponding mask to exclude land areas, reducing computational costs and improving extraction efficiency.

(b)

Mariculture area extraction

We employed the ConvNeXt-V2 Mask R-CNN instance segmentation model. Given the high resolution of remote sensing images, direct segmentation was inefficient. Thus, images were divided into 512 × 512-pixel patches with 50% overlap to maintain segmentation continuity. The model produced segmentation masks, object categories, and confidence scores. Non-Maximum Suppression (NMS) was applied to merge overlapping detections based on Intersection over Union (IoU), retaining the highest confidence targets. The final segmentation masks were converted to geometric polygons, projected onto the Universal Transverse Mercator (UTM) coordinate system for area computation, filtered using a 1 m² threshold to remove noise, and topologically corrected. The optimized data were then transformed back into the WGS 1984 coordinate system and stored in vector format for further analysis.

(c)

Data post-processing

Despite the model’s high accuracy, misclassifications persisted due to variations in resolution and coastal terrain complexity. Manual verification was conducted to remove misclassified regions. The refined vector files from 2018, 2020, and 2022 were merged to generate spatial distribution maps of offshore aquaculture facilities across six categories, serving as a data foundation for spatiotemporal change analysis and evaluating China’s marine functional zoning policies.

2.3.3. Accuracy Evaluation

To evaluate accuracy, we combined random sampling with visual interpretation [13]. A 10 km buffer was applied to offshore mariculture areas for each year to expand the assessment scope. Within these buffers, 10,000 circular samples (radius = 5 m) were randomly generated per year, serving as validation points. Each sample was spatially overlaid with the corresponding aquaculture map to determine its category. Samples falling within an aquaculture area were assigned to the respective class label; otherwise, they were marked as background. Finally, all 30,000 samples underwent manual visual interpretation to verify classification correctness. The classification matrix is presented in

Table 3. Confusion matrix between six types of mariculture and background categories.

True Prediction	TCC (0)	DWCC (1)	FRC (2)	LC (3)	RC (4)	BC (5)	Background (6)
TCC (0)	${\mathrm{T}}_0$	${\mathrm{F}}_0\to 1$	${\mathrm{F}}_0\to 2$	${\mathrm{F}}_0\to 3$	${\mathrm{F}}_0\to 4$	${\mathrm{F}}_0\to 5$	${\mathrm{F}}_0\to 6$
DWCC (1)	${\mathrm{F}}_1\to 0$	${\mathrm{T}}_1$	${\mathrm{F}}_1\to 2$	${\mathrm{F}}_1\to 3$	${\mathrm{F}}_1\to 4$	${\mathrm{F}}_1\to 5$	${\mathrm{F}}_1\to 6$
FRC (2)	${\mathrm{F}}_2\to 0$	${\mathrm{F}}_2\to 1$	${\mathrm{T}}_2$	${\mathrm{F}}_2\to 3$	${\mathrm{F}}_2\to 4$	${\mathrm{F}}_2\to 5$	${\mathrm{F}}_2\to 6$
LC (3)	${\mathrm{F}}_3\to 0$	${\mathrm{F}}_3\to 1$	${\mathrm{F}}_3\to 2$	${\mathrm{T}}_3$	${\mathrm{F}}_3\to 4$	${\mathrm{F}}_3\to 5$	${\mathrm{F}}_3\to 6$
RC (4)	${\mathrm{F}}_4\to 0$	${\mathrm{F}}_4\to 1$	${\mathrm{F}}_4\to 2$	${\mathrm{F}}_4\to 3$	${\mathrm{T}}_4$	${\mathrm{F}}_4\to 5$	${\mathrm{F}}_4\to 6$
BC (5)	${\mathrm{F}}_5\to 0$	${\mathrm{F}}_5\to 1$	${\mathrm{F}}_5\to 2$	${\mathrm{F}}_5\to 3$	${\mathrm{F}}_5\to 4$	${\mathrm{T}}_5$	${\mathrm{F}}_5\to 6$
Background (6)	${\mathrm{F}}_6\to 0$	${\mathrm{F}}_6\to 1$	${\mathrm{F}}_6\to 2$	${\mathrm{F}}_6\to 3$	${\mathrm{F}}_6\to 4$	${\mathrm{F}}_6\to 5$	${\mathrm{T}}_6$

.

Let ${\mathrm{T}}_{\mathrm{i}}$ represent cases where both the true label and the predicted category are i, and ${\mathrm{F}}_{\mathrm{j}}\to \mathrm{i}$ denote instances where a true label i is misclassified as category j.

User’s Accuracy (UA) [31] measures the accuracy of the predicted category, defined as the ratio of correctly classified samples to the total number of samples predicted for that category. It reflects the impact of false positives (FP). The calculation formula is as follows:

$${\mathrm{U}\mathrm{A}}_{{\mathrm{C}}_{\mathrm{i}}}={\displaystyle \frac{{\mathrm{T}\mathrm{P}}_{\mathrm{i}}}{{\mathrm{T}\mathrm{P}}_{\mathrm{i}}+{\mathrm{F}\mathrm{P}}_{\mathrm{i}}}},$$

(1)

where ${\mathrm{T}\mathrm{P}}_{\mathrm{i}}$ represents the number of correctly classified samples for category ${\mathrm{C}}_{\mathrm{i}}$, and ${\mathrm{F}\mathrm{P}}_{\mathrm{i}}$ denotes the number of misclassified samples for category ${\mathrm{C}}_{\mathrm{i}}$.

Producer’s Accuracy (PA) [32] measures the proportion of correctly classified samples within the actual category. It is defined as the ratio of correctly classified samples to the total number of actual samples in that category, thereby reflecting the impact of false negatives (FN). The calculation formula is as follows:

$${\mathrm{P}\mathrm{A}}_{{\mathrm{C}}_{\mathrm{i}}}={\displaystyle \frac{{\mathrm{T}\mathrm{P}}_{\mathrm{i}}}{{\mathrm{T}\mathrm{P}}_{\mathrm{i}}+{\mathrm{F}\mathrm{N}}_{\mathrm{i}}}},$$

(2)

where ${\mathrm{F}\mathrm{N}}_{\mathrm{i}}$ represents the number of samples with the true category ${\mathrm{C}}_{\mathrm{i}}$ that were misclassified.

Overall Accuracy (OA) [33] evaluates the overall performance of the classification model by calculating the proportion of correctly classified samples to the total number of samples. The calculation formula is as follows:

$$\mathrm{O}\mathrm{A}={\displaystyle \frac{\sum {\mathrm{T}\mathrm{P}}_{\mathrm{i}}}{\sum \left({\mathrm{T}\mathrm{P}}_{\mathrm{i}}+{\mathrm{F}\mathrm{N}}_{\mathrm{i}}+{\mathrm{F}\mathrm{P}}_{\mathrm{i}}\right)}}.$$

(3)

The Weighted F1-score [34] integrates User’s Accuracy (UA) and Producer’s Accuracy (PA) to balance model precision and recall while accounting for class imbalance in the final evaluation. The calculation formula is as follows:

$${\mathrm{F}1}_{{\mathrm{C}}_{\mathrm{i}}}=2\times {\displaystyle \frac{{\mathrm{U}\mathrm{A}}_{{\mathrm{C}}_{\mathrm{i}}}\times {\mathrm{P}\mathrm{A}}_{{\mathrm{C}}_{\mathrm{i}}}}{{\mathrm{U}\mathrm{A}}_{{\mathrm{C}}_{\mathrm{i}}}+{\mathrm{P}\mathrm{A}}_{{\mathrm{C}}_{\mathrm{i}}}}},$$

(4)

$${\mathrm{F}1}_{\mathrm{w}\mathrm{e}\mathrm{i}\mathrm{g}\mathrm{h}\mathrm{t}\mathrm{e}\mathrm{d}}={\displaystyle \frac{{\sum }_{\mathrm{i}=1}^{\mathrm{N}}{\mathrm{W}}_{\mathrm{i}}\times {\mathrm{F}1}_{{\mathrm{C}}_{\mathrm{i}}}}{{\sum }_{\mathrm{i}=1}^{\mathrm{N}}{\mathrm{W}}_{\mathrm{i}}}},{\mathrm{W}}_{\mathrm{i}}={\displaystyle \frac{{\mathrm{S}}_{{\mathrm{C}}_{\mathrm{i}}}}{\mathrm{T}\mathrm{S}}},{\mathrm{S}}_{{\mathrm{C}}_{\mathrm{i}}}={\mathrm{T}\mathrm{P}}_{\mathrm{i}}+{\mathrm{F}\mathrm{N}}_{\mathrm{i}}+{\mathrm{F}\mathrm{P}}_{\mathrm{i}},$$

(5)

where $\mathrm{T}\mathrm{S}$ represents the total number of samples.

The Kappa coefficient [35] evaluates whether the model’s classification performance surpasses random classification by comparing the observed accuracy with the expected accuracy under random assignment. The calculation formula is as follows:

$$\mathrm{K}\mathrm{a}\mathrm{p}\mathrm{p}\mathrm{a}={\displaystyle \frac{\mathrm{O}\mathrm{A}-{\mathrm{P}}_{\mathrm{e}}}{1-{\mathrm{P}}_{\mathrm{e}}}},$$

(6)

$$\mathrm{P}=\sum \left({\displaystyle \frac{\left({\mathrm{T}\mathrm{P}}_{\mathrm{i}}+{\mathrm{F}\mathrm{N}}_{\mathrm{i}}\right)\left({\mathrm{T}\mathrm{P}}_{\mathrm{i}}+{\mathrm{F}\mathrm{P}}_{\mathrm{i}}\right)}{{\left(\sum \left({\mathrm{T}\mathrm{P}}_{\mathrm{i}}+{\mathrm{F}\mathrm{N}}_{\mathrm{i}}+{\mathrm{F}\mathrm{P}}_{\mathrm{i}}\right)\right)}^2}}\right).$$

(7)

2.3.4. Kernel Density

This study employs the kernel density [36] tool in ArcGIS 10.8.1 to construct a continuous density surface map of offshore aquaculture distribution. This tool utilizes a non-parametric kernel density estimation method, which smooths the influence of each point within a predefined search radius, resulting in a smooth density surface. To ensure the accuracy and consistency of area units and distance measurements within the projected coordinate system, we utilized the “PLANAR” option for calculations. Additionally, the output parameter was set to “DENSITIES”, allowing the resulting raster to reflect the density per unit area. For a point located at $({\mathrm{x}}_{\mathrm{i}},{\mathrm{y}}_{\mathrm{i}})$, the density contribution at the center of the raster located at $(\mathrm{x},\mathrm{y})$ within the search radius r can be expressed as the following:

$$\mathrm{f}\left(\mathrm{d}\right)={\displaystyle \frac{3}{{\mathsf{\pi }\mathrm{r}}^2}}{\left(1-{\displaystyle \frac{{\mathrm{d}}^2}{{\mathrm{r}}^2}}\right)}^2 \mathrm{w}\mathrm{h}\mathrm{e}\mathrm{n} \mathrm{d}\le \mathrm{r}, \mathrm{a}\mathrm{n}\mathrm{d} \mathrm{f}\left(\mathrm{d}\right) \mathrm{w}\mathrm{h}\mathrm{e}\mathrm{n} \mathrm{d}>\mathrm{r}.$$

(8)

In the formula, $\mathrm{d}$ represents the Euclidean distance in the plane between the center of the grid and the specified point, while the search radius r governs the level of smoothness. We have selected a pixel size of 1000 m, with a search radius r of 15,000 m.

2.3.5. Achievement Rate and Planning Area Exceedance Ratio

This study aims to evaluate the effectiveness of offshore mariculture activities under marine functional zoning in China’s coastal regions. It uses two indicators, which are the achievement rate and planning area exceedance ratio. The study analyzes the implementation of marine functional zoning by quantifying the mariculture practices within the mariculture functional zones and the extent of mariculture expansion beyond designated boundaries.

Achievement rate (AR) indicates the proportion of actual mariculture activities within mariculture functional zones where cultivation is explicitly permitted. This metric reflects the effectiveness of mariculture functional zoning in guiding and utilizing mariculture practices. The formula is as follows:

$${\mathrm{A}\mathrm{R}}_{\mathrm{y}}={\displaystyle \frac{{\mathrm{A}}_{\mathrm{i}\mathrm{n},\mathrm{y}}}{{\mathrm{A}}_{\mathrm{p}\mathrm{l}\mathrm{a}\mathrm{n}}}}\times 100\%,$$

(9)

In the formula, ${\mathrm{A}\mathrm{R}}_{\mathrm{y}}$ represents the achievement rate for year y. ${\mathrm{A}}_{\mathrm{i}\mathrm{n},\mathrm{y}}$ denotes the mariculture area within the planned mariculture zone for year y. ${\mathrm{A}}_{\mathrm{p}\mathrm{l}\mathrm{a}\mathrm{n}}$ indicates the total area of the mariculture zone as defined in the marine functional zoning.

The planning area exceedance ratio (PAER) indicates the percentage of the mariculture area that is situated outside designated mariculture zones in all actual mariculture activities. This metric reflects the spatial expansion of mariculture activities and potential instances of boundary transgressions or violations. The formula is as follows:

$${\mathrm{P}\mathrm{A}\mathrm{E}\mathrm{R}}_{\mathrm{y}}={\displaystyle \frac{{\mathrm{A}}_{\mathrm{o}\mathrm{u}\mathrm{t},\mathrm{y}}}{{\mathrm{A}}_{\mathrm{t}\mathrm{o}\mathrm{t}\mathrm{a}\mathrm{l},\mathrm{y}}}}\times 100\%.$$

(10)

In the formula, ${\mathrm{P}\mathrm{A}\mathrm{E}\mathrm{R}}_{\mathrm{y}}$ represents the planning area exceedance ratio in year y; ${\mathrm{A}}_{\mathrm{o}\mathrm{u}\mathrm{t},\mathrm{y}}$ denotes the mariculture area that falls outside the designated mariculture functional zone in year y; and ${\mathrm{A}}_{\mathrm{t}\mathrm{o}\mathrm{t}\mathrm{a}\mathrm{l},\mathrm{y}}$ indicates the total mariculture area (${\mathrm{A}}_{\mathrm{i}\mathrm{n},\mathrm{y}}+{\mathrm{A}}_{\mathrm{o}\mathrm{u}\mathrm{t},\mathrm{y}}$) in year y.

3. Results

3.1. Quality Assessment

After visually inspecting 10,000 sample points from the offshore mariculture distribution maps of China for 2018, 2020, and 2022, we computed several accuracy metrics, including User’s Accuracy (UA), Producer’s Accuracy (PA), Overall Accuracy (OA), Weighted F1-score, and the Kappa coefficient (

Table 4. Quality Assessment Metrics for the Offshore Mariculture Distribution Map of China (2018).

	UA/%	PA/%	OA/%	F1-Score/%	Weighted F1-Score/%
TCC	92.17	93.39	94.53	92.78	94.41
DWCC	96.13	100.00		98.03
FRC	92.51	98.86		95.58
LC	86.65	76.37		81.19
RC	96.25	71.85		82.28
BC	71.48	93.45		81.00
Background	97.41	98.98		98.19

,

Table 5. Quality Assessment Metrics for the Offshore Mariculture Distribution Map of China (2020).

	UA/%	PA/%	OA/%	F1-Score/%	Weighted F1-Score/%
TCC	92.87	93.25	94.79	93.06	94.73
DWCC	90.93	100.00		95.25
FRC	91.36	94.21		92.76
LC	89.82	75.31		81.93
RC	99.37	76.42		86.40
BC	69.20	96.95		80.76
Background	97.81	99.02		98.41

and

Table 6. Quality Assessment Metrics for the Offshore Mariculture Distribution Map of China (2022).

	UA/%	PA/%	OA/%	F1-Score/%	Weighted F1-Score/%
TCC	95.03	93.87	94.55	94.44	94.47
DWCC	91.06	99.77		95.22
FRC	93.23	94.26		93.74
LC	94.27	73.07		82.32
RC	97.67	78.91		87.29
BC	65.84	94.29		77.54
Background	96.93	99.21		98.06

).

The OA and weighted F1-score of the offshore mariculture distribution maps for the three years exceeded 94%, indicating strong classification consistency. The UA and PA for traditional cage culture, deep-water cage culture, and floating raft culture remained above 90%, with F1-scores exceeding 92%, demonstrating high classification reliability. The PA of longline culture was slightly lower than other categories, recorded at 76.37%, 75.31%, and 73.07% for 2018, 2020, and 2022, respectively. In contrast, the UA of bouchot culture in 2022 was only 65.84%, but its PA consistently remained above 93.45%, indicating sufficient data certainty for practical applications. The UA and PA of background regions exceeded 96.93% and 98.98%, respectively, further confirming the overall classification stability.

In summary, the classification accuracy and consistency of the distribution maps over the three years meet high standards, providing a reliable data foundation for spatiotemporal analysis, resource management planning, and the evaluation of marine functional zoning policies in China’s offshore mariculture sector.

3.2. Spatiotemporal Distribution of Mariculture in China

Through accuracy assessment, the weighted F1 scores obtained from the 2018, 2020, and 2022 offshore mariculture zone datasets generated in this study reached 94.41%, 94.73%, and 94.47%, respectively, indicating that this dataset demonstrated high accuracy and stability. Figure 5a–c show the spatial distribution changes in mariculture zones during the three periods, respectively.

Figure 6 quantitatively compares the area changes in six mariculture types: traditional cage culture (TCC), deep-water cage culture (DWCC), floating raft culture (FRC), longline culture (LC), raft culture (RC), and bouchot culture (BC). Between 2018 and 2022, the area proportions of LC, RC, and BC exceeded 90%, while DWCC had the smallest proportion, accounting for only 2.67% in 2018. During the period from 2018 to 2022, the areas of TCC, DWCC, LC, and BC consistently expanded, with DWCC and TCC experiencing growth rates of 51.40% and 42.84%, respectively. FRC initially expanded before contracting, yet its total area showed an increase compared to 2018, whereas RC experienced continuous contraction.

Through a comparative analysis of the provincial classification statistics from 2018, 2020, and 2022 (see

Table 7. Mariculture area (km²) in coastal provinces of China in 2018.

	TCC	DWCC	FRC	LC	RC	BC	All
Liaoning	6.28	0.15	0.35	121.58	4.52	21.20	154.08
Heibei	0.02	0	0	2.62	0	0	2.64
Tianjin	0	0	0	0	0	0	0
Shandong	6.63	0.37	0.48	268.93	11.09	25.07	312.57
Jiangsu	4.53	0.02	1.38	131.31	60.37	10.54	208.14
Shanghai	0	0	0	0	0	0	0
Zhejiang	2.44	0.20	0.22	22.13	27.67	1.93	54.59
Fujian	35.23	0.49	5.70	257.95	16.83	25.46	341.66
Taiwan	0.18	0.02	2.13	0.86	0	3.18	6.36
Guangdong	5.25	0.97	8.55	51.96	0.43	34.42	101.59
Guangxi	1.05	0.29	13.39	1.78	0.01	2.89	19.40
Hainan	1.08	0.69	0.45	0.30	0	0.08	2.59
China	62.68	3.21	32.64	859.42	120.92	124.75	1203.61

,

Table 8. Mariculture area (km²) in coastal provinces of China in 2020.

	TCC	DWCC	FRC	LC	RC	BC	All
Liaoning	14.12	0.18	0.52	152.81	2.45	15.34	185.41
Heibei	0.02	0	0	0.07	0	0	0.10
Tianjin	0	0	0	0	0	0	0
Shandong	4.29	0.58	1.77	235.81	6.28	32.15	280.89
Jiangsu	2.19	0.02	0.28	90.15	57.84	34.43	184.92
Shanghai	0	0	0	0	0	0	0
Zhejiang	1.74	0.29	0.35	18.08	26.20	1.26	47.92
Fujian	34.83	0.57	6.94	307.36	29.32	26.17	405.20
Taiwan	0.04	0.20	2.29	0.84	0.11	6.88	10.36
Guangdong	5.32	1.49	7.27	76.12	0.44	24.86	115.50
Guangxi	1.26	0.61	14.53	5.26	0.01	3.56	25.23
Hainan	0.77	0.82	0.35	0.16	0	0.04	2.15
China	64.61	4.76	34.31	886.66	122.64	144.70	1257.68

and

Table 9. Mariculture area (km²) in coastal provinces of China in 2022.

	TCC	DWCC	FRC	LC	RC	BC	All
Liaoning	29.71	0.14	0.55	218.23	2.83	14.67	266.14
Heibei	0.02	0	0.01	0.20	0	0	0.23
Tianjin	0	0	0	0	0	0	0
Shandong	5.40	0.56	1.65	231.20	6.84	29.98	275.64
Jiangsu	3.07	0.07	0.20	83.96	45.50	37.83	170.64
Shanghai	0	0	0	0	0	0	0
Zhejiang	6.56	0.23	0.59	22.00	20.62	1.78	51.78
Fujian	36.87	0.99	5.64	367.11	31.59	28.19	470.39
Taiwan	0.09	0.05	2.44	0.35	0.01	6.01	8.94
Guangdong	5.47	1.39	7.94	75.35	0.31	25.46	115.92
Guangxi	1.47	0.57	14.49	3.17	0.02	2.68	22.40
Hainan	0.86	0.86	0.18	0.11	0	0.03	2.05
China	89.53	4.86	33.69	1001.68	107.73	146.63	1384.13

), it is evident that the total area of offshore mariculture in China has increased from 1203.62 km² in 2018 to 1257.68 km² in 2020, and further expanded to 1384.12 km² by 2022, reflecting a growth rate of nearly 15% over four years. These data indicate a continuous expansion of offshore mariculture in China. At the provincial level, the offshore mariculture areas have exhibited varying developmental trends during this period. For instance, the mariculture area in the Fujian province grew from 341.66 km² in 2018 to 470.39 km² in 2022, while Liaoning’s comparable area increased from 154.08 km² to 266.14 km². Conversely, traditional nearshore mariculture provinces like Shandong and Jiangsu have experienced a relative slowdown or even a slight decline in their offshore mariculture areas during this period; for example, Shandong’s area decreased from 312.57 km² in 2018 to 275.64 km² in 2022, and Jiangsu’s area fell from 208.14 km² to 170.64 km². Additionally, both Shanghai and Tianjin reported a zero mariculture area, while Hebei’s mariculture area approached zero.

3.3. Distribution of Kernel Density in China Offshore Mariculture

This study categorizes offshore mariculture into four density levels: Level I (defined as the second-density range in the legend), Level II, Level III, and Level IV. According to the data presented in Figure 7 and

Table 10. Changes in kernel density distribution levels from the coastline of offshore mariculture in China for 2018, 2020, and 2022.

	O1	O2	O3	O4	O5	O6	O7	O8	O9	O10	O11	O12	O13	O14
2018	II	II	II	I	IV	I	II	×	×	×	×	×	×	×
2020	I ↓	III ↑	II ↓	×	IV ↓	I ↓	II ↓	I	I	I	I	I	×	×
2022	II	III	II ↓	×	IV ↓	I	II ↑	I ↑	I ↑	I	×	I ↑	I	I

↑ indicates an increase in the area of the kernel density center, while ↓ indicates a decrease.

, the number of offshore mariculture centers along China’s coast increased from 7 in 2018 to 12 in 2022. Concurrently, the area of high-density mariculture zones has shown a declining trend. For instance, the Level IV density mariculture center, O5, has consistently exhibited a reduction in high-density areas throughout the study period, even indicating a trend toward downgrading. In contrast, newly established low-density mariculture centers, such as O8 and O9, are gradually expanding, suggesting significant potential for development in areas with lower mariculture density.

3.4. Tends of Deep-Sea and Far-Sea Mariculture

Deep-sea and far-sea mariculture play a crucial role in ensuring national food security, promoting high-quality economic development, and protecting the marine ecological environment. Their distinctive features lie in the utilization of modern facilities and advanced technologies to cultivate marine organisms in deep-sea areas, typically at depths exceeding 20 m, away from land-based pollution [37].

As illustrated in Figure 8, the deep-sea and far-sea mariculture area increased from 127.93 km² in 2018 to 163.99 km² in 2022, reflecting an average annual growth rate of 9%. Notably, the growth rate in 2022 (28.2%) significantly surpassed the overall growth rate of mariculture areas. This rapid expansion indicates that deep-sea and far-sea mariculture are becoming crucial drivers of growth within China’s mariculture industry. In terms of depth distribution, the 20–30 m depth range represents the primary development tier for deep-sea and far-sea mariculture, with nearly a 50% increase in the farming area within the 21–22 m range. Although the area in deeper waters (beyond 40 m) is relatively small, its expansion rate is remarkable, with a 41.8% increase observed in the 50 m depth zone.

As illustrated in Figure 9, the offshore mariculture area, located 3000 to 10,000 m from the coastline, has consistently expanded, with a particularly notable increase in 2022, further solidifying the emergence of this new paradigm. In terms of mariculture types (Figure 10), longline culture (LC) has emerged as the dominant method for deep-sea and far-sea mariculture, with an area reaching 134.35 km² in 2022, accounting for 82% of the total. Meanwhile, traditional cage culture (TCC) and floating raft culture (FRC), although smaller in scale, have demonstrated a trend of expansion; for instance, TCC increased from 8.38 km² in 2018 to 10.15 km² in 2022. The other three mariculture types are currently in a state of dynamic equilibrium.

3.5. The Results of Achievement Rate and Planning Area Exceedance Ratio

Figure 11 presents the spatial distribution of mariculture-permitted zones under the Marine Functional Zoning (MFZ) framework, outlining designated areas for mariculture development and associated restrictions along China’s coastline. This map serves as a spatial reference for assessing the alignment between planned zones and actual mariculture practices. To objectively examine the spatial relationship between MFZ plans and mariculture activities, this study analyzes the temporal changes in the achievement rate and the planning area exceedance ratio across the nation and its coastal regions from 2018 to 2022 (Figure 12 and Figure 13). Due to the absence of mariculture areas in Tianjin and Shanghai, these two regions were excluded from the analysis. In addition, Guangdong Province, Hainan Province, and Taiwan Province were not included due to the lack of publicly available MFZ maps and their relatively small share of the national mariculture area (9.18% in 2018 and 9.17% in 2022). Therefore, data from the remaining seven provinces are considered sufficient to reflect the overall spatial trends in MFZ-related mariculture distribution along China’s coast.

Research findings indicate that from 2018 to 2022, the national achievement rate of mariculture functional zones increased from 0.572% to 0.608%, reflecting a gradual rise in mariculture activities within designated zones. During the same period, the planning area exceedance ratio also rose from 58.72% to 61.79%, suggesting that mariculture expansion outside functional zones continued at a relatively higher rate. Among the coastal provinces, Fujian exhibited the highest achievement rate at 4.411% in 2022, well above the national average, along with a planning area exceedance ratio of 48.65%. In contrast, Liaoning and Hebei recorded lower achievement rates of 0.171% and 0.002%, respectively, with planning area exceedance ratios of 98.7% and 71.88%, indicating a high concentration of mariculture activities beyond the designated zones. Jiangsu and Zhejiang maintained achievement rates between 0.4% and 0.5%, while their planning area exceedance ratios in 2022 were 85.80% and 55.69%, respectively. Guangxi was the only province to show a slight decline in planning area exceedance ratio, decreasing from 63.27% in 2018 to 62.18% in 2022.

From the perspective of mariculture types, longline culture (LC) is the predominant method nationwide, accounting for over 80% of mariculture activities within designated functional zones. It also constitutes the majority of activities beyond these zones, reflecting a pattern of concurrent expansion both inside and outside the mariculture functional zones. In Jiangsu and Zhejiang, raft culture (RC) represents the main mariculture method outside the designated zones, while in Guangxi, floating raft culture (FRC) accounts for over 68% of activities beyond these zones. Deep-water cage culture (DWCC) remains limited, contributing less than 0.3% of mariculture activities both within and outside the functional zones.

4. Discussion

4.1. Summary of Key Findings

This study examines the dynamics of China’s offshore mariculture from 2017 to 2023, revealing key spatial patterns and temporal trends. The industry has gradually shifted from a concentrated coastal layout to a more dispersed configuration, with expansion into deeper and more distant waters. The total mariculture area has steadily increased, particularly in offshore regions. Spatial analysis shows a limited overlap between designated mariculture zones and actual farming locations. In 2022, only 0.608% of mariculture activities occurred within the planned zones, while 61.79% were located outside them, indicating a considerable spatial divergence. This spatial relationship varies across regions and mariculture types. Fujian Province demonstrates a relatively higher degree of alignment between planning and actual use, whereas provinces such as Liaoning and Hebei show a predominance of mariculture activities beyond designated areas. Longline culture (LC) is the dominant method both inside and outside functional zones and is the primary contributor to overall spatial expansion. In contrast, nearshore systems such as raft culture (RC) and traditional cage culture (TCC) are more commonly associated with activities outside planned zones. These findings reflect the rapid offshore growth of the mariculture industry and provide spatial insights to support future planning and management efforts.

4.2. In-Depth Interpretation of Findings

The above findings provide insights into the evolving spatial patterns of mariculture and their relevance to sustainable development. The observed shift toward offshore expansion and multi-point distribution reflects a growing trend of mariculture activities extending into deeper and more distant waters. This development may be associated with multiple factors, including policy orientation, advances in open-ocean farming technologies, and the decreasing availability of suitable nearshore sites. Offshore expansion has the potential to alleviate environmental pressure on nearshore ecosystems by reducing intensive use in enclosed bays, where issues such as water pollution and habitat degradation are more likely to occur. At the same time, the dispersed distribution of offshore farms presents new challenges for spatial monitoring and governance, particularly in terms of operational oversight and resource coordination. The long-term ecological consequences of large-scale offshore farming remain insufficiently understood, indicating the importance of continued observation and interdisciplinary research.

The observed imbalance between planned mariculture zones and actual farming locations reflects a spatial divergence that may be influenced by planning, environmental, and operational factors. With over 60% of mariculture activities occurring outside designated zones, current zoning plans appear to be only partially aligned with the spatial development of the industry. This discrepancy may result from limited zoning coverage, less favorable site conditions, or competing coastal uses. In some regions, designated mariculture zones remain unused, potentially due to overlapping marine functions or environmental constraints. For example, Shanghai and Tianjin allocated mariculture zones in their 2011–2020 plans, yet remote sensing analysis detected little to no farming activity within these areas. In Hebei, large planned zones also showed minimal use during the observation period. Additionally, technical limitations, such as the inability of satellite imagery to detect bottom-seeded species or seasonal fallow practices, may contribute to the observed spatial mismatch.

The prevalence of longline culture (LC) both within and beyond designated zones reflects its wide applicability across different marine environments. This dual pattern of spatial distribution contributes to the overall expansion of mariculture but presents complexities for spatial management, given the dispersed nature of farming sites. In contrast, deep-water cage culture (DWCC) remains limited in scale, which may be associated with higher technological requirements and operational costs. The continued development of offshore mariculture, particularly in deeper waters, has the potential to reduce pressure on nearshore ecosystems. As China’s mariculture industry undergoes rapid spatial and technological transformation, further research on governance frameworks and infrastructure capacity is essential to support balanced and sustainable growth.

4.3. Comparison with Existing Research

Our findings align with and expand on previous studies of mariculture distribution. Earlier remote sensing studies have documented mariculture growth in China, but often at broader scales or with a narrower focus. For example, Liu et al. [12] used Sentinel-1/2 imagery to map mariculture in 2020 with high classification accuracy, providing a snapshot of extensive coastal farming. Liu et al., (2022) [13] further identified a northward expansion from 2000 to 2020, which aligns with our finding that growth continued in northern regions through 2022. Similarly, Fu et al. [14] used 16 m GF-1 imagery to map nationwide mariculture, confirming the overall expansion trend. However, these studies had limitations. Many focused on single years or specific regions, lacking a long-term view of shifting mariculture patterns. Our study fills this gap by analyzing changes from 2018 to 2022, capturing the transition toward offshore and dispersed layouts that short-term studies might miss.

In contrast to previous national-scale mappings that primarily focused on the extent of mariculture expansion, this study incorporates a classification of six mariculture types. This allows for a more detailed analysis of the specific practices contributing to spatial changes, which has not been explicitly addressed in earlier works such as Fu et al. [14]. Although some regional studies have explored the spatial relationship between mariculture activities and marine functional zoning (MFZ), few have investigated this issue at the national level. By integrating mariculture distribution data with MFZ plans, this study presents the first comprehensive assessment of the spatial alignment between designated zones and actual farming locations along China’s coastline.

Overall, this research builds upon previous findings by confirming known expansion patterns and providing a broader policy-relevant spatial perspective on the relationship between mariculture development and existing planning frameworks.

4.4. Contributions and Impact

This study provides valuable contributions to the understanding of marine spatial planning and aquaculture management in China. First, by generating a high-resolution, multi-year dataset of offshore mariculture categorized by type, it offers a foundational reference for researchers and planners. This spatial dataset can inform future revisions of marine functional zoning and improve the alignment between planning and observed industry patterns. For example, the continued emergence of new mariculture sites outside designated zones reflects a spatial divergence that warrants further examination of zoning boundaries and site selection criteria. The analysis presented here offers spatial feedback to existing planning frameworks by identifying inconsistencies between designated zones and actual farming locations.

Second, the identified spatial expansion patterns provide useful insights for aquaculture development strategies. The trend toward offshore and far-sea mariculture suggests that future planning may benefit from broader spatial considerations beyond traditional coastal waters. Offshore development has the potential to alleviate pressure on nearshore ecosystems and contribute to more balanced marine resource use. However, the spatial dispersion of offshore farms also presents new challenges for monitoring and governance, which require continued methodological development and policy attention.

Third, this study explores how different mariculture practices are distributed within and beyond designated zones, offering a basis for refining spatial management strategies by culture type. For instance, the widespread presence of longline culture (LC) in both nearshore and offshore areas indicates the importance of developing technical guidelines for its sustainable use. Similarly, the notable concentration of raft culture (RC) in unplanned areas in specific provinces suggests the relevance of region-specific analysis when considering management priorities.

More broadly, the research supports the coordinated development of the marine economy and environmental protection. By integrating spatial analysis with mariculture classification, the findings contribute to a more nuanced understanding of offshore industry dynamics. This approach can support adaptive planning practices and provide a scientific basis for spatial governance, helping stakeholders better respond to the evolving patterns of mariculture expansion.

4.5. Limitations and Future Research Directions

Despite the strengths of this study, several limitations should be acknowledged to inform future research. One important limitation concerns data and detection capacity. While the remote sensing approach effectively identified surface-visible infrastructure, it likely underrepresents mariculture methods with minimal surface features, such as bottom-seeding shellfish culture and other extensive practices without fixed structures. This limitation is evident in regions such as Shanghai, Tianjin, and parts of Hebei, where official records report mariculture activities, but few facilities were identified in the imagery. Future research could explore enhanced detection techniques to improve the identification of such practices, including high-frequency temporal imaging for seasonal operations, radar or thermal sensing for subtle indicators of activity, and field surveys that incorporate local observations.

A second limitation relates to inconsistencies in spatial and temporal resolution among datasets. While 0.6-meter optical imagery was used for mariculture detection, the bathymetric data and some zoning maps had coarser resolution, which may introduce spatial uncertainty when comparing farms with depth constraints or functional zone boundaries. Furthermore, the mariculture data cover the period from 2018 to 2022, while the zoning plans were developed between 2011 and 2020. As a result, recent expansions may not correspond to the designated areas in earlier plans, which can influence spatial comparison outcomes. Future studies may benefit from incorporating more recent and higher-resolution auxiliary datasets to improve the accuracy of spatial assessments.

From a methodological perspective, although the ConvNeXt-V2 Mask R-CNN model demonstrated strong performance in mariculture classification, classification errors may still occur due to cloud cover, visual similarity with non-mariculture objects, or image acquisition timing. Continuous refinement of deep learning models and periodic manual validation will be necessary to ensure robust and consistent mapping results.

This study did not directly examine the environmental or socio-economic impacts associated with shifting mariculture patterns. While spatial patterns suggest potential ecological implications of offshore expansion and dispersal, future research could quantify these effects using field-based environmental monitoring and socio-economic data to enhance the understanding of broader impacts. In addition, the spatial divergence between planned zones and actual mariculture locations observed in this study suggests a potential need for more adaptive approaches to spatial management. Exploratory case studies in provinces such as Fujian may offer opportunities to examine how spatial adjustments based on near real-time monitoring data influence mariculture development patterns.

Finally, future research may expand beyond mainland provincial waters. As more standardized and high-quality data become available, additional regions could be included in comparative analyses. Cross-national studies may also contribute to understanding different approaches to mariculture zoning and their spatial outcomes in other major producing countries. Addressing these limitations will support more comprehensive spatial assessments and contribute to the development of improved analytical frameworks for offshore aquaculture research.

5. Conclusions

As global population growth continues and terrestrial food production faces increasing constraints, mariculture has become an important approach to ensuring a stable supply of aquatic protein while alleviating ecological pressure on coastal environments. China, as the world’s largest mariculture producer, accounts for over half of global output and offers a valuable case for understanding sustainable ocean-based food production. This study generated a high-resolution dataset of offshore mariculture facilities along China’s coastline and examined their spatial and temporal dynamics from 2017 to 2023. The results reveal a shift in mariculture patterns, with farming activities gradually expanding from nearshore zones to deeper and more distant waters. Between 2018 and 2022, the area of mariculture located in waters deeper than 50 meters increased by more than 40 percent, indicating the accelerating trend toward offshore development.

Spatial analysis shows that a large portion of mariculture activities took place outside designated functional zones. By 2022, only a small share of mariculture was located within officially planned areas, while the majority occurred beyond zoning boundaries. This spatial divergence suggests limited correspondence between planning documents and actual farming locations. Contributing factors may include the size and positioning of designated zones, environmental constraints, or competing uses in coastal areas. The dataset developed in this study provides a foundation for spatial analysis and long-term monitoring. It also offers a reference point for planners seeking to align spatial policy with observed industry distribution. The continued expansion of new farms beyond designated zones highlights the importance of regularly assessing the spatial coverage and suitability of functional zoning for mariculture.

The increasing shift toward offshore mariculture suggests that future planning approaches may benefit from broader spatial scopes that incorporate deep-sea and far-sea environments. Offshore development has the potential to reduce environmental pressure in nearshore areas and contribute to more efficient use of marine resources. However, the dispersed nature of offshore mariculture may introduce challenges for monitoring and coordination, emphasizing the need for further research on adaptive management practices. This study also provides insight into how different mariculture types are distributed relative to zoning plans. The widespread occurrence of longline culture across various environments indicates the relevance of developing specific spatial management strategies for this method. Similarly, the concentration of raft culture outside designated zones in certain provinces highlights the value of localized assessments when considering management priorities.

More broadly, this research contributes to understanding how the spatial structure of mariculture is evolving in relation to planning efforts. By combining remote sensing analysis with classification of mariculture types, the findings support improved knowledge of spatial expansion trends and their implications for sustainable marine resource use. These insights can assist policymakers, environmental managers, and industry stakeholders in enhancing spatial coordination and promoting balanced development between economic activities and ecological conservation.