Digital Clustering Method for Coastal Zone Scenes Based on Landscape Character Theory

Abstract

Aiming at the high-quality development of marine landscapes and the needs of accurate assessment and quality improvement of coastal landscapes, systematic digital analysis and quantitative research of typical coastal zone scenes have become one of the prerequisites for the in-depth research on coastal landscapes. This study, based on landscape character theory, constructs an analytical framework and technical path suitable for the digital clustering research on coastal scenes with the Gaussian Mixture Model (GMM). Taking the typical area of Taozi Bay in Yantai as an example, this study collaborates with remote sensing image interpretation and ArcGIS spatial analysis to quantitatively extract basic information of coastal landscapes, establishes a coastal zone scene characterization system, uses the GMM to form a digital clustering analysis process for scene characters, and combines Bayesian Information Criterion and expectation maximization algorithms to optimize key parameters for coastal zone scene clustering. It integrates classification and digital mapping practices for coastal zone scenes, and provides an analytical basis for the formulation of corresponding landscape and environmental management strategies. Proposing an analytical method suitable for the quantitative characterization and digital integration of coastal zone scenes, this study offers research references and practical implications for the clustering identification and collaborative management of coastal landscape resources.

Graphical abstract

Keywords

Coastal Zone Scene / Clustering Method / Gaussian Mixture Model / Landscape Character Theory / Coastal Zone / Landscape Planning and Design

Highlight

	· A classification framework for coastal scenes is constructed based on landscape character theory
	· A GMM-based clustering method is developed for digital analysis of coastal scenes
	· A case study in Taozi Bay demonstrates the classification framework with scene typologies and mapping outcomes

1 Introduction

Precise assessment and refined development of coastal zone landscapes constitute an essential means of highlighting coastal regional identity and sustaining their native landscape characters^[1]. Harnessing digital technologies to classify coastal zone scene resources and determine their characters has become the fundamental and pivotal approach in today's marine landscape upgrading initiatives. As a coordinated composite landscape with the land–sea interface, the coastal zone scene serves for purposes of ecological conservation, resource utilization, and cultural inheritance, playing an indispensable role in preserving scene authenticity, strengthening regional distinctiveness, and preventing landscape homogenization^[2]. In 2022, China explicitly set the goal of enhancing marine ecological protection and creating "beautiful bays, " further propelling coastal zone scene research from traditional macro-scale planning towards refined identification and dynamic regulation^[3]. Against the backdrop of rapidly evolving technologies such as big data and machine learning, it is imperative to develop quantitative methods and analytical techniques tailored to the environmental characters of coastal zone scenes. Such methods will assist in scene classification, character analysis, and digital representation, thereby improving the precision of graphical interpretation and the reliability of resource governance.

Contemporary research on landscape characters provides a crucial theoretical basis for scene analysis. Earlier studies dissected the scene essence from the perspectives of element composition, spatial structure, and functional relation, and their findings have been widely applied to scenic and historic areas^[4], natural reserves^[5], rural landscapes^[6], and other scenes. With the research paradigm shifting from static identification to dynamic structural analysis, current coastal zone scene studies commonly face problems such as incomplete cognition of elements^[7], dispersed sample-extraction pathways^[8], imperfect characterization systems^[9], and insufficient clustering accuracy^[10]. To overcome these limitations, this study addresses the objective need for comprehensive cognition and refined regulation of coastal zone scenes. Grounded in landscape character theory, it focuses on the multivariate characteristics and spatial differentiation patterns of coastal zone scenes, and systematically integrates the analytical workflow of "sample extraction–character representation–clustering computation–digital mapping." Accordingly, it establishes a digital clustering method and analytical mechanism suitable for studies on coastal zone scenes. The results will advance visually induced scene cognition towards an expressible and assessable quantitative description, and provide technical support for the scientific understanding and resource management of regional, native coastal zone scenes.

2 Theoretical Review of Coastal Zone Scenes Based on Landscape Character Theory

2.1 Landscape Character Theory

Landscape character is defined as "a distinct, recognizable and consistent pattern of elements in the landscape that makes one landscape different from another"^[11]. Such patterned character distinguishes one landscape type from others and endows a place with unique spatial perception and cultural identity. Offering an important lens within landscape classification, landscape character theory holds that the attribute, distribution, and morphology of landscape elements determine the statue of a landscape's character; therefore, assessing that status constitutes the landscape classification process^[12]. Since the 1990s, studies focusing on the qualification, identification, and classification of landscape characters have increasingly appeared and been widely applied in research on territory landscapes^[13] and national parks^[14], as well as meso- and micro-scale studies on cultural heritage sites^[15], settlements^[16], and rural areas^[17]. Traditional research—largely grounded in paradigms such as morphological genes and morphological genealogy—has emphasized typology-oriented reasoning: building landscape identification frameworks through graphic language, and extracting and reconstructing element composition relationships in typical environments^[18]. In recent years, with the advance of big-data and machine-learning technologies, digital clustering method has become an emerging trend in landscape character research. Digital analytical approaches centered on multi-source data acquisition and machine-learning computation have rapidly become a key technical pathway for the precise analysis of landscape characters and the optimization of spatial decision-making for landscapes.

2.2 Trends in Coastal Zone Scene Research

Scene has become one of the core topics in contemporary landscape character studies: scientifically understanding scene characters is crucial for enhancing the recognized value of a site, optimizing the allocation of spatial resources and fine-tuning landscape strategies^[19]. Drawing on existing literature^[20], this paper defines a coastal zone scene as the perceptible ensemble of marine, coastline, and land areas—a spatial form of interactions between natural and artificial elements within the land–sea transition zone—and composed of basic units such as shoreline morphology, tidal dynamics, coastal wetlands, and habitat patches. Owing to variations in geographical locations, modes of use, and other factors, coastal zone scenes have not yet converged on a common configurational model. In 2012, Natural England introduced the Seascape Character Assessment (SCA) framework, which established a local-scale mechanism for zoning identification and strategic layering based on visual perception and degrees of human intervention, and has since become an important international reference^[21]. From the perspective of geomorphic-system evolution, Carlos E. Nieto et al. devised a cartographic evaluation method suitable for diagnosing coastal zone landscape change^[22]. Ziting Bao constructed a local-scale system for landscape character identification and strategic layering that integrates visual perception, cultural intervention, and management requirements, forming a methodological framework that supports spatial identification, zonal evaluation, and management control, and provides scientific guidance for coastal zone scene assessment and resource coordination^[23].

Current studies on the landscape character of coastal zone scenes mainly address topics such as coastal style analysis, marine cultural perception, and shoreline change monitoring^[24], and have made notable progress with technical pathways such as remote-sensing image analysis and spatial statistical measurement^[25]. Yet traditional classification methods, e.g., K-means, hierarchical clustering, see limitations when confronted with complex datasets^[26]. By contrast, the Gaussian Mixture Model (GMM), with its ability to cluster high-dimensional, complex data and to assign data points to probabilistic types (which enriches interpretive information)^[27], offers more effective technical support for the identification and classification of coastal zone scenes.

In general, research on coastal zone scenes is still situated at the stage of type analysis and character identification. It is now an urgent task to explore clustering methods driven by high-dimensional data and to quantify the character factors and combinational mechanisms of typical scenes, hoping to offer scientific reference for classifying coastal zone scenes and related resource management.

3 The Workflow and Method of Digital Clustering for Coastal Zone Scenes

3.1 Workflow of Digital Clustering for Coastal Zone Scenes

Grounded in landscape character theory, this study constructs a systematic research framework that links the four stages of "sample extraction–character representation–clustering computation–digital mapping" for coastal zone scenes, thereby forming a replicable and expandable digital clustering method (Fig. 1).

1) Sample extraction. Considering the land–sea transitional nature of coastal zone landscapes, coastline vector data are first extracted as the baseline by following the SCA framework. A viewshed analysis is then performed with a digital elevation model (DEM) to delimit the initial study area^[28]. Subsequently, boundary accuracy is refined by integrating contour lines, road edges, depth contours, and watershed limits, culminating in a comprehensive sample library of coastal zone scenes.

2) Character representation. Natural and artificial landscape character elements for the study area are consolidated to establish a scene character database that covers key factors such as landforms, surface resources, urban development, coastal utilization, and culture and perception. ArcGIS spatial-analysis tools are used to disaggregate these character factors, while the MIKE 21 platform generates tidal dynamics and marine environment datasets, together forming the input for subsequent clustering statistics.

3) Clustering computation. Principal Components Analysis (PCA) is applied for dimensionality reduction and for extracting core character variables. The Bayesian Information Criterion (BIC) is invoked to determine the optimal number of clusters and hence the classification dimensionality. GMM then performs the clustering, with parameters iteratively refined via the Expectation Maximization (EM) algorithm, yielding precise classification results for coastal zone scenes.

4) Digital mapping. Naming and coding rules are scripted in Python; GIS and eCognition multi-scale segmentation are employed to enhance boundary precision and to map the gridded classification of scene types, producing an atlas of coastal zone scene characters. Finally, the hierarchical clustering is used to further classify scene types and scene clusters, supporting decision-making for the classification management and landscape control of coastal zone scenes.

3.2 Digital Clustering Methods for Coastal Zone Scenes

3.2.1 Character Representation System for Coastal Zone Scenes

Establishing a character representation system for scenes, screening core landscape elements, and performing data transformation are the key steps for accurately quantifying landscape characters^[29]. Coastal zone scenes are formed jointly by natural evolutions and human activities^[30]; hence the selection of landscape elements must accommodate the combined influence of both. Drawing on existing research^[31] and using remote-sensing interpretation, spatial analysis and hydrodynamic modelling to disaggregate element characteristics, this study—based on the hierarchical division and type combination of landscape elements—constructs a two-dimensional characterization system (natural attributes and artificial attributes) suitable for coastal zone scenes. The system contains seven landscape elements and sixteen character factors (Table 1).

3.2.2 Clustering Method for Coastal Zone Scenes

The GMM-based digital clustering method for coastal zone scenes comprises six steps: construction of factor/variable dataset; delineation of grid units; establishment of data connectivity matrix; data dimensionality reduction and BIC computation; clustering analysis with the GMM algorithm; and digital mapping of the clustering results.

Given the high dimensionality, heterogeneity, and spatially non-uniform distribution of coastal zone scene data, the GMM treats the data as a composite of several Gaussian distributions, thereby adapting dynamically to the distribution patterns of landscape elements. In addition, by coupling the model with BIC to optimize both the number of clusters and the covariance structure, the GMM gains a greater stability and classification accuracy. Iterative optimization via the EM algorithm ensures that the model converges on the optimal solution^[32]. The accuracy of a GMM depends heavily on the appropriate setting of its clustering parameters. Among these, the BIC balances model goodness-of-fit against complexity and is calculated as:

(1)BIC=-2\ln (likelihood)+k\ln (N),

where likelihood denotes the likelihood function, kln(N) is the penalty term, with N being the sample size and k the number of model parameters. The BIC is applied to compare training models that differ in covariance structure or in the number of cluster centers—the smaller the BIC value, the better the model fit.

The GMM assumes a random variable x in an unspecified (potentially high) dimensional space, whose probability-density function is defined as:

(2)p(x)=\sum _{k=1}^k{\mathrm{w}}_{k}g\left(x\mid {\theta }_k\right),

where k is the number of clusters in the Gaussian mixture; w_k is the mixing weight of the kth component (\text{with}\sum _{k=1}^k{\mathrm{w}}_k=1);g denotes the probability density function of the kth Gaussian distribution; θ_k represents the parameters of the kth Gaussian distribution—specifically its mean vector and covariance matrix.

The EM algorithm alternates between an E-step (expectation) and an M-step (maximization). These two steps are iterated to maximize the likelihood function and refine the clustering until convergence, with the parameter estimates obtained in each M-step fed back into the subsequent E-step.

E-step:

(3)w_i(k)=\frac{{\pi }_k{p}_k\left(x\mid {\mu }_k,\sum _k^2\right)}{\sum _{i=1}^k{\pi }_i{p}_i\left(x\mid {\mu }_k,\sum _i^2\right)},

where w_i(k) is posterior probability that data point x_i belongs to cluster k; π_k is the priori (mixing proportion) of cluster k; p_k (x | μ_k, ∑_k²) is Gaussian probability-density function of cluster k.

M-step:

(4){\mu }_k=\frac{\sum _{i=1}^n{W}_i(k)x_i}{n},

(5)\sum _k^2=\frac{\sum _{i=1}^n{w}_i(k)\left(x_i-{\mu }_X\right){\left(x_i-{\mu }_X\right)}^{\mathrm{T}}}{n_k},

(6)n_k=\sum _{i-1}^n{W}_i(k),

where μ_k is updated mean vector of cluster k, i.e., the weighted average of all data points in that cluster; ∑_k² is updated covariance matrix of cluster k, capturing the spread of data in each dimension; n_k is weighted sample count for cluster k, i.e., the total responsibility mass of the cluster; x_i denotes the ith observation, which captures the values of a sample across multiple dimensions; and T denotes the transpose of a vector.

3.2.3 Digital Mapping of Coastal-Zone Scenes

To enhance the spatial visualization of the clustering results, it is necessary to clarify the dominant characters and the factor structure of each scene unit. Using Python, this study established a standardized cartographic coding scheme: "LCTn" denotes a landscape character type, and "LCAn" denotes a landscape character area, where n is the serial number. The factor notation was based on the proportional area of each factor: when the factor X occupies an area ≥ 60%, it is recorded as X; when it covers between 30% and 60%, it is recorded as [X]; when it accounts for no less than 10% and no more than 30%, it is shown as (X); and it is omitted if it occupies less than 10% of the area^[33].

Because the visual output of coastal zone scene clustering often takes the form of fragmented mosaic patches, multi-level image segmentation in eCognition adopted in this study was combined with remote-sensing imagery to refine the cluster boundary and improve the precision and integrity of classified areas. The similarity among different scene types was then assessed with Pearson correlation coefficients so that compatible clusters can be merged, after which hierarchical clustering was applied to build scene clusters and to refine scene zoning. Finally, drawing on the dominant characters of each coastal zone scene, the study proposed targeted strategies for conservation and utilization, supporting fine-grained regulation of coastal zone landscape planning and management.

4 Case Study on the Digital Clustering of Coastal Zone Scenes: The Case Study on Taozi Bay, Yantai

4.1 Study Area

Taozi Bay is located in the northwestern coastal area of Yantai City, Shandong Province, China. It is a semi-enclosed secondary bay characterized by complex geomorphological conditions and a variety of coastal zone scenes, including mountainous areas, rivers, urban development areas, and marine environments^[34]. In this study, Taozi Bay was selected as the study area to verify the applicability of the proposed digital clustering method based on landscape characterization theory under conditions of diverse scene types and heterogeneous terrains.

4.2 Boundary Delimitation and Extraction of Character Factors in the Study Area

Multispectral imagery from the Gaofen-1 (GF-1) satellite was employed as the primary data source. After radiometric calibration, atmospheric correction, and orthorectification, coastline vector data were extracted. Using the ArcGIS platform, a 1-kilometer buffer was generated on both the landward and seaward sides of the coastline. Along the buffered coastline, a 300-meter interval grid of observation points was established to perform viewshed analysis, and the resulting cumulative viewshed was used to delineate the boundary of the study area, where the numerical value represents the number of times the area is seen by the observation points (Fig. 2). The final delineated area covers 770.35 km², including 264.89 km² of terrestrial areas and 505.46 km² of marine areas, with a total coastline length of 75.32 km.

Two types of grid units—500 m × 500 m for terrestrial areas and 100 m × 100 m for marine areas—were used, and a total of 108, 715 terrestrial grid units and 51, 510 marine grid units were obtained. The character factors of coastal zone scenes within the study area were derived from 3 primary data sources (Table 2): 1) vector data, including building footprints and heights, used to derive urban construction indicators such as building density and functional zones; 2) raster data, including remote sensing imagery and DEMs, used to calculate NDVI, slope, terrain relief, and other natural geographic features; 3) text data, consisting of planning documents and tidal data, from which cultural perception, land-use planning, and marine environmental factors were extracted. Specifically, marine environmental factors were derived using MIKE 21 by simulating tidal flow fields during spring and neap tides to extract current speed and direction. All measured character factors were subsequently reclassified using the Natural Breaks method in ArcGIS, and area-weighted statistics were compiled for each variable (Table 3, Fig. 3).

4.3 Clustering Analysis of Coastal Zone Scenes

Based on the above classification results, PCA was used to perform dimensionality reduction on both land and marine units. For the land units, 49 variables were reduced to 20 principal components, explaining 79.20% of the total variance. For the marine units, 24 variables were reduced to 11 principal components, with a cumulative variance explanation of 78.91%. Subsequently, the BIC was used to determine the optimal parameters for GMM—a lower BIC value indicates better model fit. The optimal BIC value for land units occurred when the covariance type was "full" and the number of clusters was 18. For marine units, the optimal BIC value was reached when the covariance type was also "full" with 7 clusters.

According to the optimal cluster numbers, the initial clustering labels generated by the GMM were further optimized via the EM algorithm. After 42 rounds of iteration, the model converged and produced the final clustering results, generating a total of 25 typical coastal zone scene types. Thus, a corresponding dataset containing these 25 scene types was then constructed (Fig. 4), including 18 land scene types (LCT1 to LCT18) and 7 marine scene types (LCT19 to LCT25). Following the preliminary classification, boundary correction was applied by adjusting parameters of scale, shape, and compactness to 140, 0.3, and 0.2, respectively. The final result yielded 161 aggregated typical sections of coastal zone scenes (Fig. 5).

4.4 Interpretation of Coastal Zone Scene Clustering Results

Based on the above analyses, it was found that the mountain–sea configuration constitutes the most prominent landscape character of the study area, with the overall spatial structure exhibiting a "mountain–city–shoreline–sea" pattern. Using Pearson correlation analysis, the aggregated coastal zone scene types were divided into two major groups: the land group (LG) and the marine group (MG). Among them, the forest–mountain scene group (LG4) is distributed as linear, fragmented patches along the urban periphery, exhibiting strong natural characteristics. The urban fringe scene group (LG2) and the urban built-up scene group (LG3) are interwoven throughout the study area. The coastal scene group (LG1) is primarily concentrated along the shoreline. MGs include nearshore shallow sea group (MG1) and offshore deep sea group (MG2) according to sea depth and current velocity. Through hierarchical clustering based on Euclidean distance and the Ward method, spatially adjacent and character-similar areas were grouped into 8 landscape character area groups (LCAG1 to LCAG8) (Figs. 6 ~ 8).

A systematic interpretation on the spatial distribution characteristics and composition of the eight resulting scene clusters was conducted from both perspectives of natural and artificial landscapes, and the following spatial differentiation patterns were identified:

1) Natural coast cluster (LCAG1): covering 5.55% of the total study area, primarily composed of MG1 and LG1; this cluster is characterized by natural shorelines and intertidal zones, with rich ecological resources and high ecological value.

2) Artificial coast cluster (LCAG2): accounting for 3.07% of the total study area, mainly consisting of MG1, LG1, and MG2; it includes urban shorelines, fishing ports, docks, and industrial coastal zones, with a high degree of functional development.

3) River and urban park cluster (LCAG3): covering 2.75% of the total study area, mainly formed by LG2 and LG3; this cluster is distributed along the urban-natural transition zones, showing corridor-like patterns but high fragmentation of landscape patches.

4) Urban built-up landscape cluster (LCAG4): representing 11.09% of the total study area, dominated by LG3 and LG2, concentrated in the core urban area; it is characterized by high building density, low greening rate, and poor spatial continuity.

5) Urban fringe landscape cluster (LCAG5): accounting for 11.24% of the total study area, scattered in block-like patterns; acting as a buffer zone between urban development and natural landscapes, it offers protective functions but suffers from insufficient green coverage.

6) Mountain–forest landscape cluster (LCAG6): occupying 8.06% of the total study area, primarily composed of LG4 and LG2; it forms a significant ecological barrier surrounding the urban areas.

7) Nearshore marine landscape cluster (LCAG7): covering 12.31% of the total study area, dominated by MG1; this cluster is concentrated in inner bay zones with low velocity and shallow water depth, exhibiting ecological sensitivity yet high biodiversity.

8) Offshore marine landscape cluster (LCAG8): the largest cluster, covering 42.78% of the total study area, primarily composed of MG2; although biological productivity is relatively low, this zone plays a critical role in regional fisheries and shipping trade.

4.5 Strategies for the Conservation and Utilization of Coastal Zone Landscapes

Based on the differentiated characteristics of typical coastal scene clusters, reference to both international practical cases, and considerations on the ecological background of the study area, this research proposed a series of conservation and utilization strategies for promoting the sustainable development of multifunctional coastal landscapes.

1) Conservation and ecological restoration of natural coasts. For natural coast clusters, strategies may draw inspiration from the Great Barrier Reef ecological conservation project in Australia^[35]. Measures such as periodic ecological monitoring, comprehensive habitat restoration, vegetation replanting, and sandbar stabilization can be adopted to strengthen the ecological buffer functions of natural shorelines.

2) Renovation and structural optimization of artificial shorelines. For artificial coast clusters, examples such like the construction experience of artificial coastlines at the Yongding New River estuary in Tianjin^[36] can be referred. Emphasis should be placed on the construction of eco-friendly embankments, and the ecological remediation of ports and industrial waterfronts to enhance the ecological functionality and aesthetic value of artificial shorelines.

3) Enhancement of urban built environment and landscape quality. For river and urban park clusters, as well as urban built-up clusters, lessons can be drawn from the Tokyo Bay wetland optimization project in Japan^[37]. Strategies include the integration of fragmented land parcels, expansion of green space coverage, and the construction of ecological greenways and pedestrian networks to improve the spatial continuity of urban green infrastructure. In addition, wetland and intertidal habitat restoration should be prioritized to strengthen local ecological resilience.

4) Protection of urban fringe green spaces and mountainous landscapes. For urban fringe landscape clusters, reference can be made to the greenbelt planning model of the Ruhr region in Germany^[38], focusing on native plant species configuration and urban heat island mitigation. For mountainous and forested clusters, strategies may refer to mountain conservation practices in the Alpine region^[39], implementing zoning-based protection and low-intervention recreation planning, balancing ecological restoration with scenic development.

5) Resource optimization and sustainable use of nearshore and offshore marine areas. For nearshore and offshore marine clusters, the intertidal ecological restoration practices of Auckland, New Zealand^[40] can provide a valuable reference. Emphasis should be placed on ecosystem protection, development threshold control, and the implementation of restricted fishing and aquaculture policies. Establishing a comprehensive environmental impact assessment mechanism is also essential to ensure the long-term sustainability of coastal resources.

5 Conclusions and Discussion

In response to the current demand for precise assessment and categorized governance of coastal zone scenes, this study constructed a digital clustering methodology grounded in landscape character theory. Taking the coastal zone of Taozi Bay in Yantai as an empirical case, the study explored the application path of this method and provided both theoretical and technical demonstrations for the fine-grained classification and governance strategies of coastal landscapes. By integrating multi-source data, this research established a representation system for coastal scenes, optimized clustering parameters using the BIC and EM algorithm, and developed a systematic clustering analysis process consisting of "sample extraction–character characterization–clustering computation–digital mapping." The feasibility of GMM-based digital clustering for coastal zones was thereby validated.

The findings indicate that the spatial distribution of coastal scenes in the Taozi Bay area is jointly shaped by natural landforms, oceanic dynamic processes, and human activities, exhibiting a typical "mountain–city–shoreline–sea" spatial pattern. Among them, scene units such as urban built-up environments, artificial shorelines, and ecological buffer zones show a clear spatial structure and well-defined boundaries, while scenes related to marine resource utilization display notable variation under different bathymetric conditions. The empirical results demonstrate that the scene classification system developed in this study effectively encompasses the representative landscape types within the study area, indicating the method's potential for further expansion and cross-regional application. China's vast coastline harbors abundant coastal landscape resources. With the rising demand for high-quality development and refined spatial governance of coastal zones, the application of digital clustering methods in coastal scene research remains in its early stages. This study, grounded in landscape character theory, explores a pathway for scene classification, analysis, and regulation in coastal zones. By digitally decoding land–marine characteristics and reshaping scene recognition through landscape structure logic, the study contributes to the establishment of a scientific evaluation framework for coastal and marine landscapes.

Nevertheless, this study primarily focused on the spatial distribution patterns of coastal scenes and did not sufficiently incorporate the influence of social and economic driving forces, which constrains the systemic scope of conservation and utilization strategies. Moreover, various coastal types exhibit significant spatial heterogeneity due to differing oceanic dynamics, climatic conditions, and anthropogenic pressures. Future research should further refine classification variables and discriminant criteria based on specific coastal contexts, enhance the methodological adaptability, and expand applications across typical coastal zones to distill universal patterns, thereby supporting the synergistic development of landscape conservation and refined spatial governance for coastal areas.

References

[1] Gonçalves, C. , & Pinho, P. (2024) A manifesto for coastal landscape governance: Reframing the relationship between coastal and landscape governance. Ambio, 53 ( 10), 1454– 1465.

[2] Yun, C., & Jiang, X. (2002). Sustainable Development and Integrated Management of Coastal Zones. Ocean Press.

[3] Ministry of Ecology and Environment of the People's Republic of China. (2022). Issuance of the 14th Five-Year Plan for Marine Ecological and Environmental Protection.

[4] Zhang, T. , & Sun, Y. (2019) Spatial characteristics and functions of national scenic spots in China. Urban Development Studies, 26 ( 8), 6– 12.

[5] Sun, Q. , & Zhang, Y. (2020) Landscape character identification and its value assessment in natural areas: A case study of Qinghai Lake Basin. Chinese Landscape Architecture, 36 ( 9), 76– 81.

[6] Zhu, C. , Li, R. , Luo, J. , Li, X. , Du, J. , Ma, J. , Hou, C. , & Zeng, W. (2024) Research on evaluating the characteristics of the rural landscape of Zhanqi Village, Chengdu, China, based on oblique aerial photography by unmanned aerial vehicles. Sustainability, 16 ( 6), 5151.

[7] Xie, Y. , Liu, D. , Li, Y. , & Xing, W. (2024) Response mechanism and path of coastal zone planning based on spatial element flow. Coastal Engineering, 43 ( 3), 235– 250.

[8] Jiang, L. , Yu, J. , & Zhang, B. (2022) Analysis of the current situation and development of coastal zone remote sensing research. Journal of Applied Oceanography, 41 ( 2), 215– 228.

[9] Qiu, S. , Ye, H. , & Liao, X. (2022) Coastline recognition algorithm based on multi-feature network fusion of multi-spectral remote sensing images. Remote Sensing, 14 ( 23), 5931.

[10] Carlier, J. , Doyle, M. , Finn, J. A. , Ó hUallacháin, D. , & Moran, J. (2021) A landscape classification map of Ireland and its potential use in national land use monitoring. Journal of Environmental Management, 289, 112498.

[11] Swanwick, C. (2002). Landscape character assessment: Guidance for England and Scotland – Edinburgh. The Countryside Agency and Scottish Natural Heritage.

[12] Tao, Y. , Xi, X. , & Zhu, M. (2018) The method of European landscape character assessment and its reference value to China. Chinese Landscape Architecture, 34 ( 8), 107– 112.

[13] Sun, Q. , Li, N. , & Zhang, Y. (2022) Research on the spatial layout of China's national parks facing the control of landscape style and features. Chinese Landscape Architecture, 38 ( 4), 26– 31.

[14] Yang, D. , Gao, C. , Li, L. , & Van Eetvelde, V. (2020) Multi-scaled identification of landscape character types and areas in Lushan National Park and its fringes, China. Landscape and Urban Planning, 201, 103844.

[15] He, D. , Lu, L. , Wang, J. , & Zhang, J. (2022) Landscape character identification in large-scale linear heritage areas: A case study of Beijing Great Wall Cultural Belt. Landscape Architecture, 29 ( 9), 99– 106.

[16] Pan, Y. , Bai, J. , & Shi, Y. (2022) Research on traditional rural landscape zoning and characteristics based on dominant land cover—A case study of the Fulao District in east Guangdong. Chinese Landscape Architecture, 38 ( 2), 48– 53.

[17] Wang, Y. , Chen, Z. , & Cheng, Y. (2021) The representation system of rural landscape character and personality in the new era. Landscape Architecture, 28 ( 7), 107– 113.

[18] Van Eetvelde, V. , & Antrop, M. (2009) A stepwise multi-scaled landscape typology and characterisation for trans-regional integration, applied on the federal state of Belgium. Landscape and Urban Planning, 91 ( 3), 160– 170.

[19] Wen, W. , & Dai, J. (2021) Theory of scenes: Paradigm transformation and its practice in China. Journal of Shandong University (Philosophy and Social Sciences), ( 1), 44– 53.

[20] Shao, Y. , Meng, Y. , & Yin, Y. (2023) Achievements and implications of the research on landscape in coastal zone from 1982 to 2022. Landscape Architecture, 30 ( 5), 10– 18.

[21] Natural England. (2012). An approach to seascape character assessment.

[22] Nieto, C. E. , Martínez-Graña, A. M. , & Merchán, L. (2024) Landscape analysis and coastal planning: Ría de Arosa (Pontevedra, Spain). Land, 13 ( 5), 645.

[23] Bao, Z. (2016). Landscape as the representation of existence and new tool for sustainability management[Doctoral dissertation]. South China University of Technology.

[24] Pan, Y. , Wu, Y. , Xu, X. , Zhang, B. , & Li, W. (2022) Identifying terrestrial landscape character types in China. Land, 11 ( 7), 1014.

[25] Ding, D. , Jiang, Y. , Wu, Y. , & Shi, T. (2020) Landscape character assessment of water–land ecotone in an island area for landscape environment promotion. Journal of Cleaner Production, 259, 120934.

[26] Lu, Y. , Xu, S. , Liu, S. , & Wu, J. (2022) An approach to urban landscape character assessment: Linking urban big data and machine learning. Sustainable Cities and Society, 83, 103983.

[27] Chen, X., & Zhang, A. Y. (2024). Achieving optimal clustering in Gaussian mixture models with anisotropic covariance structures. Proceedings of the 38th Conference on Neural Information Processing Systems (NeurIPS 2024).

[28] The Marine Management Organization. (2014). Seascape assessment for the South Marine Plan areas: Technical report.

[29] Yang, D. , & Gao, C. (2024) Analysis of five topics in multi-scale identification of landscape characters in Europe. Landscape Architecture, 31 ( 1), 14– 22.

[30] Wang, Y. , Gan, Y. , Deng, Y. , & Xie, X. (2020) Land-ocean interactions and their eco-environmental effects in the coastal zone: Current progress and future perspectives. Bulletin of Geological Science and Technology, 39 ( 1), 1– 10.

[31] Xu, L. , Li, J. , Li, W. , Zhao, S. , Yuan, Q. , Wang, M. , Yang, L. , & Lu, X. (2014) Progress in impact of human activities on coastal resource and environment. Journal of Nanjing Normal University (Natural Science Edition), 37 ( 3), 124– 131.

[32] Liu, J. , Gao, Y. , Li, Y. , Xu, Z. , Pan, Y. , Yang, J. , & Zhang, W. (2021) Soil environmental quality zoning based on the Gaussian mixture model. Journal of Agro-Environment Science, 40 ( 8), 1746– 1757.

[33] Zhang, J. , Zhao, Y. , & Gao, C. (2023) Identification of rural landscape character in Jinyuan District, Taiyuan City, Shanxi Province. Landscape Architecture, 30 ( 2), 123– 130.

[34] Sun, X. , Zhan, C. , Yu, J. , Ren, Z. , Xu, G. , & Wang, Q. (2022) Geomorphology evolution of Zhifu tombolo from 1977 to 2017. Marine Science Bulletin, 41 ( 2), 170– 178.

[35] Chan, S. C. Y. , Swearer, S. E. , & Morris, R. L. (2024) Mangrove cover and extent of protection influence lateral erosion control at hybrid mangrove living shorelines. Estuaries and Coasts, 47 ( 6), 1517– 1530.

[36] Qiao, Y. , Yin, X. , & Zhang, Z. (2021) Theoretical research and application of artificial coastal ecological restoration based on ecological restoration based on ecological construction: A case study in Tianjin Yongding New River Estuary comprehensive rehabilitation project. Coastal Engineering, 40 ( 1), 68– 75.

[37] Nakamura, K. , Tockner, K. , & Amano, K. (2006) River and wetland restoration: Lessons from Japan. BioScience, 56 ( 5), 419– 429.

[38] Robinson, K. J. (2003) Exploring urban ecosystems in the Ruhr Region of Germany: Implementation strategies. Urban Planning International, 18 ( 6), 3– 25.

[39] Chen, G. (2004) Gongga alpine ecosystem observation station of the Chinese Academy of Sciences. Journal of Mountain Science, 1 ( 2), 192.

[40] Rullens, V. , Mangan, S. , Stephenson, F. , Clark, D. E. , Bulmer, R. H. , Berthelsen, A. , Crawshaw, J. , Gladstone-Gallagher, R. V. , Thomas, S. , Ellis, J. I. , & Pilditch, C. A. (2022) Understanding the consequences of sea level rise: The ecological implications of losing intertidal habitat. New Zealand Journal of Marine and Freshwater Research, 56 ( 3), 353– 370.

RIGHTS & PERMISSIONS

© Higher Education Press 2025