Spatial variation trends of the NDVI time series were analyzed by applying Theil-Sen analysis. This is a stable trend calculation method for non-parametric statistics with high computational efficiency that is insensitive to measurement error and discrete data due to the fact that it requires a series of mutually independent data (Sen, 1968). This method can indicate changes on the trend of NDVI over a long time series at the pixel scale (Hu et al., 2021; Pei et al., 2021; Wang et al., 2022), and utilizes the following formula: S l o p e = m e d i a n x j − x i j − i , 2001 ≤ i < j ≤ 2020

In this formula, Slope refers to the slope value estimated by the Theil-Sen median; x represents the value of the NDVI for each year in the study; and i and j represent the different years between 2001 and 2020. Slope<0 indicates a decreasing trend in NDVI; in contrast, a Slope>0 indicates an NDVI with an increasing trend.

The Mann-Kendall trend test is a non-parametric statistical test that does not require the independence and normality of a time series of data and can effectively detect long-term trends (Mann, 1945; Kendall, 1975). This test method has been widely utilized in trend significance tests for long-term series of data, including meteorological data, climate data and vegetation index data (Qi et al., 2019; Feng et al., 2022); the test statistic Z was calculated as follows: S = ∑ i = 1 n − 1 ∑ j = i + 1 n s g n N D V I j − N D V I i s g n N D V I j − N D V I i = 1 , N D V I j − N D V I i > 0 0 , N D V I j − N D V I i = 0 − 1 , N D V I j − N D V I i < 0 v a r S = n n − 1 2 n + 5 / 18 Z = S − 1 v a r S , S > 0 0 , S = 0 S + 1 v a r S , S > 0

In these formulae, n represents the length of the time series, NDVI _j and NDVI _j represent data values at times j and i (j > i), and sgn (NDVI _j - NDVI _i ) represents the symbolic function. Z > 0 and Z < 0 indicate an increasing or decreasing trend, respectively. When n ≥ 8, the test statistic S was approximately normally distributed, and its mean and variance are determined, as follows. Given an α significance level, |Z|>u _(1−α)/2 indicates that the assumption that there was no trend was rejected and there was an obvious trend change in NDVI over the time series. In this study, |Z|>1.65 and |Z|>1.96 indicated that the trend passed significance tests at 90% and 95%, and symbolized more significant and significant changes, respectively.

The consistency of sequence trends can be quantitatively assessed using the Hurst Index H) (Hurst, 1951; Mandelbrot and Wallis, 1969), which is widely used for the trend analysis of vegetation indices, including NDVI, Net Primary Productivity (NPP), Fractional Vegetation Cover (FVC), and others (Li et al., 2023; Qin et al., 2016; Xu et al., 2020). The Hurst index is calculated based on the rescaling range analysis method (R/S), which can be used to judge whether a time series of data represents a random walk form or a biased random walk process. This reflects the autocorrelation of a time series, especially with regards to long-term trends that are hidden in the time series. The value of H lies between 0 and 1. An H < 0.5 indicates that the data exhibit anti-persistence within a certain time series; it is possible that the opposite trend may occur in the future. An H = 0.5 indicates that the time series of data is in the state of a random walk, and that the future trend cannot be predicted. An H > 0.5 indicates that the time series of data is persistent and will maintain the current trend in change in the future (Hurst, 1951; Mandelbrot and Wallis, 1969). In the present study, H < 0.49 is defined as anti-sustainability, 0.49 < H < 0.51 as uncertain, and H > 0.51 as a stable trend of NDVI in the future, respectively.

In this study, we used the residual analysis method to separate and quantify the anthropogenic and climatic impacts on NDVI variations at a grid-scale, so as to further identify the main driver of vegetation changes by comparing the residuals between observed NDVI values and predicted NDVI values (Geerken and Ilaiwi, 2004; Sajadi et al., 2021; Lai et al., 2023). We established a multiple regression model between annual NDVI values and corresponding climatic values (precipitation and temperature); the assumption of this model was that the anthropogenic impact on vegetation could be explained by the residual variations derived from the observed values minus the climate-based predicted values (Geerken and Ilaiwi, 2004; Huang et al., 2023). The equations for this calculation are given below. N D V I c i = a × P + b × T N D V I a i = N D V I o b − N D V I c i in which a and b represented regression coefficients; NDVI _ci represents the predicted value based on the climatic influence of precipitation (P) and temperature (T); NDVI _ai represents the residual value, which was determined by the combination of NDVI _ob (the value of NDVI time series data) and NDVI _ci . The slopes of NDVI _ci , NDVI _ai , and NDVI _ob correspond to Slope _ci , Slope _ai, and Slope _ob , respectively. The relative contributions of climatic and anthropogenic impacts on vegetation dynamics were calculated with reference to published data (Geerken and Ilaiwi, 2004).

The NDVI value of the HV between 2001 and 2020 ranged from 0.010 to 0.886, with an average of 0.589. The low value areas (NDVI <0.5) accounted for 29.86% of the study area, and were mainly distributed in high-altitude areas, including the Qilian Mountain (the Northern part of Mengyuan County) and Laji Mountain (from Guide County to the Salar autonomous county of Xunhua), and the Huangshui River Valley (HRV). The different types of land-use in high-altitude areas were mainly non-utilized land, low coverage grasslands, and permanent snow cover, while the HRV mainly featured urban and agricultural areas. Overall, the NDVI in other regions was >0.5, accounting for 70.14% of the HV with scattered low value areas which were mainly forest and grassland (Figure 2).

The NDVI time series reflected temporal changes in the characteristics of vegetation on the HV. The annual average NDVI ranged from 0.548 to 0.638 and exhibited an overall fluctuating rise of 0.0041/yr (p < 0.01) between 2001 and 2020. The increase in NDVI was relatively slow between 2001 and 2005 (0.0042/yr, p < 0.01), with a growth rate of 0.006/yr between 2016 and 2013 (p < 0.01), a decreasing trend between 2013 and 2017 (0.003/yr, p < 0.01), and a relatively fast growth rate between 2017 and 2020 (0.0108/yr, p < 0.01) (Figure 3).

Theil-Sen Median trend analysis and the Mann-Kendall test showed that the area associated with a significant increase in NDVI accounted for 75.25% of the study area, and that 5.18% of the area exhibited a medium increase; this was mainly concentrated in the region below 3700 m above sea level and a slope between 0° and 35°. This area featured a large amount of grassland, cultivated land and forest; the amount of carbon synthesized by plants increased gradually. In addition, the vegetation cover and the quality of the ecological environment improved under the interaction of human activities and obvious humidification between 2001 and 2020. The areas with a significant and more significant decline of NDVI accounted for 1.96% and 0.43% of the HV, respectively. Spatially, they were situated along the east-west orientation distributed in the HRV from Xining City to the Ledu districts, from Xining City to the north along the valley to Huzhu county, and were sporadically distributed in the remaining parts of the HV. Most of these areas were below 3000 m of the elevation and with a slope of 0°–15°, which were suitable for the layout of towns and industrial facilities, with vegetation degradation and ecological fragility between 2000 and 2020. Furthermore, 17.07% of the NDVI stabilized areas were distributed within the whole study area; 0.11% of the region was not accounted for statistically or was associated with a lack of data (Figure 4; Tables 1, 2).

Within the same period, the average annual temperature in the study area fluctuated between 0.384°C and 1.485°C, increasing at a rate of 0.0151°C/yr (p < 0.01), while the range of annual precipitation was 441–656 mm, increasing at a rate of 4.1138 mm/yr (p < 0.01); these data agreed with the overall increasing trend of NDVI (Figure 5). Firstly, ecological promotion measures, such as the 3-North Shelter Forest Program implemented since 1978 is the largest ecological restoration system project in Qinghai Province, the Greening Project of Nanshan and Beishan in Xining city and Haidong city since 1989, and since 2002, the projects of grazing bans and fencing, Pasture for Grass, livestock reduction, rodent control, ecological migration, and the establishment of nature reserves and national parks were successively implemented in Qinghai Province, were all deployed within the HV. On the other hand, For the existing forest or sand, the measures of setting apart hills (including sand) area for tree growing, forest stand quality conversion, middle-young-growth forest nurturing, and selecting the best to create modern forest farms (including such as Beishan Forest Farm Xining city, Beishan Forest Farm Huzhu county, Xianmi Forest Farm Menyuan county, etc.) were adopted by management. These measures were synergistic with warming and humidification, effectively promoted the growth of NDVI and restored the ecological environment in the HV.

According to trends in the observed NDVI, we used predicted NDVI and the residuals of NDVI to analyze the driving mechanisms of NDVI change in the HV between 2001 and 2020. Analysis showed that the proportion of NDVI improved and degraded areas in the HV were 91.80% and 5.70%, respectively, and that the area with no statistics accounted for 2.50%. Of these, 87.24% of the region showed an improvement in NDVI in response to both climate change and human activities; 23.75% and 76.25% of the contributions related to climate change and human activities respectively. These changes occurred in most parts of the study area except for cultivated land; 4.46% of the area was influenced by only human activities; this area was located in the middle and low elevation mountains and along the rivers. Only 0.10% of the area was improved by climate change; these changes were sporadically distributed across the HV. Approximately 88.20% of the area showing an increase in NDVI was concentrated below 3700 m above sea level, thus indicating that most of the area showing vegetation improvement in the HV was not higher than the upper line of the area suitable for the growth of coniferous forest. The relevant greening policy of Qinghai Province shows that the man-made greening activities are mainly distributed around the town, administrative village, water and residential, and besides of roads, as well as planting trees in the ridge to form a forest network. It includes land-use types such as forestable barren land, sandy land, non-utilized land, abandoned and damaged mountains, and degraded grassland. The increased NDVI data indicate that the Grain for Green strategy has been effective in the HV. Furthermore, 89.65% of the area showing an improvement in NDVI was distributed in areas with a slope of less than 35°; these areas featured more human activities, thus indicating that NDVI growth is closely related to human activities (Figure 6; Tables 3, 4).

Analysis showed that 4.04% of the study region was degraded due to climate change and human activities; the relative contributions of climate change and human activities were 26.04% and 73.96%, respectively. The proportions of change driven by climate change or human activities only were 0.19% and 1.47%, respectively. Spatially, the NDVI degraded region was distributed in the HRV, Yellow River Valley, and the central part of the Menyuan county; these areas were dominated by cultivated land and urban land. With regards to vertical gradient, 89.80% of the degraded areas were distributed below 3700 m, particularly between 2100 and 3000 m and between 3000 and 3700 m; furthermore, 89.95% of the NDVI degraded areas had a slope lower than 25°. This was mainly due to occupation of arable land and the expansion of towns, thus representing topographic conditions that are associated with urban development. By weighted summation, the relative contributions of climate change and human activities to NDVI changes over the past 20 years accounted for 23.85% and 76.15% of the jointly influenced area, respectively (Tables 3, 4).

Statistical analysis of land-use transfer in the HV between 2000 and 2020 showed that the area occupied by non-utilized land, cultivated land, and forest decreased were 524.67 km², 129.71 km² and 20.95 km² in size, respectively. Furthermore, the areas occupied by grassland, construction and water increased by 502.32 km², 126.61 km² and 46.47 km², respectively. Of these, the area of land that was transferred from non-utilized land to grassland was 619.61 km² in size while the areas of arable land that were degraded to grassland, occupied by construction, and converted to watershed were 165.70 km², 124.70 km², and 25.20 km² in size, respectively. The areas reclaimed from grassland and forest to cultivated land were 137.65 km² and 161.36 km² in size, respectively; the extent of transfers between the remaining land categories was generally low (Figure 8; Table 7). With regards to the area experienced by land-use transfer, the average NDVI 0.523 slightly lower than the average NDVI 0.589 in the HV; the NDVI growth area accounted for 99.16% of the area, of which 94.66% and 4.51% of the area experienced significant and more significant increases, respectively. Human activities promoted the growth of NDVI in the HV by changing the use of land (Table 8).

The trend for NDVI change in the HV between 2001 and 2020 was divided into improved areas (including obviously improved areas and mildly improved areas), stabilized areas, and degraded areas (including obviously degraded areas and mildly degraded areas), while land-use transfers were counted separately within each area. Within the NDVI improvement, stabilization and degradation zones, the proportions of reduction for non-utilized land were 94.59%, 5.02% and 0.39%, respectively. The proportions of non-utilized land conversion to grassland were 91.88%, 7.74% and 0.38%, respectively. The proportions showing the reduction of arable land were 11.73%, 28.34%, and 59.93%, respectively. Areas involving construction increased by 18.72%, 26.36%, 54.92%, respectively, while 96.95% and 3.05% of the areas showed an increase in the NDVI improved and stabilized zones (Table 9).

The area of NDVI improvement was approximately 28,342.44 km² in size, and land-use transfer was mostly distributed in the north and south of the HV. In the north of the HV (Menyuan County), land-use changes were mainly manifested by the conversion of grassland to cropland and forest land, non-utilized land to grassland, watersheds to grassland, the degradation of forest to grassland, and the degradation of grassland to non-utilized land. The south of the HV (Guide County, Jianzha County, Tongren County, Hualong County) was mainly associated with the transfer of non-utilized land and cropland being converted to grassland. The remainder of NDVI improvement only featured a small proportion of land-use transfers. These changes led to a reduction of non-utilized land by 496.33 km² and an increase of grassland by 496.08 km²; only small changes were recorded for other types of land-use. The NDVI stabilization zone was 6054.17 km² in size; the land-use changes in this area showed that cultivated land, non-utilized land and forest decreased by 36.76 km², 26.33 km² and 2.96 km², respectively, while grassland, water and construction increased by 15.59 km², 17.08 km² and 33.37 km², respectively. Overall, the extent of land-use changes was relatively small, therefore implying that the NDVI of the region remained relatively stable. The region exhibiting NDVI degradation was approximately 6841.90 km² in size, while forest, cultivated land, and non-utilized land decreased by 20.70 km², 17.55 km², and 15.52 km², respectively. Furthermore, watershed and construction increased by 146.33 km² and 94.37 km², respectively, thus leading to NDVI degradation in this zone.