Recommendations for gap-filling eddy covariance latent heat flux measurements using marginal distribution sampling

Accurate eddy covariance (EC) measurements require that the atmospheric and orographic conditions meet certain criteria. It is common that up to 60% of the original data must be rejected. In particular, a high percentage of data is often removed during nocturnal periods. Currently, the most widely used method for filling gaps in EC datasets is the tool developed at the Max Planck Institute for Biogeochemistry [as reported by Falge et al. (2001), Reichstein et al. (2005), and Wutzler et al. (2018)]. This tool has been primarily developed and tested for the gap-filling of CO2 fluxes. In this study, we provide the first detailed testing of this gap-filling tool on LE fluxes and explore alternative settings in the gap-filling procedure using different meteorological drivers. The tests were conducted using five EC data sets. Random artificial gaps of four different gap-length scenarios were used to compare the settings. Error propagation for both the default and alternative settings was computed for various time aggregations. In general, we confirm a good performance of the standard gap-filling tool with a bias error of - 0.09 and - 0.21 W m(-2) for nocturnal growing and non-growing season cases, respectively, while daytime average bias error was 0.01 W m(-2). Alternative settings produced similar results to the default settings for diurnal cases; however, the alternative settings substantially (81%) improved the performance of night-time gap-filling. At sites where night-time LE fluxes are significant, we recommend using net radiation instead of global radiation and relative air humidity instead of vapour pressure deficit to drive the LE gap-filling.