China’s deserts greening and response to climate variability and human activities

Residual analysis. NDVI variations are jointly affected by climate variability and human activities [64,66,67]. However, the impact of human activities on the NDVI of deserts is diffi- cult to assess by using only one indicator or several indicators. The residual analysis method can separate the influence of climate factors and human activities with the help of multiple lin- ear regression (Eqs 5 and 6), so it is widely used to identify the impact of human activities [23,68]. NDVIpre;n ¼ a�TMPn þ b�PREn þ c ð5Þ NDVIhuman ¼ NDVIobs NDVIpre;n ð6Þ where NDVIpre,n is the value of the NDVI predicted by temperature and precipitation in year n. TMPn and PREn are annual mean temperature and annual accumulated precipitation in year n. a and b are the coefficients of temperature and precipitation, respectively, calculated by multiple linear regression to obtain. c represents the intercept of the multiple linear regression. NDVIhuman represents the variations in NDVI affected by human activities. NDVIobs represents observed NDVI. Results Temporal variation of NDVI and climatic factors The mean values of the monthly, seasonal, and annual NDVI, temperature, and precipita- tion of the whole study area were calculated for use in the analysis of the temporal varia- tions, and the trends were identified by using the Theil-Sen estimator and Mann-Kendall test. According to the results (Fig 2a-2c), the annual NDVI showed a significant increasing trend at a rate of 0.0018/yr (|Z| > 1.96), up from 0.094 in 2000 to 0.126 in 2017. However,

temperature and precipitation showed a slight decreasing trend (-0.0010/yr) and an increas- ing trend (0.7976/yr), respectively, and both fluctuated greatly and showed insignificant trends. The seasonal differences in the NDVI were obvious. The mean values of the NDVI from 2000 to 2017 for each season were as follows: summer (0.1005)>autumn (0.0837)>spring (0.0729)>winter (0.0601). The seasonal temperature showed a different pattern and was higher in spring than in autumn. Because of the large fluctuations of seasonal precipitation in the time series, the variation curves for spring and autumn precipitation were staggered. With respect to the trends, the seasonal NDVI (Fig 2(d)) showed a significant increasing trend in all four seasons (|Z| > 1.96), with summer showing the highest rate of increase at 0.0017/yr, fol- lowed by autumn (0.0013/yr) and spring (0.0010/yr), and winter showing the lowest rate of increase (0.0008/yr). Seasonal temperature and precipitation did not show significant fluctua- tion trends. The temperature showed slight fluctuations, with increasing trends in summer (0.0054˚C/yr), and decreasing trends in winter (-0.0233˚C/yr). There were greater fluctuations in seasonal precipitation, especially in summer, with increasing trends in summer (0.4295 mm/yr) and a decreasing trend in winter (-0.0627 mm/yr). The data showed monthly NDVI variations (Fig 3). The mean values of the monthly NDVI from 2000 to 2017 showed an inverted U-shaped curve, which was the lowest in January (0.0539) and the highest in August

(0.0989). The variations in monthly temperature and precipitation were the same as the NDVI variations with one exception: the maximum values of both temperature and precipitation occurred in July. The trend of the NDVI over 12 months was positive, implying that the NDVI increased throughout the year during the study period. PLOS ONE China’s deserts greening and response to climate variability and human activities PLOS ONE | https://doi.org/10.1371/journal.pone.0256462 August 30, 2021 6 / 20