The trend shift caused by ecological restoration accelerates the vegetation greening of China's drylands since the 1980s

The total area of China's drylands is approximately 5.45 million km², with the aridity index becoming progressively wetter from west to east. China's drylands include northeastern, northwestern, northern, and part of western China, and are composed of 20 provinces or municipalities (figure S1 available online at stacks.iop.org/ERL/17/044062/mmedia). According to the Climate Change Initiative's land cover products (mean NDVI values below 0.1 were excluded), the main land cover types and areas in China in 2000 were grassland (225.88 Mha), cropland (152.19 Mha), forestland (60.28 Mha), bare land (49.01 Mha), mosaic vegetation (32.22 Mha), sparse vegetation (22.62 Mha), and urban land (2.75 Mha) (table S1). Since the late 1970s, ecological restoration projects have been implemented in China's drylands. By 2015, a total of 66.07 Mha had been afforested under the implementation of the Three-North Shelter Forest Program, the Grain for Green Program, and the Beijing-Tianjin Wind and Sand Source Control Project (China 2017). The Grain for Green Program has been implemented on the Loess Plateau of China since 1998 (figure 1). The Three-North Shelter Forest Program and Beijing-Tianjin Wind and Sand Source Control Project have mainly been implemented in the Three-North region (figure 1).

Zoom In Zoom Out Reset image size

ERA5 reanalysis data were generated by the European Climate Model ECMWF in combination with satellite data (Hersbach et al 2020). The indicators used in this study include the surface 2 m temperature, soil water (0–7 cm), relative humidity, and cloud cover probability with a monthly average temporal resolution and 0.1° spatial resolution. The data are available at the following website: https://cds.climate.copernicus.eu/cdsapp#!/search?type=dataset. CO₂ concentration data were obtained from the Hawaii Weather Monitoring Station (https://climate.nasa.gov/vital-signs/carbon-dioxide/), and these data from 1982 to 2015 were used to represent CO₂ concentration variations in China's drylands. Vegetation indices were derived from GIMMS 3.1 NDVI from 1982 to 2015, with a temporal resolution of 15 days and a spatial resolution of 1/12° (Pinzon and Tucker 2014). Monthly data were generated by calculating the monthly averages, with a spatial resolution resampled to 0.1° in R 4.1. Regions with an annual mean NDVI value below 0.1 were excluded from the analysis. Monthly averages of the climate and vegetation indices from May to September were used as annual growing season averages. Land cover data were generated using global 300 m resolution European Space Agency-generated land cover data for 1992, 2000, and 2010 from the Climate Change Initiative (Kirches et al 2014). This land cover classification system includes 22 types, which were integrated into seven types: cropland, forest, grassland, sparse vegetation, mosaic vegetation, bare area, and urban area. The land cover data were compared across the three periods, and the areas showing the conversion of land cover types were grouped as 'changed'.

2.3.1. Calculation of VPD

The VPD represents the atmospheric moisture conditions. VPD was calculated using equation (1) (Venturini et al 2011)

$$\begin{equation}{\text{VPD}} = 0.61078 \times {e^{\frac{{17.27 \times {T_a}}}{{{T_a} + 237.3}}}} \times \left( {1 - {\text{RH}}} \right)\end{equation} \tag{ 1 }$$

where T_a represents land surface temperature (°C) and RH represents the surface relative humidity.

2.3.2. CO₂ fertilization effect

By combining the results of existing CO₂ enrichment experiments, Franks et al (2013) found a pattern of relative proportional changes in plant productivity at varying CO₂ concentrations and showed that this pattern occurred across plant types and initial concentrations. As shown in equation (2), CO₂ fertilization is related to the CO₂ concentration. As CO₂ accumulates in the atmosphere, the ratio of CO₂ fertilization relative to the original effects will increase

$$\begin{equation}{\text{Re}}{{\text{l}}_{\left( t \right)}} = \left[ {\frac{{\left( {{c_{a,\left( t \right)}} - {\Gamma ^ * }} \right)\left( {{c_{a0}} + 2{\Gamma ^ * }} \right)}}{{\left( {{c_{a,\left( t \right)}} + 2{\Gamma ^ * }} \right)\left( {{c_{a0}} - {\Gamma ^ * }} \right)}}} \right]\end{equation} \tag{ 2 }$$

where Rel_(t) represents the relative CO₂ assimilation rate (%), C_a,(t) represents the atmospheric CO₂ concentration (µmol mol⁻¹) at year t, and Γ* represents the CO₂ compensation point in the absence of dark respiration (µmol mol⁻¹). We set C_a0 to the CO₂ concentration in 1980 (∼339 µmol mol⁻¹) and Γ* to 40 (µmol mol⁻¹). We assume that the increased proportion of NPP dynamics is equal to the increased proportion of gross primary productivity dynamics and the NDVI is a proxy of NPP. Thus, NDVI_adjusted can represent vegetation greenness independent of the effect of CO₂ fertilization (equation (3)). The effect of CO₂ fertilization on vegetation trends is equal to the trend of NDVI_raw minus the trend of NDVI_adjusted

$$\begin{align}{\text{NDV}}{{\text{I}}_{{\text{adjusted}}}} = {\text{NDV}}{{\text{I}}_{{\text{raw}}}}/{\text{Re}}{{\text{l}}_{\left( t \right)}}.\end{align} \tag{ 3 }$$

2.3.3. Principal component regression analysis (PCR) and RESTREND

Considering the significant coupling between the above factors, such as VPD and SM, VPD and temperature, the PCR method can be used in this study to differentiate between climatic factors (Seddon et al 2016). The VPD, SM, temperature, and cloud cover data were standardized and then converted to four independent principal components (equation (4)). Multiple linear regression was performed between the principal components and the adjusted NDVI_adjusted values (equation (5)). The residuals in the relationship between NDVI_adjusted and climate factors were considered the changes caused by anthropogenic impacts. When the relationship between the principal components and NDVI_adjusted was not significant (P < 0.1), the effect of climate on vegetation dynamics was zero. The linear trend of the residual was calculated as the anthropogenic effect

$$\begin{equation}\left\{ \begin{gathered} P1 = a1{\text{VPD}} + b1{\text{SM}} + c1{\text{TEM}} + d1{\text{CLOUD}} \hfill \\ P2 = a2{\text{VPD}} + b2{\text{SM}} + c2{\text{TEM}} + d2{\text{CLOUD}} \hfill \\ P3 = a3{\text{VPD}} + b3{\text{SM}} + c3{\text{TEM}} + d3{\text{CLOUD}} \hfill \\ P4 = a4{\text{VPD}} + b4{\text{SM}} + c4{\text{TEM}} + d4{\text{CLOUD}} \hfill \\ \end{gathered} \right.\end{equation} \tag{ 4 }$$

$$\begin{align}{\text{NDV}}{{\text{I}}_{{\text{adjusted}}}} = aP1 + bP2 + cP3 + dP4 + {\text{residual}}\end{align} \tag{ 5 }$$

where P1, P2, P3, P4 represent the principal components of climatic factors (equation (4)). A trend analysis of the residuals was conducted to obtain the anthropogenic impacts on the trend of vegetation greenness during 1982–2015.

2.3.4. Breakpoint detection

From 1982 to 2015, 41.51% of the vegetation greenness in China's drylands showed a significant increase (P < 0.05; figure 2) and 8.40% showed significant decrease (P < 0.05). The areas with greening were mainly distributed in the northwestern and central-eastern parts of the study area (e.g. Shaanxi, Shanxi, Liaoning, Gansu, Southern Inner Mongolia, and Xinjiang provinces), while the areas with decreases in vegetation greenness were mainly distributed in the eastern drylands (e.g. eastern Inner Mongolia and Heilongjiang provinces). The overall greenness of China's drylands showed an increasing trend (+0.60 × 10⁻³ yr⁻¹), and the overall greenness of the Loess Plateau and the Three-North region also showed an increasing trend with +1.70 × 10⁻³ yr⁻¹ and +0.60 × 10⁻³ yr⁻¹, respectively. Overall, the vegetation greenness in all land classes except sparse vegetation and urban areas showed a significant increase from 1982 to 2015, with the greatest increase in the greenness of cropland (+0.95 × 10⁻³ yr⁻¹), woodland (+0.60 × 10⁻³ yr⁻¹), land transformation areas (+0.75 × 10⁻³ yr⁻¹), and mosaic vegetation (+1.10 × 10⁻³ yr⁻¹) (table S1).

Zoom In Zoom Out Reset image size

Breakpoint detection at the pixel scale showed that the areas with significant breakpoints accounted for 12.85% (61.28 Mha) of the dryland area, with temporal concentrations in 1998–2000 and 2005 (figure 3(a)). For the areas with breakpoints, the vegetation showed a mean greening trend of +2.10 × 10⁻³ yr⁻¹ before the breakpoints and an enhanced greening trend of +4.20 × 10⁻³ yr⁻¹ after the breakpoints. Further statistics on trend changes at the pixel scale showed that the main types of trend shifts were maintaining the trend of greening, switching from no significant change to greening, and switching from greening to no significant change (figure 3(b)).

Zoom In Zoom Out Reset image size

The effect of CO₂ fertilization on vegetation was mainly positive (66%), with an overall effect of +0.55 × 10⁻³ yr⁻¹ (figure 4(a)). Vegetation was the most sensitive to VPD, but SM was the dominant climatic factor for vegetation change during the study period (figure S2). The effect of climate change on vegetation was mainly negative (21%), which primarily occurred in the eastern drylands (e.g. central-eastern Inner Mongolia, northeastern China) (figure 4(b)), with an overall effect of −0.08 × 10⁻³ yr⁻¹. The positive and negative anthropogenic effects on vegetation accounted for 26% and 21% of the area, respectively, with the positive effect mainly distributed in the Loess Plateau, northern Tianshan mountains, western Xinjiang, and western Liaoning Provinces, the negative effect mainly distributed in the northeastern drylands (e.g. eastern Inner Mongolia, western and southeastern Heilongjiang, central Liaoning, and eastern Jilin provinces), with an overall effect of +0.12 × 10⁻³ yr⁻¹ (figure 4(c)).

Zoom In Zoom Out Reset image size

By comparing the effects of CO₂ fertilization, climate change, and anthropogenic impact, the factor with the largest absolute value of effect was extracted as the dominant factor (figure 4(d)). The percentage of changes was dominated by anthropogenic effect was 37%, followed by CO₂ and climate change, which accounted for 28% and 5% of the total drylands in China, respectively. The anthropogenic effects varied significantly between regions, with positive effects on the Loess Plateau (+1.01 × 10⁻³ yr⁻¹) and the Three-North region (+0.24 × 10⁻³ yr⁻¹), and a significantly higher effect on the Loess Plateau than the mean value of the study area (figure 4(c)). Decreased greenness was caused by the effects of climate change and anthropogenic impacts in the southern foothills of the Daxingan Mountains. The change in VPD dominated the vegetation degradation in this region (figure S2). In the Tibet and Qinghai provinces, the effect of climate change was positive while the anthropogenic effect was negative. The negative effects of climate change and anthropogenic impacts coexisted in urban and forest areas, and in urban areas, they exceeded the positive effect of CO₂ fertilization (table S1).

The calculation procedures in 3.2 were followed to calculate the dominant factors of vegetation change before and after the breakpoint. The results are shown in figure S3. Before the breakpoint, anthropogenic impacts on vegetation were mainly observed in the southeastern part of the Three-North region (including southeastern Inner Mongolia and western Liaoning Provinces) (figure S3(e)). After the breakpoints, anthropogenic impacts were mainly observed in the Loess Plateau region (including northern Shaanxi, southern Inner Mongolia, Shanxi, Ningxia, and eastern Gansu Provinces) (figure S3(f)). For the areas with breakpoints, the area dominated by anthropogenic impacts increased from 42.32% to 52.04% (figures S3(g) and (h)). Compared to the area before the breakpoints, the area dominated by the CO₂ fertilization effect significantly decreased (from 13.38% to 6.39%), and the area dominated by climate change significantly increased (from 6.25% to 15.70%) (figures S3(g) and (h)). The results of the breakpoint analysis and RESTREND showed when anthropogenic impacts diminished to a nondominant factor, vegetation dynamics shifted from greening to no change (A-U&I-N,7.98 Mha); when anthropogenic impacts strengthened to a dominant factor, vegetation dynamics shifted from no change to greening (U-A&N-I,11.76 Mha); and when anthropogenic impacts were maintained as a dominant factor, the slope of greening increased (A-A&I-I/C-A&I-I,12.91 Mha) (figure 5(a)). The above four types account for 6.73% of the study area and 56.72% of the area where breakpoints exist, or approximately 31.64 Mha (figure 6). Figure 6(b) shows the mean trend in certain situations, and it demonstrates that the overall greening trend of vegetation can be attributed to specific-stage human-induced greening (figure 5(b)). For example, although weakened anthropogenic effects induced a shift in vegetation dynamics from greening to no change (figure 5(a), I-N&A-U), the overall trend was greening (figure 5(b), I-N&A-U).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

These analyses were further performed for the Three-North region (figures S4(c), 6(b) and (c)) and the Loess Plateau (figures S5 and 6(a)). In the Three-North Region, the main types included a shift in vegetation greenness trend from no change to greening and an acceleration of vegetation greening trend when anthropogenic impacts strengthened to a dominant factor or a shift in the vegetation dynamics from greening to no change when the anthropogenic impacts diminished to a nondominant factor (figure S4(a)). The apparent difference in the temporal distribution was between the eastern (approximately 1999, figure 6(c)) and western (approximately 2000 and 2005, figure 6(b)) Three-North region. On the Loess Plateau, the main trend was a shift in vegetation dynamics from no change to greening (approximately 1998, figure 6(a)) and acceleration of greening trends (approximately 1998 and 2003, figure 6(a)) when human activity intensified as the dominant factor (figure S4(a)).

Anthropogenic impacts dominated vegetation dynamics in 37% of China's drylands and had an overall positive effect. Areas with greater improvement were mainly observed on the Loess Plateau and in the Three-North region. In addition, the most significant anthropogenic dominance was found in grassland, cropland, sparse vegetation cover, mosaic vegetation cover, and areas with land cover change (table S1). On the one hand, the increase in the average greenness of cropland was caused by the increase in yield and the agricultural extension (Chen et al 2019). On the other hand, the improvement was also associated with ecological conservation and restoration, such as the Grain for Green Program, in which arable land on slopes >25° had been returned to the previous forest and grassland types and original disturbance to natural forests had been prohibited, and the Three-North Shelter Forest Program, in which forest degradation areas had been restored (Bryan et al 2018). Moreover, vegetation restoration had also been implemented in sparsely vegetated areas, including the Mu us and Kubuqi Deserts (Yu et al 2020, Lin et al 2021).

Compared with the work of Zhang et al (2015) and Chen et al (2021), our study promotes an understanding of the processes of intensified greening. The implementation of ecological restoration can be divided into two phases based on the implementation of the Three-North Shelter Forest Program (TNSFP), which decelerated after 2000, and implementation of the Grain for Green Program (GTGP), which accelerated after 2000 (figure 7). This corresponds to the significant increase in ecological investment (Bryan et al 2018). We found that the implementation of ecological projects has been the dominant driver of trend shifts in China's drylands. When the dominant factor shifted to anthropogenic impacts between 1999 and 2004, the vegetation greenness shifted from no change to increasing. This mapped with the spatial patterns of ecological project implementation of the Grain for Green Program (figure 7). The trend of increasing vegetation greenness was significantly intensified after the breakpoint, with anthropogenic impacts as the dominant factor, indicating an increase in the intensity and effect of ecological restoration. When the dominant factor shifted from anthropogenic impacts to unchanged, the vegetation greenness shifted from increasing to no change in approximately 2000. This suggests that the positive effect of ecological restoration was constant after 2000. These results capture the process of ecological restoration and provide an assessment of ecological projects. This result is different from the study by He et al (2019), in which the intensified greening of China after 2000 was attributed to an enhanced summer monsoon and warming hiatus. Although the degradation of planted forests was observed for 3.23 Mha of the Three-North region (National Forestry and Grassland Administration 2020) and excessive afforestation had been criticized (Cao et al 2011, Zhang et al 2018), decreased vegetation greenness was barely detected in this study. This may be because the increased greenness in the prebreakpoint period covers this declining dynamic or the hysteresis between vegetation functional degradation and greenness decline.

Zoom In Zoom Out Reset image size

The results of our study are different from those from Naeem et al (2021), who found that both climate change and anthropogenic activities determine vegetation greening. However, we found that the positive effects of CO₂ fertilization and anthropogenic impacts offset the negative effects of climate change and resulted in an overall increase in vegetation greenness (figure 4). On pixel scale, the effect of CO₂ fertilization was wide and strongly offset the negative anthropogenic effect (20.84% vs. 10.15%) and climate change (21.23% vs. 6.36%) (table S2). Similar results had been reported by Piao et al (2015) and Wang et al (2006), who found that CO₂ fertilization is the dominant factor of vegetation greening in China, and climate change has negative effects on vegetation greening.

We found similar patterns of the effects of climate change and the anthropogenic (figure 4). For the integration of anthropogenic and climate change effects (A + CC), the proportion of increases (24.02%) and decreases (24.10%) is relatively similar to the proportion under their single impact (A: 26.06%, CC: 16.63%; A: 20.84%, CC: 21.23%) (table S2). This indicated the potential synergistic effects between them. On the Loess Plateau, the positive effect of climate change and the anthropogenic effect was found, which has also been observed by Tian et al (2021) and Chen et al (2021). The main reason for this finding may be that ecological restoration improved the local environment through plant-atmospheric interactions, resulting in favorable climate conditions that enhanced the growth of restored vegetation (Zhang et al 2021). Besides, we also found the opposite effect between climate change and anthropogenic impact on vegetation greening of the Tibet province (figure 4). In this region, the ecosystem is specifically sensitive and vulnerable. Warming relieved the limitation of temperature on plant growth, and increased moisture from snowmelt facilitates plant growth on the Tibetan Plateau (Shen et al 2015, Piao et al 2020). The effect of climate change to some extent compensated for the destruction of vegetation from human activities.