Relationship between Green Space and Mortality in the Cities of the Yangtze River Delta Urban Agglomeration

: Intense work pressure and unhealthy lifestyles significantly threaten urban public health. Urban designs have quickly developed, such as the introduction of green space (GS), which has been suggested to improve public health. Prior epidemiological studies have investigated GS and mortality and have generally found potential benefits. However, these studies were primarily conducted in developed Western countries and the potential effects of GS on mortality in developing Asian countries are poorly understood. In this study, our goal was to investigate the effect of GS landscape attributes and socioeconomic conditions on all-cause mortality in 41 cities in the Yangtze River Delta urban agglomeration. Nine GS variables and seven socioeconomic variables were collected. An optimal general linear model with the selected variables was established using LASSO regression to explore the direction and relative importance of GSs and socioeconomic conditions for urban public health. The results showed that (1) socioeconomic conditions and GS jointly affect all-cause mortality, (2) people with greater personal wealth suffered less health risk, and (3) regularly shaped and highly connected GSs decreased the risk to public health. These findings suggest that reasonable GS policies and planning strategies are effective approaches for improving public health in Chinese cities.


Introduction
Urbanization has been the main trend in the era of human domination-in other words, the Anthropocene.Currently, 55% of the world's population lives in urban areas, and the urban population is expected to reach 70% by 2050 [1].With rapid urban expansion, the living environments and lifestyles of residents continue to change dramatically [2].Intense work pressure and unhealthy lifestyles significantly threaten urban public health [3,4].In response to this problem, urban health designs have quickly developed, such as the introduction of green space (GS), which has been suggested to improve public health [5].GS promotes physical activity, reduces mental stress, and increases social interaction [6].It also provides additional ecosystem services, such as air purification, heat island mitigation, and biodiversity maintenance [7].Although some studies have reported the health risks of GS, such as exposure to allergens, pesticides, or disease vectors [8,9], most evidence supports that exposure to GS benefits public health [10,11].
Most existing studies on the health effects of GS have been conducted at small scales, such as in public parks or neighborhoods, using indices that measure the surrounding greenness [12,13].The amount and accessibility of GS, plant diversity, canopy coverage, and green-looking ratio are commonly used to analyze users' behavior and mental activities in GS [14], and consequently guide the design of parks, gardens, and backyards.However, few large-scale studies have been conducted across cities, provinces, or states, the results of which may be important for the planning of GS across a wider range [15].Stakeholders, However, few large-scale studies have been conducted across cities, provinces, or states, the results of which may be important for the planning of GS across a wider range [15].Stakeholders, such as policymakers, urban managers, and urban planners, were more interested in the health effects of GS on a larger scale, but the absence of quantitative evidence from cross-city or cross-region studies hindered their implementation.Robust quantitative evidence from a cross-city analysis regarding GS and its health outcomes could facilitate reasonable and effective approaches, such as health impact assessment, and support public health policy making and application.
Mortality is a commonly used indicator of public health at city or regional scales.In China, more than 70% of deaths are caused by cardiovascular diseases and cancer [16].GS may support cardiovascular health in the general population via multiple pathways such as mitigating heat waves and pollution [17,18], improving sleep [19], and enriching outdoor activities and relative social and physical recreation [20,21].Prior epidemiological studies have investigated GS and mortality and have generally found potential benefits.However, these studies were primarily conducted in developed Western countries [10].The potential effects of GS on mortality in developing Asian and African countries are poorly understood.To fill this knowledge gap, the objectives of the present study were to analyze the relationship between the landscape attributes of GSs and the all-cause mortality of cities in the Yangtze River Delta urban agglomeration (YRDUA), China.We also explored the relative contributions of socioeconomic conditions and GS to urban mortality.Our goal was to offer data-driven recommendations for the planning and management of GS to promote public health in Chinese cities.

Study Region
The YRDUA is China's most economically developed region, with the highest urbanization level in East China.The YRDUA is located at the range of 27°09′ N~35°20′ N and 114°54′ E~122°12′ E, which comprises 41 cities of the three provinces of Jiangsu, Zhejiang, Anhui, and one municipality of Shanghai (Figure 1, Table S1).Its area is 358,000 km 2 with a more than 60% urbanization rate.The YRDUA has a subtropical monsoon climate.Its typical topographical pattern is high in the south with mountains and hills, and low in the north with plains.The total GDP of the YRDUA in 2023 exceeded 30.51 trillion (RMB), accounting for about 24% of the total national GDP.Among them, the GDP of Shanghai, Jiangsu, Zhejiang, and Anhui reached 4.72 trillion, 12.82 trillion, 8.26 trillion, and 4.71 trillion RMB, respectively.In this study, we analyzed the relationship between all-cause mortality and GSs in 41 cities of the YRDUA.[22].The raw data were reclassified into cropland, GS, water bodies, built-up areas, and bare land.Based on the reclassified LULC data, nine landscape indices of GSs (Table 1) were calculated using FRAGSTATS v4.2 (produced by McGarigal K., Cushman S.A., and Ene E.; available at https://fragstats.org/, accessed on 14 June 2014).All-cause mortality data were obtained from the China Health Statistics Yearbook.The mortality rates of the 41 cities in 2018 were selected to avoid the influence of COVID-19 from 2019 to 2023.Then the mortalities were classified into five levels: critical (≥1%), high (0.8%-1%), medium (0.6%-0.8%), low (0.4%-0.6%), and mild (<0.4%).We also collected seven socioeconomic variables (see Table 2) from the China Urban Statistical Yearbook.16 variables, we constructed two independent datasets involving socioeconomic and GS variables and applied them to a general linear model (GLM) to test their effects on all-cause mortality.GLM was mostly considered as a correlation analysis rather than a cause-effect analysis, which may misestimate the actual effect of multiple factors.However, in this study, it can be treated as a quasi-cause-effect analysis because it was confirmed that GS attributes and socioeconomic conditions were important in influencing mortality, as proved by many previous studies [4,10,12].Furthermore, the relevant mortality-related variables in the YRDUA were selected using a penalized regression, the LASSO regression by the R package {glmnet} [23,24], to optimize the model.The relative contributions of GS and socioeconomic conditions were measured by variation partitioning using the R package {vegan} [25].The analysis framework is shown in Figure 2.

Spatial Pattern of All-Cause Mortality
The results of global Moran's I showed that there was no significant autocorrelation of these cities' mortalities (I = 0.04, p = 0.543).All-cause mortality in the YRDUA showed a spatial pattern of higher levels in the northwest and lower levels in the east (Figure 3).All-cause mortalities in 16 cities in the YRDUA were at a medium level.The cities with critical and high mortality levels are located at the western and northern edges of the YR-

Spatial Pattern of All-Cause Mortality
The results of global Moran's I showed that there was no significant autocorrelation of these cities' mortalities (I = 0.04, p = 0.543).All-cause mortality in the YRDUA showed a spatial pattern of higher levels in the northwest and lower levels in the east (Figure 3).All-cause mortalities in 16 cities in the YRDUA were at a medium level.The cities with critical and high mortality levels are located at the western and northern edges of the YRDUA, mostly in Anhui Province.Cities with low and mild mortality levels are located in the eastern part of the YRDUA, which is better developed, such as Suzhou, Hangzhou, Ningbo, and Shanghai.Fuyang in Anhui Province was the city with the highest all-cause mortality (1.62%), whereas Suzhou in Jiangsu Province, which is famous for its many Chinese classical gardens, was the lowest (0.39%).

Spatial Variation of GS in the YRDUA
In terms of the proportion of LULC types, croplands were the main LULC type in the YRRUA, followed by GSs, built-up areas, and water bodies (Figure 4).A relatively higher ratio of water bodies was observed in cities in northern and southern Zhejiang than in other cities.As expected, Shanghai has the highest percentage of built-up areas.The cities in Zhejiang Province had the largest GS PLAND (Figure 5a), which also had the largest LPIs (Figure 5b).
Detailed values of the landscape attributes of GS across these cities in the YRDUA are listed in Table S1.LSI was an important indicator for assessing the shape complexity of GSs in this study.A higher LSI indicates a more irregular shape, leading to stronger edge effects that affect the GSs.The cities with higher LSIs were mainly located at the center and southeast of the YRDUA (Figure 5c) because of higher urbanization and artificial modification.The cohesion index assessed the aggregation of GS patches.A high cohesion value indicates high connectivity between GS patches and less fragmentation.The spatial pattern of the cohesion index was similar to that of PLAND, being higher in the south and lower in the north of the YRDUA (Figure 5d), which may be related to topography.The southern part of the YRDUA is a hilly area suitable for forests, and the northern part con-

Spatial Variation of GS in the YRDUA
In terms of the proportion of LULC types, croplands were the main LULC type in the YRRUA, followed by GSs, built-up areas, and water bodies (Figure 4).A relatively higher ratio of water bodies was observed in cities in northern and southern Zhejiang than in other cities.As expected, Shanghai has the highest percentage of built-up areas.The cities in Zhejiang Province had the largest GS PLAND (Figure 5a), which also had the largest LPIs (Figure 5b).
Detailed values of the landscape attributes of GS across these cities in the YRDUA are listed in Table S1.LSI was an important indicator for assessing the shape complexity of GSs in this study.A higher LSI indicates a more irregular shape, leading to stronger edge effects that affect the GSs.The cities with higher LSIs were mainly located at the center and southeast of the YRDUA (Figure 5c) because of higher urbanization and artificial modification.The cohesion index assessed the aggregation of GS patches.A high cohesion value indicates high connectivity between GS patches and less fragmentation.The spatial pattern of the cohesion index was similar to that of PLAND, being higher in the south and lower in the north of the YRDUA (Figure 5d), which may be related to topography.The southern part of the YRDUA is a hilly area suitable for forests, and the northern part consists of plains that are suitable for cropland and urban construction.

Socioeconomic Conditions of the Cities in the YRDUA
Detailed values of the socioeconomic conditions across these cities in the YRDUA are listed in Table S2.The population of the YRDUA reached 233.52 million, with 9559 km 2 of urban built-up land and an RMB 21,854.18 billion GDP.Shanghai is the largest city with the largest population and the highest GDP.The highest GDP per capita was found in Wuxi, Jiangsu (174.27thousand RMB), and the lowest was in Fuyang, Anhui (21.59 thousand RMB) (Figure 6a).Consumption expenditure per capita (CEpc) indirectly represents personal wealth.The highest CEpc was found in Shanghai (RMB 86.66 thousand), and the lowest was found in Suzhou, Anhui (RMB 8.44 thousand) (Figure 6b).
The number of participants with basic medical insurance, hospital beds, and registered doctors represented cities' public health levels.Public willingness to participate in medical insurance was high in the east and low in the west (Figure 6c).The number of participants in the cities of Anhui Province was lower than those in Shanghai, Zhejiang, and Jiangsu Provinces.However, medical resources such as hospital beds and doctors did not show a regular spatial pattern (Figure 6d).The capital cities of every province and the richest cities had more medical resources.The groups of the values in the legends were based on the natural breaks' classification.

Socioeconomic Conditions of the Cities in the YRDUA
Detailed values of the socioeconomic conditions across these cities in the YRDUA are listed in Table S2.The population of the YRDUA reached 233.52 million, with 9559 km 2 of urban built-up land and an RMB 21,854.18 billion GDP.Shanghai is the largest city with the largest population and the highest GDP.The highest GDP per capita was found in Wuxi, Jiangsu (174.27thousand RMB), and the lowest was in Fuyang, Anhui (21.59 thousand RMB) (Figure 6a).Consumption expenditure per capita (CEpc) indirectly represents personal wealth.The highest CEpc was found in Shanghai (RMB 86.66 thousand), and the lowest was found in Suzhou, Anhui (RMB 8.44 thousand) (Figure 6b).
The number of participants with basic medical insurance, hospital beds, and registered doctors represented cities' public health levels.Public willingness to participate in medical insurance was high in the east and low in the west (Figure 6c).The number of participants in the cities of Anhui Province was lower than those in Shanghai, Zhejiang, and Jiangsu Provinces.However, medical resources such as hospital beds and doctors did not show a regular spatial pattern (Figure 6d).The capital cities of every province and the richest cities had more medical resources.

Effects of GS Attributes and Socioeconomic Conditions
Pearson's correlation analysis showed that socioeconomic variables were strongly negatively correlated with all-cause mortality; however, only a few GS variables worked (Figure 7).Socioeconomic variables were also positively correlated with each other but were less correlated with GS attributes.
After standardization of the 16 variables, we constructed the original GLM involving socioeconomic and GS variables.To avoid the influence of collinearity between these variables, LASSO regression was applied to select the relevant variables driving all-cause mortality in the cities in the YRDUA for subsequent analysis (Figure 8).The final optimized GLM model contained two socioeconomic variables and two GS variables: CEpc, DOCT, LSI, and COHESION (Table 3).CEpc and COHESION had significant negative effects on all-cause mortality, whereas LSI had a significant positive effect.The DOCT effect was not statistically significant in the final model.Compared with the original model involving all 16 variables (Figure 9a), the relative contributions of socioeconomic factors and GS increased in the optimal models, as well as the overall explanatory power (Figure 9b).Socioeconomic conditions were the dominant drivers of all-cause mortality (31% independent and 14% joint contributions), as was our expectation.However, GS also played an important role in all-cause mortality (11% independent and 14% joint contributions).

Effects of GS Attributes and Socioeconomic Conditions
Pearson's correlation analysis showed that socioeconomic variables were strongly negatively correlated with all-cause mortality; however, only a few GS variables worked (Figure 7).Socioeconomic variables were also positively correlated with each other but were less correlated with GS attributes.
After standardization of the 16 variables, we constructed the original GLM involving socioeconomic and GS variables.To avoid the influence of collinearity between these variables, LASSO regression was applied to select the relevant variables driving all-cause mortality in the cities in the YRDUA for subsequent analysis (Figure 8).The final optimized GLM model contained two socioeconomic variables and two GS variables: CEpc, DOCT, LSI, and COHESION (Table 3).CEpc and COHESION had significant negative effects on all-cause mortality, whereas LSI had a significant positive effect.The DOCT effect was not statistically significant in the final model.Compared with the original model involving all 16 variables (Figure 9a), the relative contributions of socioeconomic factors and GS increased in the optimal models, as well as the overall explanatory power (Figure 9b).Socioeconomic conditions were the dominant drivers of all-cause mortality (31% independent and 14% joint contributions), as was our expectation.However, GS also played an important role in all-cause mortality (11% independent and 14% joint contributions).

Discussion
The longitudinal studies in the US, Canada, Australia, Spain, and Italy all found tha the high-quality GS reduced local mortalities [26][27][28][29][30]. Based on studies that included 41 cities in the YRDUA, the richest region in China, we found lower mortality in cities with high connective GS and richer residents.These results suggest that reasonable GS policies and planning strategies to increase GS connectivity are effective approaches for improving public health.

Discussion
The longitudinal studies in the US, Canada, Australia, Spain, and Italy all found that the high-quality GS reduced local mortalities [26][27][28][29][30]. Based on studies that included 41 cities in the YRDUA, the richest region in China, we found lower mortality in cities with high connective GS and richer residents.These results suggest that reasonable GS policies and planning strategies to increase GS connectivity are effective approaches for improving public health.

Spatial Pattern of Mortality in the YRDUA
There was obvious spatial variation in all-cause mortality in the YRDUA.The southern region had a lower health risk than the northern region.This spatial heterogeneity may be because of differences in lifestyles, socioeconomic conditions, and environmental quality.A light diet with more seafood and vegetables in southern cities decreased the risk of cardiovascular disease [31].Better economic and social development provide more nutrients, diverse foods, and sufficient medical resources.Because of the variation in topography, there were more forests in the southern cities and more croplands in the north.Forests provide better ecosystem services, such as purified air and water pollution, and regulate the climate, which improves local environmental quality [32].

Personal Wealth Is Strongly Correlated with Mortality
Socioeconomic conditions were the dominant factors influencing all-cause mortality in the cities in the YRDUA.In our study, consumption expenditure per capita, representing personal wealth, was found to drive city-level mortality.A study covering 12 provinces in China supported that the resident groups with low individual-level integrated socioeconomic status had a 65% higher risk of all-cause mortality [33].High-income residents were willing to perform physical exercise, live in greener habitats, and prefer a healthy diet and lifestyle.They also have easy access to better medical and healthcare resources.Most low-income families cannot afford long-term medications for chronic diseases, such as cardiovascular diseases.Based on this finding, the development of targeted policies should be encouraged to improve healthcare accessibility, equally benefit people, and reduce the mortality of cardiovascular disease and cancer in people with low incomes [11], especially in Anhui Province, the western part of the YRDUA.Some studies have also reported that regional-level socioeconomic conditions such as medical resources [34] and GDP [35] had no significant effects in our study.The YRDUA is the most urbanized region in China, with a well-established transportation infrastructure [36].People living in small cities and towns easily share the abundant medical resources of the surrounding large cities.However, the cities in the YRDUA are better developed than most others in China [15], although there are economic gaps between them.Public health is less limited by urban development in the YRDUA.We speculate that the sharing of medical resources and the high level of development weakened the effects of city-level socioeconomic variations.

Landscape Configuration of GS Affected Mortality
Our results also showed that GS played an important role in all-cause mortality, although personal-level socioeconomic conditions were the dominant drivers.In previous studies, most GS indices, such as the amount, NDVI, accessibility, and area ratio, were related to mortality [10,37,38].Beyond our expectations, we found that the indices of landscape configuration, LSI, and COHESION were more important for all-cause mortality than landscape composition in the YRDUA.
A higher LSI increases the health risk caused by GS overexposure.The complex boundary of the GS increases the interface between the natural and human worlds, introducing more negative effects in human society, such as epidemic pathogens, allergens, and wild animal attacks.A higher LSI indicates intense GS fragmentation.The fragmented GS patches are too small to provide sufficient ecosystem services and health benefits [39].
The COHESION values negatively correlated with all-cause mortality.Higher GS connectivity leads to a lower risk of all-cause mortality, which is also consistent with the finding of stronger fragmentation of GS associated with poor health in the Taipei metropolitan area because of the GS replacement by heavy industry [40].A study of US cities also found that neighborhoods with more aggregated GS morphologies had a lower prevalence of non-communicable diseases [41].More connected GS forms a GS network that provides more ecological services, such as improving air quality, reducing the heat island effect, and increasing opportunities for outdoor activities and social communication, which, in turn, improves physical and mental health [42].Larger connected GSs also provide sufficient space for thematic events and exhibitions to attract residents.Some previous case studies reported the importance of the amount of GS for urban public health and mortality at the neighborhood scale [10].In our studies, the CA and PLAND of GS, which represented GS amount, were not found to be significantly correlated with city-scale mortality in the YRDUA.We speculated that the users of large integrated parks and country parks, such as the Hangzhou Westlake Scenic Area, are mainly from outside of this region, which provided less health benefits to local residents.Another reason is that the large and dense GS would reduce the willingness of local users to physically exercise, especially single visitors who consider potential criminal risks [43,44].European studies also supported that neighboring GS was better for the health of local residents than the total GS amount in the city, and reduced the natural-cause and suicide mortality [45,46].

Study Limitations
Although all-cause mortality was the most accurate, objective, and accessible data for urban public health measures, the effect of green space on its components, such as cardiovascular mortality, cancer mortality, infectious disease mortality, or accidental mortality varied widely.How to distinguish the specific effects of green space on different causes of death is a key question for urban public health management that was not answered in the study.A long-term longitudinal social survey is necessary to investigate this issue [47].Another limitation was that the study did not consider non-fatal health risks, such as subhealth conditions, mental illness, and chronic diseases.These non-fatal health risks are more prevalent in the public but are difficult to identify and measure.Large-scale demographic and health censuses [10] and international and interdisciplinary collaborations [7,11] will illustrate the detailed effects of green space on overall public health.

Conclusions
The present study of the cities in the YRDUA found that the all-cause mortality showed a spatial pattern of "high in the south and low in the north".Socioeconomic conditions and GS jointly affect all-cause mortality.Regarding socioeconomic conditions, people with greater personal wealth had fewer health risks.Regarding the health effects of GSs, regularly shaped and connected GSs would decrease the risk to public health.Based on these findings, we suggest that public health policies consider the equity of healthcare resources when caring for poor families in cities.The construction of a GS network should also be included in urban planning.

Supplementary Materials:
The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/f15061066/s1,Table S1: Details of landscape attributes of green space of the cities in the YRDUA; Table S2: Details of socioeconomic condition of the cities in the YURDA.

Figure 2 .
Figure 2. Brief analysis framework of this study.

Figure 2 .
Figure 2. Brief analysis framework of this study.

Figure 3 .
Figure 3. Spatial pattern of all-cause mortalities in the YRDUA.

Figure 4 .
Figure 4. Area ratio of LULC types across the cities in the YRDUA.

Figure 4 . 16 Figure 5 .
Figure 4. Area ratio of LULC types across the cities in the YRDUA.

Figure 5 .
Figure 5. Landscape attributes of green space in the YRDUA.(a) PLAND; (b) LPI; (c) LSI; (d) COHESION.The groups of the values in the legends were based on the natural breaks' classification.

ForestsFigure 6 .
Figure 6.Socioeconomic conditions in the YRDUA.(a) GDP per capita; (b) consumption expenditure per capita; (c) number of participants in basic medical insurance per 1000 people; (d) number of doctors per 1000 people.The groups of the values in the legends were based on the natural breaks' classification.

Figure 6 .
Figure 6.Socioeconomic conditions in the YRDUA.(a) GDP per capita; (b) consumption expenditure per capita; (c) number of participants in basic medical insurance per 1000 people; (d) number of doctors per 1000 people.The groups of the values in the legends were based on the natural breaks' classification.

ForestsFigure 7 .
Figure 7. Pearson's correlation analysis between all-cause mortality and the green space and socioeconomic variables.The sizes of the squares in the grid indicated the absolute values of the correlation coefficient.* p < 0.05; ** p < 0.01; *** p < 0.001

Figure 8 .
Figure 8. Variable selection based on the LASSO penalized regression.The dashed line indicates the optimal lambda value in cross-validation.

Figure 7 . 16 Figure 7 .
Figure 7. Pearson's correlation analysis between all-cause mortality and the green space and socioeconomic variables.The sizes of the squares in the grid indicated the absolute values of the correlation coefficient.* p < 0.05; ** p < 0.01; *** p < 0.001.

Figure 8 .
Figure 8. Variable selection based on the LASSO penalized regression.The dashed line indicates the optimal lambda value in cross-validation.

Figure 8 .
Figure 8. Variable selection based on the LASSO penalized regression.The dashed line indicates the optimal lambda value in cross-validation.

Figure 9 .
Figure 9. Independent and joint effects of green space and socioeconomic conditions to all-cause mortality using variation partition.(a) The original model with 16 variables; (b) the optimal mode with four variables.

Figure 9 .
Figure 9. Independent and joint effects of green space and socioeconomic conditions to all-cause mortality using variation partition.(a) The original model with 16 variables; (b) the optimal model with four variables.

Author Contributions:
Conceptualization, Y.W. and G.H.; Methodology, M.L. and G.H.; Validation, G.H.; Formal analysis, M.L. and G.H.; Resources, Y.W. and G.H.; Data curation, M.L.; Writing-original draft preparation, M.L.; Writing-review and editing, M.L., Y.W. and G.H.; Visualization, M.L. and G.H.; Supervision, G.H.; Project administration, G.H.; Funding acquisition, Y.W. and G.H. All authors have read and agreed to the published version of the manuscript.Funding: This work was supported by the National Natural Science Foundation of China (32171570), the Medical Science and Technology Project of Zhejiang Province (2023KY1002), and the Hangzhou Key Medicine Discipline Fund for Hygienic Microbiology Laboratory.Data Availability Statement: The datasets used in this study are available in open databases listed in the article.
2.2.Data Collection 2.2.1.Landscape Attributes of GS Raw land use and land cover (LULC) data in 2018 with 30 m resolution were obtained from the annual China Land Cover Dataset

Table 1 .
Landscape indices of the green spaces of these cities in the YRDUA.

Table 2 .
Socioeconomic variables of these cities in the YRDUA.

Table 3 .
Effects of the selected variables on the all-cause mortality in the optimized GLM model.

Table 3 .
Effects of the selected variables on the all-cause mortality in the optimized GLM model.