Biases in open-path carbon dioxide flux measurements: Roles of instrument surface heat exchange and analyzer temperature sensitivity

Authors: DEVENTER, M. - University Of Minnesota; ROMAN, TYLER - Us Forest Service (FS); BOGOEV, IVAN - Campbell Scientific, Inc; KOLKA, RANDALL - Us Forest Service (FS); ERICKSON, MATT - University Of Minnesota; LEE, XUHUI - Yale University; Baker, John; MILLET, DYLAN - University Of Minnesota; GRIFFIS, TIMOTHY - University Of Minnesota

Submitted to: Agricultural and Forest Meteorology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 10/11/2020
Publication Date: 1/15/2021
Publication URL: https://handle.nal.usda.gov/10113/7182236
Citation: Deventer, M.J., Roman, T., Bogoev, I., Kolka, R.K., Erickson, M., Lee, X., Baker, J.M., Millet, D.B., Griffis, T.J. 2021. Biases in open-path carbon dioxide flux measurements: Roles of instrument surface heat exchange and analyzer temperature sensitivity. Agricultural and Forest Meteorology. 296. Article 108216. https://doi.org/10.1016/j.agrformet.2020.108216.
DOI: https://doi.org/10.1016/j.agrformet.2020.108216

Interpretive Summary: Measurements of CO2 exchange in different ecosystems provide critical data for assessing the impacts of climate change on the terrestrial carbon cycle. The primary method for making these measurements is eddy covariance (EC), and such systems have been deployed throughout the world over the past few decades, most of them using open-path gas analyzers. Unfortunately, there have been persistent reports of systematic errors in open-path systems, but the source of these errors has never been fully resolved. We conducted a long-term side-by-side comparison of measurements made with an open-path analyzer with those from a closed-path analyzer in an attempt to determine if the commonly used correction procedure is appropriate. The measurements were made in a boreal forest in northern MN, USA. We found that the uncorrected errors in the open-path system were indeed significant, resulting in a substantial overestimate of carbon fixation. We considered two possible sources of the error: heat exchange in the sensor path and temperature sensitivity of the analyzer, and we also explored potential correction procedures. Measurements of temperature gradients in the sensor path indicated that there was indeed heat exchange in the sensor path, but these heat fluxes were poorly correlated with errors in the CO2 flux measurements. Previously proposed correction methods, while they reduced the overestimation of carbon sink strength, they also resulted in either systematic overestimation or underestimation. A new approach, based on short-term measurements and a machine learning algorithm, produced improved results. Further, analysis of open-path calibration residuals as a function of temperature revealed a sensitivity of 5 µmol m-3 K-1. This temperature sensitivity can offset the observed growing season flux bias by 50%. Consequently, we call for a new OP correction framework that characterizes SPHE and temperature induced CO2 measurement errors.

Technical Abstract: Eddy covariance (EC) measurements of ecosystem-atmosphere carbon dioxide (CO2) exchange provide the most direct assessment of the terrestrial carbon cycle. Measurement biases for open-path (OP) CO2 concentration and flux measurements have been reported for over 30 years, but their origin and appropriate correction approach remain unresolved. Here, we quantify the impacts of OP biases on carbon and radiative forcing budgets for a sub-boreal wetland. Comparison with a reference closed-path (CP) system indicates that a systematic OP flux bias persists for all seasons leading to a 110% overestimate of the ecosystem CO2 sink. Two potential sources of OP biases are considered: Sensor-path heat exchange (SPHE) and analyzer temperature sensitivity. We examined potential OP correction approaches including: i) Fast temperature measurements within the measurement path and sensor surfaces; ii) Previously published parameterizations; and iii) Optimization algorithms constructed using machine learning. The measurements revealed year- round average temperature and heat flux gradients of 2.9 ' and 16 W m-2 between the bottom sensor surfaces and atmosphere, indicating SPHE induced OP bias. However, measured SPHE correlated poorly with the observed differences between OP and CP CO2 fluxes. While previously proposed “universal” corrections for SPHE reduced the cumulative OP bias, they lead to either systematic under-correction (by 51%) or to systematic over-correction (by 22-47%) depending on the methodological approach. The resulting budget errors exceeded CP random uncertainty and change the sign of the overall carbon and radiative forcing budgets. Machine learning based OP corrections, informed by representative short-term CP experiments, were superior to all other correction approaches evaluated. Further, analysis of OP calibration residuals, as a function of temperature, revealed a sensitivity of 5 µmol m-3 K-1. This temperature sensitivity can offset the observed growing season flux bias by 50%. Consequently, we call for a new OP correction framework that characterizes SPHE and temperature induced CO2 measurement errors.