Do We Need More Urban Green Space to Alleviate PM_2.5 Pollution? A Case Study in Wuhan, China

Abstract

Urban green space can help to reduce PM_2.5 concentration by absorption and deposition processes. However, few studies have focused on the historical influence of green space on PM_2.5 at a fine grid scale. Taking the central city of Wuhan as an example, this study has analyzed the spatiotemporal trend and the relationship between green space and PM_2.5 in the last two decades. The results have shown that: (1) PM_2.5 concentration reached a maximum value (139 μg/m³) in 2010 and decreased thereafter. Moran’s I index values of PM_2.5 were in a downward trend, which indicates a sparser distribution; (2) from 2000 to 2019, the total area of green space decreased by 25.83%. The reduction in larger patches, increment in land cover diversity, and less connectivity led to fragmented spatial patterns of green space; and (3) the regression results showed that large patches of green space significantly correlated with PM_2.5 concentration. The land use/cover diversity negatively correlated with the PM_2.5 concentration in the ordinary linear regression. In conclusion, preserving large native natural habitats can be a supplemental measure to enlarge the air purification function of the green space. For cities in the process of PM_2.5 reduction, enhancing the landscape patterns of green space provides a win-win solution to handle air pollution and raise human well-being.

1. Introduction

Urban green space can include street trees, private gardens, parks, and peri-urban agricultural land within the city [1,2,3]. Ecosystem benefits provided by green space significantly contribute to maintaining a livable environment for citizens. For instance, air purification [4], alleviation of urban heat islands [5,6], carbon sequestration [7], recreational service [8], and biodiversity conservation [9] can enhance the local quality of life. Small particulate matter with widths less than 2.5 μm (PM_2.5) is one of the outcomes of anthropogenic activities [10,11]. PM_2.5 damages human health in terms of the cardiovascular system [12], respiration [13,14], and mortality [15]. In contrast, green spaces directly absorb particulate matter through the retention ability of leaves [16] and slow down the spread of PM_2.5. However, green space is often affected by urban development, which exacerbates the correlations with PM_2.5 in areas where population density is already high.

The adverse effects of PM_2.5 concentrations are a global concern. According to the guidelines released by the WHO, annual exposure to PM_2.5 concentrations under 5 $\mathsf{\mu }\mathrm{g}/{\mathrm{m}}^3$ has a less negative impact on humans [17]. The guidelines and actual emissions of PM_2.5 in many countries and regions have not yet adopted or attained the same standard yet. For example, the two levels of PM_2.5 concentration identified in China are 15 $\mathsf{\mu }\mathrm{g}/{\mathrm{m}}^3$ and 35 $\mathsf{\mu }\mathrm{g}/{\mathrm{m}}^3$ (annually) [18]. According to the National Ambient Air Quality Standard (NAAQS), the PM_2.5 concentration in India should be no more than 40 $\mathsf{\mu }\mathrm{g}/{\mathrm{m}}^3$ (annually) [19]. The limit of PM_2.5 set by the European Commission is 20 $\mathsf{\mu }\mathrm{g}/{\mathrm{m}}^3$ (annually) [20]. There are gaps between the different guidelines (Appendix A,

Table A1. Examples of PM_2.5 concentration guidelines.

Organizations/Commissions/ Countries	Guidelines for PM2.5 (μg/m3, Annual Average)		Explanations	Reference
WHO	Interim target 1	35	The Air quality guidelines (AQG) level was based on the relationship between PM2.5 and non-accidental mortality in the long run. The interim targets, as incremental steps, guide progressive air quality improvement for areas with high pollution.	[17]
	Interim target 2	25
	Interim target 3	15
	Interim target 4	10
	AQG level	5
European Commissions	Target value to be met as of 1.1.2010; Limit value to be met as of 1.1.2015	25	For a target value, countries are responsible to implement measures to ensure that it is attained. Limit value relates to the maximum margin of tolerance. Stage 2 is stricter than the former standards.	[20]
	Stage 2 limit value to be met as of 1.1.2020	20
China	Level 1	15	PM2.5 of natural reserve area and other protected places meet the Level 1; Residential area, factories, and rural area meet Level 2. No specific time limitation.	[18]
	Level 2	35
India	National Ambient Air Quality Standard (NAAQS)	40	National standard released in 2009	[19]

), which indicates that many efforts need to be made in global PM_2.5 reduction. Notably, the most dramatic influence of air pollution is the effect on longevity. Ebenstein et al. (2017) claimed that a 10.35 $\mathsf{\mu }\mathrm{g}/{\mathrm{m}}^3$ increase in PM₁₀ reduced life expectancy by 7.68 months [21]. If PM_2.5 concentrations reach WHO air quality guidelines (annual 10 $\mathsf{\mu }\mathrm{g}/{\mathrm{m}}^3$, released in 2005) in all countries, then life expectancy will increase by a population-weighted median of 7.20 months [22]. Moreover, poor air quality exposure increases the rates of cardiovascular diseases [23,24]. In addition, regional air quality can affect citizens’ self-reported happiness [25].

Governments and research institutes construct PM_2.5 datasets to examine the concentration, possible origins, and the main influencing factors. Ground air quality stations observe hourly, daily, monthly, and yearly results of particulate matter measurements. These datasets are widely applied in spatial and temporal change analysis of PM_2.5 [26,27,28] and are combined with Aerosol Optical Depths (AODs) data to generate full cover maps across regional and global scales [29,30]. PM_2.5 concentrations and distributions are affected by anthropogenic and natural factors. The emissions from burning biomass and fossil fuels are the main origins of particulate and gaseous pollutants [31]. Landscape fires are a significant PM_2.5 source in many regions [32]. Meteorological factors significantly influence the distribution of PM_2.5 through wind speed and temperature [33]. Urbanization, road density, and urban morphologies correlate with PM_2.5 pollution [34,35]. Land use and cover change also contribute to the distribution of PM_2.5 [36]. Methods such as land-use regression, random forest, and the spatial panel model have been widely applied in examining the influencing factors and predicting the features of PM_2.5 concentration [37,38,39,40].

Green space is one of such factors that can influence regional PM_2.5 concentration, decreasing the concentration by deposition from leaves, affecting wind speed, and impacting air humidity [41,42]. Morphological and physiological traits of green space also make a difference in PM_2.5 reduction [43]. Higher green space core and bridge proportions significantly reduce the PM_2.5 concentration [44]. The leaf area index negatively correlates with PM_2.5 mass concentration [45]. The air purification effects of green space have a particular functional scale. Considering the circular buffers of 0.5–5 km in radius, centered in the monitoring stations, the total edge length of green space has a greater impact on PM_2.5 concentration than the green cover area. However, the influence of green cover dominates on the 3–5 km scale [46]. In addition, biogenic volatile organic compounds (BOVC) emissions of green space-induced PM_2.5 concentration are harmful to human health [47]. The dual role of green space in warm seasons is to both remove air pollution and enhance air quality [48], which needs to be pointed out in the correlation analysis of PM_2.5 concentration and green space morphology.

During the process of urbanization, urban green space is inevitably influenced by building encroachment. Decreasing its area will further undermine the ecosystem functions [49], especially for the central city area where impervious land occupies a higher proportion. Air purification, recreation, and habitat maintenance services of green spaces play significant roles in residents’ well-being. Thus, the correlation of green space landscape patterns on PM_2.5 concentration across an extended period needs to be clarified. Because PM_2.5 can migrate within the regional area within a particular time, long time-series data of PM_2.5 concentrations are more useful to develop annual full cover maps [50]. Furthermore, these maps can be fitted to examine the relationship between PM_2.5 and green space landscape patterns on a grid scale.

This study focuses on the central city of Wuhan because it has undergone rapid urbanization and drastic changes in air quality during the last 20 years. The main objectives of the present work are to (1) reveal the spatial-temporal changes in green space and PM_2.5 concentration and to (2) identify the correlations between landscape features of green space and PM_2.5 concentration. Finally, we propose strategies to enhance the positive relationship between PM_2.5 and green space.

2. Data and Methodology

2.1. Study Area

Wuhan is the capital of Hubei province, the geographical and economic center of the middle part of China, and it has a total area of 8569.15 km² (Figure 1). It is located in a humid subtropical climate zone, and the Yangtze River flows across the central city. The annual precipitation is 2012.4 mm, and the average temperature is about 17.2 °C (2020). The total population is about 12.44 million [51], which increased by 54.67% from 2000 to 2020. Wuhan has undergone rapid urbanization over the past two decades, especially in its central districts where 57.88% of the population (7.21 million) lives.

Expanding the impervious surface threatens the distribution pattern and amount of green space. Factory production, fossil fuel consumption, and private cars are all growing along with urban development, which causes urban problems such as air pollution and traffic jams. More attention has been given to these negative outputs and the importance of protecting the natural environment in the central districts. The local government has implemented a series of measures to deal with these problems, such as reducing energy consumption and unqualified fuels of private cars. The Chinese government gives priority status to the promotion of air quality. A PM_2.5 concentration reduction by 10% is promised in China’s 14th Five-Year Plan (2021–2025) [52], and rules such as restricting high-polluting sources and constant air quality monitoring will continue.

2.2. Data for Analysis

This study represented green space as areas composed of land cover, including woodland, grassland, and cropland. There was a rapid increase in construction land use in the central districts, so we also considered artificial surfaces. According to the current land use classification in China (GB/T 21010-2017) [53], the land cover types were manually interpreted from Google history remote sensing images with a resolution of 2 m (2000/2010/2020) [54]. To improve the precision of our interpretation, we took the land cover maps generated from GlobalLand30 as a reference [55]. The total accuracy of GlobalLand30 is 85.72%, and the Kappa coefficient is 0.82. We calculated landscape patterns of green space in 2020. For coherence with the following data, we described the results of green space landscape patterns as in 2019.

PM_2.5 concentration data (1 km resolution, annual average) were obtained from an open-access database [56], which includes long-term, full-coverage, and high-resolution ground-level air pollutants in China. The local government of Wuhan implemented lockdown strategies from 23 January–22 February 2020. Consequently, this study applied the PM_2.5 dataset of 2019 in the spatial panel regression to avoid the lockdown effect. Thus, this study extracted annual PM_2.5 concentration data from 2000, 2010, and 2019.

The spatial maps of gross domestic product (GDP) were drawn from the Data Center of Resources and Environmental Sciences (RESDC) [57]. The resolution of GDP maps in 2000, 2010, and 2015 was 1 km. Because the GDP for 2019 was unavailable, it was estimated by multiplying 2015 data by the growth rate from 2015 to 2019.

2.3. Methodology

2.3.1. Landscape Patterns of Green Space

The configuration and composition of green space can influence the deposition rate of particulate matter. Hence, we applied the total area (TA), contagion index (CONTAG), largest patch (LPI), and Shannon’s diversity index (SHDI) to demonstrate the spatial distribution of green space (

Table 1. Formula and explanations of landscape indexes.

Landscape Indexes	Formula	Description	References
Total area (TA)	$TA={\displaystyle \sum _{j=1}^n}{a}_{ij}$	Equals the sum of green space areas of patches.	[58,59]
Largest patch index (LPI)	$LPI=\frac{max\left(a_{ij}\right)}{A}\ast 100$	Equals the percentage of the landscape comprised by the giant patch	[60,61]
Contagion index (CONTAG)	$\begin{array}{c}CONTAG\hfill \\ =\left(1+\frac{{{\displaystyle \sum }}_i^n{{\displaystyle \sum }}_j^n(P_i\ast m-\ln \left(P_i\ast m\right))}{2\ln \left(n\right)}\right)\hfill \\ \ast 100\hfill \end{array}$	Ranging from 0 to 100. The lower the index value, the more scattered the urban landscape pattern and the higher the average degree of fragmentation.	[62,63,64]
Shannon’s diversity Index (SHDI)	$SHDI=-{\displaystyle \sum _{i=1}^n}\left(P_i\ast ln{P}_i\right)$	This represents diversity. The value increases as the number of different patch types increases.	[65,66]

Note: $a_{ij}$ is the area (m²) of patch ij; $\mathrm{A}$ is the total area of the grid (m²); n is the number of patches; $P_i$ represents the proportion of landscape occupied by patch type i; $g_{ik}$ is the number of adjacencies (joins) between pixels of patch types (classes). $m=g_{ik}/{{\displaystyle \sum }}_{k=1}^n{g}_{ik}$.

). TA aims to reflect the area of green space. LPI represents the dominating patches occupying the proportion of green space area. CONTAG, as the fractal dimension index, describes the spatial patterns of patches. Finally, SHDI can manifest the complexity of green space types.

A fishnet (grids with 1 km resolution) was created from the PM_2.5 dataset (ArcGIS 10.5, “create fishnet” tool) and applied in the extraction of green space. The landscape indexes were calculated by Fragstats (4.2).

2.3.2. Spatial Correlation Analysis

(1)

Spatial weight matrix

The weight matrix is based on the queen contiguity criterion to reflect the neighboring influence. Grids with the common vertexes and edges are assigned to 1, and otherwise 0. The formula is as follows:

$$W_{ij}=\{\begin{array}{c}1, i shares a corner or edge with j;\\ 0, i is{n}^{\prime }t adjacent to j.\end{array}$$

(1)

(2)

Moran’s I and hotspot analysis

We conducted the spatial autocorrelation test to examine the spatial interdependence of PM_2.5 concentration. Moran’s I was applied to calculate the degree of spatial autocorrelation [67,68]. The formula is expressed as follows:

$$\mathrm{I}=\frac{{{\displaystyle \sum }}_{i=1}^N{{\displaystyle \sum }}_{j=1}^N{W}_{ij}\left(x_i-\overline{x}\right)\left(x_j-\overline{x}\right)}{s^2{{\displaystyle \sum }}_{i=1}^N{{\displaystyle \sum }}_{j=1}^N{W}_{ij}}$$

(2)

$${\mathrm{S}}^2=\frac{{{\displaystyle \sum }}_{i=1}^N{\left(x_i-\overline{x}\right)}^2}{N}$$

(3)

where $\mathrm{I}$ is the univariate global spatial correlation index; $x_i,x_j$ are the attribute values of features i and j, respectively; $\overline{x}$ is the average value of features i and j; ${\mathrm{S}}^2$ is the sample variance; $W_{ij}$ is the spatial weight between features i and j; and N equals the total number of features.

For the Moran’s I analysis, the null hypothesis states that the attribute being analyzed is randomly distributed among the features in the study area. The null hypothesis may be rejected when the p-value is lower than the defined threshold (5% in this study) and the z-score is positive. The spatial distribution of high values and low values in the dataset is more spatially clustered. When the p-value is lower than the defined threshold and the z-score is negative, the null hypothesis may also be rejected. In that case, the spatial distribution of high values and low values in the dataset are more spatially dispersed.

Hotspot analysis is applied to identify the clustering attribute of factors [69,70]. This analysis creates maps of statistically significant hot and cold spots. The hotspot analysis calculates the Getis–Ord $G_i^{\ast }$ statistic for each feature in a dataset:

$$G_i^{\ast }=\frac{{{\displaystyle \sum }}_{j=1}^N{w}_{i,j}{x}_j-\overline{x}{{\displaystyle \sum }}_{j=1}^N{w}_{i,j}}{S\sqrt{\frac{(N{{\displaystyle \sum }}_{j=1}^N{w}_{i,j}^2-{({{\displaystyle \sum }}_{j=1}^N{w}_{i,j})}^2)}{N-1}}}$$

(4)

The explanations of variables are the same as in Formula (2). $G_i^{\ast }$ is a z-score that represents the standard deviation. The resultant z-scores and p-values show where features with high or low values cluster spatially. For statistically significant positive z-scores (5% in this study), the larger z-scores represent intense clustering of high values (hot spot), and vice versa. The quadrants were divided as “High-high (H-H)”, “Low-low (L-L)”, “High-low (H-L)”, and “Low-high (L-H)”.

2.3.3. Spatial Panel Model

Variance inflation factors (VIFs) were calculated to exclude the highly correlated variables from a stepwise procedure. First, we assumed that relevant factors influenced the PM_2.5 concentration without spatial dependency. A basic Ordinary Least Squares (OLS) model is expressed as follows:

$$Y_{it}={\beta }_0+\beta X_{it}+{\epsilon }_{it}$$

(5)

where $Y_{it}$ is the PM_2.5 concentration in cell $i$ at time $t$; ${\beta }_0$ is a constant term; $\beta$ is the parameter of $X_{it}$; $X_{it}$ represents the independent variables, referring to the landscape indexes of green space, GDP, and artificial surface proportion in this study; and ${\epsilon }_{it}$ is the error term.

Ignoring the spillover effect of PM_2.5 concentration might lead to bias. The spatial lag model (SLM) presumes that dependent variables interact among neighbors [71,72]. Thus, we changed Formula (5) into:

$$Y_{it}={\beta }_0+\rho {{\displaystyle \sum }}_{j=1}^N{W}_{ij}{Y}_{it}+\beta X_{it}+{\epsilon }_{it}$$

(6)

where $\rho$ is the spatial lag effect coefficient. When the error terms correlate, the random effect affects the spillover effect [73]. The Spatial Error Model (SEM) is expressed as follows:

$$Y_{it}={\beta }_0+\beta X_{it}+{\mu }_{it}$$

(7)

$${\mu }_{it}=\lambda W{\mu }_{it}+{\epsilon }_{it}$$

(8)

where $\lambda$ represents the average degree of spatial correlations of errors. Furthermore, the Spatial Durbin Model (SDM) measures the effect of neighboring dependent and independent variables [74]. In this case, PM_2.5 concentration is influenced not only by the PM_2.5 value of neighbors but also by the neighboring independent variables:

$$Y_{it}={\beta }_0+\rho {{\displaystyle \sum }}_{j=1}^N{W}_{ij}{Y}_{it}+\beta X_{it}+\theta {{\displaystyle \sum }}_{j=1}^N{W}_{ij}{X}_{ijt}+{\epsilon }_{it}$$

(9)

where $Y_{it}$ is the PM_2.5 concentration and $\theta$ is the spatial spillover effect coefficient of the independent variables.

The Likelihood Ratio test (LR test) determines whether SLM, SEM, and SDM models are more fitted for the green space and PM_2.5 data. The Hausman test was conducted to select between the fixed-effects (FE) or random-effects (RE) panel model [75]. Akaike’s Information Criteria (AIC) and Bayesian Information Criteria (BIC) supported the model selection [76]. The lower the AIC and BIC values, the better the model fits the data. The spatial panel model was processed in Stata15 (package “xsmle”).

3. Results

3.1. Temporal Change and Spatial Distribution of Green Space from 2000 to 2019

The amount of green space (TA) decreased from 381.04 km² to 282.60 km², or a decrease of 25.83%, from 2000 to 2019. Conversely, artificial surfaces increased from 290.54 km² to 387.68 km² in the same period, increasing by 33.43%. The adverse direction of transformation reflects the observation that human activities have an intervention effect on the natural environment. Consistent with the downward trend of TA, both LPI and CONTAG decreased, while SHDI increased (

Table 2. Landscape indexes of green space in the central city from 2000 to 2019.

Year/Landscape Indexes	TA (km2)	LPI (%)	CONTAG (%)	SHDI
2000	381.04	6.05	63.28	0.78
2010	323.52	5.93	54.81	0.96
2019	282.60	3.35	53.14	1.00
changing rate (%)
2000–2010	−15.10	−2.04	−13.39	23.17
2010–2019	−12.65	−43.40	−3.04	4.37
2000–2019	−25.83	−44.56	−16.03	28.55

).

The changing rates of landscape indexes in the earlier 10 years (from 2000 to 2010) were higher than those in later years (from 2010 to 2019), except for LPI. The evolution of green space resulted in an evident reduction in total area and large patches, which led to fewer connections of natural land cover. There was an apparent downward trend in green space in the urban periphery and an incremental trend in the center area (Figure 2). In addition, landscape indexes demonstrated a clustering pattern (Appendix A, Figure A1). Moran’s I for TA was in a downward trend due to the loss of ample green space around the edge of the central city. The spatial correlation of CONTAG kept increasing while LPI and SHDI decreased from 2000 to 2019. This result suggested that green space has become more fragmented and less diverse.

3.2. Spatial and Temporal Changes in PM_2.5 Concentration

From 2000 to 2019, the PM_2.5 concentration underwent a transition from an upward to a downward trend. The annual average values of PM_2.5 for 2000, 2010, and 2020 were 75.20 $\mathsf{\mu }\mathrm{g}/{\mathrm{m}}^3$, 85.87 $\mathsf{\mu }\mathrm{g}/{\mathrm{m}}^3$, and 46.10 $\mathsf{\mu }\mathrm{g}/{\mathrm{m}}^3$, respectively (Figure 3). Compared to 2000, a higher level of PM_2.5 concentration in 2010 led to an increase in the average PM_2.5 by 14.19%. The maximum value occurred in 2010. The dense PM_2.5 concentration clustered around the city center. PM_2.5 concentration decreased by 46.33% from 2010 to 2019, which corresponds with the promotion of air quality across the region. The distribution of PM_2.5 was in a clustering pattern according to the global Moran’s I value from 2000 to 2019 (see Appendix A, Figure A1).

3.3. Spatial Patterns of Green Space and PM_2.5 Concentration

The local spatial correlations among PM_2.5 concentrations and landscape indexes are shown in Figure 4. The number of grids with significant “H-H”, “H-L”, “L-H”, and “L-L” are demonstrated in the four quadrants. From 2000 to 2019, the spatial scale of “H-H” and “H-L” cluster patterns decreased, which indicates that both PM_2.5 and green space were downward. “L-L” group numbers increased on the left-hand side of the Yangtze River, while “L-H” cells clustered on the right-hand side. There were spatial correlations between PM_2.5 and green space landscape indexes. However, clarifying how the landscape patterns correlated with PM_2.5 requires further regression analysis.

3.4. Correlations of Green Space Landscape Patterns and PM_2.5 Concentration

We also considered GDP and the artificial surface area because PM_2.5 is affected by economic development and urbanization [33,34]. The VIF value of independent variables was less than 10. There was a low correlation among variables and the fit for regression analysis.

As the results have shown (

Table 3. Regression analysis of PM_2.5 concentration (dependent variable) and landscape indexes, GDP, artificial surfaces.

	Models	OLS	SLM	SEM	SDM
Coefficients
Constant		169.218	-	-	-
Artificial surfaces		0.000	−0.009	0.002	0.006
GDP		−10.204 ***	−1.601 ***	−1.630 ***	−0.486
TA		−0.15 ***	0.006	0.032 **	0.030 **
LPI		−0.072 ***	−0.023 ***	−0.027 ***	−0.028 ***
CONTAG		−0.002	−0.007	−0.010	−0.008
SHDI		−3.64 ***	−0.597	−0.510	−0.638
W * Artificial surfaces		-	-	-	−0.022
W * GDP		-	-	-	−1.324 ***
W * TA		-	-	-	−0.065 ***
W * LPI		-	-	-	0.026
W * CONTAG		-	-	-	0.023
W * SHDI		-	-	-	−0.197
$\rho$		-	0.876 ***	-	0.868 ***
$\lambda$		-	-	0.938 ***	-
Residual variance		-	30.191 ***	29.901 ***	30.145 ***
R-square		0.435	0.4254	0.2900	0.4872
AIC		20,268.721	15,859.38	15,965.58	15,853.29
BIC		20,309.385	15,905.85	16,012.05	15,934.61

Note: ***, **, * represent the significance level of p-value 1%, 5%, and 10%, respectively.

), ordinary linear regression and spatial panel models examined the correlations between landscape indexes and PM_2.5 from different aspects. The R-square of OLS (0.435) showed that the landscape patterns of green space and GDP could explain the PM_2.5 concentration to a certain degree. Meanwhile, the artificial surface demonstrated an insignificant correlation with PM_2.5. GDP and PM_2.5 concentration significantly correlated. TA and LPI was negatively correlated with PM_2.5, which indicates that the area of green space and larger patches could contribute to the air purification. CONTAG demonstrated an insignificant relationship with PM_2.5 concentration. Finally, SHDI showed a strong influence in alleviating PM_2.5 concentration.

Spatial panel models further manifested the spatial correlations of PM_2.5 and green space. LPI and GDP negatively correlated with PM_2.5 in SLM and SEM. Neighboring GDP and TA showed a negative correlation with PM_2.5 concentration in the results of SDM. To determine a model for further analysis, we examined the statistically significant LR and Hausman test, together with lower AIC and BIC values. We found that the spatial lag model (SLM) was more adequate. In detail, LPI negatively correlated with PM_2.5, which suggests the importance of protecting natural sizeable green space. Even though SHDI showed the most effective reduction in PM_2.5, in addition to the effect of GDP, the p-value was not significant. TA and CONTAG correlations with PM_2.5 were unclear.

4. Discussion

This paper has focused on the relationship between green space landscape patterns and PM_2.5 concentration in Wuhan central city over the last two decades. The significant negative correlation between LPI and PM_2.5 was examined in the regression analysis. Reserving naturally sizeable green space and enlarging present green space are promising measures to support the mitigation of PM_2.5. However, impervious land constantly encroaches on surrounding green space [77]. The present study found that almost a quarter of green space vanished in the central city of Wuhan from 2000 to 2019. The loss of marginal green space exceeds the increment in the city center (Figure 2), which reduces the total area [78].

From 2010 to 2019, PM_2.5 decreased by 46.33% in the central city of Wuhan. Policies such as the Air Pollution Prevention and the Control Action Plan, which was released by the Chinese government in 2013, significantly decreased the population-weighted annual mean PM_2.5 concentrations in China from 61.8 to 42.0 μg/m³ within five years [79]. Anthropogenic emission dominated the reduction. National policies could accelerate the efforts of local authorities to improve the air quality for sustainable development. In the central city of Wuhan, measures such as promoting green space coverage, reducing the use of coal, and green energy consumption have gradually alleviated regional PM_2.5 pollution [80]. Indeed, the average PM_2.5 concentration decreased and reached 46.1 $\mathsf{\mu }\mathrm{g}/{\mathrm{m}}^3$ in 2019. However, this is still far from the guideline (5 $\mathsf{\mu }\mathrm{g}/{\mathrm{m}}^3$) suggested by the WHO to reduce the harmful impacts on human health and longevity [17].

Although artificial surfaces did not show an evident relationship with PM_2.5 concentration, the economic activities demonstrated a statistically significant correlation with PM_2.5. According to the Environmental Kuznets Curve, a post-urbanization period of economic development can exert a less negative influence on the environment [81,82]. In our case, during the primary period of urbanization, economic development dominated society. Factories with polluting emissions near the central city center led to poor air quality. Later, advanced economic development and urban planning would reduce the disturbance to air quality. This interpretation might explain the cluster of the “L-L” group on the left-hand side of the Yangtze River, where high urbanization and a low area of green space occur.

Notably, the relationship between green space and PM_2.5 needs to be carefully interpreted. First, many anthropogenic and meteorological factors mutually affect PM_2.5 concentration. The contribution rate of the main urban area source to PM_2.5 concentration exceeds 60% in Wuhan [83]. Humidity and temperature can also contribute to the level of PM_2.5 [84]. Even though green space features showed significant correlations with PM_2.5 in spring [85], this mechanism is complicated. Meanwhile, biogenic volatile organic compound emissions from green space and other meteorological conditions might influence PM_2.5 mitigation. In the context of regional climate and air quality model simulations, vegetation emissions led to an increase in PM_2.5 concentration over 10 $\mathsf{\mu }\mathrm{g}/{\mathrm{m}}^3$, which occupied around 20% of the average observed PM_2.5 concentration [86]. Vegetation emissions can also increase the particle fluxes in urban environments [87]. Thus, the dual effects of vegetation need to be considered in increasing green space to achieve PM_2.5 mitigation.

Green space landscape patterns are stressed in the present study for two reasons. First, according to previous studies, the green space area and diversification of vegetation could exert great significance in air regulation [88]. SHDI also showed a negative correlation with PM_2.5 concentration in this study. The amount of PM_2.5 adsorption by leaves demonstrated a noticeable seasonal difference. Conifers were more effective in PM_2.5 adsorption than broadleaf trees were [89]. Green Infrastructure (GI) practices could significantly improve air quality in urban areas [90]. Vegetation structure, composition, and management are essential to optimize the capacity of green spaces to purify air quality [91]. The bio-physical attributes of green spaces directly contribute to the reduction in PM_2.5. Second, the ecosystem services from green space combined with PM_2.5 reduction could mutually affect the health of the residents [92]. Expanding the green area and reducing fragmentation can support the improvement of air quality and decrease the mortality rate of the citizens. Combined with the results of this study, large green space patches deserve more attention in urban land use planning.

This study examined the correlations between green space morphology and PM_2.5 concentration over a long period. However, there are some limitations. First, during the lockdown period, the PM_2.5 concentration of Wuhan decreased by 27 μg/m³ compared with the same period in 2019 [93]. In total, 90% of the reduction resulted from the lockdown because PM_2.5 mainly comes from factories, coal consumption, and transportation. A series of research studies in India [94], America, Italy, France [95], Southern Italy [96], the United Kingdom [97], Barcelona [98], and China [99] have shown positive outcomes of promoting air quality during the same period. Although the lower pollution level briefly appeared during the lockdown, the positive effects will guide decision-makers to make more efforts to moderate PM_2.5 concentration from the origins [100,101]. Correlations between green space and PM_2.5 concentrations should be further investigated under different policies.

Although this study showed the possible correlations between green space morphology and PM_2.5 concentration, the interpretation was limited by the absence of PM_2.5 sources and meteorology data. Moreover, the physical mechanisms of green space and PM_2.5 concentration need to be further explained. For instance, deposition on leaves might offset the emissions of biogenic volatile organic compounds, even though the net PM_2.5 deposition of green space remains. In that case, the contribution of green space to decrease the PM_2.5 concentration will be more comprehensive.

5. Conclusions

Urban green space landscape patterns have shown different correlations with PM_2.5. In the present study, LPI and SHDI negatively correlate with PM_2.5 concentrations. Hence, it is rational to implement measures to strengthen the advantageous landscape features of green spaces. However, under the pressure of urbanization, only a limited quantity of land is left for expanding green spaces. Our results show that green space has decreased by a quarter in the Wuhan central city over the last two decades. Therefore, enhancing landscape patterns of green space deserves more attention. In terms of PM_2.5 concentration, preserving larger natural patches and new increments around the existing green area are promising supplementary measures for locations with a lower proportion of green space and higher pollution emissions. These findings may serve as a reference for cities undergoing land-use conflicts and experiencing air pollution.