Interpretability of negative latent heat fluxes from Eddy Covariance measurements during dry conditions

Abstract. It is known from arid and semi-arid ecosystems that atmospheric water vapor is directly adsorbed by the soil matrix during the night. Soil water vapor adsorption was typically neglected and only recently got attention because of improvements in measurement techniques. One technique rarely explored is eddy covariance (EC). EC nighttime measurements are usually discarded, but soil water vapor adsorption may be detectable as downwards-directed EC latent heat (λE) flux measurements under dry conditions. We propose a classification method to exclude conditions of dew and fog when λE derived from EC is not trustworthy due to stable atmospheric conditions. We compare downwards-directed λE fluxes from EC with measurements from weighable lysimeters for four years in a Mediterranean Savannah ecosystem and three years in a temperate agricultural site. Our aim is to assess if overnight water inputs from soil water vapor adsorption differ between ecosystems and how well they are detectable by EC. 

At the Mediterranean site, the lysimeters measured soil water vapor adsorption each summer whereas at the temperate site soil water vapor adsorption was much rarer, and measured predominantly under extreme drought. In 30 % of nights in the four-year measurement period at the Mediterranean site, the EC technique detected downward-directed λE fluxes of which 88.8 % were confirmed to be soil water vapor adsorption by at least one lysimeter. At the temperate site, downward-directed λE fluxes were only recorded during 15 % of the nights, with only 36.8 % of half-hours matching simultaneous lysimeter measurement of soil water vapor adsorption. Although this relationship slightly improved to 60% under bare soil conditions and extreme droughts, this underlines that soil water vapor adsorption is likely a much more relevant process in arid ecosystems compared to temperate ones and that the EC method was able to capture this difference. The comparisons of the magnitudes between the two methods revealed a substantial underestimation of soil water vapor adsorption with EC. This underestimation was, however, on par with the underestimation in evaporation. Based on a random forest-based feature selection we found the mismatch between the techniques being dominantly related to the site's inherent spatiotemporal variations in soil conditions, namely soil water status, and soil (surface) temperature. 

We further demonstrate that although the water flux is very small with mean values of 0.04 or 0.06 mm per night depending on either EC or lysimeter detection it can be a substantial fraction of the diel soil water balance under dry conditions. Although the two instruments substantially differ with regard to the evaporative fraction with 64% and 25% for the lysimeter and EC methods, they are in either case substantial. Given the usefulness of EC for detecting soil water vapor adsorption as demonstrated here, there is potential for investigating adsorption in more climate regions at longer timescales thanks to the greater abundance of EC measurements compared to lysimeter observations.