2.3. Data processing
After data collection, we performed quality control on all flux and meteorological data. Since small deviations can be ignored with low canopy height in flat study sites, axial rotation and trend elimination for raw eddy data were not performed (Zhao et al. 2006, Fu et al. 2009). We use the data processing method of flux from ChinaFLUX (Yu et al. 2008). The data quality was improved by spike screening and night-time filtering. The spike was defined as the beyond 4.5 standard deviations of a 10-day moving window. The night-time flux data collected under low atmospheric turbulence conditions were screened using thresholds of friction velocity (u*), and flux data were rejected when u* < 0.15 m·s−1. Then, the missing CO2 flux data were filled by empirical non-linear regression of valid data with environmental variables (Ma et al., 2007; Li et al., 2016). Daily GPP was obtained by subtracting NEE from RES (Eq. (1)), and daily RES was the sum of nocturnal respiration (RESn) and daytime respiration (RESd), which was extrapolated the exponential regressions of RESn with nighttime soil temperature to the daytime periods. Negative and positive NEE represent the ecosystem to absorb and release CO2 (Yu et al. 2008).
GPP = RES − NEE = (RESd + RESn) − NEE (1)