Effects of Urban Landscapes on Pollutant Concentrations in Chengdu Plain Urban Agglomeration

Abstract

Frequent air pollution due to urbanization poses a severe threat to urban environments. In this study, the effects of urban landscapes on pollutant concentrations at different spatial and temporal scales are examined using “3S” technology, based on land-use (LU) classification maps for two time phases (2015 and 2018) in conjunction with monitoring data for air pollutants of the same periods. The results showed that: (1) A very high overall LU ratio was found for Chengdu Plain urban agglomeration (CPUA), with only 0.3% of the land being unused. Farmlands (43%), forests (36.45%), and grasslands (14.17%) were identified as the main landscape types. A decrease in the proportions of the total area occupied by farmlands and grasslands was 0.44% and 0.72%, respectively, and 0.10%, 0.04%, and 0.97% increased in the proportions of the total area occupied by forests, water bodies, and developed lands, respectively, were found from 2015 to 2018. (2) NO₂, PM_2.5, and PM₁₀ were mainly distributed in the central and eastern parts of the study area, while SO₂ was mainly distributed in the southwest. In 2018, compared with 2015, the maximum concentration of NO₂ decreased from 60.36 μg/m³ to 49.75 μg/m³, and the distribution range of high concentration NO₂ was reduced and concentrated in Chengdu; the concentrations of PM₁₀ and PM_2.5 decreased significantly, and the maximum concentration decreased by 20.53% and 23.93%, respectively, but the concentration of pollution in the northeast increased significantly. The scope of SO₂ pollution had shifted from the south to the southwest, and the pollution level had decreased from south to north. (3) The effects of various landscape types on pollutant concentrations were complex. At a patch-type level, increasing the area proportions of “pollution-reducing” landscape types could reduce pollutant concentrations. Specifically, increasing the area, largest patch index, and patch cohesion of forests and grasslands, as well as reducing the area, largest patch index, and patch cohesion of farmlands and developed lands, could effectively lower pollutant concentrations. From a landscape pattern perspective, high shape regularity and low diversity of landscape patches resulted in high concentrations of NO₂, PM_10, and PM_2.5. In contrast, high levels of dominance and aggregation of landscapes lead to low concentrations of SO₂.

1. Introduction

The urban ecosystem is a special nature-society-economic complex system established by human beings on the basis of transforming and adapting to the natural environment [1]. Over the past two centuries, cities have expanded rapidly around the world, bringing together people and productive activities, changing land use, and driving economic growth, but also with many pollution problems. The 2018 Environmental Status Bulletin of China published by the Ministry of Ecology and Environment of China did not give rise to much optimism for the ambient air quality (AAQ) in the cities across the country. Of the 338 Chinese cities at the prefectural level or above, 121 (35.8%) met the AAQ standard, while the remainder (217; 64.2%) failed. In addition, regional dust-haze pollution occurred frequently and was characterized primarily by large distribution areas, long durations, high-intensity levels, and rapid pollutant concentration accumulations. Apart from northern China, as well as the Yellow and Pearl River Deltas, the central-southern part of the North China Plain, the Yangtze River Delta, and the Sichuan Basin were also identified as regions where air pollution occurred with rapidly increasing frequency. CPUA is located in the central part of the Sichuan Basin. It is one of the areas with the largest urbanization rate in the west and one of the top five areas with heavy air pollution in my country [2].

Nevertheless, urbanization, energy consumption, and economic development are inevitable consequences of social progress. A large number of studies have shown that urban green space can effectively reduce air pollutants [3], but under the premise of limited urban green space, it is difficult to improve the urban ecological environment only by increasing the green space. The primary pollutant in the urban area of Chengdu is PM_2.5 [4]. The pollutant emissions are mainly concentrated in the urban area of Chengdu, the industrial areas around Chengdu, and the expressway areas with large traffic flow [5]. Air pollutants in Chengdu were affected mainly by the topographic and climatic conditions of the Sichuan Basin and were markedly more concentrated in winter and less so in summer [6], resulting in the most severe pollution in winter. The total area of green space in Chengdu is relatively large, but the per capita park green space still does not meet the standard of a garden city [7]. From 2000 to 2019, the land use change of the Chengdu Plain urban agglomeration showed a trend of “four increases and two decreases” [8]. The area of construction land, woodland, water area, and unused land increased, and the area of grassland and cultivated land decreased. Deng [9] believed that areas with small green space patches, scattered distribution, regular shapes, and poor connectivity of green space corridors in Yongzhou City have correspondingly higher PM_2.5 concentrations. Guan [10] studied the influence of the urban landscape pattern on the haze effect of Shijiazhuang and pointed out that the haze effect of the overall landscape level pattern and its change trend with the scale is consistent, it is consistent with Qiu’s [11] view that the green space patch density index is both significantly and negatively correlated with NO₂ and PM_2.5 concentration and that the correlation size increases with the total landscape area. Sun [12] believed that the high correlation of PM_2.5 concentration was the proportion of plaque types and the edge density index in the landscape, and Xie et al. [13] also believed that the composition characteristics and structural characteristics had a significant influence on PM_2.5 concentration. Han et al. [14] pointed out that Nanchang city has complete landscape types. Still, the spatial distribution is not balanced, the old city landscape is relatively more concentrated, and the diversity index and uniformity index are low. He proposed a combination of south and north blocks, three rings and eight shots, and a combination of central city belts, rings, wedges, corridors, and gardens. Zhou et al. [15] combined the principle of landscape ecology and the future development form of Shenyang city to construct the optimization scheme of urban green landscape patterns. The green space network space structure of “four belts, three rings, seven wedges, and network connection” has been formed.

To sum up, the urban landscape has an important impact on the concentration and distribution of pollutants. The rational use of land for development is an effective measure to reduce air pollution, but only studying the response of the diffusion process of air pollutants to the urban landscape pattern can no longer meet the requirements of urban air pollution control. Scientists around the world have conducted studies centering on LPs and pollutant concentrations. However, most of these studies focused on PM_2.5, and they were often limited to one city and therefore failed to account for air mobility. In order to study the impact of various ground landscapes such as forest land, buildings, and their landscape structures on various pollutants, reasonably plan the landscape structure to mitigate pollution, improve the quality of life of residents and maintain human life and health. This paper presented a case study of the Chengdu Plain urban agglomeration (CPUA). Specifically, the coupling relationships between the LP and the concentrations of main pollutants (i.e., NO₂, PM₁₀, PM_2.5, and SO₂) were analyzed at multiple spatial scales. Compared with Liu et al. [16] survey of urban LPs and air pollutant concentrations, higher-accuracy results were derived in this study from a larger number of sampling points with more details based on unified grids. The results of the presented research can be useful and used to design green areas as a filter for air pollutants.

2. Data and Methods

2.1. Study Area

Situated in the western part of the Sichuan Basin, the CPUA (28°28′–33°03′ N, 101°56′–105°43′ E)—one of the three major secondary megalopolises within the Sichuan Basin—consists primarily of Deyang (31°14′ N, 104°22′ E), Mianyang (31°28′ N, 104°48′ E), Chengdu (30°40′ N, 104°04′ E), Meishan (30°03′ N, 103°50′ E), Ya’an (29°58′ N, 103°00′ E) and Leshan (29°34′ N, 103°46′ E) and spans over 373 km from east to west and 527 km from south to north. The Chengdu Plain is located at the southwestern margin of the Sichuan Basin in the third Neocathaysian subsidence zone in China and is enclosed between the Longmenshan uplift fold belt and the Longquanshan and Wuzhongshan fault-fold belts. With its northern part being constrained by the Mianyang and Zhongxingchang vortex structures, the Chengdu Plain displays the characteristics of a faulted basin and generally inclines from the northwest to the southeast.

2.2. Data Sources

The LU data (spatial resolution: 1 km) of two periods (2015 and 2018) for the CPUA used in this study originated from the Resource and Environment Science and Data Center of the Chinese Academy of Sciences. The original data contained information for 25 secondary LU types. The hourly NO₂, SO₂, PM_2.5, and PM₁₀ data collected at the Sichuan Provincial General Environmental Monitoring Station from October to December 2015 and 2018 were used as air pollution data in this study. We used 93 monitoring sites in Sichuan Province. In the study area, each city has at least five monitoring points, which can well monitor the air quality status and the transmission and influence range of pollutants in the regional area (Figure 1). Due to the small number of monitoring sites in the study area, in order to obtain a better interpolation effect, we chose to use the data of monitoring sites around the study area. Each hour of pollution data was used to stack the daily data obtained, and then the daily data was averaged.

2.3. Research Methods

In this study, the landscapes in the study area were categorized using two supervised classification methods—maximum likelihood classification and interactive visual interpretation—into farmlands, forests, grasslands, water bodies, developed lands, and unused lands (Figure 2).

LP often refers to the spatial pattern of landscapes, that is, the spatial arrangement and combination of landscape elements of different sizes and shapes, including the types, numbers, spatial distributions, and configurations of elements that constitute landscapes (e.g., the random, uniform, or clustered distribution of patches of different types). It is a detailed reflection of landscape heterogeneity and is a result of the action of various ecological processes at different scales [17].

In this study, LP characteristics were analyzed at three levels: single patches, patch types, and the overall landscape mosaic composed of several patch types [18]. Correspondingly, LP metrics were grouped into three classes: patch-level, patch type-level, and landscape-level (

Table 1. Brief explanation of the landscape pattern metrics.

LP Metric	Description of the Equation	Ecological Meaning
CA (ha)	Sum of the areas of all the patches of a certain type	A measure of the composition of the landscape, as well as a basis for calculating other metrics
PLAND (%)	Percentage of the total landscape area occupied by a certain patch type	A measure of the composition of the landscape, as well as a basis for determining the dominant landscape elements
PD (number per ha)	Ratio of the number of patches to the landscape area	A reflection of the number of patches per unit area, which characterizes the level of fragmentation of the landscape. A high value means that the landscape is highly fragmented.
LPI (%)	Ratio of the largest patch of the corresponding type to the total landscape area	A measure of the ecological characteristics (e.g., abundance) of the dominant and internal species in the landscape. Its changes reflect the intensity and frequency of interference, as well as the direction and intensity of human activity.
PAFRAC	$\mathrm{PAFRAC}=\frac{2\ln \left(\raisebox{1ex}{$\mathrm{P}$}\!\left/ \!\raisebox{-1ex}{$4$}\right.\right)}{\ln \left(\mathrm{A}\right)}$. P is the plaque perimeter. A is the total plaque area.	A reflection of the shape complexity of the patch. Its value ranges from 1 to 2. A value close to 1 indicates that the patch has a simple boundary and a regular shape.
COHESION	$\mathrm{COHESION}=\left[1-\frac{{{\displaystyle \sum }}_{\mathrm{j}=1}^{\mathrm{n}}{\mathrm{P}}_{\mathrm{ij}}}{{{\displaystyle \sum }}_{\mathrm{j}=1}^{\mathrm{n}}{\mathrm{P}}_{\mathrm{ij}}\sqrt{{\mathrm{a}}_{\mathrm{ij}}}}\right]\times {\left[1-\frac{1}{\sqrt{\mathrm{A}}}\right]}^{-1}\times 100$. The n is the total number of patch types in the landscape. The aij refers to the area of the jth patch in the ith type of landscape. The Pij represents the perimeter of the jth patch in the ith landscape. A is the total plaque area.	A reflection of the physical connectivity of the patches of the corresponding type.
LSI	$\mathrm{LSI}=\frac{\mathrm{E}}{4\sqrt{\mathrm{A}}}$. A is the total plaque area. E is the total length of the patch boundary.	A measure of the dispersion or aggregation between all the patches of a certain type or landscape.
CONTAG	The proportion of the landscape area occupied by each patch type multiplied by the proportion of the number of adjacent grid cells in each patch type to the total number of adjacent grid cells, multiplied by the natural logarithm of the same quantity, summed over each patch type; divided by twice the natural logarithm of the number of patch types; plus 1; multiplied by 100 to convert to a percentage.	A high value suggests good connectivity for a certain dominant patch type in the landscape
SHEI	SHDI divided by the maximum possible diversity under the given landscape abundance	A low value often means a high level of dominance and can reflect that the landscape is controlled by one or a few dominant patch types. A value close to 1 means a low level of dominance and suggests that no patch type is dominant in the landscape and that all the types are evenly distributed in the landscape.

). Because the patch-level metrics were an excessively small scale and had little value to the description of the overall LP, four factors were primarily considered in this study for landscape types, namely, their area, level of fragmentation, patch shape, and patch cohesion. To conduct this study, the following six metrics were selected: the area of the landscape type (CA), the proportion of the total area occupied by the patch type (PLAND), PD, LPI, the fractal dimension (PAFRAC), and the patch cohesion index (COHESION). At the landscape level, the focus was placed on the dominance, dispersion, and shape of patches, as well as the diversity and uniformity of landscapes. To achieve this focus, the following five metrics were selected: the PD, the LPI, the LSI, the contagion index (CONTAG), and Shannon’s evenness index (SHEI).

Based on the LU-type map for the megalopolis (Figure 2), moving windows were selected in the FRAGSTATS version 4.2 to set landscapes to specific sizes and shapes. Then, the LP metrics for the LU types in the megalopolis were calculated at the patch-type and landscape levels.

Currently, China has few automatic atmospheric environment monitoring stations, which are distributed in a scattered and nonuniform manner. As a consequence, data collected at the monitoring stations cannot represent the AAQ across a region. AAQ information predicted using spatial interpolation algorithms can better reflect the spatial distribution patterns of pollutants in an unknown geographic space. In this study, pollutant concentrations were calculated using data collected from 93 monitoring points as sample points. We selected urban and regional points for ambient air quality monitoring. The representative range of urban points is generally 500 m to 4 km in radius. If the concentration of pollutants is low, it can be expanded from 4 km to tens of kilometers in radius. The representative range of regional points is generally tens of kilometers in radius. The monitoring stations in the study area are densely distributed and can cover the whole study area. Then, the distribution of the concentration of each pollutant in the study area was determined through inverse distance weighted (IDW) interpolation.

Considering that data collected at an AAQ monitoring point generally can reflect the AAQ in the area within a radius of 0.5–4 km, data were sampled in this study for the study area at multiple scales, i.e., multiple grid cell sizes, starting from 10 km × 10 km and then increasing by a factor of 2 to 20 km × 20 km and further to 40 km × 40 km. A total of 732, 185, and 47 grid cells with different side lengths and the corresponding centers were obtained using ArcGIS 10.0. The LP metrics were calculated for each grid cell at each scale, and the obtained values were assigned to the center of each grid cell using the “Extract Values to Points” tool.

Similarly, the spatial distribution of the concentration of each of the four pollutants was segmented at the same scales to determine the concentrations of each pollutant at multiple scales (Figure 3). The correlations between each LP metric and the concentrations of the pollutants at each scale were analyzed using SPSS software to examine the multiscale coupling relationships between the landscape types and structures and the concentrations of the pollutants.

3. Results

3.1. LP Analysis

3.1.1. LP Analysis at the Patch-Type Level

Analysis of the PLAND revealed a very high LU ratio for the CPUA in 2015, with farmlands as the landscape type occupying the highest proportion (43.98%) of the total land area, followed by forests (36.45%), grasslands (14.17%), developed lands (3.81%), water bodies (1.29%) and unused lands (0.3%) (

Table 2. Structural characteristics of different landscape types in the Chengdu Plain urban agglomeration.

Year	Landscape Type	CA (ha)	PLAND (%)	PD (Number/ha)	LPI (%)	PAFRAC	COHESION
2015	Grasslands	1,036,500	14.1666	0.0138	1.8998	1.6223	93.6581
	Forests	2,666,800	36.4491	0.0176	14.7024	1.6414	98.5021
	Farmlands	3,217,700	43.9787	0.0114	37.4455	1.6426	99.4743
	Water bodies	94,200	1.2875	0.0065	0.0902	1.6398	46.5739
	Developed lands	279,000	3.8133	0.0104	0.7695	1.5932	79.9526
	Unused lands	22,300	0.3048	0.0013	0.0301	1.5001	51.9069
2018	Grasslands	984,100	13.4471	0.0158	1.3774	1.6199	91.7677
	Forests	2,675,100	36.5536	0.0173	15.0650	1.6499	98.8810
	Farmlands	3,186,500	43.5415	0.0109	38.8273	1.6374	99.5633
	Water bodies	97,400	1.3309	0.0066	0.0997	1.6462	50.4470
	Developed lands	349,800	4.7798	0.0105	1.6684	1.5704	88.5883
	Unused lands	25,400	0.3471	0.0014	0.0410	1.6097	56.2936

). The same pattern could be observed for 2018, albeit with some changes in the proportions of the landscape types. Specifically, in 2018, farmlands, forests, grasslands, developed lands, and unused lands accounted for 43.54% (down from 2015), 36.55% (up from 2015), 13.45% (down from 2015), 4.78% (considerably up from 2015) and 1.68% (considerably up from 2015) of the total land area, respectively (Table 2). Analysis of the PD revealed the following results. The levels of fragmentation of the landscape types remained unchanged between 2015 and 2018, with forests being the most fragmented, followed by grasslands. A high level of patch dominance could be observed for farmlands and forests in 2015 and 2018, while an increase in the level of patch dominance was evident for farmlands, forests, and developed lands from 2015–2018. A PAFRAC value of approximately 1.6 with an insignificant change from 2015–2018 could be observed for each landscape type, suggesting the presence of patches with complex shapes and low regularity. Analysis of the COHESION revealed high (over 90%) levels of aggregation (an indicator of high connectivity) for farmlands, forests, and grasslands, a moderately high level (79.96%) of aggregation for developed lands, and low levels of aggregation (an indicator of a dispersed distribution pattern) for water bodies and unused lands. An increase of 8.62% was found in the level of aggregation of developed lands from 2015 to 2018.

3.1.2. LP at the Landscape Level

PD values of 0.0608 and 0.0624 were obtained for 2015 and 2018 at the landscape level, respectively, suggesting a low level of fragmentation and the presence of a basically intact LP on the surface. The small values of the LPI for 2015 and 2018 (37.45 and 38.83, respectively) (

Table 3. Overall landscape pattern metrics for the Chengdu Plain urban agglomeration.

Year	PD (Number/ha)	LPI (%)	LSI	CONTAG	SHEI
2015	0.0608	37.4455	46.0388	41.5530	0.6721
2018	0.0624	38.8271	46.6670	40.7144	0.6821

) indicated a small difference between the areas of the patches in each year. In 2015 and 2018, the landscape shape index (LSI) and the spread index (CONTAG) were not much different. The results show that the patches of Chengdu Plain Urban Agglomeration were relatively compact, and the overall landscape pattern was relatively clustered and distributed continuously. In 2015 and 2018, the Shannon Evenness Index (SHEI) was 0.67 and 0.68. This indicates that there is no obvious dominant type and that each patch type is relatively evenly distributed in the landscape (Table 3).

3.2. Spatial Distribution Patterns of the Concentrations of the Pollutants

In 2015 and 2018, NO₂ was concentrated primarily in the central part of the CPUA. That is, with the city of Chengdu as the core, it gradually decreases to the surrounding areas. In 2018, the maximum NO₂ concentration decreased from 60.3 μg/m³ to 49.75 μg/m³. In 2015, PM₁₀ was concentrated in Chengdu, and the concentration in the northeast and southwest regions was small. In 2018, the pollution range of PM₁₀ expanded, the concentration in the north increased significantly, and the maximum and minimum values of PM₁₀ concentrations decreased significantly, and the maximum concentration value decreased by 20.53%. PM_2.5 in 2015 had one pollution center in each of the central and southern regions, with less pollution in the northeast and southwest. The distribution pattern of PM_2.5 in 2018 remained basically unchanged, but the pollution range in the central and eastern regions was slightly reduced, and the pollution level in the northeast increased. In 2015, SO₂ was mainly distributed in the southern region of the study area and gradually decreased from south to north, and in 2018, the scope of SO₂ pollution in the southwest was expanded, and the pollution range in the central and eastern regions was reduced (Figure 4).

A correspondence was identified between developed lands and areas with high concentrations of NO₂, as well as between developed lands and farmlands and areas with high concentrations of PM₁₀ and PM_2.5. Areas with high concentrations of SO₂ were found mostly distributed in the southern part of the study area. In contrast, those with high concentrations of NO₂, PM_10, and PM_2.5 appear to correspond mainly to urban business centers and neighborhoods with several large malls on the surface. In addition, pollutants emitting from PetroChina’s oil refining and ethylene plants in Pengzhou further worsened air pollution in the study area. According to the big data of Winshang, in 2018, as a new first-tier city, Chengdu was the preferred location in southwestern China for major brands, with more consumer brand stores than any of the other new first-tier cities. Moreover, Chengdu is the trade center in southwestern China, where wholesale and retail vendors from around the country ply their trades. As a result, the streets of Chengdu are loaded with heavy traffic, which inevitably leads to substantial vehicle exhaust emissions. The considerable forests, grasslands, and water bodies in the southwestern area of the CPUA, to some extent, reduced pollutant concentrations. In 2018, high concentrations of SO₂ appeared primarily in the southern part of the study area and were lower than those in 2015, which could be attributed to government efforts, namely, the implementation of energy conservation and emission reduction policies, as well as rectification measures in plants. PetroChina is the main source of SO₂ and the distribution of pollution sources and the regional circulation field will affect the distribution of SO₂. Inland winter winds in the basin prevail in winter, with northerly winds becoming the most frequent, so the SO₂ concentration in the southern region is higher.

3.3. Multiscale Coupling Relationships between the LP and the Concentrations of the Pollutants

3.3.1. Coupling Relationships between the Concentrations of the Four Pollutants and the Overall LP in Different Years and at Different Scales

The concentrations of NO₂, PM_10, and PM_2.5 in 2015 were found to be negatively correlated with the LSI and SHEI and positively correlated with the LPI and CONTAG. The significance of these correlations increased as the scale increased. At the medium and small scales, the concentration of SO₂ was negatively correlated with the LSI and SHEI and positively correlated with the LPI and CONTAG. The absolute value of the coefficient of each of these correlations increased as the scale increased (Figure 5). Approximately similar but stronger correlations could be found between the overall LP and the concentrations of the pollutants in 2018 (

Table 4. Correlations between the concentrations of the four pollutants and the landscape-level landscape pattern metrics at different scales.

Year	Pollutant	Grid-Cell Side Length (GCSL)	PD	LSI	LPI	CONTAG	SHEI
2015		10 km	0.022	−0.042	0.208 **	0.244 **	−0.238 **
	NO2	20 km	0.063	−0.120	0.190	0.240 *	−0.437 **
		40 km	0.239	−0.711 *	0.837 **	0.703 *	−0.714 *
		10 km	−0.013	−0.066	0.236 **	0.281 **	−0.272 **
	PM10	20 km	0.058	−0.068	0.216 *	0.266 *	−0.425 **
		40 km	0.319	−0.617	0.728 *	0.511	−0.563
		10 km	−0.034	−0.127 **	0.298 **	0.346 **	−0.340 **
	PM2.5	20 km	0.064	−0.163	0.367 **	0.397 **	−0.528 **
		40 km	0.293	−0.642 *	0.720 *	0.492	−0.603
		10 km	0.038	0.245 **	−0.218 **	−0.270 **	0.270 **
	SO2	20 km	−0.042	0.427 **	−0.372 **	−0.336 **	0.338 **
		40 km	0.045	0.683 *	−0.690 *	−0.790 **	0.768 *
2018		10 km	−0.035	−0.178 **	0.262 **	0.364 **	−0.351 **
	NO2	20 km	0.019	−0.271 *	0.361 **	0.402 **	−0.587 **
		40 km	0.360	−0.840 **	0.800 **	0.779 **	−0.840 **
		10 km	−0.038	−0.240 **	0.307 **	0.435 **	−0.416 **
	PM10	20 km	0.006	−0.371 **	0.449 **	0.493 **	−0.625 **
		40 km	0.334	−0.771 **	0.776 **	0.839 **	−0.847 **
		10 km	−0.024	−0.137 **	0.211 **	0.318 **	−0.300 **
	PM2.5	20 km	0.032	−0.190	0.296 **	0.353 **	−0.459 **
		40 km	0.370	−0.720 *	0.642 *	0.769 **	−0.769 **
		10 km	0.050	0.325 **	−0.418 **	−0.552 **	0.543 **
	SO2	20 km	−0.032	0.461 **	−0.604 **	−0.614 **	0.724 **
		40 km	−0.345	0.711 *	−0.786 **	−0.726 *	0.835 **

Notes: ** and * mean significant correlations at the 0.01 and 0.05 levels (two-tailed), respectively.

).

From an overall LP perspective, high regularity, and low diversity of landscape patches, as well as high levels of landscape dominance and aggregation, led to high concentrations of NO₂, PM_10, and PM_2.5 and a low concentration of SO₂ in air. The results of this study showed the following attributes. (1) Increasing the irregularity of landscape shapes and landscape diversity, as well as reducing the levels of landscape dominance and aggregation, were effective means to reduce the atmospheric concentrations of NO₂, PM_10, and PM_2.5. (2) Increasing the levels of landscape dominance and aggregation was effective at reducing the atmospheric concentration of SO₂.

3.3.2. Coupling Relationships between the Concentrations of the Four Pollutants and the Six Landscape Types

For 2015, a significant negative correlation was found between the concentration of each of NO₂, PM_10, and PM_2.5 and the PD of farmlands, the absolute value of the coefficient of which increases with the landscape scale. In addition, significant positive correlations were apparent between the concentration of each of these three pollutants and the PLAND, LPI, and CA of farmlands, the coefficients of which increase with the landscape scale. The coefficient and significance of the correlation between the concentration of each of these three pollutants and the COHESION of farmlands decreased as the scale increased, with the coefficient turning from positive to negative at a GCSL of 40 km. The concentration of SO₂ was weakly negatively correlated with the PLAND, LPI, CA, and COHESION of farmlands, with low coefficients (

Table 5. Correlation coefficients between the patch-type-level landscape pattern metrics and the concentrations of the four pollutants at each scale.

Landscape Type	Year	Pollutant	GCSL	PD	PLAND	LPI	CA	PAFRAC	COHESION
Farmlands	2015	NO2	10 km	−0.237 **	0.235 **	0.243 **	0.280 **		0.274 **
		NO2	20 km	−0.412 **	0.262 *	0.296 **	0.329 **		0.203
		NO2	40 km	−0.473	0.774 **	0.805 **	0.337 *	−0.900	−0.477
		PM10	10 km	−0.324 **	0.366 **	0.368 **	0.432 **		0.389 **
		PM10	20 km	−0.421 **	0.319 **	0.347 **	0.425 **		0.253 *
		PM10	40 km	−0.444	0.751 **	0.780 **	0.414 *	−0.976	−0.499
		PM2.5	10 km	−0.380 **	0.448 **	0.449 **	0.476 **		0.449 **
		PM2.5	20 km	−0.472 **	0.449 **	0.476 **	0.496 **		0.309 **
		PM2.5	40 km	−0.467	0.731 *	0.765 **	0.476 **	−0.925	−0.319
		SO2	10 km	0.134 **	−0.126 **	−0.137 **	−0.138 **		−0.053
		SO2	20 km	0.171	−0.302 **	−0.295 **	−0.168 *		−0.254 *
		SO2	40 km	0.496	−0.581	−0.595	−0.205	−0.930	−0.488
Forests	2015	NO2	10 km	0.062	−0.238 **	−0.202 **	−0.247 **		−0.271 **
		NO2	20 km	0.120	−0.428 **	−0.333 **	−0.369 **	0.434	−0.556 **
		NO2	40 km	0.385	−0.667 *	−0.446	−0.439 **	0.992 **	−0.540
		PM10	10 km	0.186 **	−0.334 **	−0.312 **	−0.310 **		−0.310 **
		PM10	20 km	0.152	−0.429 **	−0.355 **	−0.354 **	−0.066	−0.460 **
		PM10	40 km	0.477	−0.663 *	−0.473	−0.324	0.913 *	−0.584
		PM2.5	10 km	0.253 **	−0.338 **	−0.324 **	−0.304 **		−0.316 **
		PM2.5	20 km	0.233 *	−0.424 **	−0.366 **	−0.353 **	0.035	−0.431 **
		PM2.5	40 km	0.475	−0.650 *	−0.453	−0.281	0.921 *	−0.650 *
		SO2	10 km	−0.058	0.109 *	0.081	0.098 *		0.249 **
		SO2	20 km	−0.160	0.167	0.232 *	−0.012	−0.382	0.225 *
		SO2	40 km	−0.188	0.574	0.432	0.037	−0.653	0.458
Grasslands	2015	NO2	10 km	−0.168 *	−0.394 **	−0.331 **	−0.333 **		−0.416 **
		NO2	20 km	−0.402 *	−0.667 **	−0.491 **	−0.491 **	−0.011	−0.679 **
		NO2	40 km	−0.536	−0.906 *	−0.770 *	−0.569 **	−0.116	−0.846 *
		PM10	10 km	−0.196 **	−0.326 **	−0.286 **	−0.260 **		−0.315 **
		PM10	20 km	−0.454 **	−0.632 **	−0.473 **	−0.402 **	0.220	−0.657 **
		PM10	40 km	−0.678	−0.944 **	−0.825 *	−0.504 **	−0.064	−0.818 *
		PM2.5	10 km	−0.188 **	−0.382 **	−0.327 **	−0.315 **		−0.371 **
		PM2.5	20 km	−0.384 *	−0.671 **	−0.504 **	−0.479 **	0.346	−0.696 **
		PM2.5	40 km	−0.701	−0.959 **	−0.824 *	−0.584 **	−0.095	−0.861 *
		SO2	10 km	0.009	0.189 **	0.156 *	0.077		0.268 **
		SO2	20 km	−0.211	0.178	0.155	0.008	−0.371	0.272
		SO2	40 km	−0.446	0.074	0.110	0.057	0.147	0.309
Water bodies	2015	NO2	10 km	−0.008	−0.271 *	−0.270 *	−0.156 *		−0.210
		NO2	20 km	0.201	0.050	−0.062	0.072		−0.009
		NO2	40 km	0.698	0.316	0.052	0.067	−1.00 **	−0.203
		PM10	10 km	0.117	−0.159	−0.231	−0.105		−0.132
		PM10	20 km	0.183	0.169	0.085	0.142		0.251
		PM10	40 km	0.521	0.226	0.200	0.265	−1.000 **	−0.051
		PM2.5	10 km	0.206	−0.109	−0.239 *	−0.056		−0.121
		PM2.5	20 km	0.245	0.353 *	0.262	0.221 *		0.418 **
		PM2.5	40 km	0.609	0.348	0.348	0.465 *	−1.000 **	0.027
		SO2	10 km	0.062	0.159	0.111	0.100		0.098
		SO2	20 km	−0.081	0.231	0.336 *	0.002		0.376 **
		SO2	40 km	−0.768 *	−0.410	0.223	0.386	1.000 **	0.497
Developed lands	2015	NO2	10 km	0.432 **	0.691 **	0.579 **	0.693 **		0.543 **
		NO2	20 km	0.677 **	0.851 **	0.783 **	0.771 **	0.058	0.609 **
		NO2	40 km	0.636	0.737 *	0.251	0.752 **	0.800	0.510
		PM10	10 km	0.444 **	0.670 **	0.553 **	0.669 **		0.532 **
		PM10	20 km	0.739 **	0.867 **	0.774 **	0.739 **	0.309	0.645 **
		PM10	40 km	0.819 *	0.772 *	0.079	0.748 **	0.617	0.367
		PM2.5	10 km	0.449 **	0.542 **	0.430 **	0.536 **		0.415 **
		PM2.5	20 km	0.706 **	0.736 **	0.632 **	0.568 **	0.534	0.600 **
		PM2.5	40 km	0.730	0.694	0.126	0.654 **	0.856	0.297
		SO2	10 km	0.255 **	0.361 **	0.283 **	0.341 **		0.314 **
		SO2	20 km	0.326 *	0.367 **	0.296 *	0.259 *	0.386	0.116
		SO2	40 km	0.129	−0.269	−0.417	0.226	−0.823	−0.483
Unused lands	2015	NO2	10 km	0.377	−0.513	−0.535	0.445		−0.356
		NO2	20 km				0.549
		NO2	40 km				1.000 **
		PM10	10 km	−0.261	−0.405	0.428	0.418		−0.211
		PM10	20 km				0.302
		PM10	40 km				−1.000 **
		PM2.5	10 km	−0.320	0.460	0.482	0.425		−0.139
		PM2.5	20 km				−0.182
		PM2.5	40 km				−1.000 **
		SO2	10 km	0.377	−0.513	−0.535	−0.504		−0.421
		SO2	20 km				−0.093
		SO2	40 km				−1.000 **
Farmlands	2018	NO2	10 km	−0.234 **	0.325 **	0.325 **	0.315 **		0.335 **
		NO2	20 km	−0.369 **	0.324 **	0.431 **	0.336 **		0.259 *
		NO2	40 km	−0.572	0.776 **	0.779 **	0.327 *	−0.974	−0.315
		PM10	10 km	−0.254 **	0.326 **	0.361 **	0.385 **		0.360 **
		PM10	20 km	−0.343 **	0.400 **	0.472 **	0.389 **		0.335 **
		PM10	40 km	−0.563	0.739 **	0.723*	0.353 *	−0.998 *	−0.024
		PM2.5	10 km	−0.215 **	0.285 **	0.278 **	0.316 **		0.355 **
		PM2.5	20 km	−0.224 *	0.241 *	0.324 **	0.327 **		0.242 *
		PM2.5	40 km	−0.517	0.608 *	0.573	0.326 *	−0.568	−0.023
		SO2	10 km	0.339 **	−0.494 **	−0.496 **	−0.470 **		−0.357 **
		SO2	20 km	0.454 **	−0.552 **	−0.592 **	−0.420 **		−0.379 **
		SO2	40 km	0.531	−0.744 **	−0.756 **	−0.282	0.733	0.043
Forestlands	2018	NO2	10 km	0.101 *	−0.286 **	−0.253 **	−0.336 **		−0.377 **
		NO2	20 km	0.007	−0.565 **	−0.461 **	−0.449 **	−0.145	−0.617 **
		NO2	40 km	0.550	−0.716 *	−0.527	−0.530 **	0.096	−0.690 *
		PM10	10 km	0.138 **	−0.303 **	−0.273 **	−0.306 **		−0.414 **
		PM10	20 km	0.088	−0.507 **	−0.419 **	−0.355 **	−0.255	−0.609 **
		PM10	40 km	0.529	−0.707 *	−0.520	−0.426 *	0.133	−0.627 *
		PM2.5	10 km	0.217 **	−0.320 **	−0.303 **	−0.270 **		−0.375 **
		PM2.5	20 km	0.058	−0.396 **	−0.328 **	−0.263 **	−0.274	−0.512 **
		PM2.5	40 km	0.508	−0.669 *	−0.489	−0.266	0.137	−0.417
		SO2	10 km	−0.210 **	0.304 **	0.281 **	0.342 **		0.435 **
		SO2	20 km	−0.266 *	0.470 **	0.409 **	0.399 **	−0.062	0.526 **
		SO2	40 km	−0.551	0.626	0.443	0.494 **	−0.101	−0.713 *
Grasslands	2018	NO2	10 km	−0.029	−0.482 **	−0.422 **	−0.410 **		−0.575 **
		NO2	20 km	−0.287	−0.680 **	−0.535 **	−0.564 **	0.193	−0.819 **
		NO2	40 km	−0.514	−0.887 **	−0.892 **	−0.676 **	−0.390	−0.850 **
		PM10	10 km	−0.060	−0.505 **	−0.437 **	−0.372 **		−0.571 **
		PM10	20 km	−0.188	−0.696 **	−0.558 **	−0.497 **	0.113	−0.789 **
		PM10	40 km	−0.413	−0.832 *	−0.839 **	−0.578 **	−0.427	−0.677 *
		PM2.5	10 km	−0.108	−0.485 **	−0.421 **	−0.315 **		−0.502 **
		PM2.5	20 km	−0.205	−0.686 **	−0.552 **	−0.392 **	0.193	−0.729 **
		PM2.5	40 km	−0.355	−0.823 *	−0.843 **	−0.447 *	−0.433	−0.293
		SO2	10 km	0.029	0.504 **	0.433 **	0.431 **		0.617 **
		SO2	20 km	0.118	0.693 **	0.554 **	0.626 **	−0.143	0.781 **
		SO2	40 km	0.480	0.883 **	0.873 **	0.766 **	0.484	0.932 **
Water bodies	2018	NO2	10 km	0.221	−0.149	−0.234	−0.410 **		−0.169
		NO2	20 km	0.344 *	0.102	−0.158	0.103		−0.038
		NO2	40 km	0.862 **	0.189	−0.366	0.141	−1.000 **	−0.413
		PM10	10 km	0.175	−0.178	−0.261 *	−0.372 **		−0.165
		PM10	20 km	0.272	0.058	−0.199	0.080		−0.029
		PM10	40 km	0.902 **	0.279	−0.366	0.107	−1.000 **	−0.332
		PM2.5	10 km	0.164	−0.037	−0.138	−0.315 **		−0.027
		PM2.5	20 km	0.248	0.149	−0.102	0.065		0.050
		PM2.5	40 km	0.911 **	0.366	−0.353	0.169	1.000 **	−0.199
		SO2	10 km	−0.240 *	0.156	0.279 *	0.431 **		0.185
		SO2	20 km	−0.377 **	−0.224	−0.032	−0.249 *		−0.080
		SO2	40 km	−0.858 **	−0.445	0.222	−0.261	−1.000 **	0.349
Developed lands	2018	NO2	10 km	0.240 **	0.689 **	0.626 **	0.662 **		0.613 **
		NO2	20 km	0.555 **	0.749 **	0.729 **	0.703 **	−0.567	0.640 **
		NO2	40 km	0.339	0.754 *	0.716 *	0.697 **	−1.000 **	0.039
		PM10	10 km	0.119	0.515 **	0.482 **	0.483 **		0.492 **
		PM10	20 km	0.328 *	0.578 **	0.559 **	0.503 **	−0.856	0.521 **
		PM10	40 km	0.105	0.661	0.767 *	0.484 *	−1.000 **	0.375
		PM2.5	10 km	0.010	0.560 **	0.544 **	0502 **		0.496 **
		PM2.5	20 km	0.230	0.631 **	0.647 **	0.514 **	−0.708	0.485 **
		PM2.5	40 km	0.048	0.617	0.797 *	0.488 *	−1.000 **	0.311
		SO2	10 km	−0.090	−0.029	−0.020	−0.092		−0.102
		SO2	20 km	−0.152	−0.097	−0.082	−0.137	0.985 *	−0.292 *
		SO2	40 km	−0.164	−0.516	−0.715 *	−0.191	1.000 **	−0.269
Unused lands	2018	NO2	10 km	0.724	0.931	0.841	0.296		−1.000 **
		NO2	20 km			0.729 **	0.265
		NO2	40 km				0.608
		PM10	10 km	0.763	0.950	0.808	0.233		−1.000 **
		PM10	20 km			0.559 **	0.127
		PM10	40 km				0.922
		PM2.5	10 km	−0.992	−0.967	−0.358	0.161		1.000 **
		PM2.5	20 km			0.647 **	−0.037
		PM2.5	40 km				0.999 *
		SO2	10 km	0.865	0.990	0.690	−0.283		−1.000 **
		SO2	20 km			−0.082	−0.090
		SO2	40 km				−0.359

Notes: ** and * mean significant correlations at the 0.01 and 0.05 levels (two-tailed), respectively.

). These results showed the following attributes. Increasing the level of fragmentation of farmlands, as well as reducing their relative area proportion and dominance (LPI), could reduce the concentrations of NO₂, PM_10, and PM_2.5. These effects strengthened as the scale increased. At a small landscape scale, reducing the level of connectivity of farmlands was also effective at reducing the concentrations of NO₂, PM_10, and PM_2.5. At a small landscape scale, reducing the level of connectivity of farmlands, as well as increasing their relative area proportion and dominance, could lower the concentration of SO₂.

For 2015, significant negative correlations were evident between the concentration of each of NO₂, PM_10, and PM_2.5 and the PLAND, LPI, CA, and COHESION of forests, the absolute values of the coefficients of which increased with the landscape scale. The concentration of each of these three pollutants was only significantly positively correlated with the PAFRAC of forests at a large scale (GCSL: 40 km). The concentration of SO₂ was negatively correlated with the PD of forests at the small landscape scale (GCSL: 10 km) and was significantly positively correlated with their PLAND, LPI, CA, and COHESION at the medium and small scales (GCSL: 10 and 20 km, respectively). These results showed the following attributes. Increasing the relative area proportion, dominance, area, and connectivity of forests, as well as reducing their complexity, could lower the concentrations of NO₂, PM_10, and PM_2.5. On a small scale, reducing the area proportion, dominance, area, and connectivity of forests could lower the concentration of SO₂.

For 2015, strong negative correlations were found between the concentration of each of NO₂, PM_10, and PM_2.5 and the PD, PLAND, LPI, CA, and COHESION of grasslands, the significance of which, however, decreased somewhat as the scale increased. At the small scales (GCSL: 10 and 20 km), the concentration of SO₂ was significantly positively correlated with the PLAND, LPI, and COHESION of grasslands, albeit with a small coefficient in each case (Table 5). These results showed the following features. Increasing the level of fragmentation, area proportion, dominance, area, and connectivity of grasslands could lower the concentrations of NO₂, PM_10, and PM_2.5. At small scales, reducing the area proportion and connectivity of grasslands could lower the concentration of SO₂.

For 2015, significant negative correlations could be found between the PLAND, LPI, CA, and COHESION of water bodies and the concentration of NO₂. At the large scale, the concentration of SO₂ was strongly negatively correlated with the PLAND of water bodies (Table 5). These results showed the following attributes. At a small scale, increasing the relative area proportion, dominance, area, and connectivity of water bodies could lower the concentration of NO₂. At a large scale (e.g., at a GCSL of 40 km), increasing the area proportion of water bodies could reduce the concentration of SO₂. The correlation between water bodies and each LP metric was very weak.

For 2015, strong positive correlations were apparent between the concentration of each of NO₂, PM_10, and PM_2.5 and the PD, PLAND, LPI, CA, and COHESION of developed lands, the significance of which was the highest at a GCSL of 20 km. At the small scale, the concentration of SO₂ was significantly positively correlated with the PD, PLAND, LPI, CA, and COHESION of developed lands. The correlations were the strongest at the medium scale (GCSL: 20 km), and their significance decreases as the scale increases (Table 5). These results showed that reducing the level of fragmentation, area proportion, dominance, area, and connectivity of developed lands were effective means to lower the concentrations of NO₂, PM₁₀, PM_2.5, and SO₂. The correlations between the concentrations of the four pollutants and each LP metric of unused lands were insignificant.

While basically similar patterns could be observed in the correlations between the concentrations of the four pollutants and each LP metric of each landscape type in 2018, the following changes in 2018 compared with 2015 were also evident. The correlation and significance of NO₂, PM_10, and PM_2.5 concentrations with various landscape indexes of cultivated land and forest land were slightly enhanced, which may be related to the conversion of cultivated land to forest land in 2010–2019 [8]. Under the policy of returning farmland to forest and grassland, the concentration of pollutants decreased rapidly while the cultivated land decreased and the forest land and grassland increased [19]. The correlation and significance between SO₂ concentration and patch density index of water area had been enhanced, indicating that the humidification effect of the water body [20] had a certain inhibitory effect on SO₂. The correlation between PM₁₀ and PM_2.5 concentrations and patch type area ratio index (Pland), maximum patch index (LPI), and landscape type area index (CA) of construction land decreased. On a large scale, the NO₂ concentration and the maximum patch index (LPI) of building land were significantly enhanced (Table 5), which may be due to the change in building density [21] which changed human activities and thus promoted the accumulation of surrounding pollutants.

4. Discussion

In this study, we have obtained the following conclusions by studying the impact of the urban landscape on pollutant concentration in different spatial and temporal scales of the CPUA.

Farmlands were the principal landscape type in the CPUA and were distributed mainly in its eastern part. Grasslands and forests were distributed primarily in the west, while developed lands were clustered predominantly in Chengdu. Water bodies accounted for an extremely low proportion of the total land area and were scattered across the study area. From 2015 to 2018, the proportion of cultivated land landscapes decreased, and the proportion of forests and developed lands increased. This change feature was contrary to the regulations on strictly controlling the conversion of farmlands to non-farmlands and minimizing the occupation of developed lands in the new regulations for the implementation of the land management law. However, the concentration of pollutants decreased. This phenomenon could be closely attributed to the efforts of the Chengdu Municipal Environment Protection Bureau during the 13th Five-Year Plan period to implement the guiding principles of the Sixth Plenary Session of the 12th National People’s Congress and adhere to green development by tackling pollution from domestic, industrial and aquaculture wastes, as well as stringently controlling emissions from key sources of pollution, with the improvement of environmental quality as the main theme. However, it was not excluded that the local complex meteorological, environmental conditions, and human activities cooperated with the surface cover to jointly affect the pollutants [20], and further research was required. In terms of the overall landscape pattern, from 2015 to 2018, the fragmentation and dominance of the CPUA increased, and the landscape pattern distribution became more discrete. This was due to the further development and outward extension of the new city at the edge of the central city, resulting in its complex internal structure and enhanced landscape heterogeneity and diversity. However, Chengdu has developed rapidly, with a high density and height of construction land, and there are many artificial emission sources, and it is not conducive to the diffusion of pollutants. Therefore, increasing the diversity and reducing the advantage and aggregation degree can reduce the concentration of NO₂, PM_10, and PM_2.5 in the air. However, SO₂ is mainly distributed in the West and the south. Firstly, the coal resources are mainly distributed in the West and the south. Secondly, it may also be that the development of the West and the south is not as fast as that of Chengdu, and the desulfurization technology in life and production is not perfect. Thirdly, there are few water areas in the West and the south, and the absorption effect of SO₂ is not obvious. Therefore, increasing the dominance and concentration of the landscape can reduce the concentration of SO₂ in the air. Guan [10] believed that increasing the regional landscape diversity and balancing the area ratio of each landscape type was an important means to reduce the concentration of pollutants, which was basically consistent with the results of this study.

The mechanism by which each landscape type affects pollutant concentrations was complex. In general, the concentrations of NO₂, PM_10, and PM_2.5 were significantly negatively correlated with the patch density index of cultivated land landscape types, which was consistent with previous research results [11], and the absolute value of the correlation coefficient increases with the increase of landscape scale, which indicates that the effect of landscape pattern optimization was more obvious in large scale than in small scale. We could reduce the concentration of NO₂, PM_10, and PM_2.5 by increasing the fragmentation degree of farmlands, reducing the relative proportion and dominance of area, increasing the area proportion, dominance, area, and connectivity of forests and grasslands, reducing the fragmentation degree of forests and reducing the area proportion, dominance, area, and connectivity of developed lands; The SO₂ concentration could be reduced by reducing the fragmentation of cultivated land, increasing the relative proportion and dominance of area and reducing the fragmentation, area ratio, dominance, area, and connectivity of developed land; The reason was that forests, grasslands, and other natural landscapes were the “sinks” of air pollution. The green patches with low fragmentation and high dispersion could cause the turbulence effect of air pollutants to drop, and the pollutants would gather in the green patches [11], thus absorbing particles through dry and wet deposition. Yang et al. [22] used the urban forest effect model to study the impact of the urban forest effect on air pollution. It showed that trees in central Beijing could remove 1261.4 t pollutants every year, most of which were particulate matter. Developed lands were the “source” of air pollution, and high-density buildings were not conducive to the diffusion of pollutants [23]. The farmland belongs to semi-natural landscape and its role in reducing pollution and dust is far inferior to the natural landscape. On a small scale (the scale side length is 10 km), the concentration of NO₂, PM_10, and PM_2.5 could be reduced by reducing the connectivity of farmlands. Because the farmland was a special land use type. On the one hand, as a part of urban green space [24,25], farmland can effectively reduce the concentration of pollutants through the dry and wet deposition of leaves. On the other hand, the dense smoke generated by straw burning after harvesting is the main reason for the high concentration of PM_2.5 and PM₁₀ [26]. Therefore, we should implement the cultivated land protection policy, protect the quantity and quality of cultivated land, reasonably control the increment of construction land [27], improve the land-use efficiency of construction land, and seriously implement the forest land and grassland protection policy and turn wasteland into high-quality ecological land. In addition, increasing the relative proportion, dominance, area, and connectivity of the water bodies could reduce the NO₂ concentration. On a large scale (the scale side length is 40 km), the water bodies had a significant effect on reducing SO₂ concentration. It may absorb NO₂ and SO₂ due to the humidification effect of the water bodies. The waters usually had an open surface, which was conducive to the diffusion of pollutants [28]. The coupling relationship between each pollutant and the landscape type index of farmlands, grasslands, forests, and developed lands was obvious, but the correlation with each landscape index of unused land was not significant, which may be because the area of the unused landscape is too small, so its effect is not obvious.

In the CPUA, natural (farmlands, grasslands, and water bodies) and semi-natural (farmlands) landscapes were all, to some extent, effective at reducing pollutant concentrations and were of the “pollution-reducing” type. Of these landscape types, grasslands were the most effective at reducing pollution. In contrast, artificial landscapes (i.e., developed lands) aggravated air pollution and were of the “pollution-promoting” type. Considering its current situation, a good landscape configuration for the CPUA could be created through the largest possible decrease in the proportion of the total area occupied by “pollution-promoting” landscape types and an increase in the proportion of the total area occupied by “pollution-reducing” landscape types. Completed developed lands could not be altered at a large scale but could be optimized through small-scale measures. For cultivated land, grassland, and forest land, large-scale optimization measures could be adopted to improve the combination of landscape types to balance the area of each landscape type, such as appropriately reducing the area of cultivated land and transforming it into forest land or water area.

This study used inverse distance interpolation to obtain the pollutant concentration value of each grid point, which is closer to the actual situation than Kriging interpolation and spline function interpolation. However, due to the limited number of automatic monitoring sites for the atmospheric environment in China, the monitoring sites were mainly concentrated in urban areas, and the interpolation results were still relatively rough. At the same time, considering the fluidity of the atmosphere and complex terrain conditions, the spatial transport of atmospheric pollutants was affected by multiple factors in the atmospheric environment, and the transport and diffusion characteristics of pollutants in different regions were also different [29]. Air pollution is regional, and more reasonable research zones need to be considered rather than a city alone. At the same time, the central city should strengthen the communication with the surrounding cities on landscape pattern planning, optimize the distribution of green landscapes through reasonable landscape configuration so as to minimize pollutant concentration, promote the sustainable development of urban landscape and maximize the benefits. In future research, we should not only increase the number of monitoring sites to improve the accuracy of existing pollution data but also conduct in-depth study on the research methods and theories of landscape ecology. Starting from the “source” and “sink” of pollution, in-depth research on rational planning of landscape to alleviate pollution in Chengdu. In addition, the height of buildings also has a great influence on the diffusion of pollutants. In future research, the height of buildings may be used as a landscape pattern.