Fig. 3.

To identify and remove data collected under low turbulence conditions, under which advective fluxes could lead to an underestimation of fluxes, filtering based on the USTAR threshold was used. In order to estimate the uncertainty in the USTAR threshold calculation, a bootstrapping approach was implemented, with a selection of values representative of the distribution included in the final data products. From the (up to) 200 thresholds from the combined bootstrapping of the two methods, 40 percentiles are extracted. All the subsequent steps of the pipeline are applied to all 40 versions. For each of the final output products (e.g., NEE, as illustrated here), seven percentiles representative of the distribution are included.