Multi-scale analysis of urbanization and gross primary productivity during 2000–2018 in Beijing, China

The study area Beijing is the capital city of China. The climate of this region is characterized as a warm-temperate, semi-humid and semi-arid monsoon climate (Beck et al 2018). The annual average air temperature in this area was ∼13.35 °C, and the annual precipitation was ∼505.02 mm during 2000–2018. Beijing is divided into 16 county-level districts. Two districts (Dongcheng, Xicheng) are the central urban core, and nine districts (Fengtai, Chaoyang, Haidian, Shijingshan, Shunyi, Tongzhou, Daxing, Fangshan, and Changping) are central urban areas, and the other five districts (Pinggu, Miyun, Huairou, Yanqing, and Mentougou) represent suburban areas far from the central urban core. As one of the world's largest metropolises, Beijing has undergone rapid urban expansion in the past two decades. The built-up area has increased from 5575.25 km² in 2000 to 7059 km² in 2018, while the urban resident population has grown from 13.636 million to 21.54 million during the same period (figure 1).

Zoom In Zoom Out Reset image size

GPP data during 2000–2018: A GPP dataset derived from a satellite-based Vegetation Photosynthesis Model (VPM) (https://doi.org/10.1594/PANGAEA.928381) (Zhang et al 2017). This dataset has a spatial resolution of 500 m and a temporal resolution of 8 d. Given the significance of the carbon cycle in the urban environment, this study utilized the 2000–2018 inter-annual GPP dataset to track the dynamic changes and growth status of vegetation in Beijing. Compared to the MOD17A2 product dataset, this dataset provides a more accurate representation of the vegetation conditions in urban areas (Running and Zhao 2015, Zhang et al 2017). At the same time, studies have shown that the dataset has spatiotemporal consistency with SIF (including Beijing), further indicating the reliability of the dataset (Cui et al 2017, 2022).

Impervious surface data (IS): Annual IS data with 30 m spatial resolution during 2000–2018 was provided by Tsinghua University (Gong et al 2020) (http://data.ess.tsinghua.edu.cn). The fine-scale IS data was used to reflect the land cover information within urban areas. To differentiate GPP between urban and non-urban areas, the IS range was regarded as the urban area, and the large-scale voids (which may be urban parks, wetlands, or enclosed fields) that are less than 2 km² in the urban area are filled. Areas larger than 2 km² do not belong to urban area (Angel 2012). Urban area change reflects the degree of urbanization. To match the spatial resolution of GPP, the IS data with 30 m resolution were aggregated to 500 m, and ISC ratio within 500 m gridcells was calculated, ranging from 0 (full vegetation) to 1 (full impervious surface) (Zhong et al 2019).

Population data: Worldpop provides population data at a resolution of 1 km since 2000 (www.worldpop.org/). The data is projected using the Geographic Coordinate System, WGS84, and is reported in terms of the number of people per grid (Tatem 2017).

GDP data: The 2000, 2005, 2010, 2015, and 2019 spatial GDP is obtained from the Resource and Environmental Science and Data Center (www.resdc.cn/DataList.aspx), and is based on the national GDP statistics by county, land use type, settlements density, and nighttime light brightness. The data has a spatial resolution of 1 km, and is reported in unit of 10 000 RMB km⁻².

In this study, we employed classical statistical methods to analyze the trends of the GPP and three urbanization factors, as well as their correlations. To further delve the intrinsic correlation between urbanization and GPP, we introduced a methodological framework of growth offset with ISC and GPP to analyze the impact of ISC and non-ISC urbanization factors on GPP.

• (i)  
  Trend analysis: The change trends of GPP (/three urbanization factors) were calculated by the slope of the least squares regression. The formula is as follows:

  $$\begin{align}{\text{Slop}}{{\text{e}}_i} = \frac{{n{{\mathop \sum \nolimits}}_{i = 1}^n\left( {i \times {\text{GP}}{{\text{P}}_i}} \right) - {{\mathop \sum \nolimits}}_{i = 1}^ni \times {{\mathop \sum \nolimits}}_{i = 1}^n{\text{GP}}{{\text{P}}_i}}}{{n{{\mathop \sum \nolimits}}_{i = 1}^n{i^2} - {{\left( {{{\mathop \sum \nolimits}}_{i = 1}^ni} \right)}^2}}} \nonumber\\ \end{align} \tag{ 1 }$$

  where i is the year; the value is corresponding to GPP (/three urbanization factors) during 2000–2018.
• (ii)  
  Correlation analysis: The Pearson correlation coefficient expresses the degree of linear relationship between two variables. Its formula is as follows:

  $$\begin{equation}r = \frac{{\mathop \sum \nolimits \left( {a - \bar a} \right)\left( {b - \bar b} \right)}}{{\sqrt {\mathop \sum \nolimits {{\left( {a - \bar a} \right)}^2}{{\left( {b - \bar b} \right)}^2}} }}\end{equation} \tag{ 2 }$$

  where $\bar a$ and $\bar b$ are the means of the variables, and r is the correlation coefficients among the variables.
• (iii)  
  A conceptual framework of growth offset of urbanization: To systematically quantify the impact of urbanization on vegetation growth, Zhao et al (2016) has proposed a theoretical framework, that is, the framework involves decomposing urban GPP into contributions from both vegetation and non-vegetation land (figure 2). The framework is calculated as follows:

  $$\begin{equation}{\text{GP}}{{\text{P}}_{zi}} = \left( {1 - {\text{ISC}}} \right) \times {\text{GP}}{{\text{P}}_v} + {\text{ISC}} \times {\text{GP}}{{\text{P}}_{nv}}\end{equation} \tag{ 3 }$$

  where GPP_zi is the theoretical GPP with 500 m gridcells (Zhong et al 2019), GPP_v is the GPP corresponding to fully vegetated areas (ISC= 0, GPP =GPP_v ), and GPP_nv is the GPP corresponding to fully urbanized areas (ISC =1, GPP =GPP_nv ). GPP_v and GPP_nv are utilized to depict idealized situations with fully vegetated and fully urbanized areas, respectively. In theory, GPP_nv tends to zero, but mixed pixels of 500 m within a city always include vegetation due to the wide distribution of small-scale green spaces, parks, etc.

Zoom In Zoom Out Reset image size

Urbanization can have an impact on vegetation growth, which differs from the direct impact of land conversion (Zhong et al 2019). The relative change in GPP from the zero-impact line can be used to measure this indirect impact. The calculation formula for indirect impact ω_i is:

$$\begin{equation}{\omega _i} = \frac{{{\text{GP}}{{\text{P}}_{obs}} - {\text{GP}}{{\text{P}}_{zi}}}}{{{\text{GP}}{{\text{P}}_{zi}}}} \times 100\% \end{equation} \tag{ 4 }$$

where GPP_obs is the actual GPP at grid scale.

Then, comparing the indirect change in GPP resulting from urbanization to the direct changes in GPP caused by land conversion. Then, we can express the concept of growth offset by coefficient τ:

$$\begin{equation}{\tau _i} = \frac{{{\text{GP}}{{\text{P}}_{obs}} - {\text{GP}}{{\text{P}}_{zi}}}}{{{\text{GP}}{{\text{P}}_v} - {\text{GP}}{{\text{P}}_{zi}}}} \times 100\% .\end{equation} \tag{ 5 }$$

If τ is positive, it indicates that urbanization has had a positive impact on vegetation growth and has partially offset the direct negative impact (GPP_v −GPP_zi ). Conversely, if τ is negative, the direct negative impact of urbanization will be exacerbated.

The inter-annual trend of GPP generally increased during 2000–2018 in Beijing (figure S1). Among these districts, the GPP of Dongcheng, Xicheng, and southern Haidian, which are in the central urban core, grew slowly, with a growth rate lower than 10 gC m⁻² yr⁻² (figure S2). As expected, in the central urban areas, especially Daxing and Tongzhou, although there were areas with decreased GPP in these districts, the overall GPP was increased, with a mean GPP growth rate of 4.70 g C m⁻² yr⁻² and 5.62 gC m⁻² yr⁻², respectively, while the GPP of Miyun and Mentougou in the suburban areas increased significantly. Overall, Beijing's GPP showed a fluctuating increased trend, with the total GPP increasing from 15.56 Tg C in 2000 to 19.38 Tg C in 2018.

Three urbanization factors showed an increased trend with varying degrees during 2000–2018 (figure S2). For ISC, the inter-annual trend of ISC at grid scale showed that the increase of ISC in the central urban core was not obvious, while the ISC increased significantly in the central urban areas, such as Chaoyang, Changping and Daxing, the annual growth rate of ISC were 2.17%, 1.98%, and 1.72%, respectively. There 57% of grids showed a significantly increased trend in both ISC and GPP trends. Notably, a phenomenon was observed in the central urban core, where ISC remained basically stable while GPP increased. As for population and GDP, they had similar spatial-temporal changes, that is, the inter-annual trends of both all showed a decreasing pattern from the central urban core to the periphery, especially in Haidian and Chaoyang, where population and GDP showed high growth trends, the annual growth trend of population was 11.41 million people yr⁻¹ and 8.2 million people yr⁻¹, and GDP is 378.69 billion RMB yr⁻¹ and 357.46 billion RMB yr⁻¹, respectively. Overall, the relationships among GPP and ISC, population, and GDP in Beijing show a positive correlation, where GPP increases with increasing three urbanization factors (figure 3).

Zoom In Zoom Out Reset image size

Here we analyzed the relationships between GPP and three urbanization factors from three scales (city-district-grid scales). Surprisingly, their relationships exhibited different or even opposite characteristics at various scales. At city scale, GPP increased linearly with three urbanization factors during 2000–2018 (p <0.05) (figure S2). At district scale, ISC, population, and GDP showed significant correlations with GPP too. The correlation coefficient (r) between GPP and ISC was equal to 0.79 (p <0.05), r between GPP and population, and GDP was equal to 0.87 and 0.85, respectively (p < 0.05) (figure 4). Unlike the single linear correlation at city scale, GPP exhibited a nonlinear decreased trend with the increase of ISC, while a U-shape with the increase of population and GDP (figure 5). Moreover, Dongcheng, Xicheng, Haidian, Fengtai, and Chaoyang, whose GPPs were highly correlated with the three urbanization factors, especially after 2008, mainly appeared in the right half of the U-shape (tables S1–S3). Simultaneously, we found that several points with GDP in the range of 200–500 billion RMB are far away from the fitting curve in figure 4(c), which are mainly Haidian and Chaoyang from 2008 to 2014 (tables S1 and S3).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

At grid scale, the correlation coefficients between GPP and three urbanization factors exhibited a high strong degree of spatial heterogeneity. GPP was significantly positively correlated with the three urbanization factors in the central urban core (r >0, P <0.01) but negatively correlated in the suburban areas (r <0, P <0.01) (figures 6(a)–(c)). The relationship between mean GPP and ISC was consistent with that at district scale (figure 6(d)). The mean GPP decreased rapidly with increased population in the range of 0–20 000 people and tended to change slowly, which was lower than 500 g C m⁻² yr⁻¹ (figure 6(e)). Similarly, the mean GPP showed a decreased trend and then tended to be flat with the increase of GDP (figure 6(f)). To sum up, the results of information transfer at city-district-grid scales are different. Although the positive linear relationships between GPP and three urbanization factors pass the significance test at city scale, considering the spatial heterogeneity of GPP and urbanization, it shows that there was a nonlinear relationship when pushed down to the district and grid scale.

Zoom In Zoom Out Reset image size

Based on the above research, finer-scale grid data in 2018 and 2000 can be directly compared to reflect the impact of urbanization. Compared with 2000, the curve relationship between GPP and ISC became more prominent in 2018 and the relationship among GPP and population and GDP gradually showed a U-shaped or slowly decreased relationship, implying a reverse countervailing effect on GPP during urbanization may exist (figure 7). To further analyze the intrinsic correlation, we used the conceptual framework of growth offset of urbanization to quantify the impacts of ISC and other non-ISC factors (population, GDP, or urban management and climate (UMC)) on GPP year by year.

Zoom In Zoom Out Reset image size

Although the negative impact always exists, the positive impact of growth offset on vegetation GPP increased with urbanization (figure S3). Especially since 2008, the impact has increased obviously. In particular, the GPP of the whole vegetation area in 2008 was the maximum before 2015, 1309 g C m⁻². The mean GPP_v increased from 988.89 g C m⁻² in 2000 to 1404.2 g C m⁻² in 2018 and the mean GPP_nv increased from 148.36 g C m⁻² in 2000 to 451.57 g C m⁻² in 2018. When ISC < 0.1, urbanization has little impact on vegetation, mainly fluctuating at 0. With the increase of ISC, the positive impact of urbanization (mean ω) on vegetation gradually increased, reaching 35.69% and 53.94% in 2000 and 2018, respectively (figure 8(a)). The growth offset of urbanization (τ) offset the direct land conversion impact was 9.9% and 35%, respectively (figure 8(b)). Additionally, this growth offset was very stable with a certain ISC range. All these show that the growth offset will compensate partly negative impact of urbanization.

Zoom In Zoom Out Reset image size

Urban vegetation plays a critical role in the infrastructural of the urban ecosystem, as noted by Yang et al (2008). In the past, it was thought that urban GPP would decline because of the continuous expansion of urban impervious surface area (Trusilova and Churkina 2008). Liu et al (2018) quantified the impact of Wuhan's urban expansion on GPP in Wuhan City from 2000 to 2013 and showed that there was a decline of 0.31 TgC in Wuhan's GPP during the study period. Similarly, a study conducted in Indonesia by Nuarsa et al (2018) indicated that urban expansion had a negative overall impact on terrestrial GPP. Additionally, Ding et al (2021) conducted a study in southwest China which revealed that urban expansion partially offset the increase of terrestrial primary productivity brought by afforestation. However, these studies tended to consider the negative impact of urban expansion on terrestrial GPP from the increase of non-vegetation areas (impervious surface areas) during urban expansion.

In the context of carbon neutrality, various cities and regions have taken significant measures to mitigate and offset anthropogenic carbon emissions. Examples of these measures include the Mayor's Climate Protection Agreement and the Regional Greenhouse Gas Initiative (RGGI) (Hutyra et al 2011). China has also launched a large-scale ecosystem restoration project (Kuang 2020), and more Chinese cities have joined the C40 City Group (www.c40.org/), including Beijing and Shanghai. These projects aim to increase carbon storage and reduce carbon emissions in urban areas (Xia et al 2018). Measures like afforestation, landscape greening, irrigation, and regular fertilization management along with urban expansion have always provided more suitable conditions for urban vegetation growth. Additionally, urbanization can create a conducive environment for vegetation growth by mitigating the impact of natural disasters and overcoming limitations imposed by higher CO₂ concentrations and urban heat islands. Therefore, urbanization provides a suitable environment for vegetation (Buyantuyev and Wu 2009, Cui et al 2017, Li et al 2020, Liu et al 2023). This study utilized the GPP to evaluate the dynamic changes and growth status of vegetation in Beijing during 2000–2018. The result showed that the GPP of vegetation in Beijing is on the rise. Cui et al (2017) analyzed and calculated the annual total GPP of ten megacities in the world, including Beijing, from 2000 to 2014 under the 0.5°× 0.5° grid unit. The results showed that the inter-annual GPP of Beijing generally showed a fluctuating increased trend (Cui et al 2017). In recent years, studies by Sun et al (2020), Yang et al (2021) have shown that there was obvious greening in the urban center of Beijing. A previous study in Shanghai also found that the enhanced vegetation index (EVI) and GPP in the suburbs and rural areas decreased while the EVI and GPP in the inner city increased from 2000 to 2016, further indicating that the urban environment is conducive to vegetation growth (Zhong et al 2019). In this study, we found that in the central urban core with a high urbanization level, GPP still maintains continuous growth. Simultaneously, GPP increased relatively slowly at Chaoyang, Haidian and Fangshan, indicating that it may have the potential to further enhance urban vegetation in these districts (figure 9).

Zoom In Zoom Out Reset image size

Given the complex urban environment, many factors may influence the urban growth offset. Many studies have indicated that vegetation growth changes are often driven by multiple interactions of natural and human factors (Xie et al 2020, Shi et al 2021). Changes in human activities, climatic conditions, and ecological restoration and protection projects all have produced changes in urban GPP (Xie et al 2020). Urban expansion causes an increase in impervious surfaces, leading to a reduction in vegetation cover and perhaps in GPP (Wang et al 2018, Zhang et al 2020). However, urbanization affects the regional environment. The urban heat island, humidity, radiation, artificial management, and tree transplanting within urban areas provide favorable conditions for the growth of urban vegetation, which extends urban phenology and promotes the growth of urban GPP (Xu 2021). At the same time, we found a close non-linear relationship between ISC and GPP, as the decline of GPP caused by land conversion also be offset by the positive impact of urbanization factors (Zhao et al 2016). The current findings align with those of Cui et al (2022), who revealed that although urban expansion has a negative impact on GPP, this impact is partially counterbalanced by the UMC worldwide. We also found that there was a dividing point between positive and negative impact at ISC = 0.1 in Beijing, that is, the growth offset of urbanization is negative at ISC < 0.1, while the negative impact caused by the direct land conversion can be partly offset at ISC > 0.1. A previous study in Shanghai obtained a similar result but the dividing point for ISC was 0.2 (Zhong et al 2019). The reason may be the different geographical environments and development styles (Zhou et al 2014, Zhao et al 2016).

The relationships between GPP and the three urbanizations are different at city-district-grid scales. At city scale, these relationships reflect the overall changes in GPP, ISC, population, and GDP, and at grid scale, they reveal the spatiotemporal variation in detail. At district scale, they reflect the vegetation growth and socio-economic situations of each district. Furthermore, our study revealed that different districts were located at various positions in the U-shaped relationship. In districts such as Changping, Shunyi, Daxing, and Fangshan, which correspond to the left half of the U-shaped relationship (figure 5) (Kuang et al 2018, Zhang et al 2021). On the other hand, in districts such as Dongcheng, Xicheng, Haidian, Fengtai, and Chaoyang appeared the right half of the U-shaped relationship. Temporally, the positive correlation between GPP and population, and GDP in these districts also confirmed the positive impacts of urbanization (the right half of the U-shaped), particularly after 2008 (tables S1–S3). The 'Green Olympics' in 2008 and the subsequent hosting of the Horticultural Expo and World Expo have contributed to significant improvements for vegetation growth in districts such as the Olympic Games venue Chaoyang, as well as the core urban districts Dongcheng, Xicheng, and Haidian (figure 9) (Beyer 2006). Our findings imply that socioeconomic development plays an increasingly crucial role with urbanization in driving urban vegetation growth. However, the cost of water consumption, management, and protection, which are often related to non-native vegetation species in urban area, are also worthy of further discussion (Yang et al 2008).