3.1. Spatiotemporal Variation in Normalized Difference Vegetation Index (NDVI)
shows the basic situation of the main counties, which shows that water bodies and built-up land accounted for a small proportion, so vegetation variations in these land-use types will not be subsequently analysed. Among the different land-use types, the proportion of grassland in the region was the highest. Among the different counties, NDVI in Zoigê had the highest values.
The spatial distribution of the NDVI in the growing season in the SRYR from 1998 to 2016 exhibited obvious regional differences. The spatial variability analysis showed an increasing gradient of NDVI from northwest to southeast in Figure 2
a. The maximum NDVI value was 0.76, which was located in the Zoigê wetland. By referring to related studies [36
], the NDVI values were classified into 5 levels. The NDVI distribution was analysed in combination with the land-use types (Figure 2
c). The multiyear average NDVI in the growing season was 0.486, of which the area where NDVI was <0.1 covered 1.17% of the total area, mainly including water bodies represented by Eling Lake, Zaling Lake, and Longyangxia Reservoir and permanent glacial snow on the Anyemaqen Mountains. The area with NDVI values between 0.1 and 0.3 covered 15.12% of the area, mainly including unused land that was dominated by sand; the Gobi Desert; the marshlands in northern Qumarlêb, northern Madoi, and western Xinghai around Longyangxia Reservoir; and the sandy area in Huangshatou. The area where the NDVI was between 0.3 and 0.6 covered 50.77% of the area and was mainly distributed in Qumarleb, Madoi, Chindu, Maqên, Xinghai and Guinan, with medium- and low-coverage grassland. In addition, this area also included cultivated land in parts of Guinan. The NDVI values between 0.6 and 0.7 covered 27.90% of the total area. These areas were mainly located in the central counties of the region, which are dominated by medium- and low-coverage grasslands. The areas where NDVI was >0.7 covered 5.04% of the total area, mainly including Aba, Maqu, Zoigê, and Hongyuan, which have high-coverage grasslands and some medium-coverage grasslands.
The spatial distribution of the mean NDVI values can represent the overall trend of the vegetation, but there were opposite changes in different regions, and they can offset each other. Therefore, based on the unitary regression model, the trend of NDVI over the 19 years was analysed at the pixel scale. According to Figure 2
b, the NDVI in the SRYR increased in most areas and decreased in some local areas. According to the statistics, from 1998 to 2016, the area where the NDVI increased covered 71.40% of the total area. Among the areas with NDVI increases, the rapidly increasing area covered 33.12% of the total area and was mainly distributed in the southeast. The NDVI values did not change significantly in 19.41% of the areas. These areas were mainly distributed in Madoi, Gadê and Huangshatou in Guinan. As a typical aeolian sand control area, the trend of NDVI remained basically unchanged, which reflected the long-term and arduous nature of sandy land management. The reduced NDVI area covered 9.19% and was mainly distributed in Qumarleb (grasslands with medium and low coverage, unused land with bare rock), Maqên, and an urban area of Gonghe. The above studies indicated that while the state of vegetation in the SRYR had improved, some areas experienced vegetation degradation.
As we can see in Figure 3
, the NDVI in the SRYR increased slowly over the past 19 years, with a slope of 0.00204/a. Before 2005, the NDVI was lower than the multiyear average values, and then it fluctuated around the average, indicating that the vegetation coverage had improved since 2005. The State Council approved and launched the “master plan for ecological protection and construction of the Three-River-Source Nature Reserve in Qinghai” in 2005 and implemented a series of engineering measures. The results of this article showed that the implementation of these projects had a certain effect on vegetation restoration and protection. From the different land-use types, the trend of the grassland NDVI was the most consistent with that of the whole region. The NDVI values for different land-use types in the region showed an upward trend. The increasing trend of cropland NDVI was the most obvious, with a linear tendency of 0.00559/a, an average NDVI value of 0.46, and a change point that occurred in approximately 2004. Both the woodland NDVI and grassland NDVI showed slow trends, with linear tendencies of 0.00248/a and 0.00214/a, respectively. The average woodland NDVI value was 0.60, and the change point was between 2009 and 2011. The average grassland NDVI value was 0.50, and the change point was approximately 2006. The unused land NDVI showed slight increasing trends, with linear tendencies of 0.00147/a, an average NDVI value of 0.40, and the abrupt point occurred in approximately 2003. In the study area, the unused land in the west mainly included sandy land and the Gobi with low NDVI values; the eastern part, namely, the Zoigê wetland, was dominated by marshland with high NDVI values. In summary, the distribution of the NDVI values in the different land-use types was woodland > grassland > cropland >> unused land. In terms of trends, the grassland NDVI contributed significantly to the annual NDVI in the study area.
The increasing trends in NDVI were obvious in many regions in China; however, the NDVI changes in the SRYR were relatively small, even though many protective measures were adopted by the government in the region at the same time, and increasing trends of climate change were faster than the average in China. Therefore, we discussed the main factors affecting NDVI in the subsequent analysis.
3.2. Impact of Meteorological Elements on the NDVI
Many studies have shown a clear response of NDVI to climate change. The impact of climate change on vegetation is mainly reflected in the hydrothermal conditions. Evapotranspiration data for long-term continuous observations are difficult to obtain. Therefore, climate change can be attributed as a cause of changes in precipitation and temperature. Temperature and precipitation are the most direct and important factors for plant growth [37
]. Figure 4
shows that the precipitation in the growing season showed an upward trend, with a linear trend of 7.17/a and an average value of 449.52 mm. The average temperature linearly increased at 0.04/a, with an average value of 6.42 °C. The precipitation and temperature were mainly below average before 2005. The precipitation had the lowest value in 2001. After 2005, the climate elements fluctuated around the mean value. The consistency between NDVI and temperature was better than the consistency between NDVI and precipitation.
Correlation coefficients between climate elements and NDVI in different land-use types are shown in Table 2
. The partial correlation coefficient between the NDVI and precipitation was 0.221 (P
= 0.289), and the coefficient between the NDVI and temperature was 0.467 (P
= 0.131). Among the NDVI values for the different land-use types, precipitation exhibited the best correlation with cropland, and passed the 5% significance level test. Cropland mainly located in the central of Guinan and north of Tongde. The driving force of precipitation on cropland was stronger than temperature. Temperature had the best correlation with grassland, followed by woodland.
shows the partial correlation between climate elements and NDVI in different counties. The correlation coefficient between precipitation and NDVI was positive in all counties. Among the different counties, at the significance level of 0.05, Guinan exhibited the highest significant correlation proportion that covered 71.08% of the total area, followed by Zêkog and Gonghe. The significant correlation proportion in these counties were all above 50%. The counties where the significant correlation proportion were between 25% and 50% were mainly distributed in Xinghai, Tongde, Madoi, Zoigê, Henan and Qumarleb. The counties where the significant correlation proportion were <25% mainly included Maqên, Hongyuan, Darlag, Maqu, Aba, Jigzhi, Gadê and Chindu.
The counties where the significant correlation proportion between NDVI and temperature were above 75% mainly included Aba, Zêkog and Hongyuan. Aba exhibited the highest significant correlation proportion that covered 83.15% of the total area. Both Aba and Zêkog correlation coefficients passed the 5% significance level test. That is, the driving force of temperature on vegetation was stronger than precipitation in these areas. The counties where the significant correlation proportion were between 50% and 75% were mainly distributed in Jigzhi, Henan, Chindu, Darlag, Maqu, Zoigê, Madoi, Qumarleb and Gadê. The counties where the significant correlation proportion were between 25% and 50% were mainly included Tongde, Maqên and Guinan. The counties where the significant correlation proportion were <25% mainly included Xinghai and Gonghe. The contents of correlation coefficients are shown in Figure 5
shows the relationship between NDVI, climatic elements and partial correlation coefficient in different counties. Precipitation and temperature were positively correlated with NDVI. That is, in different counties, the NDVI increased gradually with increasing precipitation and temperature. The higher the NDVI is, the weaker the correlation between precipitation and NDVI. The higher the NDVI is, the stronger the correlation between temperature and NDVI. Figure 5
c,d show the partial correlation coefficient between climatic elements and NDVI. As shown in the figure, precipitation, temperature and the correlation coefficient were negatively correlated; that is, with the increase in precipitation and temperature, the correlation between NDVI and climatic factors weakened. In general, the lower the precipitation and temperature of the county were, the stronger the correlation between climate factors and NDVI. Generally, the effects on vegetation were more obvious under unfavourable climate conditions than under suitable ones.
The spatial distribution of the partial correlation coefficient between the NDVI and climate elements in the growing seasons is shown in Figure 6
. At the pixel scale, the partial correlation coefficient between the NDVI and precipitation showed a significant positive correlation with precipitation that covered 27.08% of the total area in Figure 6
a. This correlation was mainly distributed in Guinan, which mainly includes grassland and sandy land, most of this region is located between 2559 and 4759 m above sea level, and the annual average precipitation is below 400 mm; significant positive correlations were also observed in Zêkog, Gonghe, Tongde, Xinghai, and northwestern Madoi. A total of 68.99% of the area was not significantly related. The areas with significant negative correlations covered 3.93% of the total area, and points were distributed in Darlag and Madoi. This region is located between 3787 and 5236 m above sea level, and the annual average temperature is below 5.5 °C. There was an increase in precipitation and widespread melting of glaciers and snows, which fed glacial lakes and wetlands, reducing the vegetation coverage in glacial snow regions to a certain extent.
The area where there was a significantly positive correlation between NDVI and temperature covered 56.34% of the total area in Figure 6
b, and was mainly distributed in Zêkog, the southern Madoi, Chindu, Darlag, and the southeastern SRYR. The area that was not significantly related covered 43.60%, and was mainly distributed in Gonghe, Xinghai, northern Guinan, Maqên and Gadê.
Overall, the NDVI exhibited a positive correlation with precipitation and temperature in the SRYR, and the correlation with temperature was higher than that with total precipitation. This result showed that the sensitivity of the NDVI to temperature was higher than that of precipitation, which showed that temperature had a greater impact on vegetation.
3.3. Impact of Human Activities on NDVI
Numerous studies have shown that alpine vegetation, which is highly sensitive to global changes [39
], has been severely affected by global climate change and human activities. The impact of human activities on vegetation changes mainly includes the promotion of increased vegetation cover (ecological engineering, etc.) and the destructive effect of reduced vegetation cover (grazing, urban expansion, etc.). Spatial distribution of the residual analysis for NDVI are shown in Figure 7
. Eight typical areas were selected in the figure, and human activities information in these areas were collected to verify the residual results.
shows that 53.58% of the residual values were negative, which mainly included the central and western regions in the SRYR, and Maqên accounted for the highest proportion. Human activities in these areas play a negative role in vegetation. The values in Maqên within the territory of the Anyemaqen Mountains, were sensitive and exhibited risk of change [41
]. These areas are high-altitude regions with the following basic characteristics: poor water-heat conditions and strong solar radiation, which are not conducive to the implementation of ecological construction projects. The ecological environment continues to deteriorate. Second, at the border of Gadê and Darlag, human activity had a negative effect on vegetation. Over the last 19 years, the desertification and environmental degradation of this region have mainly been attributed to human activities such as overgrazing, under the background of regional climate changes. Liu reported that grassland degradation was the most important land-cover change in the SRYR [42
]. Furthermore, in the central part of Gonghe, land-use changes were caused by the rapid expansion of built-up land and had a negative effect on local vegetation.
In addition, 46.42% of the area exhibited positive residuals, mainly in the Zoigê wetland and nature reserves. Human activities in these areas play a positive role in vegetation. The Zoigê wetland mainly includes Hongyuan, Aba, Zoigê, and Maqu. The residuals in core areas of nature reserves [43
] were positive, which mainly including Yoigilangleb, Eling Lake–Zaling Lake, and Zhongtie-Jungong. The NDVI values in these areas showed an increasing trend, indicating that decreasing trends of vegetation and expanding desertification were restrained, and wetland expansion and increasing vegetation cover were obvious. To a certain extent, the effects of the establishment of the Three-River-Source Nature Reserve (2000) were confirmed, and the ecological protection construction project (2005) has already achieved initial results. The establishment of the Three-River-Source National Park in 2020 indicated that the ecological protection of the Yellow River source area had reached a new level.
3.4. Trend Prediction
Multivariate linear regression equations were used to obtain the regression coefficients of the observed NDVI values and the observed climate elements (precipitation and temperature) from 1998 to 2016. The regression coefficients were fitted based on the climate forecast data from RegCM4 during the same period to simulate the pixel-based change trend of the NDVI. The comparison in Figure 8
b shows that the simulated NDVI tendency value with linear tendencies of 0.00207/a, was the same as the observed NDVI tendency value with linear tendencies of 0.00204/a. This result shows that the credibility of the simulated NDVI trend was high. For the simulated future time period, we chose 2020–2038, which was similar to the past time length and close to the present time. Based on the grid, using the established pixel-scale NDVI-climate model, NDVI change trend distribution from 2020 to 2038 was analysed at the pixel scale with MATLAB. The statistics were calculated with the equation:
NDVI (2020–2038) = Precipitation regression coefficient (1998–2016) * precipitation (2020–2038) + Temperature regression coefficient (1998–2016) * temperature (2020–2038).
According to Figure 8
, the NDVI will show a slight upward trend over the next 19 years, with a slope of 0.00096/a. From 2020 to 2038, the areas where the NDVI will basically remain unchanged and slowly increase cover 54% and 42.43% of the total area, respectively. Among these areas, the basically unchanged areas are mainly distributed in Chindu and Qumarleb, the proportions of which are 91.16% and 86.15% in each county. The slowly increasing areas are mainly concentrated in Zêkog and Tongde, covering 70.76% and 69.83% of the county, respectively. In addition, NDVI has been increasing rapidly in the areas of Guinan, Zêkog and Tongde, where there is currently a large amount of cropland and a small amount of sandy land, following a similar trend over the past 19 years. The increasing NDVI trend in Guinan is the most obvious, with a rapid growth rate of 0.00267/a, covering 83.34% of the county.
The inputs to the prediction model are mainly precipitation and temperature, so the increase in NDVI is related to global warming. Rising temperatures, melting glaciers and increasing precipitation provide a good environment for vegetation cover. New studies have found that shrubs and grasses are springing up around Mount Everest [44
], and the temperature in Antarctica exceeded 20 °C for the first time. These results suggest that Himalayan ecosystems are highly vulnerable to climate-induced shifts in vegetation, and the effects of global warming are spreading. Climate change affects vegetation growth, and vegetation change reflects climate variation. The SRYR is a sensitive area to climate, and the past and future trends of NDVI both demonstrate the warming and wetting trends of climate, which should arouse attention.
The climate simulation model was different from the weather forecast model and the short-term climate prediction model. The dates in the model were not equal to actual calendar dates. Therefore, the results of this study were only for the simulation of future NDVI trends and do not represent current NDVI values.