Use of change-point detection for friction-velocity threshold evaluation in eddy-covariance studies

The eddy-covariance method often underestimates fluxes under stable, low-wind conditions at night when turbulence is not well developed. The most common approach to resolve the problem of nighttime flux underestimation is to identify and remove the deficit periods using friction-velocity (u_*) threshold filters (u_*^Th). This study modifies an accepted method for u_*^Th evaluation by incorporating change-point-detection techniques. The original and modified methods are evaluated at 38 sites as part of the North American Carbon Program (NACP) site-level synthesis. At most sites, the modified method produced u_*^Th estimates that were higher and less variable than the original method. It also provided an objective method to identify sites that lacked a u_*^Th response. The modified u_*^Th estimates were robust and comparable among years. Inter-annual u_*^Th differences were small, so that a single u_*^Th value was warranted at most sites. No variation in the u_*^Th was observed by time of day (dusk versus mid or late night), however, a few sites showed significant u_*^Th variation with time of year. Among-site variation in the u_*^Th was strongly related to canopy height and the mean annual nighttime u_*. The modified u_*^Th estimates excluded a high fraction of nighttime data - 61% on average. However, the negative impact of the high exclusion rate on annual net ecosystem production (NEP) was small compared to the larger impact of underestimating the u_*^Th. Compared to the original method, the higher u_*^Th estimates from the modified method caused a mean 8% reduction in annual NEP across all site-years, and a mean 7% increase in total ecosystem respiration (R_e). The modified method also reduced the u_*^Th-related uncertainties in annual NEP and R_e by more than 50%. These results support the use of u_*^Th filters as a pragmatic solution to a complex problem.