The 10 Hz eddy covariance raw data were processed into 30-minute averaged fluxes using EddyPro software (version 7.0.9, LI-COR Biosciences, USA). Double coordinate rotation was applied to align the sonic anemometers with the local wind streamlines (Wilczak et al., 2001), which showed better data quality than other coordinate rotation approaches for these sites. Post-processing was applied for further quality control and data correction, including removal of poor-quality data, such as measurements did not pass the tests on steady state and developed turbulent conditions based on a quality flag of 2, data with lower instrument signal strength (<80%), and abnormal data exceeding three times the standard deviation (Chi et al., 2019). Storage correction was applied to each half hourly turbulent flux using the one-level storage estimates. Friction velocity thresholds were determined at each site to remove flux data measured under stable atmospheric conditions (Papale et al., 2006). Additionally, the REddyProc online data processing tool was used for gap-filling based on the marginal distribution sampling approach (Wutzler et al., 2018). In our manuscript, "Contrasting impacts of large-scale afforestation on cloud cover in subtropical China," we compared monthly averaged sensible and latent heat fluxes from five in-situ eddy covariance (EC) sites with corresponding data from the ERA5 and MERRA2 reanalysis products. The sites encompass a secondary karst forest (PD), a primary karst forest (HJ), a eucalyptus plantation (NN), a mixed forest (DHS), and an evergreen broadleaf forest (DJY).