Exploring the relationships between urban form metrics and the vegetation biomass loss under urban expansion in China

Abstract

Compact urban form has been applied as a strategy to reduce the loss of green space that occurs from development, but the impact of this policy on the provision of green space still presents many uncertainties. This research investigated the statistical relationship between urban form indicators and the loss of vegetation biomass to understand the response of quality green space provision to changes in urban morphology. A methodology combining multi-source data assimilation, statistical analysis, and spatial analysis was adopted for the Yangtze River Delta cities of China. First, six urban metrics were selected to describe the shape and layout of urban patches in each city, and the total biomass loss index was then introduced as a parameter. The values of urban metrics and total biomass loss index were calculated for the 50 Yangtze River Delta cities. Second, ordinary least squares regression and geographically weighted regression analyses were then used to establish a quantitative relationship between total biomass loss index and urban form indicators. The results revealed an extremely negative correlation between total biomass loss index and the three urban variables of Richard compactness, density gradient, and the Gini coefficient; moreover, the parameter estimates for the three variables in the geographically weighted regression model were local and varied over space. Third, the mechanisms by which the urban form influences biomass loss were discussed and different urban form planning strategies for particular urban areas were suggested. In conclusion, compact urban form in a clustered layout of urban areas with a dense central agglomeration was verified to be ecologically superior and conducive to green space protection. For the physical interpretation of the statistical relationship between urban morphology and vegetation loss, the interface effect of urban agglomeration on vegetation merits further study.

Keywords

Green space compact city geographically weighted regression urban metrics development

Introduction

The growth of urban areas is widely considered to cause the loss of green spaces and wildlife habitats, while a rapid increase in urbanization has occurred worldwide, particularly in developing countries (Haas et al., 2015; United Nations, 2014; Van Vliet et al., 2017). To mitigate these adverse consequences, compact urban form has been promoted as a strategy for sustainable development despite the uncertain environmental consequences (Bunker, 2014; Burton, 2002; Jenks et al., 1996; Rérat, 2012; Sushinsky et al., 2013). How to accurately assess the layout of urban agglomerations has become crucial to distinguish compact development from urban sprawl. Spatial metrics, which provide sophisticated descriptors of urban spatial heterogeneity based on the distribution of built-up structures and open areas, have been reported to quantify urban form within a GIS and remote sensing framework (Herold et al., 2005). They were applied to describe population density, urban land intensity, as well as the shape and continuity of urban land patches (Galster et al., 2001; Song and Knaap, 2004). Because of the standardized calculation method based on remote sensing images, some metrics have been adopted in comparative studies of different cities based on data availability. For example, Huang et al. (2007) used seven spatial metrics to demonstrate the more compact and dense urban agglomerations in developing countries than in their counterparts in Europe or North America.

Moreover, urban form research is often linked to environmental impact, including the loss of green spaces. First of all, in terms of the amount of green space, it is widely documented that compact urban form will reduce the occupation of natural or agricultural green space because of the controlled built-up area, but may also lead to the decline of green space provision with the consolidation of the central area (Jim and Chen, 2003; Kong and Nakagoshi, 2006). To get a more comprehensive analysis of profit and loss of urban compaction requires a great deal of sample data. The Corine Land Cover (CLC) data set has greatly facilitated this research across a large number of European cities. With green space and population data in 2000, Fuller and Gaston (2009) pointed out that an increase in population density tends to lead to a decline in green space availability per capita. However, they did not think that for an already compacted landscape, a further increase in density led to further reductions in green space provision. Using years of CLC urban green space data, Kabisch and Haase (2013) argued that there is no significant relationship between population density and green space provision. They pointed out the growth of residential areas, even in shrinking cities, and that the renovation of inner cities in Eastern Europe has resulted in a decrease in urban green space provision. Hence, the two studies show that the effect of compaction policies on the amount of green space still remains unclear and more research needs to be done. In addition, both studies emphasized the importance of urban planning and concluded that a systematic conservation plan that integrates urban development and the measures of green space enlargement is a key approach. Some research on developing cities has also demonstrated that the extent of urban green space is greatly affected by urban planning and green space policies (Sperandelli et al., 2013; Tong and Ding, 2011; Zhou and Wang, 2011).

Second, changes in the quality of green space have been more frequently reported under the term ecosystem services. Tratalos et al. (2007) surveyed the urban density, tree cover, and biodiversity potential of five UK urban areas and found that most measures of ecosystem performance declined with increasing urban density, but there was considerable variability in the relationships due to the different methods of measuring urban density. Hutyra et al. (2011) assessed carbon stocks as a function of distance to the urban core and land cover. Similar findings that carbon stocks gradually increase from the centre to the periphery have been reported in Chinese cities (Li et al., 2016a; Tao et al., 2015). These studies explored relationships of vegetation carbon stock with urban density, and distance to the urban core and land cover at the plot scale, but few studies have discussed the impact of urban form factors at the city scale such as shape and direction.

Judging from the process of urban development in a region, the quality and quantity of green space are determined by three primary factors: the original type and distribution of vegetation, the amount of urban development, and the layout and form of the urban land. A reasonable layout or form should retain as much of the existing green space as possible, while not easily causing subsequent development to occupy high-quality green space, such as a forest with high carbon stocks and other ecosystem services. For cities that are expanding, the development shape and location, in addition to the development volume, will also affect green space provision. Developments that are close to high-quality green space will result in more loss with land clearing, which should be avoided. Most previous studies on urban green space provision changes focus only on the intensity and area of development. Hence, to the authors’ knowledge, there is little research to date on the analysis of the location and the shape of urban expansion. Moreover, there is a lack of orientation analysis of the vegetation loss in urban extension areas, which cannot provide direct evidence for research on the relationships between urban spatial form and green space provision. This research attempts to find a way to optimize urban form through analysing expansion data of dozens of typical Chinese cities. An indicator was introduced to quantify the loss of green space provision and simultaneously link the urban form parameters. The vegetation biomass, which represents the total amount of organic matter (dry weight) in a unit area at a given time, is an important parameter in urban ecosystem studies (Alberti et al., 2003). Through photosynthesis, it was naturally connected to vegetation ecosystem services, such as primary productivity, reducing the temperature of the heat island, carbon sequestration, and air pollution mitigation, characterizing the comprehensive function of vegetation and the quality of green space. The objective of this paper is to determine the statistical relationship between vegetation biomass loss index and urban morphological metrics. Guided by the established quantitative relationships, the environmental performance of various urban forms was assessed and could, thus, provide beneficial planning strategies for urban agglomeration layouts to reduce vegetation loss.

Study area and methods

Study area

The Yangtze River Delta (YRD) megalopolitan region is located in the central east coast of China, which covers an area of 110,800 km² and comprises the triangular-shaped territory of Shanghai, the southern part of Jiangsu Province and the northern part of Zhejiang Province (Figure 1). Occupying less than 1% of China’s land, the YRD contributes to over 20% of China’s GDP, leading the economic development in China. After the open policy, the YRD experienced rapid urban expansion, especially in the first decade of the 21st century. Based on results from remote sensing (Chen et al., 2015), the construction land areas of 50 cities in the YRD region have increased from 4080.84 to 9914.67 km² during 2001–2007. There are 51 cities in this region, where Zhoushan city was excluded in this research due to its scattered built-up areas on islands. The samples of 50 cities included one provincial city (Shanghai), 13 prefecture-level cities with districts (e.g. Nanjing, Hangzhou, Ningbo), and 35 county-level cities without districts (e.g. Kunshan, Cixi, Jiangyin). In the context of China’s city system (National Bureau of Statistics, 2009), the selected 50 cities were categorized into four types based on the non-agricultural population (NAP): megacity (NAP > 1 million), major city (0.5 million<NAP < 1 million), medium-sized city (0.2 million<NAP < 0.5 million), and small city (NAP < 0.2 million).

Figure 1.

Location of the YRD and distribution of the urban construction land.

Biomass loss measurements

Vegetation biomass is an important indicator of the terrestrial ecosystem. Although several works have been completed to estimate vegetation biomass at the regional and global scales (Foody et al., 2001; Saatchi et al., 2011), very few are available in the literature about the spatial estimation of biomass for an ecosystem as a whole (Fang, et al., 2007), and the spatial resolution (1–8 km) of the estimation is too low to reflect changes in land use/cover (Yu et al., 2009). Based on multi-source data, including remote sensing, meteorology, land use/cover, forest inventory, and grain yield, the spatial distribution of vegetation biomass at 250 m resolution was studied with the aid of remote sensing models, a spatial downscaling technique, spatial analysis using a GIS, and mathematical statistical analysis. The 250 m resolution is the highest spatial resolution available from the MODIS satellite data that can simultaneously meet the time resolution requirements. In this study, the vegetation biomass is calculated from the net primary productivity (NPP), which is simulated on the basis of satellite data with a 15-day recurrence period. The temporal resolution of high spatial resolution satellite data such as Landsat does not meet these requirements.

First, NPP was estimated with a modified CASA model based on land use data (http://www.resdc.cn, http://lake.geodata.cn), remote sensing data (MODIS/Landsat, http://www.gscloud.cn), and meteorological data (http://data.cma.cn) from 2000 to 2010 (Li et al., 2016b; Zhu et al., 2007). Second, a statistical downscaling technique was adopted to map the forest biomass at the provincial scale by combining NPP and forest inventory data (http://www.cfsdc.org/), which has recently been used to estimate the spatial distribution of forest biomass at 1 km and achieved good results (Kindermann et al., 2008; Liu et al., 2012a, 2012b). The shrub biomass was assigned as 12.34, 33.14, and 12.34 t hm⁻² for Jiangsu, Zhejiang, and Shanghai, respectively, according to local research on biomass carbon stocks (Hu et al., 2006). Third, cropland biomass was estimated by multiplying NPP by 0.45, and the unit of NPP was g m⁻² a⁻¹. Fourth, the biomass of grassland was assigned as 7.17, 9.26, and 8.59 t hm⁻² according to local research on biomass (Piao et al., 2004). Finally, we obtained a vegetation biomass distribution map with a single land use type of biomass using a mosaic method.

Lost vegetation biomass under urban expansion is represented by a total biomass loss index (TBLI), which represents the overall influence of urban sprawl on the ecosystem. Here, the urban expansion refers to the growth of urban construction land from 2001 to 2007, which was extracted from Landsat5, Landsat7 images and corrected according to the corresponding high-resolution images of Google Earth (Chen et al., 2015). TBLI can be characterized as

TBLI = [\sum_{i = 1}^{n} x_{i} \times 250 \times 250] / 10^{9}

(1)

In equation (1), $x_{i}$ is the amount of vegetation biomass per unit area (kg m⁻²) in 2000, when the ith vegetation pixel was not occupied by urban expansion; 250 × 250 is the pixel size (m²); n is the number of vegetation pixels occupied by urban expansion in 2010; the denominator 10⁹ is present to convert the unit of TBLI from kilogram to Teragram (1 Tg = 10⁹ kg).

Urban form measurements

Urban form, which basically refers to the geometric shapes of urban land patches, depicts the attributes of the spatial structure of urban areas. When judging urban sprawl, we generally judge whether the growth rate of the surrounding area exceeds the growth rate of the central area (Brueckner and Fansler, 1983). For a compact city, indicators of shape compactness, urban density, and mixed-land use are developed for measurement (Burton, 2002). Schwarz (2010) revisited the commonly used urban form indicators and divided them into landscape metrics and socio-economic indicators. She found that several previously reported indicators were, in fact, mostly related, and selected seven of them with respect to landscape metrics and population-related indicators by correlation and factor analysis, including the area of the discontinuous urban fabric, edge density, mean patch size, number of patches, compactness index of the largest patch, population number, and population density. Considering the availability of population distribution data within the city, we mainly adopted landscape metrics in this study, focusing on the distribution of the urban land patches at three spatial levels: central urban agglomeration, the relationship between central urban agglomeration and the other urban land patches, and the spatial distribution of urban land patches at the city level. Six indicators were selected to cover the three levels while considering their efficiency and use in the literature (Galster et al., 2001; Schwarz, 2010; Tsai, 2005) and the availability of data in the YRD cities (Table 1).

Table 1.

Descriptions and formulas for the urban form indicators.

No.	Indicator	Formula	Expression	References
1	Richardson compactness	$S = \frac{2 \sqrt{π A}}{P}$	P: perimeter of the central urban agglomeration outlineA: area of central urban agglomeration	Richardson (1973)
2	Density gradient	$D_{i} = G \times d_{i} + D_{0}$	D_i: density of the urban land within the ith sphered_i: distance from the centroid of the urban area to the ith sphereG: density gradientD₀: the density of urban land in the first sphere	Galster et al. (2001) Brueckner and Fansler (1983)
3	Continuity	$R = (A′) / A$	A: total area of urban land patchesA′: area of central urban agglomeration	Cole (1964), Galster et al. (2001)
4	Galster centrality	$M = \frac{\sqrt{A}}{\sum_{i = 1}^{n} d_{i} x_{i}}$	A: urban areax_i: the proportion of construction land in grid i to the total area of construction land in the urban aread_i: distance from grid i to the nearest urban centre	Galster et al. (2001)
5	Gini coefficient	$G = 0.5 \times \sum_{i = 1}^{N} \| X_{i} - Y_{i} \|$	X_i: proportion of construction land in grid i to the total area of construction landY_i: proportion of the i grid area to the total area of gridsN: number of grids	Tsai (2005)
6	Moran's I	$I = \frac{N \sum_{i = 1}^{N} \sum_{j = 1}^{N} ω_{i j} (X_{i} - X) (X_{j} - X)}{(\sum_{i = 1}^{N} \sum_{j = 1}^{N} ω_{i j}) \sum_{i = 1}^{N} {(X_{i} - X)}^{2}}$	X: average proportion of construction land in one grid to the total construction land areaX_i/X_j: proportion of construction land in grid i/j to the total construction land areaω_ij: weighting, equal to the distance between i and j	Tsai (2005)

In this study, Richardson compactness describes the roundness of patch shape of the central urban agglomeration, with a value between 0 and 1. The closer the value is to 1, the closer the shape of the patch is to the circle, i.e. the more compact the patch is. Conversely, the closer the value is to 0, the less compact the shape.

The density gradient index represents the ratio of the density decrease to the distance from the centre. It describes the slope of the density linear attenuation curve of the construction land in this study. The value of sprawling development is smaller than that of compact cities.

Continuity describes the relative scale of the central urban agglomeration area. The value is between 0 and 1; the closer the value is to 1, the higher the proportion of the urban central agglomeration.

Galster centrality describes the distribution of construction land relative to the centre of the city. The greater the index value, the higher the degree of urban land concentration and distribution around the centre.

The Gini coefficient measures the spatial equilibrium of urban development. The larger the index, the more uneven is the distribution of construction land.

Moran’s I is used to describe the spatial autocorrelation characteristics of variables, with a value between −1 and 1. The closer the value is to 1, the higher is the degree of urban land cluster distribution.

Regression analysis

Both ordinary least squares (OLS) regression and geographically weighted regression (GWR) have been adopted with the dependent variable of TBLI and the independent variables of selected urban form indicators to investigate their statistical relationships.

First, for the dependent variable, Moran’s I approach was applied (Diniz‐Filho et al., 2003; Legendre, 1993) to examine the spatial autocorrelation and spatial heterogeneity of TBLI. Moran’s I usually varies between −1 and 1 for maximum negative and positive autocorrelation, respectively, while a zero value indicates no spatial autocorrelation. Moran’s scatterplot (Anselin, 1995) depicts four quadrants corresponding to four types of spatial association: High–High values in the upper right quadrant and Low–Low values in the lower left quadrant indicate the spatial clustering of similar values, while High–Low values in the lower right quadrant and Low–High values in the upper left quadrant represent the spatial dispersion of dissimilar values. If most observations are High–High or Low–Low values with a spatial clustering distribution, the spatial heterogeneity of observations can be verified.

Second, for the independent variables, we calculate Pearson correlation coefficients for the six urban form indicators. When the absolute value of the correlation coefficient of a variable pair was more than 0.75 at the 1% level of significance, we concluded that the two variables were highly correlated and should not be present in the same regression model (Cohen, 1988). Moreover, a partial correlation analysis of TBLI and urban form indicators was carried out when we considered the urban expansion area as the control variable, eliminating its impact on biomass loss. The independent variable with a partial correlation coefficient significant at the 1% level (two-tailed) was regarded as an influential factor that could be added to the regression model.

Third, OLS regression was implemented. It is a standard global regression, which assumes that relationships are constant across space. The OLS model can be presented as

y_{i} = β_{0} + \sum_{k} β_{k} x_{i k} + ε_{i}

(2)

where y_i is the estimated value of the dependent variable for observation i, β₀ is the intercept, β_k is the estimation parameter for variable k,

x_{i k}

is the value for the kth variable of observation i, and ε_i is the error term.

Finally, GWR was applied. It was proposed by Fotheringham et al. (1996) to explore the relationships that might vary over space. The global regression equation (2) was rewritten as

y_{i} = β_{0} (u_{i}, v_{i}) + \sum_{k} β_{k} (u_{i}, v_{i}) x_{i k} + ε_{i}

(3)

where

(u_{i}, v_{i})

is the coordinate location of the observation i (Brunsdon et al., 1998); the parameters

β_{k} (u_{i}, v_{i})

were estimated by

β_{k} (u_{i}, v_{i}) x_{i k} = {(x^{T} ω (u_{i}, v_{i}) x)}^{- 1} x^{T} ω (u_{i}, v_{i}) y

(4)

In equation (4), $ω (u_{i}, v_{i})$ is the weighting function, ensuring that the observations near the location have more influence on each model than those further away. A Gaussian function is adopted here for $ω$

ω_{i j} = exp [- {(\frac{d_{i j}}{b})}^{2}]

(5)

where

d_{i j}

is the distance between location i and location j, and b is the kernel bandwidth. A fixed kernel bandwidth was used here. It was determined by a cross-validation approach or Akaike Information Criterion (AIC) through the GWR4.0 software (Akaike, 1973; Fotheringham et al., 2002; Hurvich et al., 1998). The fixed bandwidth was set to 100 km in this case.

For the explanatory variables involved in the GWR model, an F-statistics test was conducted to identify the spatial non-stationarity of the estimation parameter (Fotheringham et al., 2002). When p was less than a certain value, such as 0.05, the hypothesis of spatial stationarity was rejected, the estimation parameter was considered as local, and the corresponding variable could be treated locally in the GWR model. Furthermore, the estimation parameter of the GWR model outputs was visualized according to a standard deviation classification in ArcGIS10.2 software. The local values that were not significant at the 2% level for a two-tailed test were hidden by a grey mask in the map to highlight the significantly distinguished areas (Clement et al., 2009; Mennis, 2006).

In addition, the goodness of fit of GWR was compared with that of OLS in three ways. First, the AICc and R² were detected. The best model generally represents a lower AICc and higher R², and the difference in AICc values between the models should be at least 3 (Fotheringham et al., 2002). Second, the residual sum of squares (RSS) was then evaluated. A lower RSS value represents a better goodness of fit for dependent variables. Third, if the approximate likelihood ratio, RSS_OLS/RSS_GWR, based on the F-test, was greater than 1.028 (significant at the 5% level), the performance of the GWR model was considered to be better than that of the OLS model.

Results

Variation in TBLI by city in the YRD

During 2000–2010, the total biomass loss in YRD cities was estimated to be 2.97 Tg by equation (1). Shanghai has the largest biomass loss of 0.529 Tg, accounting for 17.82% of the total, followed by Nanjing (0.215 Tg), Ningbo (0.213 Tg), Suzhou (0.193 Tg), Hangzhou (0.151 Tg), Wuxi (0.134 Tg), and Changzhou (0.132 Tg). As shown in Figure 2 from left to right, the amount of biomass loss decreased generally within the cities in decreasing order according to population size. The TBLIs in megacities or major cities were much larger than those in medium or small cities, indicating that the city size has a great impact on biomass loss. However, Figure 2 also demonstrates that the curve of urban expansion area was not identical to the trend of TBLI, meaning that the single factor of urban encroachment alone cannot account for the changes in biomass loss observed here. Vegetation type may also have a considerable influence on regional biomass loss (Seto et al., 2012; Song and Deng, 2015; Wu et al., 2014; Xu et al., 2007).

Figure 2.

Urban expansion area and biomass loss.

With respect to the spatial distribution, Moran’s I for TBLI (0.2021) indicates that TBLI is positively autocorrelated with the YRD region. Approximately 52% of the cities were categorized as “Low–Low “(16) and “High–High” (10) from the scatter plot of Moran’s I. The “Low–Low” and “High–High” clustered patterns were located in Shanghai, Nanjing, Suzhou, and the south-eastern YRD (e.g. Ningbo, Shaoxing, and Fuyang), respectively, highlighting the spatial heterogeneity of TBLI.

Correlations of TBLI with urban form metrics

The Pearson correlation analysis of the pairs of urban form metrics demonstrates that some were significantly correlated. For example, Galster centrality was significantly correlated with the Gini coefficient and Moran’s I at the 1% level, with Pearson correlation coefficients equal to −0.840 and 0.752, respectively (Table 2). Hence these three variables should not be present in the same regression model.

Table 2.

Correlation coefficients between urban form indicators.

Urban form indicators	Richardson compactness	Density gradient	Continuity	Galster centrality	Gini coefficient	Moran’s I
Richardson compactness	1
Density gradient	.570**	1
Continuity	−.267	−.175	1
Galster centrality	−.641**	−.545**	.468**	1
Gini coefficient	.660**	.497**	−.208	−.840**	1
Moran’s I	−.599**	−.402**	.543**	.752**	−.670**	1

**Correlation was significant at the 0.01 level (two-tailed).

Moreover, the results of the partial correlation analysis between TBLI and the urban form indicators illustrated that Richardson compactness, density gradient, and Gini coefficient were the most important urban form indicators, successively, with partial correlation coefficients equal to −0.611, −0.597, and −0.547, respectively, significant at the 1% level (two-tailed) (Table 3). Here, Galster centrality was not a significant explanatory factor for TBLI. Although continuity is distinguished from the three urban form indicators by the correlation analysis, it was not significant according to partial correlation analysis. Among the three significant indicators, both Richardson compactness and density gradient described central urban agglomeration while the Gini coefficient depicted the city region. The indicators of the relationships between the core area and peripheral areas were not significant explanatory factors for TBLI. The Kolmogorov–Smirnov test verified that all three significant indicators were normally distributed, meeting the basic requirements of OLS regression analysis.

Table 3.

Partial correlation analysis between TBLI and urban form indicators setting urban expansion area as the control variable.

	TBLI	Richardson compactness	Density gradient	Continuity	Galster centrality	Gini coefficient	Moran’s I
TBLI
Correlation	1	−0.611	−0.597	−0.186	−0.500	−0.547	0.385
Significance (two-tailed)	–	0.000	0.000	0.201	0.000	0.000	0.006

TBLI: total biomass loss index.

The results of OLS illustrate that the estimation parameters of the independent variables, including the three urban form indicators and the urban expansion area, were all significant at the 0.05 level according to the T-test (Table 4) with an R² of 0.915. In the OLS model, when the other factors were fixed, increasing the value of Richardson compactness by one unit reduced the biomass loss by 0.091 Tg, similar to the density gradient and Gini coefficient with a loss reduction of 0.113 and 0.053 Tg, respectively.

Table 4.

Estimation parameters in the two models.

Urban form indicators	OLS				GWR regression parameters
Urban form indicators	parameter	t	Sig.	VIF	Maximum	Minimum	Average	SD
Richardson compactness	−0.091	−2.702	0.000	2.192	−0.081	−0.099	−0.089	0.005
Density gradient	−0.114	−3.485	0.001	1.586	−0.103	−0.119	−0.111	0.004
Gini coefficient	−0.053	−2.339	0.024	1.984	−0.048	−0.061	−0.055	0.002
Urban expansion area	0.001	28.364	0.000	1.682	0.001	0.001	0.001	0.000
Constant	0.118	6.890	0.000		0.124	0.114	0.118	0.002
RSS	0.008				0.007
AICc	−276.40				−280.10
R square	0.915				0.976

GWR: geographically weighted regression; OLS: ordinary least squares.

GWR model of TBLI with urban form metrics

The same variables as those of OLS were involved in the GWR model; the estimation parameters for the GWR model are also shown in Table 4. When GWR was compared with OLS, the improvement in model performance was evident by the AICc, R square, and RSS as shown in Table 4. The goodness of fit of the dependent variable was slightly improved in the GWR. RSS_OLS/RSS_GWR (based on the F-test) equalled 1.188, which is greater than an absolute F-value of 1.028 (significant at the 0.05 level), indicating that the GWR model had a slightly better performance than the OLS.

The results of the F-statistics test for the GWR estimation parameters indicated that the three selected urban form indicators should mainly be treated as local, despite the significance of the F-test for Richardson compactness over 0.05 (Table 5).

Table 5.

F-statistics test for the non-stationarity of the estimation parameters.

Independent variable	F statistics	Threshold	p-value
Intercept	0.707	0.824	0.018
Richardson compactness	2.868	0.815	0.058
Density gradient	4.892	0.873	0.004
Gini coefficient	4.323	0.882	0.007

The statistical significance of local parameter estimates over space should be further examined by local t-values derived from the GWR output. We used 2.403 as the critical value to accept or reject the null hypothesis, which was obtained by finding the critical value table of a two-tailed T distribution with 50 degrees of freedom and a significance level of 0.02. Parameter estimates with a Pseudo t-value between −2.403 and 2.403 are considered non-significant. The results showed that non-significant relationships between biomass loss and urban form metrics occurred in seven south-eastern cities for Richardson compactness and 22 western cities for the Gini coefficient, respectively. In particular, t-values for density gradient estimates were all significant over space. To address the multiple hypothesis testing problem, we used 3.54 as the critical t-value to assess the significance of parameter estimates again, which approximates the Bonferroni-adjusted critical t-value employed at the same significance level 0.02. We then found that the non-significant areas were extended more than with 2.403 as the critical value. However, as Fotheringham et al. (2002) argued, since the null hypothesis will be accepted too many times, the Bonferroni correction may be a little too conservative. The authors still utilized the absolute t-values in interpreting the interesting areas for further research. The urban areas with a non-significant parameter estimate were hidden by a grey mask in the map, to highlight the significantly distinguished areas.

The visualization of local estimation parameters is shown in Figure 3(a) to (c), illustrating the spatial variation of the relationships between urban form metrics and biomass loss. Richardson compactness was negatively correlated with TBLI in all cities, and the absolute value of local estimation parameters showed a clear decrease from the western to the eastern coastal cities (Figure 3(a)), with the lowest in Shanghai (−0.082). Figure 3(b) shows that the estimation parameters of the density gradient were significantly negative in all cities, with higher absolute values largely clustered around the Taizhou–Wenling–Linhai agglomeration relative to those in the north-western part. For the Gini coefficient (Figure 3(c)), estimation parameters were only negatively significant in eastern coastal cities, with unexpected estimates in Wenlin (−0.061), Taizhou (−0.061), Ningbo (−0.060), Fenhua (−0.060), and Linhai (−0.059) noticeably higher than in other cities according to absolute values. The higher the absolute value, the stronger the explanatory power of the urban form metric for biomass loss.

Figure 3.

Spatial distribution of the local estimation parameters of GWR. (a) Richardson compactness, (b) density gradient, and (c) Gini coefficient. ① Nanjing–Jurong–Yizheng–Yangzhou; ② Shanghai; ③ Taizhou, Wenling, and Linhai; ④ Ningbo, Fenghua, and Cixi.

Discussion

This study measured the urban expansion area within 10 years for the 50 sample cities, calculated the corresponding vegetation biomass loss, and obtained statistical relationships with the urban form metrics. Although the loss of vegetation biomass largely accounted for the area of urban expansion, both the partial correlation analysis and the OLS model revealed the significant explanatory power of some urban form indicators when controlling for urban expansion area, including Richardson compactness, density gradient, and the Gini coefficient. Richardson compactness, which depicts the shape of the central agglomeration, has significant negative effects on biomass loss. The higher the value, the smaller the perimeter of the same urban area, and the lower the biomass loss per unit area of urban expansion. The density gradient, referring to the concentration of built-up patches in the central agglomeration, also negatively influences the biomass loss. That is, the larger the value, the denser the central area, and the lower the mean biomass loss. The Gini coefficient, which describes the layout of urban patches at the city level, was negatively correlated with the biomass loss. The higher the value, the more clustering of the urban patches, and the lower the biomass loss. When consolidating the effects of the three indicators, we can infer that the urban expansion from the clustered urban areas with a compact and dense central agglomeration might yield an average lower vegetation biomass loss than from a scattered, loose layout. This conclusion is based on the statistical analysis of empirical data, showing that compact urban morphology has ecological superiority and is conducive to green space protection.

The statistical relationship between morphological parameters and vegetation biomass loss can be explained to a certain extent by the interface effects between urban and rural areas. Villasenor et al. (2016) showed that the structural elements of vegetation decrease near the urban boundary because of stronger anthropogenic impacts than natural effects. Some urbanization activities, such as recreation, infrastructure construction, and the abandonment of arable lands, were also observed and resulted in vegetation degradation in the periphery of Chinese cities (Li et al., 2016a). Thus, the development of the influence circle around highly developed cities tends to consume less biomass on average than scattered development in rural areas. Moreover, these statistical relationships seem consistent with the previously discussed law that the vegetation biomass (or carbon stock) increases as the distance from the city centre increases (Hutyra et al., 2011; Tao et al., 2015). This is because, compared with complex and loosely constructed cities, a compact city allows development along the boundary to be closer to the centre of the city. Given the increase in the urban vegetation biomass from the centre outward, there is less biomass loss due to development closer to the centre of a city. This reasoning is based on the reported findings and explains, to some extent, the rationality of the statistical relationships. However, more studies are needed on the interface effects and the distance effects on the distribution of the vegetation biomass.

According to the GWR model, the urban indicators in particular areas have different influences due to the spatial autocorrelation of vegetation, but the negative relationship with biomass loss has not changed. The spatial change in the estimated parameters indicates a kind of optimization strategy for the urban form from the perspective of vegetation conservation. For the Nanjing–Jurong–Yizheng–Yangzhou area (Figure 3(a)), characterized by hilly and mountainous territory and fragmented urban landscapes, improving the region’s Richardson compactness can significantly reduce biomass loss. A strategy of compact shape is suggested for the plan of these urban areas. For Shanghai, which is located in the delta plain with a river network and has relatively few vegetation resources and a near-circular city form, the improvement in Shanghai’s Richardson compactness would have no obvious effect on biomass loss. The Taizhou city group, represented by Taizhou, Wenling, and Linhai (Figure 3(b)), is naturally divided by a mountain, and most of the construction land is scattered along the traffic lines in the long valley and is not centripetally organized. The high proportion of forests in this area accounts for the largest biomass density among the 50 YRD cities. Therefore, an improvement in the density gradient in the region can effectively reduce the average biomass loss. Thus, a densification policy is suggested for these urban areas. In the Ning-Shao plain, including Ningbo, Fenghua, and Cixi (Figure 3(c)), industrial development is prominent and scattered at a variety of sites that are cleared of vegetation; thus, increasing the regional Gini coefficient may reduce the biomass loss. A clustering layout is suggested for these urban areas. Above all, for different cities, the GWR parameter estimates of the three morphological indicators for vegetation biomass loss are different. The larger the absolute value of the coefficient is, the stronger is the role of morphological indicators. Therefore, different optimization measures such as compact shape, densification, and concentration layout can be proposed as the form of particular cities.

The method used here and results from this paper can be applied to other cities. The TBLI can be used to describe the ecological loss in vegetation biomass caused by urban expansion. The statistical analysis in this study showed that compact urban form is conducive to reducing biomass loss caused by urban expansion, providing strong evidence for the implementation of a compact city policy. The compactness here includes three attributes: shape compactness, density, and degree of regional agglomeration. The analysis of GWR shows that cities with different terrains should emphasize different compact features. For example, the cities in hilly areas with gentle slopes should draw special attention to compact shape, while mountainous cities should improve density to reduce biomass loss. The urban planning in coastal plain areas should avoid excessive decentralization and emphasize the development of central agglomeration in order to reduce biomass loss most effectively.

Within the limitations of the methods used, the accuracy of the regional biomass estimation can be improved in future studies. Although the estimation based on multi-source data sets is a commonly used method for macro-scale studies (Fang et al., 2007; Sun and Yue, 2012), the data assimilation techniques affected the accuracy of the results to varying degrees. There were errors in the absolute value estimation of the biomass due to the lack of data on forest age and types on the 250 × 250 grid scale. The use of 250 m data also has the problem that small vegetation cannot be identified, but it does not have a great impact on the results. Because all the urban parks and public green patches are protected by law, they are scarcely lost by urban expansion in the study area. Considering that the comparison of biomass loss was applied at the whole city scale, the relative error was not significant and the statistical analysis was accepted. In addition, for a single city, if the survey data of vegetation distribution are available, a detailed spatial distribution map of vegetation biomass can be formed. The TBLI index can be used to calculate the existing biomass of any planned plot and the loss of biomass under particular planning scenarios, providing an ecological loss plan for urban planning.

Conclusions

In conclusion, the TBLI was developed to assess the ecological cost of urban expansion based on 50 YRD cities in China. The urban form metrics of Richard compactness, density gradient, and the Gini coefficient were significantly negatively related with biomass loss. It was inferred that a clustered layout of urban areas with a compact and dense central agglomeration leads to low-cost urban growth according to vegetation biomass. Various specific urban form strategies were proposed for different cities based on the GWR model parameters. The approach presented here provides an alternative tool for assessing urban form changes that occur with additional or extensive development. To further explore the mechanisms by which the urban form influences biomass loss, the interface effects and the distance effects of urban expansion on vegetation should be investigated in the near future. In addition, more sampled cities should be involved in the statistical analysis to look for common relationships between biomass loss and urban form indicators. It is noteworthy that vegetation is just one of the key criteria, and healthy and comfortable living conditions with moderate density are also very important for a sustainable development pattern.

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work is supported by the National Natural Science Foundation of China (Grant Nos. 41371179 and 41771140).

Tong Zhang graduated from the Normal University of South China and received her Master degree of GIS application to urban modelling in 2015. Now she is at the third year of study for a PhD in the University of Chinese Academy of Sciences. Her research focuses on the evolution and reconstruction of urban ecological space in the perspective of socio-ecological system.

Sophia Shuang Chen is a Professor of urban ecology and environmental planning in Key Laboratory of Watershed Geographic Sciences at Nanjing Institute of Geography and Limnology of CAS. She received her PhD at University of Hong Kong in 2002. She directs Master and PhD Programs of human geography at University of Chinese Academy of Sciences. Chen’s research interests are in geography and urban ecology. Her research focuses on dynamics of urban expansion in developing countries, interactions between urban development patterns and ecosystem function, and nature-based solutions for urban development. She leads research on ecological impacts assessment of urban growth in the Yangtze River Delta of China and in the East African cities.

Guangyu Li is a Scientist in Nanjing Institute of Environmental Sciences, Ministry of Environmental Protection of China. He received his PhD at Key Laboratory of Watershed Geographic Sciences at Nanjing Institute of Geography and Limnology of CAS in 2015. His research focuses on methods to assess terrestrial ecosystem services based on remote sensing data.

References

Akaike

(1973) Information theory and an extension of the likelihood principle. International Symposium on Information Theory 1: 610–624.

Alberti

Marzluff

Shulenberger

, et al. (2003) Integrating humans into ecology: Opportunities and challenges for studying urban ecosystems. Bioscience 53(12): 1169–1179.

Anselin

(1995) Local indicators of spatial association – LISA. Geographical Analysis 27(2): 93–115.

Bekara

Hafidi

Fleury

, et al. (2005) Smoothing parameter selection in nonparametric regression using an improved Kullback Information Criterion. In: Proceedings – 8th International Symposium on Signal Processing and its Applications, vol. 2: 887–890.

Brown

Gillespie

AJR

Lugo

(1991) Biomass of tropical forests of South and Southeast Asia. Canadian Journal of Forest Research 21(1): 111–117.

Brueckner

Fansler

(1983) The economics of urban sprawl: Theory and evidence on the spatial sizes of cities. The Review of Economics and Statistics 65(3): 479–482.

Brunsdon C, Fotheringham A S and Charlton M (1998) Geographically weighted regression-modelling spatial non-stationarity. Journal of Royal Statistical Society 47(3): 431–443.

Bunker

(2014) How is the compact city faring in Australia? Planning Practice & Research 29: 449–460.

Burton

(2002) Measuring urban compactness in UK towns and cities. Environment and Planning B: Planning and Design 29(2): 219–250.

10.

Chen

Yan

Gao

, et al. (2015) Quantifying circular urban expansion patterns of compact Chinese cities: The case of Yangtze River Delta, China. Environment and Planning B: Planning and Design 42(2): 279–299.

11.

Clement

Orange

Williams

, et al. (2009) Drivers of afforestation in Northern Vietnam: Assessing local variations using geographically weighted regression. Applied Geography 29(4): 561–576.

12.

Cohen

(1988) Statistical Power Analysis for the Behavioral Sciences. Hillsdale, NJ: Lawrence Erlbaum Associates.

13.

Cole

(1964) Study of major and minor civil divisions in political geography. In: The 20th international geographical congress.

14.

Diniz ‐ Filho

JAF

Bini

Hawkins

(2003) Spatial autocorrelation and red herrings in geographical ecology. Global Ecology and Biogeography 12(1): 53–64.

15.

Fang

Guo

Piao

, et al. (2007) Estimation of carbon sequestration of land vegetation in China 1981–2000. Science China D – Earth Science 6: 804–812 (in Chinese).

16.

Foody

Cutler

McMorrow

, et al. (2001) Mapping the biomass of Bornean tropical rain forest from remotely sensed data. Global Ecology and Biogeography 10(4): 379–387.

17.

Fotheringham

Charlton

Brunsdon

(1996) The geography of parameter space: An investigation of spatial non-stationarity. International Journal of Geographical Information Systems 10: 605–627.

18.

Fotheringham

Brunsdon

Charlton

(2002) Geographically Weighted Regression: The Analysis of Spatially Varying Relationships. West Sussex: John Wiley & Sons Ltd.

19.

Fuller

Gaston

(2009) The scaling of green space coverage in European cities. Biology Letters 5(3): 352–355.

20.

Galster

Hanson

Ratcliffe

, et al. (2001) Wrestling sprawl to the ground: Defining and measuring an elusive concept. Housing Policy Debate 12: 681–717.

21.

Haas

Furberg

Ban

(2015) Satellite monitoring of urbanization and environmental impacts – A comparison of Stockholm and Shanghai. International Journal of Applied Earth Observation and Geoinformation 38: 138–149.

22.

Herold

Couclelis

Clarke

(2005) The role of spatial metrics in the analysis and modeling of urban land use change. Computers Environment & Urban Systems 29(4): 369–399.

23.

Wang Z

Liu

, et al. (2006) Vegetation carbon storage of major shrub lands in China. Journal of Plant Ecology 30(4): 539–544 (in Chinese).

24.

Huang

Sellers

(2007) A global comparative analysis of urban form: Applying spatial metrics and remote sensing. Landscape and Urban Planning 82(4): 184–197.

25.

Hurvich C M, Simonoff J S and Tsai C (1998) Smoothing parameter selection in nonparametric regression using an improved akaike information criterion. Journal of Royal Statistical Society 60(2): 271–293.

26.

Hutyra

Yoon

Alberti

(2011) Terrestrial carbon stocks across a gradient of urbanization: A study of the Seattle, WA region. Global Change Biology 17(2): 783–797.

27.

Jenks

Burton

Williams

(1996) The Compact City – A Sustainable Urban Form? London: Spon Press.

28.

Jim

Chen

(2003) Comprehensive greenspace planning based on landscape ecology principles in compact Nanjing city, China. Landscape and Urban Planning 65(3): 95–116.

29.

Kabisch

Haase

(2013) Green spaces of European cities revisited for 1990–2006. Landscape and Urban Planning 110: 113–122.

30.

Kindermann GE, McCallum I, Fritz S, et al.(2008) A global forest growing stock, biomass and carbon map based on FAO statistics. Silva Fennica 42(3): 387–396.

31.

Kong

Nakagoshi

(2006) Spatial-temporal gradient analysis of urban green spaces in Jinan, China. Landscape and Urban Planning 78(3): 147–164.

32.

Legendre

(1993) Spatial autocorrelation: Trouble or new paradigm? Ecology 74(6): 1659–1673.

33.

Li B, Chen D and Wu S. (2016a) Spatio-temporal assessment of urbanization impacts on ecosystem services: Case study of Nanjing City, China. Ecological Indicators 71: 416–427.

34.

Zhang

Chen

, et al. (2016b) Assessing the impact of urban development on net primary productivity during 2000–2010 in Taihu Basin. Environmental Earth Sciences 75(18): 1266.

35.

Liu

Zhou

Wei

, et al. (2012a) The spatial distribution of forest carbon sinks and sources in China. Chinese Science Bulletin 57(14): 1699–1707.

36.

Liu

Zhou

Shu

, et al. (2012b) The estimating of the spatial distribution of forest biomass in china based on remote sensing and downscaling techniques. Acta Ecologica Sinica 32(8): 2320–2330 (in Chinese).

37.

Mennis

(2006) Mapping the results of geographically weighted regression. The Cartographic Journal 43(2): 171–179.

38.

National Bureau of Statistics (2009) China Statistical Yearbook. Beijing: China Statistics Press.

39.

Noori

Hassan

Mustafa

(2014) Spatial estimation of rainfall distribution and its classification in Duhok governorate using GIS. Journal of Water Resource and Protection 6(2): 8.

40.

Piao

Fang

, et al. (2004) Spatial distribution of grassland biomass in China. Chinese Journal of Plant Ecology 28(4): 491–498 (in Chinese).

41.

Rérat

(2012) Housing, the compact city and sustainable development: Some insights from recent urban trends in Switzerland. International Journal of Housing Policy 12: 115–136.

42.

Richardson

(1973) The Economics of Urban Size. Farnborough and Lexington, Mass: Saxon House.

43.

Saatchi

Harris

Brown

, et al. (2011) Benchmark map of forest carbon stocks in tropical regions across three continents. Proceedings of the National Academy of Sciences of the United States of America 108(24): 9899–9904.

44.

Schwarz

(2010) Urban form revisited-selecting indicators for characterising European cities. Landscape and Urban Planning 96(1): 29–47.

45.

Seto

Gueneralp

Hutyra

(2012) Global forecasts of urban expansion to 2030 and direct impacts on biodiversity and carbon pools. Proceedings of the National Academy of Sciences of the United States of America 109(40): 16083–16088.

46.

Song

Deng

(2015) Effects of urbanization-induced cultivated land loss on ecosystem services in the North China Plain. Energies 8(6): 5678–5693.

47.

Song

Knaap

G-J

(2004) Measuring urban form: Is Portland winning the war on sprawl? Journal of the American Planning Association 70(2): 210–225.

48.

Sperandelli

Dupas

Dias Pons

(2013) Dynamics of urban sprawl, vacant land, and green spaces on the metropolitan fringe of Sao Paulo, Brazil. Journal of Urban Planning and Development 139(4): 274–279.

49.

Sun

Chen

(2017) Effects of green space dynamics on urban heat islands: Mitigation and diversification. Ecosystem Services 23: 38–46.

50.

Sun

Yue

(2012) Effects of China future land use change on aboveground vegetation biomass. Chinese Journal of Applied Ecology 23(8): 2225–2232 (in Chinese).

51.

Sushinsky

Rhodes

Possingham

, et al. (2013) How should we grow cities to minimize their biodiversity impacts? Global Chang Biology 19: 401–410.

52.

Tao

Liu

, et al. (2015) Variation in ecosystem services across an urbanization gradient: A study of terrestrial carbon stocks from Changzhou, China. Ecological Modelling 318: 210–216.

53.

Tong

Ding

(2011) A method for planning mandatory green in China. Computers Environment & Urban Systems 35(5): 378–387.

54.

Tratalos

Fuller

Warren

, et al. (2007) Urban form, biodiversity potential and ecosystem services. Landscape and Urban Planning 83(4): 308–317.

55.

Tsai

(2005) Quantifying urban form: Compactness versus ‘sprawl’. Urban Studies 42(1): 141–161.

56.

United Nations (2014) World Urbanization Prospects. New York: United Nations.

57.

van Vliet

Eitelberg

Verburg

(2017) A global analysis of land take in cropland areas and production displacement from urbanization. Global Environmental Change 43: 107–115.

58.

Villasenor

Blanchard

Lindenmayer

(2016) Decline of forest structural elements across forest urban interfaces is stronger with high rather than low residential density. Basic and Applied Ecology 17(5): 418–427.

59.

Zhou

Chen

, et al. (2014) Determining the contributions of urbanisation and climate change to NPP variations over the last decade in the Yangtze River Delta, China. Science of the Total Environment 472: 397–406.

60.

Liu

, et al. (2007) Assessing the impact of urbanization on regional net primary productivity in Jiangyln County, China. Journal of Environmental Management 85(3): 597–606.

61.

Shao

Shi

, et al. (2009) How does the conversion of land cover to urban use affect net primary productivity? A case study in Shenzhen city, China. Agricultural and Forest Meteorology 149(11): 2054–2060.

62.

Zhou

Huang

Cadenasso

(2011) Does spatial configuration matter? Understanding the effects of land cover pattern on land surface temperature in urban landscapes. Landscape and Urban Planning 102(1): 54–63.

63.

Zhou

Wang

(2011) Spatial-temporal dynamics of urban green space in response to rapid urbanization and greening policies. Landscape and Urban Planning 100(3): 268–277.

64.

Zhu

Pan

Zhang

(2007) Estimation of net primary productivity of Chinese terrestrial vegetation based on remote sensing. Chinese Journal of Plant Ecology 31(3): 413–424 (in Chinese).