How Does Vegetation Landscape Structure of Urban Green Spaces Affect Cultural Ecosystem Services at Multiscale: Based on PLS-SEM Model

Abstract

Benefits of cultural ecosystem services (CESs) of urban green spaces (UGSs) for human health and general well-being have been widely recognized. Optimizing the landscape structure of green vegetation and improving CES are essential to reduce environmental inequality, and detecting the determinant landscape features that influence CES at multi-scale is the first step. Using partial least squares structural equation modeling (PLS-SEM), we evaluated multiscale effects of vegetation landscape structure of UGS on residents’ perceptions of CES in 40 residential communities in Zhengzhou city, China. According to our results, at the micro-scale level in a single residential community, public activity spaces within green spaces, particularly large size of open spaces, was the most critical factor affecting residents’ perceptions of recreational services, which provided a multifunctional landscape, with opportunities for multiple recreational leisure activities and entertainment. Then, the percentage of vegetation coverage in green spaces, and large vegetation patches that can significantly improve residents’ perceptions of CES and were identified in the natural landscape. At the ecosystem level and species community level, although vegetation structure contributed little to the level of CES, an open vegetation structure with a large area of grass cover was particularly beneficial to increase aesthetic services, and both richness of flowers and ornamental trees improved residents’ spiritual perception. However, our findings suggest that improving the management of green space vegetation is the most effective and direct way of improving CES and resolving environmental inequities between residential communities with different vegetation coverage and infrastructure, and we suggest that future research should explore residents’ subjective perceptions of both vegetation and landscape structure of UGS at larger spatial scale.

1. Introduction

There is universal agreement about the benefits of urban green spaces (UGSs) for residents’ health and general well-being [1,2]. Such spaces are known to contribute greatly to residents’ mental health [3] and reduce the occurrence probability of obesity [4], heart disease [5], cancer, type-2 diabetes [6], and other chronic diseases. We assume that the pathways through which UGS influence health behaviors and health outcomes are mainly via regulatory and cultural ecosystem services (CESs) of UGS (Figure 1) [7,8]. CES involves the intangible advantages humans receive from ecosystems, particularly the spiritual enrichment, cognitive growth, and the use of ecosystems for recreation and enjoyment as cultural services. However, the emphasis in discussions has been on recreational, aesthetic, and amenity aspects due to limited comprehension regarding the effect of cultural services on well-being [9]. CES can also be defined as the non-material benefits that people obtain from nature, including spiritual, aesthetical, educational, and recreational values [10,11] are closely linked to residents’ mental health [12]. Scholars have called for research on how to improve CES and their health-related effects [13].

The landscape structure of UGS refers to the spatial components of the spaces and their spatial layout [17]. Research on the impact of landscape structure in improving ecosystem functions and services has long been a focus in multidisciplinary fields, such as ecology [1], sociology [18], and urban planning [19]. The effects of landscape structure of UGS on supporting services and regulating services, such as maintaining biodiversity [20], regulating heat island effects [21], sequestrating carbon [17], and reducing soil erosion have been widely studied. However, research on the relationship between CES and landscape structure is relatively scarce.

At the regional scale, vegetation coverage of UGS is one of most important determinants of CES [22,23]. Inequity in UGS has prompted studies on the relationship between landscape structure and CES. It has been widely demonstrated that residents prefer spaces with green vegetation cover compared with bare land [24,25] and that the percentage of vegetation cover of UGS [26], Normalized Difference Vegetation Index (NDVI) [27], and street views [28] are usually significantly associated with CES and residents’ health [29,30]. Moreover, some studies have shown that residents tend to have higher perceptions of landscape aesthetics when the landscape includes large open areas and diverse green cover types [31], such as urban forests, urban parks, and wetland landscapes. For example, Lee found that residents preferred more complex green landscapes with connected tree patches with a high degree of complexity in terms of shape and variability [32]. Thus, residents’ preferences in relation to vegetation landscape structure directly affect their use of green space and the value of CES. Others also detected a significant positive correlation between the size of green space patches, patch connectivity, and residents’ health and well-being [25]. In addition, there is broad agreement that the greater the accessibility of UGS, the higher the value of CES [31,33], and the better the health of the residents [5]. However, the higher land prices and population density in urban versus rural areas often result in large environmental inequities in terms of the spatial distribution of UGS. Therefore, the impact of the spatial structure of the vegetation cover on CES versus that of the amount of vegetation cover needs to be further explored.

As shown by studies on CES, at the ecosystem scale, two variables, openness and naturalness, are important [34]. Most research has confirmed that access to natural landscapes contributes to recreational activities and improves the psychological well-being of urban residents [35,36]. It is generally assumed that more tree canopy, water bodies, and moderately dense vegetation increase the recreational value of UGS and interactions with nature [18,37]. In previous research, neighborhood well-being was positively related to a range of natural features, including species richness, bird abundance, and vegetation cover [38]. However, land cover in UGS is subject to human interference due to management by the residents themselves, the government, and planning bodies. Thus, the landscape characteristics of UGS are very different from those in a natural landscape. Numerous studies have found that urban residents tend to prefer more homogeneous and well-managed UGS [39,40]. Therefore, identifying the impact of naturalness and management of UGS on CES at ecosystem scale and balancing the trade-off between residents’ need for natural vegetation and landscape management are important for future landscape planning and design of UGS.

At the species community scale, there is a high degree of uncertainty regarding the relationship between vegetation structure and CES. In terms of vegetation land cover types, several studies have shown that residents prefer a mix of trees and shrubs rather than shrub cover only [41,42,43]. However, the impact of plant diversity on ecosystem aesthetic services is more controversial, with studies reporting positive [23], negative [44], and no associations [45]. Furthermore, at the species level, it seems that residents tend to prefer ornamental plants [46], such as flowers [47] and fruit trees [48]. Nevertheless, it is unclear how much the vegetation community structure contributes to CES compared with green space vegetation cover and the landscape structure of UGS.

Different from ecosystem regulating services and supporting services, CESs are closely related not only to landscape features but also to landscape functions, including the provision of space for activities (e.g., sports and fitness), children’s play areas, sanitation facilities, and cultural facilities. Infrastructure plays an important role in the quality of UGS and then the quality of CES. Recreational spaces in UGS (e.g., stadiums, playgrounds, plazas, and water features) and their supporting infrastructure provide opportunities for residents to participate in recreational activities [49,50]. Research in the field of sociology has widely demonstrated the relationship between public open spaces and human health [51,52]. Compared with the contribution of natural vegetation to CES, the contribution of infrastructure of UGS is less clear. When studying the relationship between landscape structure and CES, the infrastructure within the vegetation landscape of UGS is normally ignored. Therefore, we need to explore how we can optimize the infrastructure to maximize the benefits of CES and compensate for the lack of green space coverage. Resolving this issue can have practical significance for landscape planning and design of UGS.

There has been extensive research on the relationship between the vegetation landscape structure of UGS and CES, with such research focusing on different scales. Based on this research, the influence of the vegetation landscape on CES involves various indicators, such as vegetation coverage, NDVI, percentage of green views, green space size and type, accessibility, openness, and vegetation density at medium scales and community structure, plant diversity, and species characteristics at microscales. We suggest that research on the relationship between vegetation landscape and CES at multi-spatial scales can help to detect key landscape elements that affect CES.

Urban residents are most likely to utilize green spaces closest to where they live. Thus, green spaces close to urban residents’ homes can compensate for a lack of UGSs. Previous studies reported an association between the amount of green space within 500 m of a residential area and the health of the residents in the area [26,53]. In our previous research, we found that the less frequently residents visited surrounding parks, the higher their subjective perceptions of CES of neighborhood green spaces. Despite this, residential green spaces are often overlooked. In addition, the urban vegetation structure of residential green spaces in China is characterized by fragmented green coverage with complex patch shapes and high patch size variations, and are complex ecosystems that integrate social, ecological, and economic categories [54], which can better represent the fragmented vegetation landscape of the urban ecosystem. Therefore, to explore the response of CES to vegetation landscape, we need to include landscape structure in studies on the impact of UGS on CES. More attention should be paid to the study of CES of residential green spaces. We believe improvements in the quality and vegetation landscape structure of UGS can improve residents’ satisfaction of a quiet environment, stress relief, sense of belonging, and general well-being. Our study intends to answer the following questions: How does the amount of UGS in residential areas and its landscape layout affect CES at the microscale level? How is infrastructure in UGS of residential areas related to CES? Does vegetation structure at the ecosystem level and species community level have a significant impact on residents’ satisfaction with CES? What indicators contribute greatly to residents’ perception of CES in residential communities? Answering these questions will not only enrich our understanding of the relationship between urban landscape patterns and CES but also provide a basis for ecological planning and landscape design of vegetation cover oriented toward the enhancement of CES of UGS.

2. Methods

2.1. Study Area

Our study area is in Jinshui District, Zhengzhou City, Henan Province, China. Zhengzhou is located at 112°42′–114°14′ E, 34°16′–34°58′ N. It is the capital city of Henan Province, which is one of China’s national central cities and an important transportation hub. As of 2021, Zhengzhou had 12 administrative districts covering a total area of 7446 square km. The built-up area in the center of the city covers 651.4 km². The total population is 12.6 million, with an urban population of 9.879 million. The urbanization rate is 78.40%. The annual GDP is RMB 1200.4 billion, and the GDP per capita is RMB 109,652 (USD 15,336.47) [55]. The population density of Zhengzhou is 1665/km², ranking seventh among 33 major cities in China. Zhengzhou is located in a transition zone of the north temperate zone and has a subtropical climate and a semiarid and semihumid continental monsoon climate with four distinct seasons, long hours of sunshine, high temperatures, and low precipitation. The average annual temperature is 14.8 °C, and rainfall is 586.1 mm. Zhengzhou tends to be dry and windy in spring; hot and rainy in summer; cool in the fall, with long hours of sunshine; and dry in winter, with more wind than in the fall and little snow.

Zhengzhou is currently experiencing rapid economic development and urban expansion. At present, its green space resources are relatively poor and far from meeting the needs of the city’s residents. As of 2022, the percentage of green coverage in built-up areas in Zhengzhou was 41.5% (28,208 ha), the percentage of land for green space was 36.07% (24,517 ha), and the percentage of park green area was 6.8% (4628 ha). In terms of per capita green space, Zhengzhou ranks 26th out of 36 major cities in China (20.3 m²/person) and 17th of 36 major cities in terms of parkland space (8.36 m²/person). These data suggest that Zhengzhou’s green space resources, as well as its parkland resources, lag behind those of other cities in China. Ensuring the continued economic development of Zhengzhou, which is one of China’s central cities, while addressing the needs of its residents for green spaces is a challenge. Although large parkland areas have the highest level of ecosystem services, limited land resources make it difficult to develop large park. The green space closest to residents can help to compensate for the shortage of parkland. In this study, we selected 40 residential communities from the Jinshui District in the central urban area of Zhengzhou and analyzed the relationship between residents’ satisfaction with the CES of residential green space and vegetation landscape structure at multiscale levels and explored the key landscape features affecting CES of residential green spaces at different scales.

The population density of Zhengzhou is ranked high among the 30 major cities in China, but its per capita park and green space area is seriously inadequate, which is a common situation in most cities in China. Therefore, we take Zhengzhou as our study area, and attempt to explore the planning and design strategies for inadequate park and green space in residential areas through the relationship between green landscape and cultural ecosystem services in this city.

2.2. Evaluation of CES

In our previous study [56], we used a face-to-face questionnaire to determine residents’ satisfaction with aesthetic services, spiritual services, and recreational services of CES in Zhengzhou’s 40 residential communities. Most of our interviewees were residents who were participating in recreational activities within green space. As a supplement, a household survey was conducted to ensure an appropriate sample size. Finally, 4519 participants were surveyed, with a range of 93 to 135 individuals from each of the 40 communities. The detailed method as well as descriptive statistics can be found in our original study [56]. The residents’ satisfaction levels were scored from 1 to 10 points, and we defined the satisfaction level with CES < 4 as low, 4–7 as medium, and >7 as high. Finally, we characterized the level of CES based on three types of indicators, including recreational, aesthetic, spiritual services, which have been widely used to estimate CES. Here, we assessed spiritual services by the stress relieving, quietness, and sense of belonging.

2.3. Quantification of Landscape Structure of Green Vegetation at Microscales

In this study, UGS within the residential areas were extracted using high-resolution Google Earth imagery [57]. The extracted elements were divided into buildings, impervious layers, water, and green vegetation. The landscape pattern of UGS is usually quantified by landscape ecology indexes, such as the vegetation coverage area index, Maximum patch index, Patch density (PD), Patch shape index, and Patch aggregation degree [54,58]. FRAGSTAT 4.0 software was then used to quantify the vegetation landscape structure of residential green spaces. Metrics were calculated at the patch level for all vegetated areas in the residential landscape, and non-vegetated patches were treated as a background matrix and excluded from the analysis [17]. Data were collected and summarized (mean ± standard deviation) for each of the 40 residential areas. The metrics chosen are shown in

Table 1. Landscape metrics of green spaces used in the study at minor scale [17,32,54].

Metrics	Calculations and Implication
Percent of landscape (PLAND)	PLAND equals the sum of the areas (m2) of all patches of the corresponding patch type, divided by total landscape area (m2), multiplied by 100 (to convert to a percentage), which quantifies the proportional abundance of each patch type in the landscape (%).
Largest patch index (LPI)	Largest Patch Index (LPI) quantifies the percentage of total vegetation landscape area comprised by the largest patch, which is a simple measure of dominance (no unit).
Edge density (ED)	ED equals the sum of the lengths (m) of all edge segments in the landscape, divided by the total landscape area (m2), multiplied by 10,000 (to convert to hectares).
Patch density (PD)	The number of patches of the corresponding patch type divided by the total landscape area (km2).
Mean patch size (MPS)	MPS equals the sum of the areas of all patches of the corresponding patch type divided by the number of patches of the same type (m2).
Standard deviation of patch area (SDPA)	The standard deviation of the vegetation patch size.
Euclidian nearest neighbor distance (ENN_MN)	ENN equals the distance to the nearest neighboring patch of the same type, based on shortest edge-to-edge distance, Euclidean nearest-neighbor distance is perhaps the simplest measure of patch context and has been used extensively to quantify patch isolation (m).
Cohesion index (COHE)	COHESION equals 1 minus the sum of patch perimeter divided by the sum of patch perimeter times the square root of patch area for patches of the corresponding patch type, divided by 1 minus 1 over the square root of the total number of cells in the landscape, multiplied by 100 to convert to a percentage (no unit).
Shape index (SHAPE)	SHAPE equals patch perimeter (m) divided by the square root of patch area (m2), adjusted by a constant to adjust for a square standard, and is the simplest and perhaps most straightforward measure of shape complexity.

and the represented measures are believed to be of greatest interest to researchers and planners due to their common use and relative ease of interpretation. These metrics have also been generally used to explore the relationship between the vegetation structure of residential green spaces and ecosystem services. To determine the quantity of green vegetation, we used only one metric (the percentage of green landscape (PLAND)). The selected landscape metrics to measure the landscape characteristics considered included the largest patch index (LPI), Patch density (PD), edge density (ED), and mean patch area (MPA) for patch fragmentation. The Euclidian nearest neighbor (ENN) was selected for measuring aggregation, and the cohesion (COHE) index was used to measure patch permeability. Finally, the shape characteristics of the patches were measured by the shape index distribution (SHAPE) and fractal dimension (FRAC).

2.4. Quantification of Vegetation Structure at the Ecosystem and Community Scale

The vegetation survey was conducted using a two-stage systematic sampling procedure [47]. In the first stage, trees and shrub vegetation data were collected within a 40 m × 10 m area. Grass cover in six square subplots of 0.25 m² (50 cm × 50 cm) established within the plot was then calculated, and the height of herbaceous plants was measured. Unidentified herbaceous plants were collected for later identification in the six subplots. Finally, 298 subplots were established in the 40 selected residential communities. Considering the huge difference of area and vegetation structure between the residential areas, the number of plots in each residential area ranged from 7 to 12. The size of the largest residential community was 173,420 m² (17 ha), and the size of the smallest area was 51,877 m² (2.2 ha). The residential areas with the least vegetation richness were generally sites with low-rise residential buildings (5–7 floors), and the residential areas with the most vegetation richness were generally sites with high-rise residential buildings (20–30 floors). The vegetation transect survey consisted of a species inventory and recording of the tree height and diameter at breast height of all trees with diameter at breast height (dbh) over 5 cm and the basal diameter for all trees under 5 cm in dbh. Each stem was identified in the field to the species level, and the data were recorded. Both the height, numbers, and coverage ratio of shrubs and herbage in each plot were recorded. For species that were difficult to identify in the field, samples were collected, and photographs were taken for later identification. For each plot, we calculated three plant quantitative metrics to describe the vegetation structure, including vegetation types and vegetation density in ecosystem scale, and plant diversity in plant community scale (

Table 2. Indicators of vegetation composition.

Metrics	Description	Unit	Scale
Types of vegetation structure	Coverage of tree	%	Ecosystem
	Coverage of shrub	%	Ecosystem
	Coverage of herb	%	Ecosystem
Vegetation density	Abundance of tree	n/400 m2	Ecosystem
	Abundance of herbage	n/400 m2	Ecosystem
	Abundance of shrub	n/400 m2	Ecosystem
Richness	Richness of tree	n	Species community
	Richness of shrub	n	Species community
	Richness of herb	n	Species community
Shannon’s index	Diversity index based on species frequencies, including tree, shrub, and herb	/	Species community
Flowering species	Richness of flowering tree	%	Species community
Fruit tree	Richness of fruiting tree	%	Species community

). Here, plant diversity was quantified by the species richness and Shannon’s diversity index.

2.5. Quantification of Infrastructure

While conducting the sample survey of the vegetation community, we also investigated the infrastructure within the green spaces in the residential communities of Zhengzhou (

Table 3. Indicators of infrastructure at minor scale.

Types of Facilities	Description	Unit
Public activity spaces (PAS)	Richness of public activity spaces	n
	Total number of public activity spaces	n
	Total area of public activity spaces	m2
	Area of water body	m2
Rest facilities	Number of chairs for resting	n
Art installation	Number of sculptures	n
Cleaning facilities	Number of trashes	n
Sporting facilities	Fitness for number of equipment	n
Cultural and Promotional Facilities	Number of express cabinets, newspaper column, signboard, publicity column	n
Management	Level of management	Very neat and clean, relatively neat and clean, generally neat and clean, relatively messy and dirty, very messy and dirty.

). These included public activity spaces (PASs, e.g., fitness fields, plazas, children’s play spaces, and gardens), picnic areas (e.g., leisure mounts, stone tables, and benches), fitness facilities (e.g., gymnasiums and sports venues), ancillary service facilities (e.g., public notice boards, letter boxes, express cabinets, and road signs) and art facilities (e.g., sculpture, cultural walls, and art modeling). We recorded the type, number, and measured area of each PAS in the residential areas. In addition, we recorded the management level of the green space in the plots, which were divided into very neat and clean, relatively neat and clean, generally neat and clean, relatively messy and dirty, and very messy and dirty.

2.6. Partial Least Squares Structural Equation Modeling (PLS-SEM)

To identify the critical factors that influence CES at different spatial scales in urban settlements, we used structural equation modeling (SEM) to examine the relationship between the landscape structure of UGS and CES. SEM can contain both explicit variables that can be observed and latent variables that cannot be directly observed. Unlike traditional linear regression models, SEM allows researchers to test a batch of regression equations simultaneously. As these regression equations are very different from traditional regression in terms of model form, variable settings, and equation assumptions, they have a more diverse range of applications than traditional regression analysis. Thus, SEM is a method for building, estimating, and testing causal models. SEM can be used as an alternative to multiple regression, pass-through analysis, factor analysis, and analysis of covariance to analyze explicitly the effects of independent variables on dependent variables and the interrelationships between the respective variables.

Covariance-based structural equation modeling (CB-SEM) and partial least squares structural equation modeling (PLS-SEM) are the most frequently used structural equation models [59]. CB-SEM analyzes the covariance structure of variables and solves them by using the maximum likelihood estimation under a normal probability model. Compared with CB-SEM, PLS-SEM is based on PLS and analyzes the variance structure of the variables. Although earlier studies commonly used CB-SEM, many studies now use PLS-SEM, primarily because PLS-SEM has the following advantages [60,61]:

(1)

PLS-SEM is a regression-based modeling method that utilizes a principal component-centric approach to describe the direct dependencies between a range of variables [61]. Compared with CB-SEM, PLS-SEM is well-suited for most current predictive research or causal analysis of relationship studies as it facilitates the discovery of predictors that affect the results. For example, Li and Wen used PLS-SEM to study the drivers of national and regional CO₂ emissions in China [62]. Zhu et al. measured the efficiency and driving factors of urban land use in China based on PLS-SEM [63].

(2)

PLS-SEM does not require a normalized distribution of the data and is quite robust to skewness.

(3)

PLS-SEM does not assume a large sample size and achieves higher levels of statistical power. A minimum requirement of 30 samples is accept [64].

(4)

PLS-SEM is robust to quiets complex models containing hundreds of observed variables, and rarely faces convergence problems.

(5)

CB-SEM minimizes the divergence between estimated data and the sample covariance matrix by estimating the model parameters. The PLS-SEM approach estimates some of the model relationships in an iterative series of ordinary least squares regression. In this way, the explained variance of the endogenous latent variables can be maximized [60].

(6)

PLS-SEM does not use the model fit as an evaluation metric when obtaining the structural model solution. Instead, the investigators rely on a different set of indices, including structural reliability and validity, and in-sample and out-of-sample predictive indices. The final focuses on the soundness of the model structure and is therefore suitable for path analysis and explanatory research.

(7)

PLS-SEM utilizes a regression, component-centric approach to identify direct dependencies among a set of variables [61], and the results can also reflect the contribution of each indicator to the dependent variable, which is perfectly acceptable for most research predictions. In this study, we expect the results to identify vegetation landscape factors that influence CES at multi-scale.

In view of the above advantages, PLS-SEM (Smart-PLS 3.2.7.) is increasingly used in academic research. The analysis software commonly used in PLS-SEM includes LISREL, AMOS, EQS, Mplus, and Smart-PLS. In this study, we used Smart-PLS 3.2.7.

3. Results

3.1. Indicator Screening

Due to the large number of indicators involved, based on univariate linear regression analysis, we included only factors that had a significant impact (p < 0.05) on CES of green space in the SEM model. The final selected 14 indicators are shown in

Table 4. Relationship between residents’ satisfaction on different CES and landscape indicators with the univariate linear regression analysis.

Indicators	Recreational (Beta, p)	Aesthetic (Beta, p)	Spiritual Services
			Quietness (Beta, p)	Stress Relieving (Beta, p)	Sense of Belonging (Beta, p)
LPI	0.420 (0.012)	0.524 (0.001)	0.315 (0.066)	0.493 (0.003)	0.5 (0.002)
MPS	0.397 (0.018)	0.528 (0.001)	0.516 (0.002)	0.490 (0.003)	0.451 (0.006)
PLAND a	0.440 (0.005)	0.534 (p < 0.001)	0.569 (p < 0.001)	0.534 (p < 0.001)	0.525 (0.001)
Area of PAS	0.472 (0.002)	0.53 (p < 0.001)	0.474 (0.002)	0.492 (0.001)	0.495 (0.001)
Area of Water_PAS	0.343 (0.03)	0.409 (0.009)	0.359 (0.023)	0.368 (0.02)	0.328 (0.038)
Number of PAS a	0.421 (0.007)	0.507 (0.001)	0.438 (0.005)	0.458 (0.003)	0.468 (0.002)
Richness of PAS	0.374 (0.017)	0.371 (0.019)	0.331 (0.037)	0.305 (0.056)	0.318 (0.045)
Herb canopy	0.373 (0.018)	0.528 (p < 0.001)	0.396 (0.012)	0.369 (0.019)	0.27 (0.092)
Richness of flowering tree	0.305 (0.056)	0.385 (0.014)	0.330 (0.037)	0.338 (0.033)	0.3 (0.06)
Richness of fruiting tree	0.32 (0.044)	0.35 (0.027)	0.397 (0.011)	0.359 (0.023)	0.374 (0.018)
Shannon_herb	0.338 (0.033)	0.308 (0.053)	0.337 (0.034)	0.285 (0.075)	0.321 (0.044)
Shannon_tree	0.331 (0.037)	0.299 (0.061)	0.331 (0.037)	0.275 (0.086)	0.303 (0.057)
Management a	0.526 (p < 0.001)	0.610 (p < 0.001)	0.566 (p < 0.001)	0.503 (0.001)	0.528 (0.001)

^a: These data come from our previous study [56].

.

The indicators of landscape structure of residential green areas at minor scale had a significant impact on CES, as shown by the significant correlation of the LPI (p = 0.001) and MPS (p = 0.001) with residents’ aesthetic perceptions of green areas and the highly significant impact of the PLAND on all types of CES (p < 0.001). The effects of infrastructure within green spaces on CES were attributed mainly to the distribution of public activity space (PAS), and the area, quantity, and richness (p = 0.028) of PAS had significant effects on both of recreational and aesthetic services. Additionally, the area of waterbody space improved residents’ aesthetic perceptions (p = 0.021). The diversity of PAS was associated mainly with recreational (p = 0.017) and aesthetic services (p = 0.017).

At the ecosystem level, we found herbaceous cover had a significant effect on residents’ aesthetic perceptions of green spaces. Moreover, at the species community level, we found the Shannon’s diversity index for trees was significantly associated with residents’ satisfaction with the recreational services of CES (p = 0.037). In particular, the species richness of ornamental flowering and fruit trees significantly increased multiple categories of CES (p < 0.05). In addition, the level of management of green spaces had a highly significant effect on the different types of CES (p < 0.001). Therefore, to explore the response of CES to various indicators from different spatial scales of landscape structure of residential green spaces, we first excluded management factors in the PLS-SEM model.

3.2. Assessing the Outer Model

The classic two-step evaluation method was used in our study, which is recommended by the developers of PLS-SEM [61]. First, the measurement model was evaluated (outer model), which allowed us to evaluate whether the theoretical concepts or constructs were measured correctly through the observed variables. Second, the structural model was evaluated (inner model), from which the magnitude and significance of the causal relations between the different variables were assessed.

(1)

Estimate the loadings and their p-values: Using Bias-Corrected and Accelerated Bootstrap, which is the most stable confidence interval method, a complete bootstrapping procedure was first used to estimated indicator loadings in PLS-SEM and the p-values. Our results showed that all indicator loadings are higher 0.7 (

Table 5. Outer loadings.

	Aesthetic Services	Landscape Structure at Minor Scale	PAS	Recreational Services	Spiritual Services	Vegetation Structure	Vegetation Structure at Species Community Scale
			at Minor Scale			at Ecosystem Scale
Aesthetic	1
Area of PAS			0.837
Area of WPAS			0.691
Coverage of herb						1
LPI		0.944
MPS		0.912
Number of PAS			0.856
PLAND		0.925
Quietness					0.979
Recreational				1
Richness of PAS			0.724
Richness of _flower-bearing tree							0.945
Richness of _fruit-bearing tree							0.942
Sense of belonging					0.977
Stress relieving					0.979

), and statistically significant below 0.01.

(2)

Examine construct internal consistency reliability (

Table 6. Construct reliability and validity of outer model.

	Cronhach’s Alpha (CA)	rho_A	Composite Reliability (CR)	Average Variance Extracted (AVE)
Aesthetic services	1.000	1.000	1.000	1.000
Landscape structure at minor scale	0.918	0.923	0.948	0.859
PAS at minor scale	0.781	0.807	0.860	0.607
Recreational services	1.000	1.000	1.000	1.000
Spiritual services	0.978	0.979	0.985	0.957
Vegetation structure at ecosystem scale	1.000	1.000	1.000	1.000
Vegetation structure at species community scale	0.877	0.884	0.942	0.890

): Cronbach’s alpha (α) and composite Reliability (CR) were used to assess the internal consistency reliability. The cut-off value of 0.70 for both measures can be acceptable and is widely applied in PLS-SEM research, and we found both of them are higher than 0.7 in our model.

(3)

Obtain the Average Variance Extracted (AVE) (Table 6): AVE is a measure of the amount of variance captured by a construct in relation to the amount of variance resulting from measurement error. In general, values of 0.50 or higher indicate the construct’s convergent validity. In our study, we excluded indicators with AVE less than 0.5, including Shannon_herbage and Shannon_tree. Other variables in our study that AVE exceeded 0.5 were then included in the PLS model (Table 5), which established the convergent validity of the reflective.

(4)

Check the discriminant validity through HTMT (

Table 7. Discriminant validity of outer model.

	CES	Landscape Structure at Minor Scale	PAS at Minor Scale	Recreational	Spiritual	Vegetation Structure at Ecosystem Scale	Vegetation Structure at Species Community Scale
Aesthetic	1
Landscape structure at minor scale	0.543	0.927
PAS at minor scale	0.590	0.456	0.779
Recreational	0.927		0.521	1.000
Spiritual	0.944		0.543	0.943	0.978
Vegetation structure at ecosystem scale	0.384	0.432	0.371	0.373	0.354	1.000
Vegetation structure at species community scale	0.379	0.149	0.208	0.331	0.380	0.262	0.943

): The discriminant validity was checked through heterotrait-monotrait ratio (HTMT) of the correlations. The measure indicates the degree to which a construct is conceptually distinct from other constructs in the study. There is a conservative cut-off value of 0.85 for HTMT, and for our study the value of HTMT was below 0.85, showing discriminant validity between the constructs.

3.3. Assessing the Inner Model

Examine indicator multicollinearity. Unlike reflective constructs, formative constructs ask for items that are not highly correlated, and therefore they cannot be interchanged. The suggested measure of collinearity is the variance inflation factor (VIF), and VIFs between landscape structures at minor scale, PAS, vegetation at ecosystem, and plant community level were below 3, which imply the correlations between constructs can be accept in the PLS-SEM.

3.4. Model Results

We constructed Model A and Model B, with Model B incorporating management factors. The results showed that Model A had R² value of 0.377 for recreational services, and 0.466 for spiritual services, suggesting that indicators selected for this study had a moderate effect in explaining these services of the green spaces in residential areas. Model A had an R² of 0.549 for aesthetic services, and results showed all indicators selected for this study had a strong ability to explain aesthetic services. Hair et al. noted that R² values between 0 and 0.10, 0.11 and 0.30, 0.30 and 0.50, and >0.50 are indicative of weak, modest, moderate, and strong explanatory power, respectively [61].

The path coefficient characterizes the relationship between two latent variables. The path coefficient was at least 0.1, indicating that there was a significant correlation between the two latent variables, and path coefficients in the structural model ranging from 0 to 0.10, 0.11 to 0.30, 0.30 to 0.5, and > 0.50 are indicative of weak, modest, moderate, and strong effect sizes.

In our study, we found that Model A demonstrated that only the PAS of residential green spaces at a minor scale had moderate positive effects on the level of recreational services. The path coefficients between PAS and recreational services were 0.348 (P=0.010) Meanwhile, landscape structure at minor scales and vegetation structure at the ecosystem and species community level showed no significant effect on recreational services, with path coefficients of 0.206 (p = 0.179), 0.102 (p = 0.445), and 0.204 (p = 0.167), respectively. PAS, vegetation structures at the minor scale, and vegetation structure at the ecosystem scale all showed significant effects on aesthetic services, with path coefficients of 0.343 (p = 0.003), 0.253 (p = 0.036), and 0.233 (p = 0.036), respectively. In addition, we found both PAS and landscape structure also had moderate effects on the level of spiritual services, with a path coefficient of 0.317 (p = 0.015), 0.353 (p = 0.021).

When the management factor was then included in the PLS-SEM (Figure 2, Model B), the results of the final PLS-SEM model showed that the R² of the model B fit for recreational services, aesthetic services and spiritual services were equal to 0.511, 0.677, and 0.604, respectively, suggesting that the management level particularly had a stronger power of explaining the CES of the green spaces than other indicators.

The results indicate that management is the most crucial factor influencing residents’ perceptions of CES of residential green spaces, with a path coefficient of 0.509 (p = 0.004) between management and recreational services, a path coefficient of 0.497 (p < 0.001) between management and aesthetic services, and a path coefficient of 0.516 (p < 0.001) between management and spiritual services. However, none of the other indicators had a significant effect on residents’ perceptions of recreational services. Additionally, we found aesthetic services were only significantly associated with PAS, with a path coefficient of 0.266 (p = 0.017). Except for PAS, we found that all other indicators were significantly correlated with residents’ perception of spiritual services.

3.5. Model Fit

Due to the fact that PLS-SEM is a predictive method similar to multiple regression analysis, the model structure presents uncertainties and various possibilities. Therefore, when conducting explanatory or path-oriented research using PLS-SEM models, we should pay more attention to the predictive structure rather than the goodness of fit, and we should not solely reject or accept PLS-SEM based on model fitness indices [61,65]. In this study, we construct relationships between green vegetation landscape indicators at multiple scales and CES, with the main objective of detecting the impacts of different indicators on CES. Through stability tests of both external and internal models, we find that our structural equation model is relatively stable. Nevertheless, in recent years, some scholars have conducted performance evaluations to reject mismatched models, such as the standardized root mean square residual (SRMR), the squared Euclidean distance (d_ULS), and the geodesic distance (d_G), which are normally used to evaluate the model fit of PLS-SEM [61,62,65,66].

SRMR is an index that measures the average of standardized residuals between observed and hypothesized covariance matrices, which is considered a goodness of fit measure for PLS-SEM in the detection of model error. According to Hu and Bentler, a fitting threshold of 0.10 or, more conservatively, 0.08 is considered good, with a lower SRMR being a better fit [67]. In our study, the SRMR of model A and model B was 0.070 each, showing that the final model specifications were within thresholds considered satisfactory. Moreover, Model A’s d_ULS and d_G were 0.596 and 0.592 respectively, while Model B’s d_ULS and d_G were 0.665 and 0.688 respectively, and a fitting threshold of 0.95 each for d_ULS and d_G is considered good, with a lower value being a better fit [66].

4. Discussion

By assessing residents’ perception of CES in 40 residential communities in Zhengzhou, this study explored the impact of vegetation landscape structure on CES of residential green spaces at different landscape scales: vegetation landscape structure and infrastructure at minor-scale within the single residential communities, ecosystem structure at ecosystem scale, and plant diversity at species community scale. The main finding of our study is the identification of the key factors affecting residents’ perceptions of CES of UGS. UGS in residential communities, which is the closest type of green space available to city dwellers, tends to be fragmented. Knowledge of the effects of vegetation landscape structure at multi-scale is important for improving landscape planning, design, and management of UGS.

4.1. The Critical Role of Large and Diverse PAS to Provide Various Types of Leisure and Entertainment Services

PAS within residential green spaces contribute most to improve CES. Our study found that infrastructure made a greater contribution to residents’ perceptions of CES than vegetation coverage and vegetation structure’s effects on CES. It seems that quality of CES in residential community is more important than the quantity of UGS to improve the level of CES.

Amounts of studies have shown that improving the quality of green spaces [68,69] can effectively compensate for the lack of green space quantity. Quantitative indicators of UGS usually include the total area of green space coverage, the number, and the total area of the parks. Qualitative indicators of UGS include green space maintenance, the presence of facilities for relaxation, and the quality of the planted landscape [70,71]. The results of this study suggest that PAS are the most critical elements of vegetation landscape quality for improving CES in residential green spaces. We speculate that the main reason for this finding is that PAS inside green spaces provide residents with various opportunities for leisure and entertainment [72,73], and respectively contributes much to the total level of CES. Nevertheless, we found no significant correlation between facilities (e.g., sports, culture, and recreation) and CES. This finding differs from the literature. In a study by Kulczyk, in small-scale natural landscapes, facilities (equipment rental, information centers, trails, mooring places, and picnic areas) were more strongly related to recreational services than natural landscape to provide various types of recreational services [49]. We assume that the main reason for this finding was that those living in natural landscapes have different needs from those living in artificial landscapes. In a natural landscape, there is a sufficient supply of vacant space. Nevertheless, in a residential landscape with a more limited land area, excluding building sites, road sites, and green vegetation sites, the demand for natural environment and space for recreational activities of residents comes first, and then facilities come second. Moreover, these are very different activities with different needs for infrastructures, and this can also explain the correlation between CES and facilities.

Consequently, in landscape design of UGS, we conclude that it is important to combine the needs of residents for different types of green space and arrange infrastructure to maximize the environmental quality of UGS. Moreover, a growing body of research suggests that improving the quality of green space has a more positive effect on improving residents’ health than increasing the quantity of green space [3,5,74,75]. The relationship between PAS and CES in our study showed that creating more and larger PAS can play an important role in improving the quality of UGS of residential communities.

Residents’ perceptions of recreational services of residential green spaces mainly stem from the various types of leisure and entertainment services provided by large PAS. The area (T = 14.813) and quantity (T = 12.601) of public activity space contributed more to the level of CES than the diversity. Recreational services are generally considered to be the most important type of CES [76,77], and PAS are the main venues for residents’ leisure and entertainment activities. We previously investigated the relationship between the size of PAS and the number of people participating in recreational activities in 26 of 40 residential communities. We found that the larger the size of the public activity space, the higher the number of people engaged in recreational activities within the space due to the multifunctional nature of the space (Mao, 2021 #1103). In this study, the PLAND of residential community “a” was 24.8% (Figure 3), but the score for overall satisfaction with recreational services was 8.18 (average level: 6.6). The relatively high score can be attributed to the presence of a large size of plazas for residents’ activities. In contrast, in residential community “b”, which had 36.7% of vegetation coverage but no functional larger public spaces and a small area of PAS allocated to parking, the overall satisfaction score for recreational services was only 4.6. Therefore, we believe that increasing the area of PAS remains the most effective landscape design approach for improving residents’ satisfaction of a quiet environment, stress relief, and sense of belonging, particularly in high-density residential areas in developing countries.

Creating rich PAS can provide a more diverse activity space for different groups of people. In residential community “c”, which had 41% vegetation coverage, pavilions, stadiums, gymnasiums, children’s playgrounds, miniature squares, cultural squares, and water bodies were set up internally (Figure 4c). The leisure and entertainment services, aesthetics services, and spiritual services in c were 7.25, 7.86, and 7.53, respectively. In residential community “d”, which had 39% vegetation coverage, there was almost no PAS (Figure 4d), and its leisure and entertainment services, aesthetics services, and spiritual services were 5.99, 6.30, and 5.89, respectively. Furthermore, designing more PAS for residents to participate in fitness activities is an effective channel to enhance recreational services of CES. We discovered that there is an obvious connection between the residents’ satisfaction with CES and their involvement in fitness activities (R² = 0.1736, p < 0.01). Additionally (Figure 5), we found these recreational activities have a direct correlation with the richness of PAS (R² = 0.1295, p < 0.01). Despite this, the fitness infrastructure in 40 residential communities were mainly designed for the elderly, and there is a severe lack of public space suitable for sports and fitness activities for young people. We speculate that this may also be one of the reasons why the satisfaction of the 21–29 age group with green spaces and CES is lowest [56], and we found a significant negative correlation between the proportion of residents aged 21–29 and the proportion participating in fitness activities (Figure 5, R² = 0.208, p < 0.01). Young people are more likely to prefer sports, particularly in sports fields, we suggest that in the future, more sports fields and sports facilities suitable for teenagers should be further planned in the green landscape of residential communities.

There is a synergistic relationship between size of vegetation patch and PAS. In a large-sized PAS, there is a higher percentage of an impermeable cover layer for residents’ recreational activities. As shown in the previous section (Table 4), the positive significant correlation was detected between residents’ perception of CES and size of vegetation patches. A key issue for landscape planning and design is how to improve residents’ leisure and entertainment services, while maintaining their aesthetic perceptions of the vegetation component of UGS. Numerous ecological studies have confirmed that the larger the vegetation coverage of UGS, the larger the average patch area, and the higher the ecosystem services, such as cooling and humidification [21], biodiversity protection [78], and soil and water conservation [79] in UGS. Therefore, there is a trade-off between various CES of residential green spaces, which are dependent on PAS, and supporting and regulating services, which are mainly dependent on natural vegetation. This trade-off is an urgent problem that needs to be addressed in studies of the ecological effects of landscape patterns of UGS. Nevertheless, in our study, we found that large-sized vegetation patches in the 40 residential green spaces were commonly accompanied by large PAS, indicating that there was no trade-off relationship between the vegetation coverage and PAS. We also found that residential areas with a high level of CES usually had large areas of uninterrupted greenery and vegetation in front of and behind buildings and a variety of small-sized PAS inside (Figure 6). Larger PAS were located in the center of the residential community to provide a variety of leisure and recreational opportunities for the residents. The best way to balance the tradeoff between the different types of ecosystem services through landscape optimization has long been a focus in research on ecosystem services [80,81]. This study demonstrates that both supporting services and CES can be synergistic in residential green spaces and our findings can serve as a basis for landscape design aimed at improving multiple ecosystem services of UGS.

4.2. Increasing Vegetation Coverage and Creating Large Size Vegetation Patches Could Improve Residents’ Perception to Nature of Green Spaces

In 40 residential communities, we found the vegetation cover of residential green spaces is obviously correlated with the residents’ perceptions of CES. Based on our results, within a residential community, which covered a radius of about 200–300 m, when management was not considered, the effects of vegetation landscape structure at minor scale on CES was second to PAS.

Currently, residents’ perception and satisfaction of PLAND of residential communities in Zhengzhou is much stronger than PAS, and the amount of green space is not a determining factor in residents’ perception of green space. In our previous study, we found that vegetation cover was the most important indicator of residents’ needs for UGS in a residential community, then the infrastructure and management. However, residents were more satisfied with the green space coverage of their neighborhoods than with PAS (Mao, 2020 #1007). Therefore, we suppose that the critical determinants of residents’ use of neighborhoods green spaces and their perceptions of CES are PAS. Nevertheless, we noticed a significant correlation between different landscape structural features on improving residents’ perceptions of CES.

PLAND was extremely related to residents’ aesthetic and spiritual perceptions of green spaces. This finding is consistent with that of previous studies that suggested that the amount of vegetation greenery was a key determinant of residents’ perceptions of CES, regardless of whether the vegetation structure was complex or simple, natural or artificial [5,47,68]. We assume that the main reason for this finding is that the natural vegetation of UGS is the source of CES and the only access to nature for urban residents. Numerous studies have reported the positive effect of vegetation coverage percentage or greenery in residential areas on promoting residents’ physical and mental health [82,83]. Therefore, we suggest that increasing the amount of green space in residential areas is an effective approach for improving residents’ satisfaction of a quiet environment, stress relief, and sense of belonging.

First, in terms of aesthetics, urban residents normally tend to place more value on larger than smaller patches of vegetation, with larger patches motivating them to participate in multiple entertainment activities. As shown in previous studies, urban residents perceived large patches of uninterrupted arboreal vegetation as more beautiful than fragmented patches [32], and they tended to prefer large patches of vegetation in UGS [84,85] leading to more recreational activities taking place in such space [86,87]. For example, the two residential communities investigated in our study, E and F, had the same amount of green vegetation cover (36%) (Figure 6). “e” had larger gardens, and “f” had no functional vegetation spaces. The scores for aesthetic services were 6.64 (e) and 5.01 (f), respectively, recreational services were 5.01 (e) and 4.6 (f), respectively, quietness were 6.25 (e) and 4.81 (f), respectively, stress relieving were 6.51 (e) and 5.59 (f), respectively, and sense of belonging was 6.21 (e) and 5.56 (f), respectively.

Second, larger areas of green vegetated landscape tend to have an abundance of activity space within them. For example, plot G has a PLAND of 40.92% (Figure 6g), with large areas of greenery, PAS, and infrastructure between the groups of buildings in the residential community, and a recreational services score of 7.32. Plot H has a PLAND of 41.98%, with vegetation scattered in front and behind the buildings and trees planted along the roads in the residential community, in addition to a high degree of fragmentation in the green landscape (Figure 6h). The recreational services score of plot H was 6.31.

Third, people living in residential areas with large areas of vegetation patches may have a stronger sense of identification triggered by the iconicity of the neighborhood, such as an iconic feature of a large garden in the residential community. Moreover, we found a significant correlation between LPI and residents ‘perception of sense of belonging’. Therefore, the amount of green vegetation cover alone is not the only variable that is important in CES. Including large areas of green space is necessary to enhance CES.

Previous research has also shown the presence of large-sized patches of green areas have a stronger effect on residents’ mental well-being than the total green coverage area overall [52]. Other research demonstrated that residents of urban environments with fragmented landscapes had a lower level of mental health than other residents lived in continuous landscape [88]. Increasingly, research on the relationship between UGS and the well-being of residents calls for more attention to be paid to the quality of the landscape or the structure of the landscape [30,69,75,89]. Therefore, we suggest future research needs to explore the relationships between the vegetation landscape structure of UGS, CES, and residents’ health.

In our study, other landscape indexes of residential green spaces, including SHAPE, ED, PD, COHE, and ENN_MN, contributed less to the level of green space CES. A previous study found that the morphology and diversity of vegetation patch in UGS were helpful in increasing residents’ satisfaction with the neighborhood landscape in the city of College Station, Texas [32]. In our study, we found no obvious correlation between UGS structure (patch shape and patch aggregation) and CES. The discord between the findings of our study and those of the earlier study may be because it measured only the impact of tree patch size and morphology on residents’ satisfaction. In contrast, we measured the effect of all green vegetation patches on CES. In addition, compared with the urban landscape in Lee’s study, the tree patches within the residential green spaces in the city of Zhengzhou were more fragmented, with almost no large areas of tree cover, while the residential areas in the study by Lee had large areas of trees planted in front and behind the houses, and the residents normally favored high tree canopy coverage. Additionally, other researchers have pointed out that SHAPE, ED, PD, COHE, and ENN_MN, have no particular ecological importance in terms of CES, but only a mathematical statistic [90,91].

4.3. The Impact of Large Area of Grass on Residents’ Aestheric Perception at Ecosystem Level

The vegetation structure of residential green spaces contributed little to residents’ satisfaction with CES. Previous studies have shown vegetation structure affects the frequency of visits to green space, the residence time [92,93], and the health of residents [94]. Nevertheless, the results from PLS-SEM in our study showed that compared with the effects of vegetation coverage and infrastructure, vegetation structure only had a modest effect on aesthetic services. We suspect the main reason for this was that the vegetation structure of the green space in our study was based on a quadratic survey, which is an objective description of the vegetation community of green spaces. In contrast, other studies on the relationship between vegetation structure and CES focused mainly on the residents’ subjective evaluations or perceptions of the plant community [47,95], and in these studies the residents’ evaluations of vegetation structure were more targeted. Additionally, the level of CES we obtained in this study stemmed from the residents’ perceptions of the overall landscape, while residents’ attitude on vegetation community in other studies was based on their direct perception of green spaces. Therefore, does the scale or level have to be considered when studying the relationship between CES and vegetation structure in the future? In addition, residents’ subjective perceptions of vegetation structure at the micro-scale varied greatly between different socio-economic groups of residents. Previous studies found that older people and females prefer flowering plants [47,96]. Other studies found differences in plant preferences and choices of those with different religious beliefs [97]. In our study, we did not distinguish differences in residents’ subjective perceptions of vegetation structure according to specific characteristics (e.g., sex or religion). Therefore, the exploration of the relationship between vegetation structure and CES at the ecosystem scale should be combined with the subjective perceptions of different socio-economic groups on vegetation landscapes.

An open vegetation structure with a large area of grass cover was helpful to improve the satisfaction of residents’ aesthetic perception of green spaces. We found there was a significant correlation between the amount of grass area and aesthetic services scores (Figure 2 and Figure 7), which means higher residents’ perception of aesthetic with larger lawn areas [98]. Residential green spaces with large areas of grass widely distributed between buildings and around public spaces are more open and accessible than vegetation with dense shrub coverage. Large areas of grass always companied various PAS, and which allow for a range of activities, result in increased use of the space and improved residents’ perceptions of CES. Previous studies have shown that residents prefer green space types that are open and accessible [45,68]. The old residential communities included in our study were built before 2010 and have low real estate prices. Rather than grass, most of these communities had trees and shrubs planted in front of and behind the buildings and on both sides of the roads, with the main function being shade rather than recreational use by the residents. The high-water consumption and frequency of maintenance of managed grass areas, which need to be repaired or replanted every year, mean that most residential areas in Zhengzhou tend not to have large areas of lawn. Instead, low-cost, low-maintenance, easy-to-manage shrubs are more popular with property developers. In terms of open green space, the requirements of management and residents need to be reconciled. Future landscape design should reconcile the trade-offs between the current open, lawn-dominated vegetated spaces and the financial investment in management. For example, we can design open green spaces with big tree coverage rather than higher grass cover.

4.4. Positive Role of Fruit-Bearing and Flower-Bearing Trees in Improving Residents’ Spiritual Perception of Residential Green Spaces

Increasing woody plant diversity had a positive effect on improving residents’ perceptions of spiritual services. Although the contribution of the vegetation structure of residential green spaces to CES in residential areas was small, we found a significant correlation between tree diversity and quietness (Table 4). We suspect that the main reasons for this are as follows: First, more diverse tree species may have a greater sensory impact. Previous studies have shown that urban dwellers tend to favor green spaces featuring trees due to the resulting landscape’s naturalistic appearance [37,42,43], and we found a significant correlation between Shannon_tree and residents’ perception of quietness. Second, the hot summers in Zhengzhou make shade a key consideration in vegetation landscape planning. Good shade from trees can significantly influence residents’ use and subjective evaluation of green spaces. Third, the higher the species richness of the tree layer, which contains more flowers and fruit-bearing plants, the stronger residents’ spiritual perceptions of the area.

In our study, compared with other indicators, both of Model A and Model B showed species community of the green space has almost no impact on CES. Despite this, we cannot underestimate the role that ornamental plants and fruit-bearing plants in providing aesthetic and spiritual services in the green space (Table 4). Previous amounts of studies noted that fruit-bearing trees not only have ornamental value but also practical value, especially for children, who enjoy picking the fruit, a result that has been confirmed in numerous studies [46,48,99,100]. In a study done in the north of England, fruit-bearing plants in residential areas had a positive effect on residents’ ability to regulate stress [26]. They also led to more positive emotions, increased motivation, and a sense of belonging. In our study, significant correlations were found between the richness of fruit-bearing trees and residents’ satisfaction of stress relief, aesthetic, and the sense of belonging of residential green spaces. Common fruit-bearing trees planted in the residential areas in Zhengzhou in this study included Ginkgo biloba L., Eriobotrya japonica (Thunb.) Lindl, Carica papaya L., Punica granatum L., Diospyros kaki Thunb., Malus pumila Mill., Prunus armeniaca L., Ficus carica Linn., Ziziphus jujuba Mill., and Pyrus spp.

Numerous previous studies have noted that residents favor a plant community with a variety of colorful flowering plants in UGS [101,102,103], this can explain an obvious relationship between higher richness of flower-bearing trees and higher resident satisfaction scores for aesthetic and spiritual services. Flower-bearing trees widely distributed in Zhengzhou’s residential communities include Malus spectabilis, Prunus subg., Cerasus sp., Yulania denudata (Desr.) D.L. Fu, Prunus persica ‘Atropurpurea’, Osmanthus sp., Punica granatum L., Hibiscus syriacus Linn., Syringa oblata Lindl., and Cercis chinensis Bunge. Therefore, to improve residents’ perceptions of CES of green spaces, we suggest that when planning and designing green space areas, diverse vegetation species, especially tree species with flowers and fruits, should be considered. The specific types of flowering or fruiting plants preferred by residents need to be studied, in conjunction with the subjective choices of residents, especially among different groups of people (e.g., age, sex, income level, race, and ethnicity).

4.5. Improving the Management of Green Space to Address Environmental Inequity

The level (high vs. low) of management of residential green spaces in Zhengzhou is currently the most important factor affecting CES. In our study, a high level of management of residential green spaces was characterized by well-established infrastructure, clean, and attractive vegetation, with a homogenous and well-kept plant community. Oppositely, a low level of management of residential green spaces was characterized by dirty, disorderly and an absence of infrastructure. We found that when management factors were included in the PLS-SEM, of four factors, management was the most critical factor affecting CES (Figure 2, Model B). In China, the companies that construct residential developments are responsible for landscape design, including vegetation structure and diversity. Upon completion, dedicated management companies oversee these developments. The residents themselves won’t involve in the design and planning stages. In addition, the residents of poorly managed areas tended to utilize the green space as they saw fit. The residents’ interference in the green space resulted in a mix of plant species and a lack of community structure. It also resulted in poor scores for the areas in terms of aesthetic perceptions. In Sweden, Qiu (2013) also noted that the presence of human intervention in residential parks led to residents’ negative preference for UGS [45].

Other European studies also pointed out that residents favored managed/designed over wild/naturalistic landscapes [44,104]. Furthermore, it has been demonstrated that unlike large parks, in residential green spaces and small parks, a high level of landscape management, including plant management, is the most important factor influencing CES [87]. In our previous study, only 5% of residents, on average, were satisfied with the management of green spaces in their residential neighborhoods, while it ranked third for residents’ concern about residential green spaces and was second only to vegetation coverage ratio and public activity space [56]. It is a pity that the current attitude of residents toward the management of residential communities is so poor. Given the lack of green space in residential communities in China today, we propose that enhancing day-to-day management (clean and neat) is currently the most effective way of improving CES of residential green space.

We suggest that the creation of older residential communities in China’s urban settlements should pay more attention to improving the quality and management of green space. Current planning of green space in China’s residential areas is gradually shifting from the construction and development of large-scale new residential areas to the improvement of the quality of the existing housing stock [105]. The main characteristics of these older residential neighborhoods are low-rise housing with a low percentage of green cover, few parking facilities, and lack of infrastructure. Renovation projects focusing on these residential communities mainly include infrastructure improvements. These include improvements to roads, facades, and buildings, adding better insulation and replacing existing coal, electricity, and gas heating systems with energy-efficient alternatives. We believe that infrastructure maintenance and management are the most effective ways of maintaining the vitality of residential communities and enhancing residents’ sense of well-being, which is currently the main problem with the green space landscape in residential areas in China.

In developing countries, low levels of CES and resident well-being are mainly attributed to inadequate green space coverage in high-density urban areas. We recommend that this deficiency be addressed by good management and the creation of multifunctional, multitype PAS, particularly as we believe this approach is suitable for the renovation of old and rundown residential areas currently underway in China.

4.6. Limitations

Our study has several major limitations. First, given the difficulty of obtaining face-to-face questionnaire data in residential communities, our structural equation model included sample data from only 40 residential communities. Although this complied with the minimum sample size of 30 required for SEM, future studies should increase the sample size to improve the accuracy of the model. Second, this study explored only the impact of the objective perspectives of vegetation landscape of residential communities on residents’ perceptions of CES. However, as the perceptions of CES of UGS are directly linked to residents’ subjective perceptions, studies show that gaps exist between objective and subjective perceptions of vegetation structure, finding that the latter are more important in influencing residents’ use of UGS [83,106,107]. Therefore, we suggest that future research should analyze the level of CES provision of landscapes based on residents’ subjective perceptions of the landscape structure of UGS. Moreover, such research should include residents’ evaluations of the importance of different scales of landscape configuration and residents’ preferences in terms of vegetation structure, ecosystem structure, plant diversity, and tree species. Clarifying these preferences would aid the planning and design of green spaces from the perspective of residents’ needs, which is also an effective means to address inequities in urban environments. Another limitation of our study was that we explored the relationship between the landscape and vegetation structure of UGS and CES at three scales only in the residential areas: micro-scale (approximately 200 m from the residential areas), ecosystem scale, and community scale. In the future, the impact of the landscape and vegetation structure of UGS on CES should be explored at a larger spatial scale, considering, for example, how green space inside and outside the neighborhood affects residents’ perceptions and their use of green space in Chinese cities. Additionally, future research should also focus on the impact of green space at different scales on residents’ satisfaction of a quiet environment, stress relief, and sense of belonging. The latter is an important issue for sociology, planning, and ecology. Finally, the needs and demands of different residents can vary. In our study, we did not consider the impact of specific residents’ characteristics (e.g., age, sex, income level) on their perceptions of vegetation landscape. Knowledge of such issues is important to gain an in-depth understanding of the factors driving residents’ perceptions of UGS and CES.

5. Conclusions

Based on PLS-SEM, we explored the effects of vegetation landscape structure in 40 residential communities in Zhengzhou at multiple spatial scales. We found that large, multifunctional PAS were the primary contributors to recreational services, and both the size (area) and number of PAS were significantly correlated with the level of CES of green spaces. The percentage of vegetation coverage and size of vegetation patches in residential areas can increase the level of CES, mainly due to the naturalness of large areas of vegetation and residents’ aesthetic appreciation of these areas. Vegetation structure contributed less to CES than did PAS and vegetation coverage. Nevertheless, we found an association between CES and the presence of large areas of managed lawn grass. In addition, greater tree species diversity, particularly flowering and fruiting trees, was associated with higher scores of aesthetic and spiritual services. Finally, we found that good management was more important than landscape structure in terms of CES. Therefore, this study suggests that in the presence of insufficient green spaces, diverse PAS and high-level management, which were identified by well-established infrastructure, clean and attractive vegetation, with a homogenous and well-kept plant community, can optimize the CES level of residential green spaces, thereby reducing environmental inequity in access to green space.