Attribution of growing season vegetation activity to climate change and human activities in the Three-River Headwaters Region, China

Over the past century, vegetation change has been reported at global, national, and regional scales, accompanied by signi ﬁ cant climate change and intensi ﬁ ed human activities. Among the regions is the rangeland of the Three-River Headwaters Region (TRHR) in China. However, which factor dominates in causing vegetation change in this region is still under considerable debate, and how would the grasslands adapt to the changing environment is largely unknown. To address these issues, we attribute growing season vegetation activity to climate change and human activities, investigate the interactions among different driving variables, and explore the dynamic relationship between vegetation activity and the driving variables. We perform Mann – Kendall trend analysis, Pearson correlation analysis, and partial correlation analysis. The results indicate that the dominant factor for vegetation growth, during the period 1995 – 2014, was temperature for the southeastern and southern parts of the TRHR, precipitation for the western part, and solar radiation for the northeastern part. The regulation effects of temperature on precipitation and cloud cover contributed to vegetation growth, while grazing activity and population activity offset the positive contribution of climate change. The dynamic relationship between vegetation activity and the driving variables re ﬂ ected the acclimatization and adaption processes of vegetation, which needs further investigation.


INTRODUCTION
In the terrestrial biosphere, vegetation plays a vital role in providing food and habitats for humankind and animals  land changes: climatic driver, non-climatic driver, the combination of climatic and anthropogenic drivers, or the alternation of climatic and anthropogenic drivers. They concluded that spatiotemporal heterogeneity of human activity intensity should be taken into account in the models for attribution. We address this debate and discuss the spatiotemporal heterogeneity of anthropogenic drivers in the present study.
In recent decades, the Qinghai-Tibet Plateau has undergone significant climate warming, with the magnitude of warming reported for the region to be larger than that for the Northern Hemisphere and for the entire globe (Kuang & Jiao ). The increase in precipitation has not been as pronounced as the increase in temperature (Xie et al. ). Nevertheless, it has shown more complicated spatially heterogeneous pattern (Li et al. ), which has resulted in different impacts on vegetation growth. Generally, grassland activity is extremely sensitive to temperature and precipitation changes, and also to non-climatic factors, including grazing, fires, nitrogen deposition, and rising CO 2 levels (Zelikova et  Over the past century, numerous methods have been developed to detect and assess the influence of variables/ components in dynamic systems. In terms of vegetation ecology, the application of the partial correlation analysis to study the effects of climatic factors on vegetation activity has been attempted by a few studies, with some promising outcomes. Piao et al. () argued that the weakening relationship between inter-annual temperature variability and northern sphere vegetation activity is induced by the effect of drought trends and increasing occurrence of extreme hot days. Cong et al. () found that the impact of warming on vegetation activity in the Tibetan Plateau is more positive during periods with more precipitation. In this study, we attempt to properly attribute the growing season vegetation activity to both climatic factors (precipitation, temperature, and cloud cover) and human activities (grazing effect and population pressure) in the Three-River Headwaters Region (TRHR), the hinderland of the Qinghai-Tibet Plateau. For this purpose, we perform Pearson correlation analysis and partial correlation analysis.

Study area
The TRHR (Figure 1) is the source region for three major rivers in China, namely the Yangtze River, the Yellow River, and the Lancang River. It covers an area of 35,000 km 2 and features a fluctuating terrain, dense river networks, extensive snowy mountains, and criss-crossing glaciers (Zhang et al. ). The region has a plateau continental climate and belongs to climate zones varying from semi-arid to semi-humid to humid areas from the northwest to the southeast. Along with Tibet, the region is regarded as the Asian Water Tower (Immerzeel et al. ), with altitude ranging from 2,000 m to nearly 6,800 m. Mainly covered with alpine meadow and alpine steppe, the region plays a key role in the ecological security of China due to its sensitive and fragile ecological environment (Zhang et al.

).
Over the past half a century, the TRHR has experienced

Climate data
The climate data used in this study are precipitation, temperature, and cloud cover. They are obtained from CHIRPS, ERA-Interim, and CRU TS 4.01, respectively. Monthly

Trend analysis
The non-parametric Mann-Kendall test, which is widely used in meteorology, hydrology and ecology studies The assumption of the Mann-Kendall test is that the data are serially independent (Liuzzo et al. ). Therefore, the trend-free pre-whitening method is applied to the data to eliminate the influence of autocorrelation before the application of Mann-Kendall test (Yue & Wang ).
Considering that the coefficients obtained for lag-2 and lag-3 autocorrelation are small, only the lag-1 autocorrelation signal is removed from the original signal.
After the application of the Mann-Kendall trend test, the magnitude of trend is estimated using the Thiel-Sen slope method (Hirsch et al. ): where β is the estimate of the slope for change rate, and x i refers to the observation for the i-th year.

Pearson correlation analysis
To check the relationship between the vegetation and the driving variables, the Pearson correlation analysis is performed. The Pearson correlation coefficient is defined as follows: where ρ X,Y is the Pearson correlation coefficient beween X and Y, cov(X,Y) is the covariance, and σ X and σ Y are the standard deiation of X and Y, respectively.

Partial correlation analysis
where R 12,34 … n refers to partial correlation coefficient between variables 1 and 2 while removing the effect of variables 3, 4, …, n. For example, if we denote 1 for NDVI, 2 for precipitation, 3 for temperature, 4 for cloud cover, 5 for livestock production, and 6 for population, then R 12,3456 represents the partial correlation coefficient between NDVI and precipitation when the effects of temperature, cloud cover, livestock production, and population are partialed out.
In this study, we conduct the partial correlation analysis between NDVI GS and only any one of the six driving variables at a time, controlling the other four variables. We try to disentangle the coordination effects between different variables in performing the partial correlation analysis.

Spatiotemporal change in vegetation
First, we apply the trend analysis to investigate the regional vegetation change in the TRHR. Overall, NDVI GS in the TRHR experienced, during 1982-2014, a slightly greening (Figure 2(b)). This result is in agreement with that reported by some earlier studies   We conduct a similar analysis at the county scale for the period 1995-2014. Figure 6 presents the partial correlation coefficients between NDVI GS and the driving factors. The partial correlation coefficients between NDVI GS and GP, controlling the effects of GT, GC, LP, and POP, showed a remarkable spatial pattern (Figure 6(a)). The results suggest that NDVI GS in the northeastern and western parts of the TRHR was positively correlated with GP over the period 1995-2014, but the link was weak (0.1< p < 0.2). However, NDVI GS in the southeastern and southern parts of the TRHR was negatively correlated with GP, but again the correlation was not significant (p > 0.2). The relationship between NDVI GS and GT also showed a spatial pattern ( Figure 6(b)). A significant positive partial correlation between NDVI GS and GT existed in the southeastern and southern parts of the TRHR, while a weak negative correlation between NDVI GS and GT existed in the parts. For counties where GP was relatively low in the western part, the primary driving factor was precipitation.
Besides, the results also reveal that vegetation activity in northeastern part was mainly driven by cloud cover (solar radiation). Vegetation in the western and northeastern parts responded to GP more sensitively, while it responded negatively in the southeastern part (Figure 6(a)). This heterogeneity could have been mainly due to temperature difference. As temperature is suggested as the main con- 7(h), and 7(i)), but the relationship between vegetation activity and temperature was weakening with an increase in temperature (Figure 7(d)). This means that the role of climate warming has become less dominant, or even negative.

DISCUSSION
Moreover, the continuously increasing grazing and population have posed pressure on vegetation activity, which might offset the contribution of climate warming further.

Dynamic relationship between vegetation activity and the driving factors
We assess the dynamic relationships between vegetation activity and the driving variables by performing the partial correlation analysis with an 11-year moving window.   Figure S5c). That is to say, the magnitude of the response of vegetation to changes was to alleviate the pressure that these changes put on vegetation. This function might be fulfilled by adjusting respiration, photosynthesis, or evaporation (Shen et al. b). Under intensified climate change and human activities, species must acclimate, adapt, move, or die (Corlett & Westcott ). The mechanism of this acclimatization and adaption needs further investigation.

CONCLUSIONS
In this study, we analyzed the changes in the growing season vegetation, growing season climate change, and annual human activities in the TRHR. We attributed the growing season vegetation activity to climate change and human activities by applying the Pearson correlation analysis and the partial correlation analysis. We also investigated the interactions among the climatic variables as well as between climatic variables and anthropogenic variables, and explored the dynamic relationship between vegetation activity and the driving factors. The results indicated that  (d) change in RNDVI-LP; and (e) change in RNDVI-POP. The change means the Pearson correlation coefficient between partial correlation coefficient and time. The labels on the colour bar, r ¼ ±0.77, r ¼ ±0.63, r ¼ ±0.55, and r ¼ ±0.44, correspond to the 1%, 5%, 10%, and 20% significance levels. The dots indicate significance at p < 0.05.