Net ecosystem exchange measured at Naarasneva by eddy covariance: gapfilling models and results

Supporting scripts and data to:

Buzacott et al. (2025), Afforestation-related fertilisation quickly turns barren cutaway peatland into a carbon dioxide sink. Global Change Biology.

This repo contains XGBoost eddy covariance gapfilling scripts and results. The gapfilling scripts are modified from the gapfilling scripts of Irvin et al. (2021) (https://dx.doi.org/10.1016/j.agrformet.2021.108528, https://github.com/stanfordmlgroup/methane-gapfill-ml). The package is found in the fluxgapfill folder. Modifications include to change the target gapfilling output to NEE in g C m-2 yr-1, and some other small fixes to bugs that arose due to new versions of dependencies.

The data folder contains the main gapfilling results and is run with the nsv_co2gapfill.py script. The data_ust folder contains the results u* threshold uncertainty runs, where for 40 quantiles the threshold was propagated through the gapfilling pipeline as in oneflux (Pastorello. et al. 2020; https://dx.doi.org/10.1038/s41597-020-0534-3) and was run with the nsv_co2gapfill_v2_ust.py script. For the final uncertainty, the variance of the budget means from the u* runs were added to the variance of the gapfilling uncertainty in the main run.