Quantitative analysis of vegetation restoration and potential driving factors in a typical subalpine region of the Eastern Tibet Plateau

Study area

Our study area is sited in the Muru Basin (30.6342°∼30.8132°N, 101.1160°∼101.2772°E), Daofu County, Garze Tibetan Autonomous Prefecture, Sichuan Province, on the ETP (Fig. 1). It has an area of approximately 268.40 km², the elevation range is 2,860∼4,932 m. This area experienced a long period of logging until the implementation of the NFPP and RCFGP around 2000, and the vegetation in this area was gradually restored. The species types in this area were essentially the same before and after the restoration, as native species were used for all vegetation restoration. The field survey found that the study area is dominated by subalpine evergreen coniferous forests, with the main species being Picea likiangensis and Abies squamata, mixed with Betula albosinensis and Populus davidiana at lower altitudes regions, with a clear community stratification, the main species in the shrub layer are Rhododendron decorum, Lonicera tangutica, Sorbus rehderiana, and Rosa sweginzowii. Some areas of evergreen coniferous forest have degraded into scrub after logging, with the main established species being Rhododendron decorum, Rhododendron lapponicum, Quercus monimotricha, etc. According to the nearest Daofu meteorological station data (101.11667°E, 30.98333°N; 2957.2 m a.s.l, 1982-2018), the average annual temperature is 8.23 °C. The warmest month is July, with an average temperature of 16.12 °C. The coldest month is January, with an average temperature of −1.51 °C. The land cover types are mainly forests, shrublands, and meadows. The majority of the basin is forest and shrub, with meadows mainly found at higher elevations along the basin margins. Traces of human activities such as cultivated land, roads, and residential areas are mainly located in the valleys. Soil taxonomy great groups (USDA system, https://www.openlandmap.org) mainly include Cryoboralfs, Cryoborolls, Argicryolls, Haplustalfs, and Cryumbrepts.

Data sources

Landsat satellite images at 30 m spatial resolution were obtained from Google Earth Engine (GEE, https://developers.google.com/earth-engine/), which provides uninterrupted global multispectral surface imagery every 16 days. We used Landsat 5 from 2000 to 2011, with 178 images, and Landsat 7 in 2012, with 14 images, and Landsat 8 from 2013 to 2020, with 145 images. All images were radiometrically calibrated, atmospherically corrected, and geo-referenced to reduce interference and improve image quality. In addition, cloud masks were used for all images to eliminate the effect of clouds. We used median calculations to composite each year’s images into one image without clouds in the GEE. This method is more resistant to extreme values and can be representative of the period studied (Bunting, Munson & Bradford, 2019).

Elevation, slope, and aspect were generated from DEM at 30 m spatial resolution obtained via the Geospatial Data Cloud (http://www.gscloud.cn/). The microtopographic type uses the 90 m resolution Global ALOS landforms dataset (https://www.sciencebase.gov/catalog/item/564b4bb0e4b0ebfbef0d31d2), which takes full consideration of Continuous Heat-Insolation Load Index and the multi-scale Topographic Position Index (Theobald et al., 2015), the study area has been classified into 13 microtopographic types (lower slope, lower slope (cool), lower slope (warm), upper slope, upper slope (cool), upper slope (warm), valley, valley (narrow), cliff, mountain/divide, peak/ridge (warm), peak/ridge (cool), and peak/ridge).

Soil taxonomy great groups, soil pH, and soil water content with a spatial resolution of 250 m were obtained from OpenLandMap (https://www.openlandmap.org/). Soil taxonomy great groups use the Soil Texture Class (USDA system) at 0 cm depth, soil pH uses the soil pH in H₂O at 0 cm depth, and soil water content at 0 to 200 cm uses all bands in the Soil water content dataset at 33 kPa (field capacity).

Climatic datasets were obtained from WorldClim version 2.1 (https://www.worldclim.org/data/worldclim21.html), released in January 2020, spatial resolution is 30 s. The datasets included monthly values of the minimum temperature, maximum temperature, average temperature, precipitation, solar radiation, wind speed, and water vapor pressure. Based on these dataset, the annual average temperature, annual maximum temperature, annual minimum temperature, annual precipitation, and annual average water vapor pressure were calculated.

In addition to natural factors such as climate, topography, and soil, vegetation recovery is also influenced by human activities (Zhou et al., 2021). Roads and residences are the concrete embodiment of human activities in the study area, so we used distance to roads and residences to quantify the intensity of human activity, other studies have also used these two indicators (Liu et al., 2021). Roads and residences information was obtained manually in Google Earth Pro, then the distances of roads and distance to residences were calculated in ArcGIS 10.7 using the euclidean distance tool and finally resampled to a spatial resolution of 30 m.

To avoid serious multicollinearity, the Variance Inflation Factor (VIF) was calculated in ArcGIS 10.7 using the Ordinary Least Squares tool. Usually, VIF > 10 implies the possibility of serious multicollinearity (Marcoulides & Raykov, 2019). The potential driving factors with VIF > 10 were excluded; the final 12 factors were applied to the next step of the analysis (Table 1).

Vegetation cover calculation

Plant leaves have different absorption and reflectivity of red and near-infrared light wavelengths (Xu et al., 2019). The multi-spectral sensor’s spectrum bands contain both visible and infra-red wavelengths, which combine to produce vegetation indices (Brown et al., 2006; Wang, Liu & Huete, 2003). NDVI is calculated as the proportion of the difference between the near-infrared and red bands to the sum (Bianchi, Villalba & Solarte, 2019). NDVI ranges between −1.0 to +1.0, where a value less than 0 corresponds to without vegetation and greater than 0 corresponds to vegetation (Brehaut & Danby, 2018). The FVC is the proportion of area with vegetation cover to the whole area (Hu et al., 2020). The FVC is usually calculated from NDVI data in an image element dichotomous model with the following equation (He, Shi & Fu, 2021): (1)${\displaystyle FVC=\frac{NDVI-NDV{I}_n}{NDV{I}_v-NDV{I}_n}}$

where NDVI is the NDVI value of the image pixel, NDVI_v and NDVI_n represent the NDVI values of the vegetation and bare ground image pixel, respectively. In the actual calculation, the NDVI values of 5% and 95% of the cumulative frequency of the histogram are taken as NDVI_n and NDVI_v. To avoid the FVC results being outside the range of 0 to 1, the FVC was set to 0 for pixels with NDVI less than or equal to NDVI_n and 1 for pixels with NDVI greater than or equal to NDVI_v.

FVC change trend and sustainability analysis

Linear regression equation fitting methods can accurately measure the spatial patterns of dynamic changes in FVC and identify trends in individual pixels. This study used this method to simulate spatial trends in FVC from 2000∼2020. The formula is as follows He, Shi & Fu (2021) and Zhang, Ge & Zhang (2020): (2)${\displaystyle SLOPE=\left[n\times \sum _{i=1}^{n}i\times FV{C}_i-\sum _{i=1}^{n}i\sum _{i=1}^{n}FV{C}_i\right]/\left[n\times \sum _{i=1}^n{i}^2-{\left(\sum _{i=1}^{n}i\right)}^2\right]}$

where SLOPE is the trend of FVC, n is the number of years in a time series, i is the annual variable. SLOPE > 0 indicates a tendency for vegetation cover to improve, while the opposite implies a tendency to deteriorate. For determining the significance of FVC tendency, a T-test has been applied and per pixel’s P-value was computed. The results were classified into five categories: extremely significant increase (SLOPE > 0, P < 0.01), significant increase (SLOPE > 0, 0.01 <P < 0.05), extremely significant decrease (SLOPE < 0, P < 0.01), significant decrease (SLOPE < 0, 0.01 < P < 0.05), and insignificant change (P > 0.05).

The Hurst exponent is a useful approach to predict future trends in FVC (Emamian et al., 2021). The Hurst exponent utilizes the benefits of self-covariance and is usually applied at measuring the stationarity of big data sequences in the natural world through the Rescaled Range Series Analysis (Hurst, 1951; Zhang, Ge & Zhang, 2020). Based on the FVC from 2000 to 2020, the Hurst exponent has been adopted in our research to predict the future tendency. Provided a series of time sequences x_i, i = 1, 2, 3, …, n, the Hurst exponent was computed as follows Emamian et al. (2021):

First, split the time series into different segments.

Second, calculate the mean value of each segment: (3)${\displaystyle m=\frac{1}{n}\sum _{i=1}^n{x}_i}$

Third, calculate the series of deviations: (4)${\displaystyle y_i=x_i-m}$

Fourth, calculate the widest difference: (5)${\displaystyle R_i=\mathit{max}\left(y_1,y_2,y_3,\dots ,y_i\right)-\mathit{min}\left(y_1,y_2,y_3,\dots ,y_i\right)}$

Finally, calculate the standard deviation: (6)${\displaystyle S_i={\left[\frac{1}{n}\sum _{i=1}^n{\left(x_i-m\right)}^2\right]}^{\frac{1}{2}}}$

With an increase in i, various values of R/S were derived and the below expressions were obeyed: (7)${\displaystyle R_i/S_i\propto i^H.}$

In which H is the Hurst exponent. 0 < H < 0.5 means a tendency for future variation to be the reverse direction of the history (opposite); H = 0.5 indicates random fluctuation (random); 0.5 < H < 1 represents a tendency for future variation is same as the history (continuous).

Combining the values of the SLOPE and H, we have divided the future trends of FVC into six groups: random fluctuation (H = 0.5), continuous decrease (0.5 < H < 1, SLOPE < 0, P > 0.05), continuous increase (0.5 < H < 1, SLOPE > 0, P > 0.05), continuous significant decrease (0.5 <H < 1, SLOPE < 0, P < 0.05), continuous significant increase (0.5 < H < 1, SLOPE > 0, P < 0.05), from degradation to improvement (0 < H < 0.5, SLOPE < 0), and from improvement to degradation (0 < H < 0.5, SLOPE > 0).

Geographical detector

Geodetector (http://www.geodetector.cn/) incorporates a range of spatial statistical methodologies, which explore the explanatory variables affecting the dependent variable through a spatial variance analysis (Wang, Zhang & Fu, 2016; Wang et al., 2010). Geodetector enables convenient and accurate exploration of spatial variance and quantification of drivers and is widely used to quantify the drivers and their interactions affecting vegetation cover change (Liu et al., 2021).

(1) Factor detector

The ability of an arbitrary factor to explain the FVC change is measured using the factor detector, and q-statistic was used to measure stratified heterogeneity in FVC change (Wang, Zhang & Fu, 2016). The q takes on a range of values from 0 to 1, the closer the value of q is to 1, indicates greater explanatory power of factor x for FVC change. The corresponding formulas are Chen et al. (2020b) and Liu et al. (2021): (8)${\displaystyle q=1-\sum _{h=1}^L{N}_h{\sigma }_h^2/N{\sigma }^2}$

where q is the ability of the factor to explain the FVC change. N is the sample size, N_h is the sample size of factor x in zone h, σ² is the variance of the regional FVC change, and ${\sigma }_h^2$ is the variance of factor x in zone h.

(2) Interaction detector

This module is used to determine the explanatory power of the FVC change when the two factors interact. First, the q-values were calculated by the two factors as q (x1) and q (x2). Secondly, the q-value reflecting the interaction of the two factors is calculated as q (x1 ∩x2) and compared with q (x1) and q (x2), indicating the type of interaction between the two variables (Peng, Kuang & Tao, 2019). The types of interaction were classified into five groups (Table 2).

(3) Risk detector.

The risk detector was utilized to analyze whether the mean values of the drivers of FVC change differed significantly across sub-regions and to identify the range or type of factor that was most favorable. Risk detection is based on the t-statistic (Liu et al., 2021): (9)${\displaystyle t=\frac{{\bar{Y}}_{h=1}-{\bar{Y}}_{h=2}}{\sqrt{\frac{Var\left({\bar{Y}}_{h=1}\right)}{N_{h=1}}}+\frac{Var\left({\bar{Y}}_{h=2}\right)}{N_{h=1}}}}$

where ${\bar{Y}}_h$ is the mean of the FVC change in zone h; N_h is the sample size of factor x in zone h, and Var is the variance.

Temporal variation characteristics of FVC change

Figure 2 shows the fluctuating trend of the average FVC from 2000 to 2020. The FVC ranged from 50.26 to 63.45%, with an average of 56.10%. Overall, FVC variation had a rising pattern, with a growth rate of + 0.26/year (P = 0.028), indicating that remarkable progress has been accomplished by implementing ecological protection projects during the last 21 years. The FVC change can be divided into three stages: 2000 to 2005 shows a slight upward trend, 2005 to 2010 shows a downward trend, since 2010 the quality of vegetation has been good but fluctuation is large.

From 2000 to 2020, the FVC of each microtopography in the Muru Basin changed significantly (Fig. 3). The highest FVC is 83.46% for the lower slope (cool), which is significantly higher than other microtopographic types (Fig. 3C). The lowest FVC is 0, distributed in the peak/ridge (cool) (Fig. 3A). The mean FVC for different microtopographic types 2000–2020 is ranked from highest to lowest as follows (Table 3): lower slope (cool) (69.62%), lower slope (62.90%), lower slope (warm) (58.42%), valley (57.39%), upper slope (warm) (54.42%), upper slope (54.15%), valley (narrow) (53.31%), upper slope (cool) (47.13%), cliff (44.42%), mountain/divide (32.77%), peak/ridge (warm) (23.61%), peak/ridge (17.92%), peak/ridge (cool) (2.74%). As shows in Table 4, from 2000 to 2020, the FVC of microtopography types: peak/ridge (warm), cliff, upper slope, lower slope, and lower slope (cool) increased significantly (P < 0.05), while other microtopographic types did not. The microtopography with the fastest growth in FVC is the cliff, lower slope, upper slope, and lower slope (cool), with growth rates of +0.79/year, +0.53/year, +0.52/year, and +0.49/year, respectively (Table 4).

Spatial variation characteristics of FVC change

Figure 4 shows the spatial distribution of microtopographic types and FVC changes in the study area. The areas with insignificant change, extremely significant increase, significant increase, extremely significant decrease, and significant decrease of FVC accounted for 70.42%, 13.08%, 8.62%, 4.04%, and 3.84% of the whole study region, respectively (Fig. 4B). Most regions with significant changes in FVC showed an extremely significant and significant increasing trend, only a small number of areas have shown extremely significant or significant decreasing trends (Fig. 4B). Figure 5A shows that the study area has the largest number of areas with Hurst exponent less than 0.5 (opposite) (67%), followed by greater than 0.5 (continuous) (19.22%) and equal to 0.5 (random) (13.76%). To further determine the sustainability of vegetation restoration in the study area, the FVC change trend and Hurst exponent were spatially overlaid to analyze the future tendency of FVC change (Fig. 5B). The results showed that the areas from degradation to improvement, continuous increase, and continuous significant increase accounted for 33.35% of the whole area. Random fluctuation areas mostly located at the margin of the basin accounted for 19.17% of the whole area. The areas from improvement to degradation, continuous decrease, and continuous significant decrease accounted for 47.48% of the whole area. This indicates that vegetation degradation is likely to occur in the future.

It is clear from Fig. 6A that the highest percentage of the extremely significant increase in FVC was the cliff, accounting for 15.57%, followed by the lower slope (warm), lower slope, upper slope, and lower slope (cool), accounting for 15.41%, 14.43%, 14.18%, and 13.73, respectively. It is also clear that the significant increase of FVC was highest in the lower slope, accounting for 12.53%, followed by the upper slope, cliff, and lower slope (cool), accounting for 12.21%, 10.81%, and 10.03% (Fig. 6B). The proportion of extremely significant decrease and significant decrease in FVC on the upper slope (cool) and lower slope (cool) was much less than on the upper slope (warm) and lower slope (warm) (Figs. 6C and 6D). This is likely due to the lower evaporation and better water conditions on the upper slope (cool) and lower slope (cool) than on the upper slope (warm) and lower slope (warm), so the growth of vegetation is stronger.

Future variation trends FVC are disparate in each microtopographic type (Fig. 7). The proportion of the continuous significant increase in FVC is greatest on the lower slope (warm) (Fig. 7A). The proportion of the continuous increase in FVC was greater on the valley (narrow) and valley than on other microtopographic types (Fig. 7C). The area with the largest proportion of FVC from degradation to improvement was the upper slope (warm) (36.78%), accounting for 7.3% of the total study area (Fig. 7E). Compared with other microtopographic types, valley (narrow) FVC showed the highest proportion of continuous significant decrease and continuous decrease (Figs. 7B and 7D). In the lower slope (cool) (69.41%) and lower slope (58.20%), the proportion of FVC coverage from improvement to degradation was larger, accounting for 10.23% of the total study area (Fig. 7F). The future variation trend was a continuous significant increase, continuous increase, or from degradation to improvement indicating vegetation improved. On the contrary, continuous significant decrease, continuous decrease, and from improvement to degradation indicate vegetation degradation. This means that although there is a greater proportion of continuous improvement in FVC in the lower slope (warm), valley (narrow), and upper slope (warm), there is still a risk of vegetation degradation. The proportion of random fluctuation was more extensive in the peak/ridge area (include peak/ridge (cool), peak/ridge, and peak/ridge (warm)) (Fig. 7G), indicating that afforestation in the peak/ridge area was difficult and had a principal impact on the survival of vegetation.

Identification of driving forces

(1) Factor detector analysis

We quantified the effect of each factor on the FVC change using Geodetector (Fig. 8). The effect of each factor in FVC change was significant (P < 0.05), except for the slope (P = 0.273). The explanatory power (q-statistic value) of aspect (q = 21.28%, P < 0.001), microtopographic type (q = 8.61%, P < 0.001), soil taxonomy great groups (q = 8.12%, P < 0.001), and soil water content at 200 cm (q = 5.75%, P < 0.001) exceeded 5%, are the main drivers of FVC variation in the study area. These are followed by soil water content at 30 cm (q = 3.77%, P < 0.001), distance to roads (q = 2.90%, P < 0.001), annual average temperature (q = 2.57%, P < 0.001), soil water content at 0 cm (q = 1.71%, P = 0.005), distance to residences (q = 1.57%, P = 0.010), soil pH (q = 1.53%, P = 0.010), soil water content at 10 cm (q = 1.41%, P = 0.019), and slope (q = 0.66%, P = 0.273). This indicates that aspect, microtopographic type, and soil taxonomy great groups are important factors influencing FVC change in the study area. Although human activities such as roads and residences are important factors, in our study area they are much less important than natural factors such as aspect. This probably is due to the implementation of policies such as the NFPP and RCFGP have greatly reduced the disturbance of vegetation by human activities. In addition, the vegetation in the study area is mainly deep-rooted vegetation such as Abies squamata, Picea likiangensis. Therefore, soil water content at 0–30 cm did not show a significant explanatory power to the variation of FVC.

(2) Interaction detector analysis

Figure 9 shows that the interactions between the two factors all have stronger explanatory power for the FVC changes than the individual factors and that the majority of the interaction effects are non-linearly enhanced. Specifically, the interactions between aspect and microtopographic type, soil taxonomy great groups, soil water content at 200 cm; and the microtopographic type and soil water content at 200 cm; and soil water content at 30 cm and soil water content at 200 cm demonstrated bivariate enhancement, whereas the interaction between other factors showed a nonlinear enhancement. According to the results, the q-statistic value of the interaction between any factor and aspect was larger than that of the other two factors, indicating that aspect played an important role in FVC change. This was consistent with the conclusion in factor detector analysis, the q-statistic value of aspect was 21.28%. In addition, amongst these interactions, aspect and annual average temperature (q = 29.20%) were the strongest. Although the explanatory power of annual average temperature for FVC change is less than 3% (q = 2.57%), its interaction with aspect has larger explanatory power than the sum of each and is much larger than the interaction between the other factors. It can be seen that annual average temperature also has a significant effect on the change of FVC, mainly in the interaction with aspect.

(3) Risk detector analysis

The risk detector can detect the most suitable level of vegetation restoration for each factor. A higher value of FVC slope for a given level represents a more favorable level of this factor for vegetation restoration, with different factors presenting significant differences in the value of FVC change (Fig. 10). The response of the FVC slope to the aspect of different levels first decreases and then increases (Fig. 10A). The FVC slope was the highest when the aspect was 292.5∼337.5° (northwest), 337.5∼22.5° (north), and 22.5∼67.5° (northeast) (Fig. 10A), indicating that the northern slope (292.5∼67.5°) was the most suitable for vegetation restoration. Concerning the microtopographic types, the lower slope (cool) was conducive to vegetation restoration, with the largest FVC slope (Fig. 10B). The soil taxonomy great group most conducive to vegetation restoration was Cryumbrepts (Fig. 10E). Cryumbrepts are well- or moderately drained soil common in subalpine areas that support forest, grassy, and tundra vegetation (Rieger, 1983). The value of the FVC slope increases with the increase of soil water content at 200 cm depth (Fig. 10I). When the water content at 200 cm was 15∼20%, it was the most suitable for vegetation restoration (Fig. 10I). The FVC slope was highest at an annual average temperature of 1.52∼3.14 °C (Fig. 10J), which was the most favorable level for vegetation restoration. In terms of the impact of human activities on vegetation recovery, the FVC slope shows an increase followed by a decrease with increasing distance to roads and residential areas, with the greatest FVC slope in the range of 1500∼2000 m from roads (Fig. 10K), and the greatest FVC slope in the range of 1000∼2000 m from residences (Fig. 10M).

Vegetation change features and the impact of microtopographic type

Overall, there has been an increasing trend in FVC on the Muru Basin from 2000 to 2020, which is consistent with a positive trend in vegetation cover on the TP (Song, Jin & Wang, 2018; Zhang et al., 2018). However, it must not be overlooked that some of the microtopographic types are degraded (Figs. 5 and 7), where vegetation restoration is not sustainable. Due to extreme climatic conditions and ecosystem fragility in the subalpine areas of the ETP, the ecology is highly sensitive to logging activities, and post-harvest ecological restoration faces enormous challenges (Xiong et al., 2021). Even some achievements have been made in forest restoration by government support, but the survival rate of afforestation is still low in some areas (He, Shi & Fu, 2021), and vegetation is degraded in some places (Wang et al., 2019). The uncertainty in vegetation restoration is a major challenge for sustainable forest management and conservation in subalpine areas.

We found that the lower slope was the microtopographic type with the best vegetation cover (Fig. 3). The ridge/peak areas had the greatest proportion of FVC decrease and random fluctuation (Figs. 6C, 6D and 7G). Topographic factors (elevation, slope, aspect, and microtopographic type) play an essential role in the dynamics of vegetation cover (Jia et al., 2020; Li, Shi & Wu, 2020). They influence the climate and soil moisture conditions during the year by controlling precipitation, solar radiation, and temperature (Metzen et al., 2019). The lower slopes have a non-native water input from upper slope areas in addition to groundwater (Thompson et al., 2011), which is more beneficial for plant growth. The lower slope has better hydrological conditions compared to the other microtopographic types (Thompson et al., 2011), and in study areas where Abies squamata and Picea likiangensis are common forest species, moist soil are more conducive to their growth. The ridge/peak areas soils are relatively shallow, soil moisture and nutrients are lacking (Gao et al., 2017a), so afforestation is difficult. In general, relatively humid lower slopes are the most suitable for vegetation restoration, and peak/ridge should be avoided as much as possible.

Key factors influencing vegetation restoration

The morphology and function of watersheds result from long-term co-evolution between water, soil, landforms, and ecosystems (Thompson et al., 2011), with solid feedback between vegetation, topography, climate, and soil (Kirkpatrick et al., 2014). Our study used Geodetector to determine the key drivers of FVC change in subalpine regions. In contrast to other statistical approaches, it can measure Spatial Stratified Heterogeneity (SSH) and analyze the effects of factors and their interactions on FVC (Liu et al., 2021).

Our research shows that aspect is the most critical factor influencing FVC change, followed by the microtopographic type and soil taxonomy great groups (Fig. 8). Previous studies have also shown that topography and soils are important drivers of vegetation restoration (He, Shi & Fu, 2021; Zhong et al., 2022). The northern slope (292.5∼67.5°) was the most suitable for vegetation restoration (Fig. 10A); they are more humid than the southern slopes (Qingkong et al., 2020; Sun et al., 2019). Predecessors’ research supports our conclusion that northern slopes have adequate soil moisture and higher vegetation cover than southern slopes (Louhaichi et al., 2021; Metzen et al., 2019). The lower slopes are recharged by both groundwater and non-native water from the upper slopes (Thompson et al., 2011), and are more suitable for vegetation growth. In addition, soil texture also influences FVC change, with the most suitable soil taxonomy great groups for vegetation restoration being Cryumbrepts. Soil is an important determinant of vegetation restoration (Mills et al., 2021), affects the growth and composition of vegetation (Fernández-García et al., 2021). Cryumbrepts are consistently well- or moderately drained soil (Rieger, 1983), well-drained soils are more conducive to vegetation root expansion, respiration, and growth (Crowell & Lane, 2001). In vegetation restoration, the drainage of the planting site is essential for the health and growth of seedlings.

Roads and residences are the concrete embodiment of human activities in the study area, they are mainly located in lower elevations and valleys areas, and the residences are mostly located along the roads. The field survey found that areas close to roads and residences in the study area are mostly cultivated and heavily influenced by human activities, which is not conducive to vegetation restoration. Further away from the roads and residences are located at the edges of the basin and are meadows where the vegetation is more affected by grazing and self-growth restrictions. Accordingly, the results of the study show that areas moderately close to roads and residences are the most favorable for vegetation restoration. However, the contribution of distance from roads and distance from residents to FVC change is only 2.90% and 1.57%, respectively, which is much less than 21.28% for aspect and 8.61% for microtopographic types. Since 1998, a series of ecological protection projects have been implemented, such as NFPP and RCFGP, which have greatly limited human disturbance and damage to the local ecology. Overall, the impact of human activities on vegetation restoration in the study area is much less than the natural factors.

Besides, the diversity of geographical processes shows that the interplay of various factors influences changes in vegetation cover (Huo & Sun, 2021). Interaction between the two factors had stronger explanatory power for the FVC changes, which is similar to the idea that the interaction between various factors has a markedly stronger effect on the vegetation than the individual factors (Liu et al., 2021). We explored that annual average temperature also has a significant effect on the change of FVC, mainly in the interaction with aspect. In particular, the interaction between temperature and aspect has the most excellent effect on FVC change, probably because aspect variation leads to differences in temperature (Pepin et al., 2017), which affects the vegetations’ spatial distribution and growth.

Limitations and uncertainties

Vegetation restoration is influenced by a combination of topography, climate, soils, and human activity (Chen et al., 2020b; Meng et al., 2019; Peng, Kuang & Tao, 2019). Our study used Landsat imagery with a resolution of 30 m to capture the dynamics of vegetation restoration. Unlike MODIS, Landsat with higher resolution captures more spatial detail (Seto et al., 2004). However, the 16-day revisit time and frequent cloud cover make it difficult to obtain sufficient high-quality data, which may reduce its performance in detecting rapid ecosystem change (Liao et al., 2016). A spatial–temporal fusion approach to mix MODIS and Landsat into new composite data seems to be a good solution. In this study, the main drivers of vegetation restoration in subalpine areas were identified and analyzed using Geodetector, which incorporates a range of spatial statistical methodologies to explore the explanatory variables affecting the dependent variable through a spatial variance analysis (Wang, Zhang & Fu, 2016; Wang et al., 2010). Nevertheless, some of the factors such as precipitation and elevation, which have a strong influence on vegetation restoration (He, Shi & Fu, 2021; Zhang et al., 2021), are excluded due to severe multicollinearity. This is most likely caused by the small scale of our study area and the need to consider a more detailed study on a larger scale such as the ETP.

In addition, previous research has shown that the impact of human activity on vegetation change is crucial (Cheng et al., 2021; Meng et al., 2019). Whereas, our study found that distance from roads and distance from residents contributed only 2.90% and 1.57% respectively to the change in vegetation cover. This indicates that the impact of human activity on vegetation in our study area is much lower than other natural factors. On one hand, the implementation of ecological projects limited the impact of human activities. On the other hand, it would be because the roads and residents could not cover all human activities in the study area. Some studies use land-use data to measure the intensity of human activities and have achieved good results (Chen et al., 2020b). We can combine roads, residents, and land-use data to quantify the intensity of human activities for further research.