Chengdu, with a total size of 14,335 square kilometers, is located in central Sichuan Province, in the western portion of the Sichuan Basin, between longitudes 102˚54'E and 104˚53'E and latitudes 30˚05'N and 31˚26'N. In the Yangtze River’s higher reaches, Chengdu serves as an important species gene pool and green ecological barrier. Its flat topographic conditions are suitable for large-scale urban development and expansion, and commercial building construction is an important part of urban planning, while the rich plant species and suitable climate in Sichuan Province provide favorable conditions for green roof diversification.

The scope of this research includes Chengdu’s five old urban areas: Jinjiang District, Qing yang District, Jinniu District, Wu hou District, and Cheng hua District (Figure 1). Before October 2005, the five urban areas (including the High-tech District) had achieved 300,000 m² of new green roofs (50,000 m² for each district), 120,000 m² of vertical greening (20,000 m² for each district), and 60,000 m² of green wall construction (10,000 m² for each district). In comparison to the new city.

The five urban areas are more densely populated and have more environmental problems. This is due to a lack of land resources in the oldareas, which rarely builds 12-story or 40-meter-lower buildings.

The survey selected 10 rooftops in the region that were built after 2000 (because

the technical guidelines for green roofs and vertical greening in Chengdu (for trial implementation) were promulgated in 2001), with a commercial area of 100,000 square meters or more, with external green roof space and a variety of community types: Sample 1 (Joy City), Sample 2 (Yintai Center in 99), Sample 3 (Fuli Shopping Plaza), Sample 4 (Jinniu CapitaLand), Sample 5 (Jinsha International Mall), Sample 6 (Longhu-Xichen Tianjie), Sample 7 (Perennial Qingyang Plaza), Sample 8 (Renhe New Town Mall), Sample 9 (Wanxiang City), and Sample 10 (Youfang Shopping Center) were used as the study subjects (Figure 1).

The current situation of the site and the characteristics of the plant community arrangement were combined based on a comprehensive survey of commercial rooftops in Chengdu, and from May to June 2022, a typical sampling method was used to set up 10 m by 10 m sample plots with an area of 100 m² [5] , eight for each roof, and the 80 sample plots were numbered in the order of LA1 to LA8, LB1 to LB8, and LJ1 to LJ8. The ten roofs were clustered into three clusters based on the ratio of green space to pavement, the complexity of the plant community structure, and the main functions of the roofs (Figure 2): within the boundaries of the “5” scale, the ten sample sites were divided into three clusters, with sample 1, 8, 3, and 5 clustered into cluster 1. It is observed that Cluster 1 is rich in plant layers, mostly “tree-irrigation-grass” complex structures, the sample square green area is more than 55%, and the main function of the roof is plant landscape; Cluster 2 includes Sample 4 and 7, and it is observed that Cluster 2 sports facilities occupy 70% or more of the area, with more than three

sports items and fewer plant species; Cluster 3 consists of samples 9, 10, 2, and 6, and it is observed that Cluster 3 is dominated by large paved areas, and the vertical structure of plants is primarily a “irrigation-grass” double-layer structure. These four roofs’ primary functions are recreation and dining. As a result, the ten roofs were divided into three categories based on the vertical structure of plants, the proportion of green space, and the main functions of the roofs: Sports and fitness roof, ornamental plants, and recreation, with the area and samples of each roof distributed as follows (Table 1).

Numerous studies have found no significant difference in results obtained by using photographs taken in the field as a medium for evaluating landscape quality and those obtained by being in the field [6] . Photographs of sample plant communities were taken with the same camera in clear, well-lit weather from 9:00 to 16:00 in May-June 2022. Daniel and Boster proposed that the number of photographs chosen for evaluation is determined by the diversity of the community landscape, and that one photograph of a single landscape is sufficient [7] . After removing photographs with obvious non-landscape factors, insufficient information, or obvious light spots or backlighting from each sample site, one photograph that could reflect the entire landscape of the plant community in the sample site was selected as the evaluation object, totaling 80 photographs.

Web-based questionnaires have become the most commonly used type of questionnaire due to the rapid development of the Internet in recent years, yielding more scientific and reliable results [8] . In this study, a web-based questionnaire was sent to the raters in the form of a beauty rating questionnaire via the

Note: The sample roof area is derived from studio drawing measurements.

Questionnaire Star platform. A 7-point rating scale from 1 to 7 (representing extremely dislike, very dislike, dislike, average, like, like very much, and like very much, respectively) was used as a measure, with higher values indicating more beautiful and popular scenery and lower values indicating less beautiful and popular scenery. Judges score the questionnaires, and finally, the questionnaires are collected and sorted.

To eliminate differences in evaluation criteria among different judging groups, the standardized Z value of the plant community was obtained by averaging the standardized values of all evaluators for each photograph using the standardized beauty formula, which is the standardized beauty value (SBE value) of the plant community, thus reflecting the aesthetic characteristics of the landscape of each plant community and the judges’ aesthetic tendency. The following is the calculation formula:

Z x y = ( R x y − R y ) / S x y

Z x = ∑ y Z x y / N y

where Z_xy is the yth rater’s evaluation standardized value of the xth photo, and R_xy is the yth rater’s rating value of the xth photo. Ry is the mean of all the rating values determined by the yth rater; S_xy is the standard deviation of all the rating values determined by the yth rater; Z_x is the landscape standardized score of the xth photo; and N_y is the total number of raters [9] .

Through a large number of online questionnaires, 397 rating forms were finally collected and collated; 7 invalid forms were excluded, and a total of 390 valid questionnaires were finally obtained. The distribution of the sample shows that the proportion of males is 51.1% and the proportion of females is 48.9%, which is basically in line with the results of the 7th census of Chengdu: the male population is 10,522,641, accounting for 50.26%; the female population is 10,415,116, accounting for 49.74%; in terms of age, 20 years old to 40 years old has the highest proportion of 79.50%; The proportion of people 40 years old and 60 years old is 15.80%; the proportion of people under 20 years old and over 60 years old is 3.20% and 1.50%, respectively; the proportion of landscape architecture, urban and rural planning, related design majors, and other majors is 46.80% and 53.20%, respectively.

The reliability and validity analysis using SPSS 24 reveals (Table 2, Table 3) that the standardized Cronbach’s alpha is 0.979, indicating that the questionnaire’s overall reliability is very high. According to the findings of the following exploratory factor analysis, the coefficient result of the KMO test is 0.938, and the coefficient of the KMO test takes values ranging from 0 to 1. The higher the value is, the more reliable the questionnaire is. This test’s significance is infinitely close to zero. Because the original hypothesis was rejected, the questionnaire has high validity.

According to the analysis of the plant communities based on the SBE values of all the judges, there were 43 communities with SBE values of greater than or equal to 0 (relatively good landscape quality), accounting for 53.75%, and 37 communities with SBE values of less than 0 (relatively poor landscape quality), accounting for 46.25%. The top five communities were as follows: LH3 > LA4 > LH2 > LH6 > LA1, primarily in Sample 1 and Sample 8 roofs; the bottom five communities were as follows: LG5 > LG4 > LG1 > LG8 > LD7, primarily in Sample 7. LH3, with an SBE value of 0.798, is the plant community with the highest beauty degree value. This is a mixed evergreen deciduous broad-leaved community of tree, shrub, and grass species (Figure 3), with the dominant species being purple-leaved plum, chickweed maple, and golden string willow. Throughout the plant space landscape, the canopy line and seasonal phase changes were implemented. The community with the lowest LD7 beauty degree value of −0.543 is an irrigation-grass broad-leaved evergreen community (Figure 4), with a single tree species composition, primarily moonflower and calyx distance flowers, and a small percentage of green space.

The average SBE value of the three types of roof communities was 0.635 for plant ornamental roofs, 0.080 for recreational roofs, and −0.555 for sports and fitness roofs (Table 4), with sports and fitness roofs having the lowest average SBE value of the communities. There were 16 sport and fitness roofs, with the top five plant communities being LD1 > LD6 > LD2 > LD5 > LD3. Sport and fitness roof communities all had SBE values less than 0. There was one tree-irrigation-grass type structure plant community, one tree-irrigation, two irrigation-grass, and one tree-grass among these five plant communities. The top five plant communities were LH3 > LA4 > LH2 > LH6 > LA1, which was the same as the overall SBE value ranking, demonstrating that plant ornamental roofs had the largest effect on the average SBE value. There were 21 plant communities with SBE values greater than 0, and all five of these plant communities had the “Trees-Shrubs-Groundcovers” structure. There were 32 recreational rooftops,

Note: SBE value > 0 represents relatively good landscape quality, SBE value < 0 represents relatively poor landscape quality; the higher the SBE value the better the landscape quality.

and the top five communities were: LI6 > LB5 > LJ5 > LJ2 > LJ1, with SBE values greater than 0. One of these five plant communities had a “Trees-Shrubs-Groundcovers” structure, whereas the other four had an “Shrubs-Groundcovers” structure.

In terms of different community types (Table 5), mixed coniferous communities had the highest mean SBE value (0.330), followed by bamboo mixed communities (0.094), and pure bamboo forests had the lowest mean SBE value (−0.716), and the mean SBE values of different community types were ranked as follows: mixed coniferous communities > bamboo mixed communities > evergreen deciduous broad-leaved mixed communities > pure shrub communities > deciduous broad-leaved communities > evergreen broad-leaved communities > pure bamboo forests.

Correlation analysis of the beauty SBE values with roof type (X1), community growth type (X2), community vertical structure (X3), number of community plants (X4), community mean diameter at breast height (X5), community mean height (X6), and community tree to shrub ratio (X7) was carried out in SPSS 24.0. It was found that the beauty was correlated with X1 (r = 0.448, P < 0.01), X2 (r = 0.320, P < 0.01), X3 (r = 0.295, P < 0.01), and X4 (r = 0.386, P < 0.01) (Table 6, Table 7); there was no correlation with the mean diameter at breast height (X5), mean height (X6), and tree to shrub ratio (X7).

Regression analysis using the beauty SBE value as the dependent variable, roof type (X1), community growth type (X2), community vertical structure (X3), number of community plants (X4), community mean diameter at breast height (X5), community mean height (X6) and community tree to shrub ratio (X7) as independent variables showed (Table 8) that: roof type (β = 0.383 > 0, P < 0.05), community vertical structure (β = 0.284 > 0, P < 0.05) and community mean height (β = 0.290 > 0, P < 0.05) could significantly and positively affect SBE. the following regression equation was derived between the variables: SBE = −1.114 + 0.221 × X1 + 0.090 × X3 + 0.112 × X6.

Dependent variable: SBE.