Introduction

China has undergone rapid urbanization following economic reform policies since the 1980s. Statistics indicate that ten million migrants a year have moved into large cities from the rural areas of China since then, which may be the largest migration in human history (NBSC 2014). However, problems from environmental pollution have emerged as a result of economic and population growth in cities. According to the literature, financial expenditure on degraded air quality in China added up to roughly 6.5% of the Chinese GDP annually from 2000 to 2010 (Crane and Mao 2015). In addition, the rapidly diversifying demands of urban residents for public landscapes are ever increasing (Fry 2001). Urban green spaces, which refer to non-private, freely accessible outdoor areas with amenities within urban limits (Wendel et al. 2012), have attracted attention for their important and diverse roles in providing ecosystem services and recreational opportunities. Urban green spaces are capable of filtering air, attenuating noise, cooling temperatures, and replenishing groundwater, which contribute to pollution abatement and livability in metropolitan areas. Research has also suggested that the existence of urban green spaces can deliver places for recreation, leisure, and promote social interactions (Li and Liu 2016). In China, the Gardening and Greening Bureau adopted urban green space system planning since 2002 (MHUDC 2002a, b) and a series of policies to introduce green elements into built-up areas. The Ministry of Housing and Urban-Rural Development of China proposed the construction of National Garden City and National Ecological Garden City to encourage urban greening. The corresponding standards were issued and improved to establish the base requirement of urban green space ratio of 31% and 35% (MHUDC 2016). These planning and policies have significant effects in China.

There has been discussion in the literature about the connection between the economic development and urban green spaces. Many scholars have studied the potential impacts of per capita GDP (Richards et al. 2017), built-up area (Yang et al. 2014), percentage of industry (Wang et al. 2016a, b) and urbanization rate (Liu et al. 2016), and population densities (Li et al. 2017) on urban greening. However, academic opinions are divided on whether the correlation is positive or negative. For instance, some surveys conclude that the rapidly advancing process of urbanization and economic growth has brought about the gradual diminishment of green spaces (Vos and Meekes 1999). Nevertheless, we can also find suggestions that the expansion of built-up areas caused by economic development may indirectly lead to the increase of urban green coverage (Yang et al. 2014).

The idea that economic growth promotes the development of urban green spaces can be embodied in the relationship between indexes such as GDP per capita (Richards et al. 2017) and urban green coverage rate (Fuller and Gaston 2009). Numerous studies have illustrated that there may be a positive correlation between GDP per capita and the area of urban green space. For instance, it was found that cities with a higher GDP per capita had a higher percentage area of green space in Singapore (Richards et al. 2017), because wealthier cities have tended to adopt innovative architectural techniques for enhancing and protecting urban green spaces (Jim 2004). Well-planned cities like Tampines in Singapore integrate high-density housing with surrounding parks, guaranteeing easier access to urban green spaces despite the current high population densities (Seik 2001). Moreover, many studies illustrate that urban area is positively correlated with the development of green spaces. It can be inferred from spatial statistics that more green spaces have been constructed in cities with a higher level of economic development during the last two decades in China. The major driving force behind this is support from the national government (Zhao et al. 2013). Similar phenomenon can also be found in Europe (Fuller and Gaston 2009).

Other researchers contend that economic growth restrains the development of urban green spaces, which can be reflected in the negative correlation between built-up areas, population densities, and urban green coverage rates. For instance, intensive land use change was associated with a decrease in green coverage in Kunming City. This trend occurred in the 1990s, resulting in the particularly rapid loss of green spaces in the northeast, east, and west in Kunming (Zhou and Wang 2011). Recently, further research has been undertaken on the relationship between economic growth and loss of green spaces. It was observed that the decrease of green space was greatly influenced by the amount of built-up area around it. Thus, a shortage of available land caused a reduction in urban public green spaces (Bardhan et al. 2016). In metropolitan areas, infill development of urbanization caused by economic growth was found to be another major reason for green space loss in 11 case study sites in Merseyside, UK (Pauleit et al. 2005). Surveys have also suggested a negative correlation between population densities and the development of green spaces. Evidence has shown that cities with higher population densities have less green space and less public green area per capita in Europe, because more people have been housed in the urban matrix and buildings have replaced former green areas (Fuller and Gaston 2009).

However, Chinese cities are quite different from European and American cities in terms of urban development and public green spaces. Due to limited resources, most of the previous studies have been conducted in a few cities, such as Beijing (Liu et al. 2011), Hangzhou (Zhang et al. 2013), Catania (Rosa et al. 2017), Metropolitan Area of Santiago (Barrera et al. 2016), and Greater Dhaka (Byomkesh et al. 2012). From these studies, we can conclude that the relationship between economic growth and urban green spaces differs at different scales. Nevertheless, current studies have concentrated on quantitative relationships, ignoring spatial imbalances and consequently failing to address spatial dependence and heterogeneity (Liu et al. 2017). China is a very large country, leading to spatial heterogeneity between the different regions. Therefore, it is of value to consider the association between economic growth and urban green spaces in different regions of China, combining spatially varying relationships with quantitative relationships.

Previous studies have tended to be highly specific, considering only limited aspects of the relationship between economic growth and urban public green spaces. A comprehensive evaluation of this kind of relationship is required. Thus, in the present study, considering the management and control efforts, the accessibility to data, and the diversity of urban green development in the perspective of China statistics condition, we collected records on the urban green ratio, urban green coverage rate, and public green area per capita and indexes concerned with economic growth for 289 Chinese prefecture-level cities in 2014. We posed the following research questions in relation to these data: (1) what is the spatial pattern of urban greening in China at the city level; (2) how can we comprehensively evaluate the influence of economic growth and quantitatively identify the impact of significant variables on green space indexes; and (3) to what extent does the spatial contribution made by various economic factors account for variations in green space indexes? The results from this study, we argue, could constitute a valuable reference for mid-to-long-term green space planning policy in diverse parts of China, and could further assist in coordinating the development of urban green spaces and economic growth.

Data

Globally, different countries may use a variety of indexes to embody urban greening at a national or urban scale. For instance, research in North America uses the quantity of green space as an index, which focuses on the specific area of green spaces (Beninde et al. 2015). Countries in Southeast Asia, such as Singapore, prefer to aggregate green spaces to measure their development, which focuses on the total area of green spaces regionally (Richards et al. 2017). In China, the green coverage rate, public green area per capita, and urban green space ratio in urban areas are important indexes in urban green space system planning, management and indicating the difference in green space development (Zhou and Wang 2011). Urban green space ratio is defined as the percentage of built-up areas covered by vegetation, while the green coverage rate refers to the ratio of all growing planting projection to the built-up areas. Besides, public green area per capita represents the ratio of urban public green area to quantity of urban population. The municipal cities in China often consist of urban area and large areas of countryside, and the urban area covers the built-up area and large part of natural landscape (Fig. 1). The built-up area of the city was statistical boundary in the calculation of urban greening by the National Bureau of Statistics of China (MHUDC 2002a, b). Recreational use of green spaces is extremely important problems for urban areas according to Classification Standard of Urban Green Space in China; thus, all the public green spaces in built-up area and only few areas of urban forests and farms in natural landscapes could be counted into green space calculation. Therefore, green spaces in the study are mostly constructed and modified such as comprehensive parks, botanical gardens, community parks, sporting fields, or golf courses in built-up area of the urban city.

Fig. 1
figure 1

Structure of a city in China: Qian’an as an example. The boundary of urban area in municipal Qian’an (a) and green space in urban Qian’an (b)

Full size image

Based on previous research, we take GDP per capita as the explanatory variable and use the other economic indexes as control variables to study the relationship between economic indicators and urban green spaces. Five economic indicators, namely, the urbanization rate, per capita GDP, percentage of secondary industry, population density, and urban land area, were selected as variables. To explore the interactive relationships between city greening and economic performance, per capita GDP was defined as the explanatory variable, while the others were control variables (Table 1). Data is all based on 2014 emission inventory data from 289 cities.

Table 1 Descriptive statistics of the independent variables
Full size table

Multicollinearity in a regression model represents a state of very high association among the independent variables which may lead to inaccurate analysis, generating misleading estimates of coefficients and standard errors (Lozano and Gutiérrez 2008). We used the variance inflation factor (VIF) to detect whether multicollinearity and correlation existed between the variables. The VIF between the urban green space ratio and urban greening coverage was above 10, indicating a strong correlation. Six variables were finally selected. The urbanization rate, per capita GDP, percentage of secondary industry, and urban land area were the explanatory variables representing the urban economic condition, while the urban green space ratio and public green area per capita were explanatory variables for the urban greening situation. Most of the Spearman correlations were significant at the 5% significance level (Table 2).

Table 2 Spearman’s correlation of the independent variables
Full size table

Methods

We have proposed and applied a set of methods to quantify variables that explain economic growth and urban public green spaces, including a spatial error model (SEM), spatial autoregression (SAR), and geographically weighted regression (GWR). These allow us to estimate the impact and spatial variations of China’s urbanization process on urban greening. The use of GWR is a suitable approach to investigate the relationships between the urban public green spaces and economic growth, since it can generate a set of local parameter estimates to explore the relationships between variables and spatial patterns (Fotheringham 2001).

Global and local Moran’s I

As the basis of linear regression in spatial econometrics, spatial autocorrelation is explained by Tobler’s first law of geography. Therefore, as a spatial variable, the greening index is somewhat dependent to the surrounding units and tends to concentrate in specific spaces.

Such spatial correlation can be identified by the spatial autocorrelation test, a method to validate the effectiveness of spatial econometrics models. Here, global Moran’s I and local Moran’s I were selected to test whether the greening index exhibited significant spatial autocorrelation overall and locally. Global Moran’s I is formally given by Eqs. 1 and 2:

$$ Global\kern0.5em I=\frac{\sum \limits_{i=1}^n\sum \limits_{j=1}^n{W}_{ij}\left({x}_i-\overline{x}\right)\left({x}_j-\overline{x}\right)}{S^2\sum \limits_{i=1}^n\sum \limits_{j=1}^n{W}_{ij}}=\frac{n\sum \limits_{i=1}^n\sum \limits_{j=1}^n{W}_{ij}\left({x}_i-\overline{x}\right)\left({x}_j-\overline{x}\right)}{\sum \limits_{i=1}^n\sum \limits_{j=1}^n{W}_{ij}\sum \limits_{i=1}^n{\left({x}_i-\overline{x}\right)}^2} $$
(1)
$$ Z=\frac{I-E(I)}{\sqrt{VAR(I)}}=\frac{\sum \limits_{j\ne i}^n{W}_{ij}(d)\left({x}_j-\overline{x}\right)}{S_i\sqrt{W_i\left(n-1-{W}_i\right)}/n-2}j\ne i $$
(2)

where n is the number of cities; xi and xj are the greening indexes of spatial locations i, j, respectively; and S is the standard deviation of samples. Wij denotes the means of a spatial weights matrix, representing the topology or spatial arrangement of the data and our understanding of spatial association among all unit areas (Fischer and Getis 2011).

Global Moran’s I ranges from − 1 to 1. A higher positive value of I refers to direct spatial correlation, while a significantly negative value indicates inverse spatial correlation. If the value of I is close to zero, this represents the absence of spatial correlation.

The significance of spatial autocorrelation can be detected by the value of Z, representing its standard deviations. When a Z score is more than 2.58, it indicates a 99% confidence interval of positive autocorrelation among the variables. Similarly, a 1.96 Z score indicates a 95% confidence level and 1.65 for 90%. A significant and negative Z score usually indicates a negative autocorrelation, e.g., a high-low cluster.

In addition to measuring the spatial autocorrelation of each individual observation (the greening index of each city in this case) to reveal spatial heterogeneity and identify outliers among all the observations, we used Anselin’s LISA (local indicators of spatial association) statistic to decompose global Moran’s I into local Moran’s I. Local Moran’s I at location i can be expressed as Eq. 3:

$$ Local\kern0.5em I=\frac{\left({x}_i-\overline{x}\right)}{S^2}\sum \limits_j{W}_{ij}\left({x}_j-\overline{x}\right) $$
(3)

Four types of result can be shown: high-high clusters where higher values are grouped, low-low clusters where low values are grouped, low-high clusters where high values group around low values, and high-low clusters where low values group around high values. The results indicate the spatial discrepancy of greening locally and individually, which permits the spatial clusters, hot spots, and outliers to be visualized with ArcGIS 10.2 (Zhang et al. 2008).

Ordinary least square model

To identify the relationship between explanatory variables and dependent variables in space, namely how economic factors affect the greening index in this case, several spatial autoregressive models were considered.

Globally, the ordinary least squares (OLS) model was used as conventional regression model to estimate geographical issues. OLS is formally given by Eq. 4:

$$ y={\beta}_0+\sum \limits_{i=1}^k{\beta}_i{x}_i+\varepsilon, \kern0.5em \varepsilon \sim N\left(0,{\delta}^2\right) $$
(4)

where xi represents the independent variables and y refers to the dependent; β0 is the intercept; k is the number of independent variables; βi is the parameter estimate (coefficient) corresponding to the independent variable xi; and ɛ is the error term. The parameter estimates βi are assumed to be spatially stationary.

Spatial autoregression model and spatial error model

However, the traditional OLS is a spatial constant coefficient model that requires variables to be independent, and violations of this can cause biased and inconsistent estimates. We, therefore, adopted two other spatially constant coefficient models, SAR and SEM, as spatial extensions of the conventional global regression model, OLS. They are similar in fundamental concepts, but vary in the way they control spatial dependency.

The SAR model can test the impact of nearby units on a local unit, with a spatially lagged dependent variable. It is widely used when the spatial dependence is likely to be found in the dependent variable, correcting autoregressive problems among variables. The SAR model can be expressed as Eq. 5:

$$ y=\rho Wy+ X\beta +\varepsilon, \varepsilon \sim N\left(0,{\delta}^2\right) $$
(5)

where ρ is the autoregressive parameter that expresses the spatial dependence in the lattice, spatial lag is a smoothing term for the weighted average of neighboring values, and the error ε follows a normal distribution.

The SEM is used when spatial dependence rests upon unforeseeable factors and incorporates a spatial autoregressive process in the error term. The SEM can be expressed as Eq. 6:

$$ y= X\beta +\lambda Wu+\varepsilon, \varepsilon \sim N\left(0,{\delta}^2\right) $$
(6)

where y is a N × 1 vector of the dependent variable and X denotes an N × K matrix of explanatory variables; β is a K × 1 vector of parameters; u is a N × 1 vector of residuals; the scalar λ is the spatial autocorrelation coefficient of errors; W represents a N × N spatial weight matrix containing contiguity relations or functions of distance; and ε is a vector of normally distributed errors.

Lagrange multiplier

To verify which model more accurately captured the spatial effect of the green index distribution, the Lagrange multipliers (LM) test was adopted, including two robust tests: LM (lag) test and LM (error) test, which were used to test an autoregressive spatial lag variable and the spatial autocorrelation of errors, respectively. The two forms are alternatively used to test whether the hypothesis of spatial autocorrelation is a spatial error model or a spatial lag model.

Geographically weighted regression

Although spatial autocorrelation was considered in the OLS, SAR, and SEM spatial methods, all these were spatially nonparametric approaches. Spatial heterogeneity and the effects of independent variables were not able to be shown. Accordingly, a GWR model was used to identify the effects of independent variables on the green index locally. Initially suggested by Brunsdon et al. (1999), the GWR model is widely used as an effective spatially varying coefficient model, in both socioeconomic and environmental fields. It allows parameters to differ in value depending on spatial location, instead of employing the same values across all space units. In this way, the spatial distribution of local coefficients can be depicted and mapped visually in GIS, yielding comparative and representative results. GWR is formally given by Eq. 7:

$$ {y}_j={\beta}_0\left({u}_j,{v}_j\right)+\sum \limits_{i=1}^p{\beta}_i\left({u}_j,{v}_j\right){x}_{ij}+{\varepsilon}_j $$
(7)

where j denotes the regression point where the model is calibrated; uj and vj represent the coordinates of location j; β0 (uj, vj) acts as intercept for location j; and βi (uj, vj) is the local estimated coefficient for independent variable xi.

A distance decay weighting function is constructed that places relatively more weight on sample observations from neighboring observations than distant ones. Accordingly, GWR is calibrated by weighting all observations around a sample point (Tu and Xia 2008). A Gaussian distance decay weighting can be a powerful approach to present the weight function, as in Eq. 8:

$$ {W}_{ij}=\exp \left(-{d}_{ij}^2/{h}^2\right) $$
(8)

where Wij is the weight for observation j adjacent to the position of i; dij denotes the distance between observations i and j; and h is a decay parameter or kernel bandwidth, producing estimates that vary more or less over space. GWR is sensitive to kernel bandwidth, and the optimal bandwidth can be chosen by minimizing the corrected Akaike information criterion or AIC (Liu et al. 2016).

Results

Spatial distribution of greening indexes

Figures 2 and 3 present the spatial distribution of the urban green space ratio and public green area per capita across China. It can be clearly seen that the urban green space ratio clusters highly in southeastern China, namely, in the Guangxi, Guangdong, Fujian, Zhejiang, and Jiangxi provinces, with an urban green space ratio of more than 40%. Several cities scattered on the North China Plain also perform well, such as Beijing and the cities in Liaoning, Shandong, and Hebei provinces. A relatively low rate is shown in Heilongjiang, Gansu, Guangxi, and Guizhou provinces. With regard to the public green area per capita, southeastern cities in Hainan, Zhejiang, Guizhou, and Guangdong provinces showed high values for public green area per capita that exceeded 21 as might be expected, but some cities in the north of China also showed relatively high values including in Inner Mongolia, Gansu, Ningxia, and Hebei provinces. A figure of less than 10 mainly appeared in Sichuan, Heilongjiang, southern Gansu, Hunan, Guizhou, and Guangxi provinces.

Fig. 2
figure 2

Spatial distribution of the urban green space ratio

Full size image
Fig. 3
figure 3

Spatial distribution of the public green area per capita

Full size image

Regarding the spatial autocorrelation of the urban green space ratio and public green area per capita, the results of global Moran’s I and local Moran’s I presented a convincing conclusion. Overall, the value of global Moran’s I in the urban green space ratio was 0.12, with a Z value of 14.99. Public green area per capita showed a global Moran’s I of 0.07 and a Z value of 9.41. The Z value of both urban greening indexes showed the result to be substantially over 2.56, which indicated positive autocorrelation (i.e., high-high clusters and low-low clusters) at a confidence level of 99% and demonstrated the existence of significant clustering in urban green space ratio and public green area per capita.

Locally, in Fig. 4, the result of LISA analysis depicts the spatial clusters and outliers of the urban green space ratio and public green area per capita, respectively. High-high clusters, referred to as hot spot regions, emerged mainly along the eastern coastal area of China, indicating the grouping of high urban green space ratios, while low-low clusters showed in Gansu, Sichuan, Shanxi, and Guizhou provinces, located mostly at the middle longitudes in China. As for the public green area per capita, which related strongly to population, high-high clusters appeared only in Inner Mongolia, Gansu Province, and the eastern part of Guangdong Province. Low-low clusters, or cool spots, were found scattered in Sichuan Province and Yunnan Province. Low-high clusters and high-low clusters also existed both in the spatial distribution of the urban green space ratio and public green area per capita, suggesting high spatial autocorrelation of greening indexes in Chinese cities, both globally and locally.

Fig. 4
figure 4

Clusters and outliers of urban green space ratio and public green area per capita

Full size image

Constant coefficient model estimation and comparison

Since the spatial dependence of greening indexes was confirmed, the study compared three spatial models apart from the ordinary linear regression, namely, SAR, SEM, and GWR, so as to correct the estimation error brought by spatial relationships.

OLS estimation

Initial OLS estimation was carried out separately for the urban green space ratio and public green area per capita and both passed the significance test. Tables 3 and 4 show the results for each variable with regard to the urban green space ratio. Per capita GDP and the percentage of secondary industry showed a significance value of 5%, suggesting their important impact on urban green space ratio. For the public green area per capita, factors such as per capita GDP and urban land were significant at a level of 1%, and population density reached a 5% significance level. The R2 of the urban green space ratio and public green area per capita OLS estimation, indicating the model’s goodness of fit, were 0.13 and 0.20; and the F-statistics were 8.51 and 14.18, respectively. The low values of both the F test and R2 test indicate a poor fit in the OLS estimation for both two indexes.

Table 3 Parameter summary of GWR, SEM, and SAR for urban green space ratio
Full size table
Table 4 Parameter summary of GWR, SEM, SAR for public green area per capita
Full size table

Furthermore, the inaccuracy of OLS model was proved by the diagnostic test for spatial dependence. With regard to both the urban green space ratio and public green area per capita, Moran’s I of error passed a significance of 1%, suggesting large spatial autocorrelation on the residuals. Therefore, the model estimations through OLS may lead to a biased result. According to the Lagrange multiplier test for the urban green space ratio from OLS estimation, the LM (lag) and LM (error) both passed the significance test at a 1% level. The value of LM (lag) was slightly larger than LM (error), suggesting that the spatial dependence was more likely to lie in the variables rather than other nonmodeled effects. With regard to public green area per capita, LM (lag) also showed significance at the 1% level, but LM (error) failed the significance test.

SAR and SEM estimation

SAR and SEM estimation was carried out to offer more precise information on spatial dependence. Tables 3 and 4 present the results and the comparison between the SAR and SEM outputs. As expected, in the urban green space ratio, the R2 goodness of fit for SAR estimation of 0.31 showed the model performed better than that of SEM estimation (R2 = 0.29). This was supported by a smaller AIC further indicating that the SAR depicts more precisely the relationship between the urban green space ratio and urban socioeconomic factors. The percentage of secondary industry, urbanization rate, and per capita GDP in the SAR model all passed the significance test of 5%, verifying their relevance to the urban green space ratio. Secondary industry was the most relevant variable among the three with a probability of 0.17.

In the estimations of public green area per capita, the R2 in the SAR estimation was similar to the SEM, while the AIC for SAR was slightly larger than SEM. This indicates that both models show similar probability estimations for the variables. Per capita GDP and urban land passed the significance test at the 1% level; and population density was found to be significant at the 5% level, while secondary industry and urbanization rate failed the significant test at 5%.

Spatial distribution of the GWR estimation

Urban green space ratio

The local R2 for the GWR estimates are visually described in Fig. 5. The value of the local R2 ranged from 0.09 to 0.5. Compared with the OLS estimation (0.21) and SAR estimation (0.31), the local model did not generally fit better than the coefficient variable models. Yet there is an obvious spatial pattern: from the northwest to the southeast of China, R2 decreased gradually, suggesting that the urban green space ratio is more strongly related with urban socioeconomic factors in northwestern China.

Fig. 5
figure 5

The spatial distribution of GWR local R-squared of urban green space ratio

Full size image

Some significant economic factors were used to generate coefficients with two greening indicators locally by the GWR model. Figure 6 demonstrates the spatial distribution of the relationship between per capita GDP, urbanization rate, and percentage of secondary industry with the urban green space ratio. These were the three main variables affecting the urban green space ratio as shown above. It can be observed from the estimation of local per capita GDP that every location had a positive coefficient, indicating their positive effect on urban greening. Spatially, like the local R2 pattern, the urban green space ratio was more closely related with GDP per capita from the southeast to northwest, with cities in southeastern China, especially Guangdong, Fujian, and Zhejiang provinces and Shanghai, maintaining coefficient estimates near zero. This may be led by a different percentage of investment spent on improving urban landscape construction in these areas. The spatial distribution of the urbanization rate coefficients indicates a relatively greater negative influence on the cities in the northeastern region, while very little impact existed in the western cities, such as in Gansu Province. This might be related to their different development strategies and base industries for urbanization, as well ignorance of greening in urban construction which would result in a significant negative effect in the northwestern cities. The percentage of secondary industry had a positive effect on the urban green space ratio in the central part of China and along the Yangtze valley, especially Gansu, Ningxia, Shaanxi, and Sichuan provinces, while the coefficient estimates were neutral or even slightly negative in northeastern and southwestern China. This might be caused by their dependence on secondary industry for economic development, resulting in limited investment for urban greening.

Fig. 6
figure 6

The spatial distribution of GWR local coefficients of urban green space ratio

Full size image

Public green area per capita

As mapped in Fig. 7, the local R2 of public green area per capita varied from 0.07 to 0.31 with relatively well-distributed range. Compared with the R2 in OLS, SAR, and SEM estimations, the GWR model better depicted the relationships than constant coefficient models in some areas, but was worse in others. Specifically, in the southern coastal areas and in the middle of the northern China, especially in Inner Mongolia, Guangdong, Guangxi, Ningxia, and Shaanxi provinces, cities had relatively high R2 values, demonstrating a closer spatial correlation between public green area per capita and urban socioeconomic factors.

Fig. 7
figure 7

The spatial distribution of GWR Local R-squared of public green area per capita

Full size image

Regarding the three factors influencing public green area per capita identified above, the spatial distribution of coefficient estimates in the local GWR model showed a distinct spatial pattern (Fig. 8). The urbanization rate coefficients were negative overall, suggesting the opposite effect of urbanization on public green area per capita than on the urban green space ratio. Notably, a larger impact was found in the cities of northwestern China, namely, in Inner Mongolia, Xinjiang, Gansu, and Ningxia provinces, which implies a higher price for public green area per capita was paid in the process of urbanization. These cities are mostly still at the early stage of urbanization, focusing greatly on basic urban construction, and have not fully considered the recreational functions of urban greening. In contrast, public green area per capita was hardly influenced by urbanization in cities in Yunnan Province. Per capita GDP was found to be the least effective factor among the three, yet cities in the central part of China, especially in Gansu Province and the central area of Inner Mongolia, showed relatively high coefficients, demonstrating the significant and positive influence of per capita GDP on public green area per capita in these cities. This may be because of the limited populations here. Coefficients of population density presented a similar spatial pattern. Cities in western China generally had negative parameters suggesting that public green area per capita decreased with the growth of urban population. This may be partly because of relatively slow economic development and the lack of effective and far-sighted urban planning, resulting in their susceptibility to population growth. A coefficient value of close to zero showed the lack of relevance of population density elsewhere in China.

Fig. 8
figure 8

The spatial distribution of GWR local coefficients of public green area per capita

Full size image

Overall, the local coefficients between greening indicators and economic indicators varied greatly in space, but presented distinct regional distribution patterns. With regard to the explanatory power for the greening indicators, local estimation of economic indicators generally performed better than OLS, SAR, and SEM methods.

Comparison of parameter estimation among GWR, OLS, SAR, and SEM

Tables 3 and 4 show the results of parameter estimation in selected models across the complete sample of 289 cities. The minimum, maximum, and quartile values are presented to show the general outcomes of GWR model’s estimation. Overall, the parameter estimates by the OLS model were numerically larger than those of SAR and SEM. Further, if the mean values of the coefficients generated from the GWR model are compared with the other models, similar results are shown. In summary:

  1. (1)

    Among the five explanatory variables, per capita GDP had a positive effect on both greening indices, while urbanization rate, secondary industry, urban land, and population density showed opposite effects on the urban green space ratio and public green area per capita, respectively, each with a positive impact on the urban green space ratio and negative impact on public green area per capita.

  2. (2)

    Synthesizing the estimations of the various models, per capita GDP, population density and urban land area had similar impact on the urban green space ratio, as all the coefficients had quite small values.

  3. (3)

    Compared with the parameters estimated for the urban green space ratio, parameters for public green area per capita showed more significant results, indicating a closer relationship between economic factors and public green area per capita.

  4. (4)

    The parameter estimate of the urbanization rate on public green area per capita differed between the constant coefficient models and the GWR model, reaching − 0.02 in the OLS, SEM, and SAR models, yet failing to pass the significant test at the 5% level. It had a mean of − 0.05 in the GWR model.

  5. (5)

    Regarding the impact of the economic factors on greening indices, the urban green space ratio was positively influenced by all the factors while public green area per capita was adversely affected by all except per capita GDP.

Discussion and policy recommendation

As shown above, a spatial differentiation of urban greening existed among cities in China. Generally, the urban green space ratio in northwestern China was much lower than in the southeast. As demonstrated by this study, the influence of the economy on greening was large, because green spaces in built-up areas are mostly specifically constructed and few reflecting preserved natural ecosystems. Where and how urban green space is influenced by economic factors are crucial for urban development because urban green space significantly influences the ecoenvironment in built-up areas. The relationship between the urban greening situation and public green area per capita and economic development in China is demonstrated in the study through a spatial regression model.

The study indicates that the relationship between urban greening and economic development in China is similar to Singapore (Seik 2001), but differs from some European countries, such as the UK (Pauleit et al. 2005). This difference may be partly due to the variation of urban expansion patterns in developed and developing countries. Most cities develop by expanding into surrounding countryside in China. Within the built-up area, green space construction is encouraged, supported and managed by the government to enhance the city, and thereby urban space is expanded through the encroachment of urban suburbs into rural land. Although the urban greening situation is improved, the surrounding ecological space is reduced through urbanization.

Obvious spatial variation was found regarding the influence of economic factors on urban greening in China. In general, the cities showed less socioeconomic development in the northwestern region and had less urban greening, while economically well-developed cities elsewhere had higher amounts of urban greening. This is chiefly because governments had more available greening investment funding to support urban greening in the economically developed regions. The cities of the northwestern region are relatively undeveloped in economic terms, and their urban greening situation is limited by lack of available funds. Developing cities rely on secondary industry and their proportion of greening investment is thus limited. Therefore, the level of urban greening is more constrained in the northwestern region, which is dominated by industrial development. During industrial development, rapid urbanization leads to the encroachment of new construction on existing urban green space, as shown by the negative relationship between urbanization and urban greening. In addition, the northwestern China is much lower in plant productivity, with less annual precipitation and arid or semi-arid vegetation dominates (Sun et al. 2011). Public green area per capita represents the recreational function of urban green space. On the whole, the public green area per capita was negatively related to economic development in China. Population density had a significant negative effect on the recreational function of urban green space at a national scale. This demonstrates that the construction of green space lags behind the growth of the urban population.

The study is limited to data from 2014. Further research is needed to focus on time series analysis of urban greening in China. Examining the temporal trend of urban greening in China and its relationship with economy could help understand the processes of urban development in China and green space. Although the economic development was demonstrated to positively contribute to urban greening, the driving force and mechanism behind this need further study. Besides, the study neglected the effects of such natural factors as temperature, precipitation, atmospheric pressure, and elevation on urban greening; thus, further research could be conducted including both natural and anthropogenic factors to better understand the change mechanism of urban greening in China. Furthermore, the limitation of Chinese statistical mode also brings some defects; the future research orientation of Chinese statistical data should be geared to international standards.

Policies concerning urban greening have had a significant impact on the planning, implementation and management of green spaces (Zhou and Wang 2011). Based on our findings, we made the following policy recommendations. First, it suggests the need for greater consideration on the regional imbalance of basic public services. The central government should strengthen the assessment of local leaders and realize the right equality of different cities to enjoy green space. The urban planning should attach importance to the economy development of cities with lower level of green space and limit the expansion of urban land into existing ecological spaces. Second, policies should be used to encourage multicenter urban spatial structure and greenbelt construction aimed at increase public green area per capita in the face of development pressures (Williams et al. 2004). Population urbanization and land urbanization should not be aggressive, considering the green carrying capacity of a city. Third, economy had a greater influence on urban greening in the northwestern region than in the southeastern region. This enlightens that accelerating the economic development of northwest China will bring greater effect to the construction of green space for northwestern cities. In other words, the urban greening in northwest China has a late-development advantage. On the other hand, the climate in northwest China is dry, which increase the cost of urban greening. Thus, green space construction in northwestern cities should involve more policy support and greening investment.

Conclusion

This study first analyzed the spatial pattern of urban greening in 289 cities in China. Using OLS, SAR, SEM, and GWR models, we found that urban greening in the built-up area of Chinese cities was shown to have a significant spatial dependence and heterogeneity. Economic development was demonstrated to positively contribute to urban greening. However, the urbanization rate, secondary industry, urban land, and population density showed significant opposing impacts on the urban green space ratio and public green area per capita, respectively. Per capita GDP was shown to have a greater influence on urban greening in the northwestern region than in the southeastern region. To study the driving force and mechanism systematically and integrally, investigation into the effects of both economic and natural factors may be a vital step.