Chamber measurement of surface‐atmosphere trace gas exchange: Numerical evaluation of dependence on soil, interfacial layer, and source/sink properties

Abstract

Abstract. We employed a three-dimensional finite difference gas diffusion model to simulate the performance of chambers used to measure surface-atmosphere trace gas exchange. We found that systematic errors often result from conventional chamber design and deployment protocols, as well as key assumptions behind the estimation of trace gas exchange rates from observed concentration data. Specifically, our simulations showed that (1) when a chamber significantly alters atmospheric mixing processes operating near the soil surface, it also nearly instantaneously enhances or suppresses the postdeployment gas exchange rate, (2) any change resulting in greater soil gas diffusivity, or greater partitioning of the diffusing gas to solid or liquid soil fractions, increases the potential for chamber-induced measurement error, and (3) all such errors are independent of the magnitude, kinetics, and/or distribution of trace gas sources, but greater for trace gas sinks with the same initial absolute flux. Finally, and most importantly, we found that our results apply to steady state as well as non-steady-state chambers, because the slow rate of gas diffusion in soil inhibits recovery of the former from their initial non-steady-state condition. Over a range of representative conditions, the error in steady state chamber estimates of the trace gas flux varied from -30 to +32%, while estimates computed by linear regression from non-steadystate chamber concentrations were 2 to 31% too small. Although such errors are relatively small in comparison to the temporal and spatial variability characteristic of trace gas exchange, they bias the summary statistics for each experiment as well as larger scale trace gas flux estimates based on them.

CO<sub>2</sub> flux determination by closed-chamber methods can be seriously biased by inappropriate application of linear regression

“…Changes of light, temperature and humidity would change plant physiology and thus complicate the form of the c(t) curve whereas artificial changes of pressure, wind and turbulence may additionally impact transport processes and thus even compromise the assumption that the initial slope of the c(t) is the best estimator of the predeployment CO 2 flux (Hutchinson et al, 2000;Hutchinson and Livingston, 2001). …”

“…An additional reason for the unexplainable curvature could have been small positive pressure perturbations during chamber placement (Hutchinson and Livingston, 2001). Although the possible disturbing effects of altering turbulence or pressure conditions by closed chambers were discussed previously by several studies (Hanson et al, 1993;Le Dantec et al, 1999;Hutchinson et al, 2000;Livingston and Hutchinson, 2001;Denmead and Reicosky, 2003;Reicosky, 2003;Livingston et al, 2006), additional investigations are certainly needed concerning these issues.…”

“…If initial noise is obviously present, the data from the respective time period has to be discarded, and the starting time of the experiment t 0 =0 should be delayed accordingly. It has to be stressed that this initial data discarding would lead to inherent underestimation of fluxes because the slope of the c(t) evolution curve is expected to be greatest and changing most strongly at the start of the chamber closure time (Hutchinson et al, 2000;Livingston et al, 2006, this study). Still, this underestimation would be less when applying a nonlinear model compared to the use of linear regression.…”

“…An exponential function developed from simplified diffusion theory was best suited for the flux estimate with underestimation of the fluxes of maximal 11%. In the following years, further theoretical and numerical studies came to the same conclusion that the use of linear regression can lead to serious underestimation of gas fluxes between soils and atmosphere (Hutchinson and Mosier, 1981;Healy et al, 1996;Hutchinson et al, 2000;Pedersen, 2000;Pedersen et al, 2001;Welles et al, 2001;Hutchinson and Livingston, 2001). The serious underestimation bias of the linear regression method as predicted by the theoretical and numerical studies was confirmed by Nakano et al (2004) by measurements of CO 2 release and CH 4 consumption from soils under actual field conditions.…”

“…On the other hand, it is prone to a variety of potential errors (Livingston and Hutchinson, 1995;Welles et al, 2001;Davidson et al, 2002) which the investigator has to consider and to minimise by careful experiment planning and chamber design. Sources of errors are (1) inaccurate determination of the headspace volume (Livingston and Hutchinson, 1995), (2) leakage directly at the chamber components or via the underlying soil pore space (Hutchinson and Livingston, 2001;Livingston et al, 2006), (3) temperature changes of the soil and the atmosphere beneath the chamber (Wagner and Reicosky, 1992;Drösler, 2005), (4) artificial water vapour accumulation which depletes the CO 2 concentration and might influence the stomata regulation of plants (Welles et al, 2001), (5) disturbance of pressure gradients across the soil-atmosphere interface by soil compression or insufficient pressure relief during chamber setting (Hutchinson and Livingston, 2001;Livingston et al, 2006), (6) suppression of the natural pressure fluctuations (Hutchinson and Mosier, 1981;Conen and Smith, 1998;Hutchinson and Livingston, 2001), (7) alteration or even elimination of advection and turbulence and thus modification of the diffusion resistance of the soil-or plant-atmosphere boundary layer (Hanson et al, 1993;Le Dantec et al, 1999, Hutchinson et al, 2000Denmead and Reicosky, 2003;Reicosky, 2003), and (8) the concentration build-up or reduction within the chamber headspace that inherently disturbs the underlying concentration gradients that were in effect prior to chamber deployment (e.g. Matthias et al, 1978;Hutchinson et al, 2000;Livingston et al, 2006).…”