Analysis methods

In this study, the attribution analysis of the contribution of each driving variable to the changes in the trend of the NEE was based on the upscaling results obtained for case 2. The driving factors of the NEE included the PPT, NDVI, R_d, T_mean, and T_max. A control experiment (Expert_CON) and sensitivity experiment (Expert_SEN) were set to attribute the regional interannual variation trend of the NEE. Taking 1982–2018 as an example, the actual values of each explanatory variable in the control experiment remained unchanged during 1982–2018, and five sensitivity experiments were conducted on the explanatory variables, namely, exper_PPT, exper_NDVI, exper_R_d, exper_T_mean, and exper_T_max. In each sensitivity test, the measured explanatory variables were kept constant at their mean values from 1982 to 2018, and the other variables were kept the same as in the control test. The difference in the NEE trends of the control test and the sensitivity test was considered to be the contribution of this explanatory variable to the NEE trend. However, there are interactions between these drivers, which may introduce some uncertainty into the estimation of each factor’s individual contribution to the change in the NEE. Therefore, in this study, the separation method proposed by Sun et al. (65) was used, which can minimize the error caused by the interactions between the explanatory variables and has been better applied in several studies (66, 67)

where ∑k≠inCk is the total contribution of all the remaining explanatory variables to the NEE trend of the control experiment, except the ith factor; C_k is the contribution of the kth factor to the NEE trend; n is the number of sensitivity experiments (n = 5 in this study); and E_{exper_i} is the NEE trend of each sensitivity experiment. By solving the abovementioned equations, we can obtain the contribution of each driver to the NEE trend C_i

Similarly, following the same analysis for 1982–2018, we also analyzed the contribution of each explanatory variable to the trend of the NEE during the two time periods of 1982–1999 and 2000–2018. The Mann-Kendall test is a nonparametric trend test method that has been widely used in hydrometeorological time series analysis. Compared with the linear regression method, it can provide a more accurate estimation of the skewness of the data (68). The Theil-Sen method was used to calculate the magnitude of the trend, and the nonparametric Mann-Kendall method was used to determine the level of significance of the interannual trends in the NEE and climate variables.