Disentangling the roles of land-use-related drivers on vegetation greenness across China

2.1.1. Satellite-based LAI

2.1.2. Ancillary data

2.2.1. Trend analysis

2.2.2. Optimizing the integration of model ensemble

2.2.3. Quantifying the contributions of LURDs at different scales

To investigate the trend in greenness changes that could be attributed to LURDs, we estimated the trend in LAI from the IPM optimized model ensemble based on the factorial simulations (cf sections 2.2.2 and 2.2.3) on a per-pixel basis, at the national and regional levels, and various different vegetation cover types (figure 1). The eight economic regions with significant characterization have been defined by the development research center of the state council in China (Liu et al 2012b). The eight economic regions include Northwest China (NWC), Southwest China (SWC), the Middle Reaches of the Yangtze River (MRYTR), the Middle Reaches of the Yellow River (MRYLR), Southern Coastal China (SCC), Eastern Coastal China (ECC), Northern Coastal China (NCC), and Northeast China (NEC).

Zoom In Zoom Out Reset image size

The land cover information came from the Moderate Resolution Image Spectroradiometer product (MCD12Q1 version 6) at 500 m × 500 m spatial resolution (Friedl and Sulla-Menashe 2015). It was also resampled to a common (1/12)° grid using the land cover type that the most pixels were. The International Geosphere-Biosphere Programme (IGBP) land cover classification provided by the MCD12Q1 product is used in this study. The IGBP classification types are aggregated into five broad land types-forests, grasslands, croplands, urban, and other nonvegetated types. Forests include evergreen/deciduous needleleaf forests, evergreen/deciduous broadleaf forests, and mixed forests. Besides, the closed/open shrublands and woody savannas/savannas are also aggregated into forests due to the small areas of shrublands dominated by woody perennials with 1–2 m height and the woody savanna with tree cover 10%–60% (canopy >2 m). Grasslands are just the grasslands in MCD12Q1 IGBP products. Croplands include croplands and mosaics of natural vegetation and croplands. Urban include urban and built-up lands. Other nonvegetated types include permanent wetlands, permanent snow and ice, barren and water bodies. The LURDs effects on vegetation greenness in the above-mentioned vegetated types were discussed because the obvious LURDs mainly occurred in these areas.

2.2.4. Disentangling the roles of LURDs with the BMA method

The BMA method was used to confirm the roles of LURDs on vegetation greenness. It has been generally used to address the problem of statistical model uncertainty and identify the explanatory variables, which are most likely to determine the dependent variables in many fields (Fernández et al 2001, Horváth 2013, Osada et al 2015, Duc et al 2017). The idea of the BMA lies in estimating the posterior model probabilities that result from Bayes' theorem. Then, the posterior inclusion probability (PIP) and the posterior mean (PM) of explanatory variables were estimated to determine the relative importance and the action directions of explanatory variables based on the posterior model probabilities. Specifically, let us assume that the total number of explanatory variables is K, and the total number of possible individual models is ${2^K}$. We consider numerous different models ${M_1}$,...${M_j}$, where $j \in \left[ {1,{2^K}} \right]$, grouped in the model space M. Thus, the posterior model probability of ${M_j}$ is obtained as follows (equation (1)):

$$\begin{align}P\left( {{M_j}{\text{|}}y,Z} \right) &= \frac{{P\left( {y|{M_j},Z} \right)P\left( {{M_j}} \right)}}{{P\left( {y,Z} \right)}} \nonumber\\ &= \frac{{P\left( {y|{M_j},Z} \right)P\left( {{M_j}} \right)}}{{\mathop \sum \nolimits_{i = 1}^{{2^k}} P\left( {y|{M_i},Z} \right)P\left( {{M_i}} \right)}}\end{align} \tag{ 1 }$$

where y is a dependent variable, Z is the matrix of explanatory variables, $P\left( {y|{M_j},Z} \right)$ is the marginal likelihood of model ${M_j}$, $P\left( {y,Z} \right)$ is the integrated likelihood, and $P\left( {{M_j}} \right)$ is the prior model probability. The unit information prior for coefficient priors, and the uniform model priors were used in this work, according to previous studies (Duc et al 2017). Then, the $P\left( {{M_j}{\text{|}}y,Z} \right)$ is obtained by using the Markov chain Monte Carlo Model Composition method. The detailed calculations of $P\left( {{M_j}{\text{|}}y,Z} \right)$ can be found in Fernández et al (2001) and Zeugner and Feldkircher (2015).

According to the $P({M_j}|y,Z)$ in equation (1), we compute the PIP as the sum of the models' posterior model probabilities, including the variable k, which denotes the probability that a particular regressor $\theta$ is involved in the true model (equation (2)):

$$\begin{align}{\text{PIP}} &= P\left( {{\beta _\theta } \ge 0|y,Z} \right) \nonumber\\ &= \sum\limits_{j = 1}^{{2^k}} P\left( {{\beta _{j,\theta }}|{M_j},y,Z} \right)P({M_j}|y,Z)\end{align} \tag{ 2 }$$

where ${\beta _\theta }$ is the ${\text{coefficient}}$ ($\beta$) of a particular regressor $\theta$, $P\left( {{\beta _{j,\theta }}|{M_j},y,Z} \right)$ is the posterior probability of ${\beta _{j,\theta }}$ in ${M_j}$.

The PM of the distribution of $\beta$ is as follows (equation (3)):

$$\begin{equation}E\,\left( {\beta {\text{|}}y,\,Z} \right) = \mathop \sum \limits_{j = 1}^{{2^k}} E\left( {{\beta _j}|{M_j},y,Z} \right)P({M_j}|y,Z)\end{equation} \tag{ 3 }$$

where $E\,\left( {\beta {\text{|}}y,\,Z} \right)\,$ is the averaged coefficient, and $E\left( {{\beta _j}|{M_j},y,Z} \right)$ is the evaluation of the ${\beta _j}$ from model ${M_j}$.

Eicher et al (2011) and Feldkircher et al (2014) have provided the detailed interpretation of the PIP and PM. The PIP of an explanatory variable with less than 0.5 is characterized as the least weak evidence, while a PIP with larger than 0.5 is characterized as effective evidence. Moreover, it is regarded as weak (0.5–0.75), substantial (0.75–0.95), strong (0.95–0.99), and decisive (0.99+) evidence. In addition, the sign of PM is positive, indicating that the increase in explanatory variables has a positive effect on vegetation greenness, and vice versa.

The contribution of LURDs on the trends in LAIgs (cf section 2.2.2) at the provincial level (31 main provincial administrative regions) was regarded as the dependent variable for the BMA method. Then, we selected the trends in nine potential land-change processes as the explanatory variables in 31 main provincial administrative regions of China from 2000 to 2014, according to the statistical data on land-change processes. The nine statistical land-change processes were detailed descriptions in table S1 (available online at stacks.iop.org/ERL/16/124033/mmedia).

Vegetation greenness, as characterized by the satellite-observed LAIgs, exhibited significant (P < 0.05) increasing trends over 35% of the land area of China from 2000 to 2014 (figure 2). It indicated that a greening pattern was strikingly prominent across China. The notable increase (larger than 0.05 ${{\text{m}}^2}\,{{\text{m}}^{ - 2}}\,{\text{y}}{{\text{r}}^{ - 1}}$) in greenness (greening) was generally observed in forested areas of four economic regions: SWC, MRYLR, MRYTR, and SCC (figures 2 and 3).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

In contrast, the satellite-observed LAIgs trend showed that the LAIgs significantly decrease (P < 0.05) in less than 6% of the land area of China and are sparsely distributed in the two economic regions NWC and ECC (figure 2). This indicated that the areas of vegetation greenness decreases (browning) were relatively small. NWC exhibited that the LAIgs decreased by ⩽−0.025 m² m⁻² yr⁻¹ in its northern and southern parts with the most rural areas. Compared to NWC, a notable decrease in the LAIgs (less than −0.05 m² m⁻² yr⁻¹) occurred in the central part of ECC with a lot of urban land covers (figures 2 and 3).

The nationally averaged LURDs resulted in a significant (P < 0.05) increase in China's LAIgs (figure 3). The increased magnitude was $0.7 \times {10^{ - 3}}$ m² m⁻² yr⁻¹, which was approximately 7.9% of the nationally averaged LAIgs trend estimated by satellite-observed datasets ($8.9 \times {10^{ - 3}}\,{{\text{m}}^2}\,{{\text{m}}^{ - 2}}\,{\text{y}}{{\text{r}}^{ - 1}}$) (figures 2 and 4).

Zoom In Zoom Out Reset image size

The trends in vegetation greenness that were attributed to LURDs were then further estimated at the eight economic regions and on a per-pixel basis (table 3 and figure 3). The positive and negative differences between scenario S3 and scenario S2 indicated the positive and negative effects of LURDs on greenness, respectively. We then classified the model differences in three levels using the following values as threshold values: the average LURDs contribution in the trend of the LAIgs $\left( {0.7 \times {{10}^{ - 3}}\,{{\text{m}}^2}\,{{\text{m}}^{ - 2}}\,{\text{y}}{{\text{r}}^{ - 1}}} \right)$ and the observed average LAIgs trend $\left( {8.9 \times {{10}^{ - 3}}\,{{\text{m}}^2}\,{{\text{m}}^{ - 2}}\,{\text{y}}{{\text{r}}^{ - 1}}} \right)$. The absolute difference between scenario S3 and scenario S2 was higher than the observed average LAIgs trend, which was regarded as a highly significant effect. The absolute difference was less than the average LURDs contribution, which was characterized as a mildly significant effect. The absolute difference value between $0.7 \times {10^{ - 3}}$ and $8.9 \times {10^{ - 3}}\,{{\text{m}}^2}\,{{\text{m}}^{ - 2}}\,{\text{y}}{{\text{r}}^{ - 1}}$was characterized as a moderately significant effect. Generally, the LURDs effects on the LAIgs showed a strong spatial variation across the mainland of China (table 3 and figure 3).

Our results suggested that LURDs had led to significant (P < 0.05) increases of the LAIgs over 19.11% of the land area of China (table 3). The significantly positive LURDs effects were mainly distributed in SCC, SWC, MRYTR, MRYLR, and NEC, which had dense forest cover with part cropland cover (figures 2, 4, and S1(a)). SCC had the greatest area fractions (up to 35.82% of its corresponding area) of significant positive effects of LURDs on the LAIgs trends, following by SWC (32.47% of its corresponding area) (table 3). However, SWC and NEC were dominated by the high significant positive effects of LURDs on the LAIgs trends. SCC, MRYTR, and MRYLR were dominated by the moderate significant positive effects of LURDs on LAIgs trends. Meanwhile, the statistical data on land-change processes showed that SWC had increasing trends in forest areas with a rate of 0.0090 ${\text{y}}{{\text{r}}^{ - 1}}$, which was higher than SCC, MRYTR, and MRYLR (figure 4(a)). NEC experienced the largest significant increase in crop yield (figure 4(b)). These suggested that the effects of LURDs on vegetation greening were primarily related to the increases in forest area (afforestation) and improved crop yield (advanced agriculture), but the effect of the latter was just observed in parts of cropland-dominated areas, which was far less than that of the former.

Compared to the positive effects of LURDs on the LAIgs, the area of LURDs negative effects on LAIgs trends was relatively small at a national scale. It had led to significant decreases in LAIgs over 11.83% of China's land (table 3). The decreases in LAIgs trends due to LURDs mainly occurred in ECC and NCC with much urban cover, in parts of NWC with much grassland cover, and sporadically occurred in MRYTR and SWC with forest cover (figures 2 and 4). NWC (18.78% of its corresponding area) had greater area fractions of significant negative effects of LURDs on LAIgs trends than other economic regions, following by ECC and NCC (9.14% and 8.62% of its corresponding area, respectively). However, ECC and NCC were dominated by large significantly negative effects of LURDs on LAIgs trends, while NWC was dominated by the moderate and mild significant negative effect of LURDs on LAIgs trends. Besides, MRYTR and SWC had the area fractions of significant negative effects of LURDs on LAIgs trends with 8.11% and 7.29%, respectively, in which large significant negative effects of LURDs on the LAIgs trends were dominated. Meanwhile, the statistical data on land-change processes showed that ECC had significantly increasing trends in urban areas, with the largest rates of 0.00145 yr⁻¹, followed by NCC (figure 4(a)). MRYTR experienced the largest significantly increasing trend (0.00059 yr⁻¹) in areas of forest biodisasters and fire disasters (figure 4(a)), whereas SWC had the affected grassland areas by biodisasters and fire disasters with the largest significantly increasing trend (0.00195 yr⁻¹). Besides, NWC had the largest increasing trend in total livestock (170.5 $\times {10^4}\,{\text{heads}}\,{\text{y}}{{\text{r}}^{ - 1}}$) (figure 4(b)). These results indicated that LURDs effects on vegetation browning were mainly associated with the increased urban area (urbanization), biodisasters and fire disasters of forests, biodisasters and fire disasters of grasslands, and increased total livestock (overgrazing). However, the magnitude of urban expansion on vegetation browning was larger than that of these latter factors, although the urban area in China was relatively small, accounting for approximately 1.4% of the land area of China.

To disentangle the roles of the main LURDs on the observed greenness changes in China, we introduced a BMA method based on two scenario simulations and statistical data on land-change processes. As shown in table 4, forest area change, urban green space area, crop yield, grassland area change, and crop area change showed that the signs of PM are positive. It indicated that the five abovementioned changes had positive effects on vegetation greenness. Moreover, the PIP of forest area change was 0.998, and China exhibited significant and large increasing trends in forest areas in eight economic regions (figures 2 and 5(a)). This indicated that afforestation had a decisive role in explaining vegetation greening in China. The PIPs of urban green space area, crop yield, grassland area change, and crop area change were less than 0.5, and they had significant increasing trends in most economic regions (figure 4). This indicated that urban greening, advanced agriculture, and grassland restoration had the least effective roles in explaining vegetation greening.

Zoom In Zoom Out Reset image size

In contrast, the PMs of urban area, the biodisasters and fire disasters area of forests and grasslands, and the total livestock indicated that the increases in these drivers had negative effects on vegetation greenness. Moreover, the PIP of urban area and the area of forest biodisasters and fire disasters were larger than 0.5, indicating that they were effective land–change processes that explain vegetation browning. The PIP of urban area change was 0.997, and the urban area of China showed significantly increasing trends. This indicated that urbanization had a decisive role in vegetation browning. The forest biodisasters and fire disasters had a substantial role in vegetation browning, with a PIP = 0.848, resulting from significant increasing trends in the area of forest biodisasters and fire disasters across forest-dominated regions such as MRYTR, SWC, and MRYLR (figures 2 and 5(a)). The PIPs of the total livestock and the area of grass biodisasters and fire disasters were less than 0.5, and they had significant increasing trends in some regions with excessive grassland cover such as NWC (figures 2 and 5). The phenomenon suggested that biodisasters and fire disasters or overgrazing in grassland areas had the least effective roles in explaining vegetation browning of the whole China mainland.

Our results showed that the spatial pattern of vegetation greenness change based on satellite measurements from 2000 to 2014, which was consistent with previous studies suggesting considerable local effects of LURDs on vegetation greenness (Liu et al 2016, Hua et al 2017, Chen et al 2019b, Miao et al 2020). The most notable vegetation greening was generally distributed in parts of SWC, MRYLR, MRYTR, and SCC. These regions are important forest-dominated areas with a series of large-scale conservation programs (figures 2 and S2) and have suitable climates for vegetation growth (Feng et al 2016, Li et al 2017, Deng et al 2017a). For example, two unprecedented conservation actions—the Natural Forest Conservation Program (NFCP) and the Grain to Green Program (GTGP) have been taken in China since 1999, which cover four out of five provincial administrative regions (Liu et al 2008). The notable vegetation browning was generally distributed in parts of NWC with rural areas characterized as ecological vulnerabilities (Liu et al 2008) and in parts of ECC with high urbanization. Additionally, significantly increased trends (0.0089 m² m⁻² yr⁻¹) in vegetation greenness were observed at the national scale from 2000 to 2014. Such general trends at the national scale were slightly larger than that (0.0070 m² m⁻² yr⁻¹) of the period from 1982 to 2009 (Piao et al 2015). Meanwhile, the patterns of satellite-observed LAIgs trends were consistent with the previous study (Piao et al 2015). It suggested that the trend in vegetation greening has continued in China during 2010–2014.

The IPM factorial simulations showed that the general effect of LURDs on vegetation greening was relatively limited if spatially average for China. There was almost identical evidence that the average effects of LURDs on observed LAI trends are relatively small on hemispherical or global scales (Zhu et al 2017, Chen et al 2019b). Moreover, such a positive effect was just 9.6% of that caused by both ${{C}}{{{O}}_2} + {{CLI}}$ (figure 5). It was consistent with previous studies that the enhanced ${\text{C}}{{\text{O}}_2}$ fertilization and climate change are the dominant causes of vegetation greening at a global or regional level (Piao et al 2015, Mao et al 2016, Zhu et al 2016, 2017, Chu et al 2019).

The effects of LURDs on vegetation greenness exhibited a strong spatial heterogeneity on a per-pixel basis (figure 3). The area with positive LURDs effects on greenness was overall greater than that with negative effects. Afforestation had a notable positive effect on vegetation greenness, widely distributed in densely forested areas in SCC, SWC, MRYTR, and MRYLR. It was consistent with previous results highlighting that the greening trend is the largest in southern China due to increased forest cover (Li et al 2018c). A series of regional and national afforestation projects have been developed and implemented across China to restore the degraded environment (Li et al 2019). At least three afforestation projects are implemented in the above-mentioned four economic regions, including GTGP, NFCP, and the corresponding regional forest projects (figure S2). Two of the four economic regions, SWC and MRYLR, have the largest cumulative afforestation area (figure S1(b)). Besides, timber forests (such as eucalyptus) are widely planted in the afforestation projects of the four regions. This has notable effects on the vegetation greening due to its specific characteristics as being fast-growing, highly adaptable, and fruitful (Liu et al 2017). Moreover, large significantly positive effects of LURDs were mainly distributed in parts of SWC with forest cover where the cumulative afforested areas of timber forests were larger than other economic regions. Meanwhile, large significantly negative effects of LURDs were sporadically distributed, mainly because of the negative effects of biodisasters and fire disasters of forests. It means that forest biodisasters and fire disasters generally pose some serious threats to vegetation growth (Wen et al 2016). However, the area and magnitude of afforestation effects were far greater than that of forest biodisasters and fire disaster effects (figure 3). Advanced agriculture could also lead to vegetation greening, mainly distributed in parts of cropland-dominated areas in NEC (figures 2 and 4). Based on the China Statistical Yearbooks from 2000 to 2014, the total crop yield had increased by 31.4%, and the total cropland area only increased by 3.9%. However, the effect of advanced agriculture on vegetation greening was not notable, probably resulting from the inconsistency of the various crops growing season in different regions (Wang et al 2017).

In contrast, urbanization had an important effect on vegetation browning. It was mainly distributed in densely urban areas in ECC and NCC, where the significantly increasing trends in urban areas were experienced from 2000 to 2014 (figure 4(a)). Although these regions had generally significantly increasing trends in the urban greenness space area, the negative effects of urbanization were far outweighed by such positive effects (figure 4(a)). This result is generally consistent with the findings of prior studies (Wu et al 2014, Li et al 2018b). Additionally, the biodisasters and fire disasters of grasslands and overgrazing also had local effects on greenness in grassland-dominated areas, but it had limited effects if spatially average for China (table 3 and figure 3). It was likely that the negative effects of grassland biodisasters and fire disasters, and overgrazing might be outweighed by the implementations of grassland restoration projects in China over recent years (Guo et al 2018). For example, the implementation of the Returning Grazing Land to Grassland Program (RGGP), which aims to alleviate grazing pressure through grazing exclusion, has obtained some achievements over the past two decades (Deng et al 2017b, Zhang et al 2018).

To further appreciate the contrasting effects of afforestation and urbanization on vegetation greenness, we compared the effects of LURDs and both ${{C}}{{{O}}_2} + {CLI}$ on observed LAIgs trends at four broad land types at the national (figure 5) and economic region (figure S3) scales. The magnitudes of the LURDs effects on vegetation greenness varied in different areas. However, LURDs related to afforestation and urbanization largely altered vegetation greenness at the national and regional levels. Compared to the regional scale, afforestation and urbanization made noticeable effects on vegetation greenness at the national scale. The residual LURDs effects were relatively limited, accounted for less than 8% of the corresponding observed greenness changes at the national scale (figure 5).

As shown in figure 5, LURDs related to increased forest area had a larger effect on observed greening in the forest areas than that in the cropland areas and the other land cover areas. Like previous studies of (Li et al 2018c, Chen et al 2019b), the increased forest area associated with large-scale afforestation should effectively result in vegetation greening. However, our study gave out a quantitative assessment of such a positive greening effect of afforestation in the forest area as 27%. And it only accounted for about half of the positive effects of both $C{O_2} + CLI$. It was likely that the negative effects of the damaged forest by fire, pests, and rodents may partly offset the positive effects of afforestation.

Compared to forest areas, LURDs in the urban areas exhibited the largest negative effect on vegetation greenness, while the magnitude of such negative effects was slightly less than the positive effects of afforestation on greenness (figure 5). Because the satellite-observed average LAIgs trend was negative in the urban area, the positive $C{O_2} + CLI$ effect had a 70.5% offset to the vegetation browning. If there was not the impact of $C{O_2} + CLI$ and other factors, the LURDs could cause approximately two times vegetation browning relatively to the satellite-observed average LAIgs trends in the urban area. This suggests that the magnitude of the negative effects of LURDs is about three times the positive effects caused by the $C{O_2} + CLI$. It highlights that urbanization has important effects on vegetation browning relative to the dramatic urbanization undergone in China over the last 15 years (Li et al 2018a, 2018b).

The BMA evidence made up for the deficiency of the IPM factorial simulations. It confirmed that afforestation had a decisive role in leading to vegetation greening, although the biodisasters and fire disasters of forests have substantial effects on vegetation browning. China experienced significant increasing trends in forest areas due to the widespread implementation of several afforestation projects (figure 4(a)). In contrast, urbanization had decisive effects on vegetation browning. The reason was likely because China experienced significant increasing trends in urban area changes related to rapid urban expansion since the 1990s (Liu et al 2014, Zhou et al 2016, He et al 2017, Sun and Zhao 2018). Additionally, advanced agriculture, urban greening, grassland restoration, overgrazing, and grass biodisasters and fire disaster made relatively small contributions to China's vegetation greenness at the national scale. It was likely that vegetation greenness was more susceptible to both $C{O_2} + CLI$ or other factors in the cropland and grassland areas (figure 5).

In summary, this study highlighted the importance of the combination between process-based ecosystem models and the BMA statistical method. The former was effective in separating the contributions of LURDs and biogeochemical factors. They could quantify the individual contributions of LURDs to the observed greening trend on a per-pixel basis, at the national and regional levels, and various vegetation cover types. Meanwhile, the latter based on the statistical data could characterize the roles of LURDs such as afforestation, urbanization, and agricultural management at statistical levels.

The results indicated that afforestation and urbanization as the two main types of LURDs had decisive roles in vegetation greening and browning in China, respectively. On the one hand, a series of ambitious revegetation programs have been successfully implemented in China to restore degraded ecosystems and mitigating climate change. Already 62 million hectares of current forests have been until 2008, occurring for the largest percentage of the world's afforestation, according to the seventh national forest resource inventory report (2004–2008) (Deng and Shangguan 2017, Li et al 2018c). The growth of newly planted forests could make further greening more significant. For example, Chen et al (2019b) demonstrated that China is one of the countries that lead in the greening of the world through human land-use management from 2000 to 2017. Other recent studies attest to the success of these revegetation programs in terms of restoring degraded ecosystem (Feng et al 2016), lowering the surface temperature (Yu et al 2020), improving carbon sequestration (He et al 2020), and providing biofuel for bioenergy production (Menz et al 2013), but a strain on water resources in arid and semi-arid environments has also noted (Tang et al 2018, Bai et al 2020). Furthermore, the Chinese government is projected to expand afforestation programs (Chen et al 2015). It can be expected that the afforestation would have significant effects on vegetation greening over the next several years. In addition to China, other regions of the world have also promoted afforestation in the present and future (Duque-Lazo et al 2018, Kemena et al 2018, Odoulami et al 2018, Yosef et al 2018, Zhang et al 2021). It is therefore believed that afforestation is a profound factor affecting vegetation greening and should be highlighted in the design and implementation of local ecosystem management plans.

On the other hand, the urbanization rate (urban population in China as a percentage of the total population) raised from 36.22% in 2000 to 54.77% in 2014, according to the National Bureau of Statistics of China. China's urbanization rate is expected to hit 80% by 2050 (United Nations and Social 2018). It suggested that the rapid urban expansion of China cannot be ignored by ecosystem management. Other current studies demonstrate that urban expansion reduces regional net primary productivity (NPP) due to the LURDs-induced direct effect (Guan et al 2019, Wen et al 2019). The decreases in NPP due to the urban expansion from cropland accounted for 63.02% of the total loss in NPP in mainland China, which may threaten the food security of China (Li et al 2018b). Meanwhile, other countries across the world have also experienced growing urbanization rates and their effects on vegetation dynamics over the last few decades (Cai et al 2019, Yao et al 2019, Nathaniel et al 2020). Large urbanization is projected to continue in the future across these countries, particularly in developing countries (United Nations and Social 2018). Thus, the local effects of urbanization on vegetation browning become greater with the accelerated process of urbanization and should be given increased attention at regional to global scales.

There are some uncertainties and limitations regarding the performance of the ecosystem model results in simulating the LURDs effects on vegetation greenness, despite the great efforts that have been undertaken to optimize the robustness of the ecosystem models. Future work needs to further improve the reliability of our results through the comparative analysis of our methodology and other novel methodology. For example, the novel approach to estimate LURDs effects is directly based on the difference between historic and current land cover maps and could also be considered in vegetation change attribution analyses (Pei et al 2013, Li et al 2018b). The simulation scenes of forest age, agricultural irrigation, and grazing management also need to be considered in future ecosystem models to explore the roles of LURDs on vegetation greenness.