Relationship Between Landscape Character and Public Preferences in Urban Landscapes: A Case Study from the East–West Mountain Region in Wuhan, China

Abstract

The East–West Mountain Region (EWMR) of Wuhan is a vital natural and cultdural asset, characterized by its scenic nature landscapes and rich historical and cultural heritage. This study aims to address the problems of landscape character degradation and weakened public preferences caused by rapid urbanization and proposes a research framework integrating landscape character assessment and public preferences. Initially, we utilize K-means cluster analysis to identify landscape character types based on six landscape elements, resulting in a landscape character map with 20 types. Subsequently, we employ emotion analysis based on Natural Language Processing (NLP) techniques to analyze user-generated content (UGC) from Weibo check-in data to establish perception characteristic indicators reflecting public preferences. Finally, we quantitatively identify the environmental factors influencing public preferences through the SoIVES model and compare and integrate the landscape character map with the public emotion value map. The results show that (1) public preferences hotspots are concentrated in three types: (a) urban construction-driven types, including areas dominated by commercial service functions and those characterized by mixed-function residential areas; (b) natural terrain-dominated types with well-developed supporting facilities; and (c) hybrid transition types predominated by educational and scientific research land uses. These areas generally feature a high degree of functional diversity and good transportation accessibility. (2) Landscapes eliciting stronger emotional responses integrate moderate slopes, multifunctional spaces, and robust public services, whereas areas with weaker responses are characterized by single-function use or excessive urbanization. (3) The emotional variations within categories could be influenced by (a) functional hybridity through enhanced environmental exploration; (b) spatial usage frequency through place attachment formation; and (c) visual harmony through cognitive overload prevention. These findings provide critical insights for formulating zoning optimization plans aimed at the refined conservation and utilization of urban landscape resources, as well as offering guidance for improving landscape planning and management in the EWMR.

1. Introduction

Urban landscapes refer to complex systems with ecological, social, and cultural functions that emerge during the process of urbanization [1]. These landscapes comprise both natural elements, such as water bodies, green spaces [2], and built components, including buildings, roads, and public spaces [3]. Urban landscapes are the result of dynamic interactions between human activities and the natural environment. From a global perspective, the fragmentation of urban landscapes caused by rapid urbanization has become a widespread environmental challenge. Studies have documented its impacts in cities ranging from Phoenix in the United States [4] and Berlin, Brandenburg, in Germany [5] to 16 major metropolises spanning multiple continents, all revealing consistent patterns of habitat degradation and spatial discontinuity [6]. Rapid urbanization has led to significant declines in landscape quality, reshaping both physical environments and public preferences [7]. In addition to landscape fragmentation, rapid urbanization has also led to significant declines in urban landscape quality, reshaping both natural environments and public preferences. Recent findings indicate that as natural lands are developed and transformed, the value of ecosystem services in the urban landscape declines [8]. Meanwhile, studies on high-density urban areas show that due to enhanced thermal effects and reduced green spaces, residents’ environmental quality is declining [9]. The homogenization of urban landscapes resulting from rapid urbanization poses a significant threat to sustainability, not only weakening public identification [10], but also directly jeopardizing the implementation of UN Sustainable Development Goal 11 [11]. Consequently, there is an urgent need to develop effective strategies for urban landscape planning and management that incorporate public subjective perceptions, thereby informing appropriate spatial planning measures to revitalize urban landscapes.

Landscape Character Assessment (LCA), as a tool that integrates natural and cultural landscapes along with human perceptions, provides a holistic framework for the implementation of the European Landscape Convention [12,13]. Rooted in geographical typology, LCA aims to capture the holistic of landscape identity for future protection, planning and management, thereby establishing types and spatial units [14]. This typological approach also constructs a comprehensive urban planning framework [13,15]. The methodology identifies characteristic patterns in urban structures, specifically examining street networks, architectural forms, and public spaces, to provide practical guidance for land use allocation decisions and urban renewal projects [16]. LCA emphasizes a transition from resource assessment to character assessment based on a relational perspective [17]. However, the characterization process, which involves identifying, classifying, mapping, and describing landscape character [14], often neglecting public perception dimension. This trend can be attributed to advancements in GIS and spatial analytical techniques, which facilitate the efficient identification of landscape character through the selection of biophysical landscape indicators and the application of methods such as cluster analysis [18,19,20]. However, these quantitative analyses often show discrepancies with the public’s actual perceptual experiences. Human perception plays a crucial role in identifying landscape character. It is essential to integrate the landscape’s biophysical attributes with public perception [21].

Landscape preference results from the public’s assessment based on their perception of landscape characters [22,23,24]. Public preference pertains to the visual esthetic value and the subjective experiences elicited by landscape character within esthetic activities [21]. Environmental psychology research indicates that such preferences stem from an innate human inclination toward environments that balance coherence and complexity [25], as systematically explained by Kaplan’s preference matrix theory [26]. Different methodologies have been employed to study public perception and preference [27,28,29,30,31], yet traditional questionnaire surveys exhibit notable limitations. These surveys typically restrict respondents to evaluating preselected images, which hampers the ability to capture genuine and nuanced environmental experiences [32,33]. Moreover, they are inherently constrained by temporal and spatial boundaries, limiting the scope and representativeness of the data collected. In contrast, User-Generated Content (UGC) derived from social media platforms offers a promising alternative as a dynamic and readily accessible source of collective public perception [34,35]. UGC benefits from continuous accumulation over extended periods and encompasses vast volumes of data, enhancing both temporal depth and spatial breadth [36]. Leveraging parametric analysis techniques on UGC enables researchers to accurately identify and characterize perceptions that are specific to particular places [37,38]. Additionally, the integration of advanced emotion analysis within Natural Language Processing (NLP) frameworks facilitates the creation of emotional mapping tools, which effectively bridge the gap between passively acquired public emotion and the physical and cultural attributes of landscape character [39,40]. This multi-dimensional approach substantially enhances the understanding of how individuals perceive, interpret, and emotionally relate to landscapes.

Currently, LCA is predominantly conducted using quantitative parametric methods. However, approaches focusing on the biophysical attributes of landscape, emphasize “mapping” rather than a comprehensive “character assessment”. Building on the Landscape Planning–Character–Services framework, which advocates using landscape character as a bridge to integrate multidimensional human perception and values into planning practice [41], it is necessary to establish a connection between biophysical landscape characteristics and people’s subjective perceptual preferences [42]. In particular, the differences in how the public perceives various landscape character types have yet to be fully investigated [43]. Further research is needed to systematically examine these perceptual variations associated with different landscape characters.

Urban natural mountains and lakes, as critical resources in urban environments, have been extensively studied for their ecological services, cultural heritage values, and recreational benefits across landscape planning, ecosystem service assessment, and historic conservation [44]. The East–West Mountain Region (EWMR) of Wuhan, as a crucial “urban–nature transition area”, integrates the dual attributes of ecological barrier and cultural heritage preservation. It serves as Wuhan’s primary ecological backbone, connecting fragmented habitats through the Greenway network to enhance biodiversity [45], while also preserving historical landmarks such as the Yangtze River Bridge (China’s first bridge to span the Yangtze River) and ancient structures on Guishan and Sheshan Mountains. The East Lake Scenic Area, a core component of the EWMR, exemplifies this synergy: its cultural service value surged to CNY 2.09 billion by 2019, tripling since 2011 due to eco-tourism development [46]. However, high-speed urban construction driven by economic priorities has strained the urban landscape, caused ecological degradation, and also degraded collective esthetic perception [45]. Since 2019, Wuhan’s Cultural Greenway planning has been emphasizing holistic integration of natural and cultural resources, necessitating a bottom-up approach to align landscape management with public perspectives.

This study aims to link objective landscape character with subjective public preference. The specific research objectives are as follows: (1) to quantitatively assess the landscape character of urban landscapes and analyze their spatial distribution patterns; (2) to identify public perception hotspots and their spatial characteristics based on UGC net-work texts and NLP emotion analysis; (3) to investigate the associations between landscape character and public perception using quantitative analysis with the SoIVES model and ArcGIS Pro 3.0, and provide a basis for urban landscape planning and management.

2. Materials and Methods

The proposed methodology encompasses three principal stages (Figure 1): (1) Landscape character identification: This stage involved the use of ArcGIS Pro 3.0 to delineate landscape character types, and to support spatial representation. (2) Public preference assessment: This stage employed Weibo check-in data with emotion analysis based on NLP to generate an emotional heatmap, which served to evaluate the characteristics of public preferences. (3) Correlation element quantitative analysis: This stage utilized the SoIVES model for statistical analysis and ArcGIS Pro 3.0 for the visual analysis of correlation elements and influencing factors in the relationship between landscape character and public preferences. The following sections provide a detailed exposition of the research methods and procedures associated with each stage.

2.1. Study Area

The EWMR (114°8′–114°32′ E, 30°27′–30°34′ N) is located in the Wudang and Dahong Mountains in the northwest of Wuhan, traversing the Yangtze River at the Guishan and Sheshan Mountains and extending southeastward across the city (Figure 2). It intersects the Yangtze River in a cross-shaped pattern, forming a prominent natural and geographical axis. The EWMR is notable for its rich natural and cultural resources, featuring a 24.5 km cultural greenway with an 11.5 km main route for cycling and walking and a 13 km walking branch. As a key recreational area of Wuhan’s urban landscape, the EWMR greatly enhanced the city’s livability.

2.2. Landscape Character Identification

The selection of elements for the identification of landscape character should take into account regional variations, scale, and other relevant attributes [47]. This study synthesized policy data, element maps, historical atlases, and planning drawings, while referencing the LCA guidelines [17] to classify the elements into six primary indicators and forty elements (

Table 1. Classification table of landscape character identification elements.

Level Indicators	Elements	Code
Slope A	Flat Slope ≤ 2°	A1
	Gentle Slope (2–6°]	A2
	Moderately gentle Slope (6–15°]	A3
	Moderately steep Slope (15–25°]	A4
	Steep Slope > 25°	A5
Aspect B	No Aspect (−1°)	B1
	Shady Slope (0–45°, 315–360°)	B2
	Semi-Shady Slope (45–135°)	B3
	Sunny Slope (135–225°)	B4
	Semi-Sunny Slope (225–315°)	B5
Land Use C	Farmland	C1
	Forest	C2
	Shrubland	C3
	Grassland	C4
	Water Body	C5
	Unused Land	C6
	Urban Construction Land	C7
Scenic Spot Type D	3A-Level Scenic area	D1
	4A-Level Scenic area	D2
	5A-Level Scenic area	D3
	Cultural Relics Protection Units	D4
	Park	D5
Surface Functionality E	Residential Land	E1
	Commercial Office Land	E2
	Commercial Service Land	E3
	Industrial Land	E4
	Transportation Hub Land	E5
	Airport Facility Land	E6
	Administrative Office Land	E7
	Educational and Research Land	E8
	Medical and Health Land	E9
	Sports and Cultural Land	E10
	Park and Green Space Land	E11
	Non-Construction Land	E12
Time Depth F	Urban Coverage Before 1990	F1
	Urban Coverage Before 2000	F2
	Urban Coverage Before 2010	F3
	Urban Coverage Before 2015	F4
	Urban Coverage Before 2020	F5
	Non-Built-Up area	F6

). The Digital Elevation Model (DEM) data used in this study were obtained from the Geospatial Data Cloud of the Chinese Academy of Sciences (http://www.gscloud.cn/, accessed on 1 November 2024), and used ArcGIS Pro 3.0 to process and obtain the two primary indicators of slope and aspect: the land use data were derived from Esri, the Impact Observatory and Microsoft jointly produced and released the data based on the 10-m resolution satellite data of Sentinel-2 (https://livingatlas.arcgis.com/landcoverexplorer/#mapCenter=81.18780%2C6.76158%2C9.61111111111111&mode=step&timeExtent=2017%2C2023&year=2023, accessed on 1 November 2024) released to the public by Impact Observatory and Esri; the Scenic Spot Type data were sourced from the Point of Interest (POI) data provided by AutoNavi Map (https://ditu.amap.com/, accessed on 1 November 2024); the surface functionality data were obtained from Mapping Essential Urban Land Use Categories in China (EULUC-China) (https://doi.org/10.1016/j.scib.2019.12.007, accessed on 1 November 2024) released to the public by the team of Peng Gong and Bin Chen; and the time depth data (reflecting changes in Urban Coverage over a span of 30 years) were collected through dataset of built-up areas in Chinese cities for the year 2020 (https://doi.org/10.11922/sciencedb.j00001.00332, accessed on 1 November 2024).

The Create Fishnet tool in ArcGIS Pro 3.0 was employed to generate grid cells measuring 200 m × 200 m. Subsequently, the Intersect Table tool was utilized to calculate the intersections of slope, aspect, land use, scenic spot type, surface functionality, time depth and grid (Figure 3). Ultimately, the data were organized in accordance with the landscape character identification elements classification table, resulting in an intersection table that served as the basis for further analysis.

2.3. Public Preference Assessment

In this study, public preferences were characterized by the polarity of text emotion, drawing on UGC from Weibo check-in data to reflect public preferences. The workflow comprised three steps: (1) preprocessing raw comments into valid text phrases; (2) quantifying emotion polarity using NLP-based emotion analysis methods (with values ranging from 0 to 1); and (3) integrating the resulting data with geographic information, and employing the Coordinate Transformation tool in ArcGIS Pro 3.0 to construct a visual emotion map (Figure 4).

2.3.1. Data Collection and Preprocessing

First, the study area was defined as the spatial scope for collecting Weibo check-in data. Using keywords such as “the EWMR in Wuhan” and “Ma’anshan”, a web crawler was employed to retrieve raw data from the Weibo mobile platform, including user IDs, usernames, geographic locations (latitude and longitude), timestamps, and textual content. These records were then used to construct the initial dataset.

During the data cleaning phase, it was essential to remove invalid information using Excel, with special attention paid to three categories of noise: (1) news advertisements and explanatory texts disseminated by public accounts; (2) excessively brief texts consisting solely of locations, names, or pure emoticons; and (3) comments that mentioned the research region but were irrelevant to the study’s content. A random sample of 200 entries was retained for manual verification. If any invalid data were identified, the cleaning process was repeated until all samples were deemed valid, resulting in a text dataset suitable for research purposes.

Due to the lack of explicit word delimiters in Chinese vocabulary, an initial word segmentation tool was employed to divide the continuous text into individual words [48]. In this study, the Jieba Segmentation tool was used to process the relevant text data by performing word segmentation and removing stop words. Subsequently, a word frequency analysis was conducted to statistically identify high-frequency valid terms, which were then compiled to form the database for subsequent emotion analysis.

Previous qualitative research indicated that each sentence within the interview corpus could serve as a natural unit, potentially encapsulating public preferences. Moreover, longer comments often contained multiple dimensions of emotional perception. Therefore, the valid text dataset was further segmented into shorter sentences using the split function, which divided text at punctuation marks such as “.”, “?”, “!”, and “^”. The resulting segments were then imported into Excel, where excessively brief sentences, sentences lacking emotional content, and extraneous spaces were removed, yielding a refined set of valid textual information.

2.3.2. NLP Emotion Analysis

For emotion analysis, texts containing vocabulary related to sensory perception were selected as sample data for the evaluation of emotion polarity. The processed data were subsequently classified for emotion polarity using the Tencent Cloud API, which provided a foundation for assessing public preferences. This method classified emotion as either positive or negative, with both polarities represented as real numbers ranging from 0 to 1, where higher values indicated more intense emotional expression.

By applying emotion polarity analysis to Weibo text, coordinates reflecting public emotional perceptions were identified. Subsequently, the emotional values associated with each POI (n = 20,002) were calculated. These coordinates, together with their respective emotion classification results, were imported into the ArcGIS Pro 3.0 spatial analysis software using the Coordinate Transformation tool. The emotion analysis outcomes were then visualized using kernel density analysis, resulting in the generation of an emotion map based on user preference assessments.

2.4. Correlation Element Quantitative Analysis

The SoIVES model was employed to characterize and map public preferences by statistically integrating NLP-processed emotional value point data to generate spatial maps of value indices [49,50,51]. Average nearest neighbor analysis was conducted on the value points to calculate the average nearest neighbor ratio (R-value) and its standard deviation (Z-value), thereby determining the spatial clustering of point data. Based on the identification of landscape character, the geographic database constructed for the SoIVES process included six environmental variables: slope, aspect, land use, scenic resource points, surface functionality, and time depth. A multicollinearity test of the environmental variables using Variance Inflation Factor (VIF) analysis revealed that the VIF value of time depth was 16.89, while all other variables had VIFs below 10 (Appendix A 

Table A1. Initial environmental variable VIF detection.

Variable	VIF	Judgment
Slope	1.95455	Acceptable
Aspect	3.42724	Acceptable
Land Use	2.90524	Acceptable
Scenic Spot Type	8.15787	Acceptable
Surface Functionality	4.79203	Acceptable
Time Depth	16.8917	Needs to be addressed

), indicating a severe multicollinearity issue associated with time depth [52]. To address potential multicollinearity, we first conducted VIF test and correlation analyses, which confirmed significant multicollinearity among predictors (VIF > 5), particularly a moderate positive correlation between time depth and surface functionality (r = 0.44, p < 0.01). Consequently, principal component analysis (PCA) was applied to reduce dimensionality while retaining essential variance (Appendix A, Figure A1). The principal component derived from this process accounted for a cumulative variance contribution of 70.7%, reducing the VIF to 2.54 (Appendix A,

Table A2. The variable VIF after dimension reduction.

Variable	VIF	Judgment
Slope	1.87146	Acceptable
Aspect	2.904	Acceptable
Land Use	2.43719	Acceptable
Scenic Spot Type	4.32232	Acceptable
PCA_Component	2.54133	Acceptable

), and effectively eliminating collinearity interference. Finally, the point data along with the processed five environmental variables were imported into MaxEnt 3.4.1 for modeling. The MaxEnt parameters reserved 30% of emotional value points as test data. This process was repeated five times to optimize model results and generate value index maps. Model reliability was evaluated using the Area Under the Receiver Operating Characteristic Curve (AUC), with AUC values exceeding 0.70 indicating valid model results [53].

Upon importing the results of the training model into ArcGIS Pro 3.0, it was recommended to use the Define Projection and Copy Raster tools to generate a map of emotional values. Subsequently, the landscape character elements were linked using the Zonal Statistics and Spatial Join tools within ArcGIS Pro 3.0. This approach facilitated the creation of a correlation table and a mean value map, which together illustrated the relationship between landscape character and public preferences.

3. Results

3.1. Landscape Character Identification Results

The K-means clustering algorithm, while being a conventional and efficient clustering approach, faces inherent challenges in determining the optimal K value due to its random initialization nature [54]. To address this, multiple validation methods have been developed, including the Silhouette Score [55], the Dunn Index [56], as well as the Elbow Method, which is widely applied in LCA [57]. In LCA studies, researchers typically combine the Elbow Method with visual interpretation to ensure the clustering results correspond to identifiable geographical entities [58]. Following this established practice, our study employs this combined approach. Through comprehensive analysis that integrates quantitative assessment from the elbow method with cartographic validation, we identified an optimal clustering range of K = 20–30 based on inertia stabilization (Appendix B Figure A2). Subsequent visual interpretation of clustering results at K = 5, 10, 15, 20, 25, and 30 (Appendix B Figure A3) confirmed that K = 20 most effectively represents characteristic geographical entities such as East Lake, Guishan Mountain, and Sheshan Mountain while preserving typological coherence. This outcome aligns with the core principle of “discernible geographical units” in landscape typology [20,58] while addressing the operational requirements for classification systems in spatial planning. For cluster identification integrating numerous elements, naming is often carried out in the form of coding. Previous studies have often conducted it based on the proportion of coding [20,59]. We take the element X as an example, when the proportion of X is greater than 60%, it will be directly expressed as X; when X is between 30% and 60%, it will be expressed as {X}; when X is between 10% and 30%, it will be expressed as (X); when X is less than 10%, it will be ignored. We identified 20 landscape character types based on the principles of K-value selection and encoding (

Table 2. Identification of landscape character types.

Landscape Characters	Encodings	Key Natural Elements	Key Cultural Elements
1	{A3}(A1A2). {B4B5}. C7. D3. (E3E8E9). F4. F6	moderately gentle slope, flat slope, gentle slope terrain; sunny slope, semi-sunny slope terrain; predominantly urban construction land; with small amounts of commercial service, educational and research, medical and health land; mostly located within the 2015 urban coverage region, non-built-up areas also distributed	5A-class tourist attraction
2	A4(A3). B4. C5. D2. E12 (E1E11). F6{F5}	moderately steep slope, moderately gentle slope terrain; sunny slope terrain; large water body region; mostly in non-construction land, with residential land, park and green space land; mostly in non-built-up area, partly located within the 2020 urban coverage area	4A-class tourist attraction
3	A2(A1). B4{B3}. C5. D2. E1(E12). F1F2	gentle slope, flat slope terrain; sunny slope, semi-shady slope terrain; predominantly water body area; large proportion of residential land and non-construction land; covering 1990–2020 stages of urban coverage area	4A-class tourist attraction
4	{A1A2}. {B4}(B2B3B5). C5(C1C7). D3{D4}. E12(E4). F1F2F3F4F5	flat slope, gentle slope; sunny slope terrain; mostly sunny slope, with shady slope, semi-shady slope and semi-sunny slope terrain; water body, farmland and urban construction land coexist; predominantly non-construction land and industrial land; covering 1990–2020 stages of urban coverage area	5A-class tourist attraction, major historical and cultural sites
5	(A1). (B2). {C1}(C5C7). (D2D3). (E12). (F5F6)	flat slope terrain; shady slope terrain; farmland, water body and urban construction land coexist; predominantly non-construction land; covering 2020 stages of urban coverage area and non-built-up area	4A and 5A-class tourist attraction
6	{A1A2}. {B2}(B3B5). C7. D4. E8. F1F2F3F4F5	flat slope, gentle slope terrain; mostly shady slope, with semi-shady slope and semi-sunny slope terrain; predominately urban construction land, with educational and research land; covering 1990–2020 stages of urban coverage area	Major historical and cultural sites
7	A3(A2). B2. C7{C1}. D2. E12{E11}(E1). F4F5F6. {F1F2F3}	moderately gentle slope, gentle slope terrain; shady slope terrain; predominately urban construction land and farmland; non-construction land, park and green space land, residential land are predominant; mostly located in the 2015–2020 urban coverage region, partly located in the 1990–2010 urban coverage area	4A-class tourist attraction
8	A1. B1. C5. D2. E7E12. F4F6	flat slope terrain; no aspect terrain; predominately water body; administrative office, non-construction land are predominant; mostly located within the 2015 urban coverage region, partly located in non-built-up area	4A-class tourist attraction
9	A2(A1). {B3B5}(B2). C7. D3. E7. F6(F5)	gentle slope, flat slope terrain; semi-shady slope, semi-sunny slope, shady slope terrain; predominately urban construction land; administrative office land is dominant; mostly located in non-built-up area, partly located within the 2020 urban coverage area	5A-class tourist attraction
10	{A3A4}. B4. C2. D4. E1. F6	moderately gentle slope, moderately steep slope terrain; sunny slope terrain; predominately forest; residential land is dominant; mostly located in non-built-up area	Major historical and cultural sites
11	{A1A2}. (B2B5). C1. D5. E12	flat slope, gentle slope terrain; shady slope, semi-sunny slope terrain; predominately farmland; mostly located in non-built-up area.	Park (for public recreation)
12	A1{A2}. {B3B4}. C1. E10E12. F6	flat slope, gentle slope terrain; semi-shady slope, sunny slope terrain; predominately farmland; sports and cultural, non-construction land predominates; mostly located in non-built-up area	-
13	A2(A1). {B2B5}. C1. D1. E10. F1F2F3F4F5(F6)	gentle slope, flat slope terrain; shady slope, semi-sunny slope terrain; predominately farmland; sports and cultural land dominated; covering 1990–2020 stages of urban coverage area, partly in non-built-up area	3A-class tourist attraction
14	A2{A1}. {B4}(B2B3B5). C1. D1. E12{E3}. F3F4F5F6	gentle slope, flat slope; sunny slope terrain; mostly sunny slope, with shady slope, semi-shady slope and semi-sunny slope terrain; predominately farmland; non-construction land, commercial service land are dominant; covering the 2010–2020 stages of urban coverage area	3A-class tourist attraction
15	A2(A1). B4(B5). C1. D3. E12. F6	gentle slope, flat slope terrain; sunny slope, semi-sunny slope terrain; predominately farmland; non-construction land is dominant; mostly in non-built-up area	5A-class tourist attraction
16	A3(A4). B2(B5). C2. D4. E12. F6	moderately gentle slope, moderately steep slope terrain; shady slope, semi-sunny slope terrain; predominately forest; non-construction land is dominant; mostly in non-built-up area	Major historical and cultural sites
17	A2{A1}. B4. C5{C1}. D3. E1. F3F5F6	gentle slope, flat slope terrain; sunny slope terrain; predominately water body, farmland; residential land is dominant; covering 2010–2020 stages of urban coverage region and non-built-up area	5A-class tourist attraction
18	A1. {B1}(B2B3). C1. D4. E12. F4F5F6	flat slope terrain; sunny slope terrain; mostly no aspect, with shady slope and semi-shady slope terrain; predominately farmland; non-construction land is dominant; covering 2015–2020 stages of urban coverage region and non-built-up area	Major historical and cultural sites
19	A1. B5(B4). C5. D1. E10E12. F4F5F6(F3)	flat slope terrain, semi-sunny slope, sunny slope terrain; predominately water body; sports and cultural, non-construction land are dominant; covering 2015–2020 stages of urban coverage area, partly located within the 2010 urban coverage area	3A-class tourist attraction
20	A2{A1}. B5(B1). C1. D2. E8. F5F6	gentle slope, flat slope terrain; semi-sunny slope, no aspect terrain; predominately farmland; educational and research land is dominant; located within the 2020 urban coverage area and non-built-up area	4A-class tourist attraction

). In ArcGIS Pro 3.0, the landscape character types are visualized, each type exhibits distinct natural and cultural characteristics (Figure 5). For instance, Type 1 is predominantly characterized by gentle slopes with minimal natural land use elements. Its cultural attributes are mainly represented by urban construction areas, which include limited commercial, educational, and medical land, primarily situated within the urban coverage established in 2015 as well as in some non-built-up areas. This type also encompasses a significant cultural attribute, a 5A-class tourist attraction.

In the landscape character map (Figure 5), each landscape type is composed of composite elements, which are formed by the superimposition of natural elements, such as slope, aspect, and land use, and cultural elements, including scenic spot type, surface functionality, and time depth. This integrated approach enables a nuanced classification and better captures the complexity of real-world landscapes. The spatial differentiation characteristics of various landscape elements can be summarized as follows:

(1)

Natural Terrain-Dominated Types: This group encompasses six distinct landscape types: type 2 characterized by gently sloping farmland areas, type 8 featuring flat water body systems, type 12 representing agroforestry systems on semi-sunny slopes, type 15 consisting of flat farmland expanses, type 16 dominated by woodland ecosystems, and type 18 primarily composed of farmland-dominated terrain. Their spatial distribution is primarily controlled by topographic factors such as slope and aspect, and they generally exhibit a continuous, patch-like distribution pattern. These areas are less affected by urban development pressures and tend to maintain ecological integrity, providing critical ecosystem services such as habitat connectivity and hydrological regulation.

(2)

Urban Construction-Driven Types: These landscape types comprise five distinct categories: type 1, characterized by high-density built-up urban areas; type 3, dominated by commercial service areas; type 6, primarily consisting of science and education hubs; type 7, featuring cultural and sports facilities as key elements; and type 13, representing multifunctional residential areas. These types are mainly distributed within urban expansion areas from 2015 to 2020, demonstrating a strong concordance with the phases of city growth. The spatial configuration of these landscapes reflects socio-economic forces and land demand changes associated with urbanization. As cities expand, these areas often experience significant land transformation, increased anthropogenic disturbance, and a reduction in natural surface cover.

(3)

Hybrid Transition Types: This category encompasses nine distinct landscape types: type 4 dominated by park green spaces, type 5 featuring a water-residential mixed pattern, type 9 representing built-up areas on semi-shaded slopes, type 10 combining science, education and residential functions, type 11 exhibiting farmland-built-up transitional areas, type 14 characterized by water-farmland mosaics, type 17 integrating water, residential and farmland elements, type 19 blending cultural/sports facilities with green spaces, and type 20 merging science, education and green space components. These types combine both natural and artificial elements, displaying pronounced fragmentation in spatial distribution.

3.2. Public Preferences Results and Visualization

A total of 36,716 original data entries were collected from Weibo platforms. Following the removal of 16,624 instances of non-personal user data or irrelevant information through noise data identification and filtering in Excel, 20,002 valid Weibo blog check-in entries were retained for analysis. Emotion analysis was subsequently performed using NLP techniques to extract public perceptions related to both negative and positive emotions associated with passive participation (Figure 6 and Figure 7). Based on the processed coordinates and data overlays, the public emotional responses map (Figure 8) was visualized in ArcGIS Pro 3.0.

The analysis of the negative emotion data (Figure 6) indicates that 3% of the POI negative index in the study area exceeds 0.5. The arithmetic mean of the negative index is 0.09153, and the median is 0.07315. The spatial distribution characteristics show that the negative index presents a relatively uniform but discontinuous distribution pattern, and the distribution range of extreme values is wide, mainly concentrated in sparsely populated areas. In contrast, the analysis results of the positive emotion data (Figure 7) show that 8% of the POI positive index is higher than 0.5, and both the arithmetic mean (0.16282) and the median (0.13938) are significantly higher than the negative emotion index. The spatial distribution of the positive index shows obvious agglomeration characteristics and is significantly positively correlated with population density and the diversity of landscape types. The specific manifestations are as follows: The positive index in urban central areas and transportation hub areas is generally high, while in sparsely populated areas, it shows a relatively low positive index. It is notable that the occurrence frequency of extreme positive emotional values is significantly higher than that of negative emotional values.

The intensity of public emotions is significantly influenced by the characteristics of different sites, as illustrated in Figure 8. Emotional intensity in areas such as transportation hubs, universities, and tourist attractions is markedly higher than residential types, demonstrating a distinct degree of spatial coherence. Significantly, three locations exhibiting “very strong” emotional responses are situated near prominent attractions, including Hubu Alley, Wuhan University, and Guanggu Roundabout. Additionally, one site characterized by a “strong” emotional response is located in the East Lake Ecological Tourism Scenic Area, a nationally identified 5A-level tourist attraction. In contrast, larger residential regions do not display strong emotional responses. Overall, the intensity of emotional perception reveals a distinct spatial distribution pattern, indicating that the emotional response intensity associated with tourist attractions is higher than that of residential areas, while urban landscapes elicit stronger emotional reactions compared to natural landscapes.

3.3. Relationship Between Landscape Character and Public Preferences

3.3.1. Spatial Analysis of Emotional Value Point Based on the SoIVES Model

The results from the MaxEnt 3.4.1 model analysis were visualized, resulting in the spatial distribution map of emotional value index (Figure 9). Average nearest neighbor analysis shows an R value of 0.64 and a Z value of −14.99, indicating that the points display significant spatial clustering and are therefore suitable for further examination. The value index map shows that regions dominated by low-middle emotional values are the most extensive within the study area. The majority of the remaining regions are classified as neutral emotional value areas, which are primarily arranged in planar formations, while some smaller areas are represented as points or lines. The effectiveness of the SoIVES model was assessed by AUC, which achieved a value of 0.893 (well above the 0.7 threshold), indicating a high level of reliability in the model’s outcomes.

In the Hongshan area in the eastern part of the study area, the regions with relatively low emotional value indices are mainly distributed in the south, north and east (the spatial coverage of Weibo check-in data in this area is relatively low). The central urban area is mainly composed of residential areas, commercial and office areas, and public service facilities. The emotional value index is mostly at a neutral level. Especially along major traffic arteries such as the Third Ring Road and the East Lake Tunnel, it shows a strip-shaped distribution feature. The positive emotional value areas are distributed in a scattered pattern, mainly concentrated in the Guanggu Roundabout Pedestrian Street, the intersection of Huquan Street and Xiongchu Avenue, as well as along the Second Ring Road areas such as Sakura Building.

The emotional response in the central area of Wuchang District is mainly neutral, concentrated in educational institutions, medical institutions and their surrounding residential areas. The areas with a relatively low emotional value index are mainly adjacent to Renmin Hospital, Shouyi Plaza and the areas along the Yangtze River, while the positive emotional areas are distributed in the core area of Yellow Crane Tower Park and the surrounding commercial clusters. Analysis shows that the spatial differentiation of the emotional value index is significantly correlated with the regional population density gradient, the accessibility of public service facilities, and the complexity of landscape features.

3.3.2. Correlation Analysis Between Emotional Value Index and Landscape Character Types

By associating the emotional value index with landscape character elements through ArcGIS Pro 3.0, an average emotion value map was produced (Figure 10). This was further developed into a comparative chart of average emotion values across different landscape types (Figure 11). These results show that there is a significant correspondence between the spatial distribution of emotional value and the types of landscape character. The resulting spatial patterns reveal that average emotion values display a clear consistency with the spatial distribution of the previously calculated emotional value index.

A comparison of average emotion values among different landscape character categories exhibits notable differences. The four categories with the lowest average emotion values, all around 0.1 (types 8, 15, 16, and 18), are typically characterized by gentle terrain gradients, dominance of non-built-up land covers including farmland and woodland, and peripheral locations either in non-urbanized areas or after 2015 urban expansions. These types lack functional diversity and temporal depth in development. Conversely, the four highest-scoring categories, beyond 0.4 (types 3, 6, 17, and 20), combine flat terrain with intensive urban construction. Their elevated emotion values correlate with multifunctional land uses integrating educational, cultural, and commercial spaces, as well as longer historical development within mature urban areas.

The analysis results indicate that the average emotional values in areas with different landscape character types exhibit significant spatial differentiation. From a spatial perspective, areas driven by urban construction tend to exhibit low emotional values, which are typically characterized by excessively high building density and a deficiency of green open spaces. Within nature-dominated areas, featuring water bodies and mountains with well-developed supporting facilities tend to have higher emotional values. In contrast, natural areas lacking necessary supporting infrastructure show lower emotional values. The emotional values in composite transition regions are at an intermediate level. These areas display moderate development intensity, retain more natural elements, and achieve a relative balance between multifunctional development and environmental conservation. This spatial pattern of emotional values directly reflects the combined influence of factors such as human activity intensity and functional diversity on public emotional perception.

4. Discussion

4.1. Comparison and Integration Between Landscape Characterization and Public Preferences

Rapid urbanization in the EWMR of Wuhan has resulted in significant changes in urban landscape heterogeneity. This transformation has led to a reduction in regional character and a decline in public landscape perception. LCA is commonly used to quantitatively describe various spatial scenarios and provides fundamental units for landscape protection and management [59,60,61]. By measuring public preferences for different landscape character units, a solid foundation can be established for the effective utilization of landscape resources.

In this study, a more holistic approach was adopted to explore the relationship between public preferences and landscape character. We employed semantic analysis of big data from Weibo to quantitatively assess public emotion within the study area [62,63]. The application of UGC data overcomes many limitations inherent in traditional survey-based approaches. The real-time, high-volume, and geographically extensive nature of social media data enables a more dynamic and spatially sensitive understanding of public landscape preferences [64]. The SoIVES model was used to spatially map the emotion value index, thereby facilitating an intuitive exploration of the spatial clustering of public landscape preferences [65,66]. Ultimately, the emotion value index was correlated with different types of landscape characters, enabling the creation of an average emotion value map. This map visually elucidates the relationship between landscape character and public preferences, thereby identifying spatial matching patterns between landscape character and public preferences.

The EWMR is characterized by a large population and an extensive area, encompassing numerous mountain ranges, water bodies, and other natural landscapes. In addition, the area includes various land uses such as commercial areas, schools, hospitals, tourist attractions, and residential neighborhoods, resulting in a diverse array of landscape character types. This study conducts a comparative analysis of the relationship between landscape character types and public preferences, exploring how different landscape character types influence the emotional values of the public.

(1) The research finds the areas with high emotional perception are mainly distributed in three typical areas: The first type is the urban commercial center area, such as type 3 dominated by commercial service functions and type 13 characterized by dense residential areas, represented by Guanggu Square and Chuhehan Street. These areas have achieved functional mixing and efficient space utilization through three-dimensional traffic organization and complete public service facilities. Secondly, there is the science, education and culture area. For example, type 6 with the distribution of science and education land, taking university clusters such as Wuhan University and Huazhong University of Science and Technology as examples, its success lies in the organic combination of campus functional layout and natural terrain, forming a richly layered landscape sequence. The third category is high-quality natural landscape areas, such as the regions along the East Lake Greenway. Although they are mainly characterized by flat water body landscapes, the public’s perception and experience have been significantly enhanced through the lakeside greenway system and the supporting cultural and tourism facilities.

(2) The analysis of the integration between landscape characterization and public preferences indicates that areas with high emotional values generally have the characteristic of the organic integration of natural and humanistic elements. In addition to the aforementioned types, type 7, which mainly consists of sports and cultural land, also exhibits similar character. This area not only retains some natural mountains but also integrates modern facilities such as sports venues and commercial complexes. In contrast, pure natural landscape areas such as type 16 dominated by forest land and type 18 dominated by farmland have relatively weak public emotion perception due to insufficient supporting facilities and lower accessibility. However, the overly urbanized commercial areas of type 1, mainly composed of densely built-up areas, are also difficult to obtain continuous high emotional feedback due to the lack of natural elements. Particularly worth noting is the Yellow Crane Tower Park area in type 4, which belongs to the park green space. Although it is located in the city center, it has formed a unique landscape pattern due to the good preservation of the natural landform and historical and cultural relics of Sheshan Mountain. The intensity of emotional perception is significantly higher than that of the surrounding commercial areas.

(3) Based on the theory of environmental psychology [67], the research finds that the synergistic effect of function, space, and vision is key to forming secondary differences within similar landscape features. (a) Functional mixing effect: type 6, characterized by educational and scientific land use, exhibits markedly higher emotional value than Type 1, which represents pure commercial areas. This phenomenon can be attributed to the integrated layering of natural features such as Luojia Mountain vegetation, educational infrastructure, and historically preserved architecture. Previous studies have exhibited that diverse landscapes contribute to enhanced esthetic visual experiences [68]. In other words, functional diversity stimulates positive emotions by enhancing environmental exploration. (b) Usage intensity effect: In the university area, emotional accumulation arises from frequent daily use by teachers and students, consistent with findings that repeated exposure enhances local attachment [69]. (c) Visual management effect: Yellow Crane Tower Park, classified as type 4, maintains scale harmony with Sheshan Mountain through building height control, whereas the high-density development of type 1 leads to visual overload, supporting the concept of a “visual complexity threshold” [70].

4.2. Insights on the Landscape Planning and Regulation of the Urban Landscapes in Wuhan

The Urban Landscapes in the EWMR, identified as a vital water and mountain resource, plays a significant role in urban ecology, culture, and the lives of residents during the urban development process. This study elucidates the intricate relationship between landscape character and public preferences, which serves as a foundational basis for sub-regional control planning within the urban landscapes. The research findings indicate that distinct landscape character types elicit varying public preferences, with a general tendency for higher emotional values in highly urbanized areas that have well-maintained accessibility compared to natural-dominated areas. For instance, although type 19, characterized as a cultural and sports green space mix, achieves moderate emotional values through optimized trail networks and viewpoint design, type 8, defined by extensive water bodies, exhibits lower values due to limited accessibility despite its high naturalness. In contrast, landscape type 1, marked by a high degree of urbanization and commercialization, coupled with a lack of natural scenery, receives a lower average public emotion. Consequently, these differences should be thoroughly considered in sub-regional control planning, necessitating the adoption of differentiated planning strategies for various regions.

Based on the research findings, it is recommended that differentiated planning and management strategies be adopted for areas with varying landscape character types.

(1) In urban construction-driven types, such as type 3 dominated by commercial services and type 13 characterized by multifunctional residential areas, it is crucial to balance development intensity with environmental quality. For instance, in the Optics Valley business area representing type 3, strategies could include increasing vertical green spaces along the corridor from Luxiang Square to Optics Valley Pedestrian Street, regulating the density of high-rise buildings, and preserving visual corridors to maintain the distinctive mountain silhouette of Yujia Mountain [71]. Drawing lessons from historic cities like Nanjing, urban authorities could establish view protection committees composed of developers and community representatives to mediate conflicts between skyline preservation and development needs [72].

(2) For natural terrain-dominated types, such as type 8 which features flat water landscapes and type 18 which is dominated by woodlands, ecological protection should be prioritized, with a particular focus on safeguarding aquatic environments and mountain vegetation [73]. Taking East Lake, which represents type 8 characteristics as an example, efforts should be directed toward preserving the natural features of its water bodies and gentle slopes. Building heights around the lake should be strictly regulated, and continuous viewing platforms should be planned along the Luojia Mountain–Moshan corridor, establishing an integrated vantage point for lake and mountain landscapes. The design workshop approach employed in the conservation of Hangzhou’s West Lake could serve as a model, facilitating collaboration between planners and local stakeholders to delineate mutually acceptable development boundaries for ecologically sensitive areas [74].

(3) In hybrid transition areas, such as type 4 which is primarily composed of park green spaces, planning and management should emphasize coordination between the built and natural environments. For example, in Huanghelou Park which represents type 4 characteristics, consideration is needed to harmonize the relationship between historical architecture and the surrounding environment, while protecting the natural landform of She Mountain. The ongoing Wuhan East–West Mountain Cultural Greenway Project can be leveraged to connect footpaths between Shouyi Square and the Yangtze River Bridge via a 24.5 km network (comprising an 11.5 km main cycling and walking route, and a 13 km walking branch), thereby enhancing the transition and integration of urban and natural areas. This greenway system not only improves landscape accessibility but also provides high-quality recreational spaces for the public. They also serve as educational corridors, raising public awareness of cultural heritage and ecological conservation through interpretive programs and interactive landscapes. The community participation mechanism implemented in Chengdu’s greenway development could be adopted, involving neighborhood councils in route selection and facility design to ensure that local needs are met [75].

In summary, the planning and management of all landscape area types should account for the public’s visual experience and activity needs. By optimizing view corridors and enhancing non-motorized transport systems, both the accessibility and esthetic appeal of landscapes can be improved, facilitating the organic integration of diverse landscape elements.

4.3. Limitations and Future Work

Nevertheless, this study is subject to several limitations. Firstly, the analysis of public preferences based on social media data is constrained by the absence of key user metadata such as age, gender, and educational background. This lack of information hinders the ability to segment and interpret preferences across different demographic groups [76]. Moreover, since Weibo users tend to be younger, certain demographics, notably older adults, are likely underrepresented. This issue may introduce bias in capturing the emotional perspectives of the population as a whole [77]. To address these issues, future research should consider stratifying public perception data with demographic information, possibly by employing multimodal approaches or by integrating data from multiple platforms.

Secondly, emotion data derived from NLP exhibits uneven spatial distribution, with higher concentrations in densely populated urban areas and relatively limited coverage in regions dominated by natural landscapes. This method is well-suited for analyzing urban landscapes; however, in broader natural environments, the quantity of available data tends to be relatively sparse, which may affect the representativeness and accuracy of emotional assessments in those regions. Particularly in areas with limited accessibility, including mountainous landscapes and core water bodies like Jiufeng Mountain and East Lake’s central area, the lack of check-in data prevents their evaluation through big data, resulting in artificially depressed emotion value scores at the edges of these regions on the emotional map.

Temporally, while our analysis reveals spatial patterns of emotional intensity, it fails to capture seasonal variations in public perception, despite existing studies demonstrating their significant impact on landscape preferences in temperate regions [78]. This imbalance restricts a comprehensive understanding of the emotional drivers within diverse environmental contexts [79]. Additionally, while emotion analysis provides an overarching view of public attitudes, it does not facilitate in-depth exploration of the underlying reasons for specific preferences, which is a capacity more typically associated with questionnaires or qualitative interviews. Future studies should adopt interdisciplinary approaches: (1) integrating eye-tracking and VR simulation to quantify visual–physiological responses [80,81]; (2) incorporating spatiotemporal dynamics to account for seasonal and holiday variations; and (3) combining sentiment analysis of UGC with field surveys to validate results and uncover complex motivational drivers [82,83]. Addressing these constraints will help build a more comprehensive and framework for understanding the relationship between urban landscapes and public preferences, thereby contributing to more equitable and effective landscape management strategies.

5. Conclusions

With the rapid advancement of urbanization, urban landscapes are facing challenges such as the blurring of landscape character and the reduction in green open spaces. Taking the EWMR of Wuhan as a case study, we developed a public perspective-based landscape character identification method by integrating objective landscape features with subjective public preferences. By combining emotional big data derived from social media with parametric identification of landscape features, this approach overcomes the traditional LCA regarding insufficient quantification of public perception. We identified 20 landscape character types, with emotional hotspots concentrated in specific types: urban construction-driven types featuring commercial and mixed-use functions, well-equipped natural terrain types, and hybrid transition types dominated by educational uses. These specific types exhibit notable functional diversity and accessibility. Landscape types with high emotional value consistently feature gentle terrain, multifunctional spaces, and comprehensive public services, whereas those with low value show either functional oversimplification or excessive urbanization. Variations within categories are further influenced by factors such as the degree of functional mixing, the frequency of spatial usage, and visual coherence and integrity. By enhancing the perceptual dimension of LCA, this method provides spatially explicit evidence to support urban landscape management. Future research could validate the generalizability of this approach through comparative studies across different cities, such as those within the urban agglomeration of the middle reaches of the Yangtze River. Additionally, integrating multiple data sources, such as street view imagery and eye tracking experiments, could further enhance its explanatory power.