Agreement of multiple night- and daytime filtering approaches of eddy covariance-derived net ecosystem CO2 exchange over a mountain forest

Abstract

The assessment of net ecosystem CO${}_2$ exchange often relies on eddy covariance measurements. Under stable, low-turbulence conditions, the measured flux may not be representative of the net ecosystem exchange (NEE), as unmeasured fluxes (e.g., advection) can become relevant. Consequently, such periods need to be filtered out for robust flux calculations. Typically, the focus lies on nighttime filtering alone, yet daytime flux measurements can also be unrepresentative. This study evaluates well-established and novel filtering methods applied both at nighttime and daytime at a mountain forest site in Tyrol, Austria (Forest-Atmosphere-Interaction-Research (FAIR) site, AT-Mmg). Established methods, including friction velocity filtering, its counterpart using the standard deviation of vertical velocity fluctuations (${\sigma }_w$) and an after-sunset flux maxima approach (commonly referred to as van Gorsel method), are compared. Additionally, we use a more recent approach with a physically-derived measure of flow decoupling for filtering. Moreover, we introduce a novel K-means clustering approach that groups flow situations into clusters based on vertical profiles of temperature, ${\sigma }_w$ and wind speed. Clusters in which the measured flux is expected to be a reasonable NEE estimate are retained. Such scenarios are Foehn periods, early-night situations with high turbulence and low stability, or well-mixed afternoon conditions. Despite being based on widely differing assumptions, the various filtering approaches yielded similar carbon budget estimates over 14 months of measurements (-224 to -286 g C m⁻² for nighttime filtering and -440 to -382 g C m⁻² for all-day filtering), in contrast to the unfiltered budget of -534 g C m⁻². Nighttime filtering results in higher respiration rates throughout the night, while daytime filtering suggests increased morning carbon uptake compared to unfiltered data.