Flux tower re-deployments associated with a new project located at the WSU Long Term Agricultural Research (LTAR) station have included upgraded firmware and data logging software (called EasyFlux-DL) that includes capture of the high frequency temperature fluctuations. The goal of this work is to determine the significance the CO2 flux bias and the software update has for both historical and future carbon budgets using 3 months of the new LTAR data from the summer of 2017. This required a comparison of the high and low frequency data outputs, a comparison of the EddyPro (previously used) and EasyFlux software outputs, and finally, a comparison of the original REACCH carbon flux results to the LTAR carbon flux results.
To get a sense of the high frequency/low frequency and EddyPro/EasyFlux relationships, correlations between latent heat, kinematic sensible heat, and CO2 flux were characterized between data sets. Analyzing the data in terms of those variables made it possible to narrow down where inconsistencies may occur, while also allowing for the analysis of the carbon budget over time. Three measures of characterization were used: an ordinary least squares linear regression, Pearson correlation coefficient, and fractional bias. These characterizations will also be used to error correct historic data as needed.
The change from the low to high frequency thermistor became evident in the data when the magnitude of the difference between the CO2 flux values was plotted against the actual values of the CO2 flux data. As the actual values increased at each site, so did the magnitude of the difference between the values in the data sets. The resulting difference between the cumulative carbon calculations was within 1 gm^-2 until late June, which then increased dramatically over a short time period to 6 gm^-2. This is approximately 2% of the total annual carbon flux.
The analysis of the software showed that EasyFlux regularly returns lower sensible, latent, and carbon dioxide fluxes. The resulting cumulative carbon fluxes calculated using EasyFlux were 12.7%, 10.4%, and 12.9% lower than fluxes calculated from EddyPro for the Boyd North, Boyd South, and Cook East sites respectively. The data for Cook West was insufficient to complete the analysis.
During the three-month period of measurement at the Boyd sites where winter wheat was planted, the wheat was ending its growing stage and entering senescence. When combined with the results, this suggests that the thermistor bias is small enough in magnitude that it can only be observed at times of lower carbon exchange, and furthermore, historical data may only need to be corrected using the linear regression at certain times of the year. The timing of applying the correction needs to be quantified in future work. It will also be important to validate the results of the low and high frequency comparison at the Cook sites using multiple growing seasons.
Although the differences between EddyPro and EasyFlux were within the overall uncertainty of eddy covariance measurements overall, they were still substantial. Because EddyPro and EasyFlux were coded on different platforms, it is likely that an in-depth comparison of the two software codes could be used to evaluate why EddyPro is returning greater flux values when using the same corrections. Going forward, it will be important to quantify these differences in order to have an accurate record of carbon exchange over agricultural sites as we go into the future.