Evaluation of uncertainty in eddy covariance measurements within Fluxnet-Canada

Over the course of the last three years, the Fluxnet-Canada Research Network (FCRN) conducted a series of intercomparison experiments to establish common quality standards for flux measurements across the network. A standard eddy covariance (EC) system consisting of a GillR3 sonic anemometer and a LI-COR LI-7500 open-path as well as an LI-7000 closed-path infra-red gas analyser (IRGA) was installed at the main tower sites of all FCRN flux stations for up to four weeks at a time. During the field experiments, we followed a protocol for daily data analysis consisting of 1) visual inspection of high frequency data, 2) check of diagnostic variables, 3) analysis of power and co-spectra, 4) regression analysis of processed fluxes and 5) calibration and delay-time check of the closed-path system. Any problems discovered in the course of this daily analysis were solved and the measurements were continued until enough data points for a representative regression analysis were collected. Only measurement periods with good wind direction, no rain and air temperatures >5 ^oC were used in this analysis. Measurements were usually continued until at least 100 half-hourly periods were collected.

Over the course of the three field seasons that the XSITE project has been in operation, three lessons have been learned: 1) Correct positioning of EC sensors with respect to the tower and dominant wind direction is essential to achieving good agreement between EC systems; 2) Sonic anemometers are the most common source of disagreement between site and intercomparison EC systems; 3) The LI-7500 open-path IRGA, when carefully calibrated in the laboratory and used in favourable conditions (T_air > 5^oC, no rain), is reliable enough to detect most faulty IRGA behaviour. We did not attempt to correct the calibration parameters of the LI-7500 open-path IRGA in the field since changing exposure to wind and sunlight as well as temperature variations made it extremely difficult to obtain sufficiently accurate and stable calibrations even under favourable environmental conditions.

Generally, the differences between fluxes of sensible heat, latent heat, CO₂ and friction velocity (u_*) from the XSITE and site system were (with both systems operating properly) less then 5 %, the threshold value for the flux comparison. This value was chosen because instantaneous errors for anemometers (based on Gill and CSAT3 manuals) can be up to 6 % of the measured value depending on the attack angle, noise levels and calibrations. Gain errors in the IRGA can be up to 1 % of the measurement and depend on instrument temperature. Based on these estimates and our experience with instrument comparisons in the field, we decided to use the 5 % threshold value to identify cases in which an attempt to improve measurements is justified. Comparisons where agreement better than 5% could not be achieved, even after some attempt at improvement of the site EC system, were those where the criteria for the number of data points and weather conditions could not be met. The agreement between flux measurements from the two systems at all FCRN stations gives good confidence in the reliability of the results produced by the network.