Вы находитесь на странице: 1из 3

DIVISION S-2-NOTES

CHAMBER MEASUREMENT OF the exchange process depends on conditions that vary


with each chamber deployment. In some situations, such
SOIL-ATMOSPHERE GAS EXCHANGE: as during measurement of small exchange rates from
LINEAR VS. DIFFUSION-BASED highly porous soils, limits imposed by sampling logistics
FLUX MODELS and analytical precision often preclude shortening the
sampling period enough to avoid significant nonlinearity.
W. H. ANTHONY, G. L. HUTCHINSON,* AND Application of a linear model to such data may seriously
G. P. LIVINGSTON underestimate the unperturbed flux (Hutchinson and Liv-
ingston, 1993).
Abstract Alternatives to the linear model are few, and they
typically lack many of its advantages. Some investigators
We compared linear regression with a diffusion-based model for
N2O flux estimation using non-steady-state chamber gas concentrations
(e.g., Matthias et al., 1978; de Mello and Hines, 1994)
from a long-term study of N cycling in a managed grass pasture on have adopted an exponential model based partly on diffu-
sandy soil in southern Texas. Of 2224 chamber deployments, 449 met sion theory to account for the diminishing rate of change
criteria established for using the diffusion-based model, which yielded in headspace concentration as a function of time (r). The
flux estimates that averaged 54% larger than linear regression (n = 3). first derivative of that relation evaluated at t = 0 is then
Although they represented only about 20% of all chamber deployments, used to predict the exchange rate at the moment of
this group included most of the data with greatest influence on the deployment, when the chamber's impact is minimal.
magnitude and dynamics of total N2O exchange at our site; e.g., of the Hutchinson and Mosier (1981) proposed a special case
263 fluxes >10 g N ha"' d' 1 ,192 (or 73%) were included. Apparently, of the exponential model that is applicable only when
application of a linear model to nonlinear chamber concentration data
represents a potentially serious source of measurement bias that may
the change in concentration (C) is measured for two
influence not only summary statistics for the experiment, but also successive periods of equal length:
larger scale budgets based partially or wholly on those data. V(Ci - Co)2 - C
-In
At\(2C\ 2 Co) C2 -

G AS EXCHANGE across the soil-atmosphere boundary


depends largely on the diffusivity and concentration
gradient of each species between the soil surface and
for ?2 = 2t\ and
C\ Co
C2 -
[1]

subsurface sites of production or consumption (Ghildyal where / is gas flux, V is chamber volume, A is the
and Tripathi, 1987). Rates of exchange are commonly covered soil area, and subscripts identify the sampling
estimated from observed changes of the gas concentra- event (Event 0 represents ambient conditions).
tions within a non-steady-state chamber, so called be- We compared fluxes computed from Eq. [1] with those
cause the gradient of each species in underlying soil estimated by linear regression using a large N2O data
continually adjusts to the changing concentrations in the set from a long-term N cycling study. Our objective was
chamber headspace (Livingston and Hutchinson, 1995). simply to demonstrate the potential for large differences
Diffusion theory describes this feedback process, pre- in the flux estimates that result from applying the two
dicting a decreasing rate of change in headspace concen- models to nonlinear chamber concentration data.
tration with time (Matthias et al., 1978; Hutchinson and
Mosier, 1981). Nevertheless, a linear model is often Materials and Methods
used to approximate the relation between observed con-
centrations and time under the assumption that for short Soil-atmosphere N2O exchange was measured periodically
deployment periods, the rate of change is nearly constant. for 1 yr at four locations within each of four replicate plots
Adopting a linear regression approach offers many advan- under either minimal or intensive management treatments. The
study site is a grass pasture on well-drained, very uniform
tages, including that it accommodates measurement vari- Kenney sandy loam (a member of the loamy, siliceous, thermic
ability and facilitates testing both the model's goodness Grossarenic Paleudalfs) in a humid, subtropical region of south-
of fit to the observed concentration data and whether ern Texas. Field site characteristics, treatment variables, mete-
each observed exchange rate is significantly different orological data, and chamber design were described in detail
from zero (Livingston and Hutchinson, 1995). by Hutchinson et al. (1993). Accumulation of N2O within each
The validity of assuming a linear model to describe chamber after 10 and 20 min (C\ and C2, respectively) was
determined by drawing 30-mL air samples through the cham-
bers' sampling ports using 60-mL polypropylene syringes fitted
W.H. Anthony, Cooperative Agricultural Research Center, Prairie View with nylon stopcocks. The initial N2O concentration of each
AM Univ., P.O. Drawer U, Prairie View, TX 77446; G.L. Hutchinson, chamber (Co) was assumed equal to that of a single ambient
USDA-ARS, Soil-Plant Nutrient Research Unit, P.O. Box E, Fort Collins, air sample collected at the field site each day in a 600-mL gas
CO 80522; and G.P. Livingston, Earth Systems Science Division, NASA
Araes Research Center, SGE: 239-20, Moffett Field, CA 94035. Received
sampling bag. All samples were transported to the laboratory
13 Oct. 1994. *Corresponding author (glhutch@lamar.colostate.edu). in an insulated container for analysis within 12 h by gas
chromatography using an electron capture detector (Mosier
Published in Soil Sci. Soc. Am. J. 59:1308-1310 (1995). and Mack, 1980). Measured exchange rates at this very sandy
1308
NOTES 1309

Fig. 1. The defining inequality and an example of each of the six possible curve shapes for N2O accumulation in a non-steady-state chamber
during two successive equal periods beginning with chamber deployment.

site were generally small compared with other subtropical of diffusive gas exchange, i.e., Type 1 curves for which
grasslands except for a brief period following fertilization the N2O concentration change during both sampling periods
or precipitation, and especially a combination of these two exceeded the MDCD. Type 4 cases were eliminated from this
(Hutchinson et ah, 1993). subset because for the two cases that met the criteria for
The underlying assumption in our analysis of these data inclusion, ancillary data did not support the assumption of
was that soil-atmosphere N2O exchange is driven primarily N2O consumption. Model comparisons and evaluation of the
by diffusion, which should generate N2O accumulation curves significance of each linear regression slope were performed
like the Type 1 and Type 4 curves in Fig. 1 in production using simple f-tests at a significance level of a = 0.05. Statisti-
and consumption situations, respectively. It follows that the cal significance of the Eq. [1] flux estimates could not be
remaining four Fig. 1 curve types should be observed only determined because both degrees of freedom were used to
when the concentration changes during each 10-min period characterize the nonlinear relation of chamber concentration
are small compared with measurement variability inherent in vs. time.
the data (i.e., the random variability associated with collection,
handling, and analysis of the samples). Alternatively, occur- Results and Discussion
rence of curve Types 2 , 3 , 5 , and 6 could suggest that produc-
tion-consumption was not constant during the deployment Data for the 2224 chamber deployments at our experi-
period or that diffusion was not the mechanism governing gas mental site are summarized in Table 1. For the 449 Type
transport, but neither interpretation was supported by ancillary 1 cases with both NaO concentration changes greater
data collected during the study. than our MDCD, Eq. fl] yielded flux estimates that
The availability of only three observations for each chamber averaged 54% larger than linear regression (P < 0.001).
deployment precluded using a nonlinear regression approach The much larger difference (81 %, P < 0.001) computed
to partition our measurement variability, so we adopted an for all 1172 Type 1 cases reflects, in part, that Eq. [1]
arbitrary estimate of this parameter. We defined our minimum is highly sensitive to measurement error, especially when
detectable concentration difference (MDCD) on a daily basis as
twice the mean absolute difference between successive replicate the flux is small, because it can take advantage of only
analyses of the large ambient air sample that we routinely three observations. Although absence of a statistical mea-
performed following every sixth chamber sample injection. sure of measurement variability limits interpretation of
We present results for both the entire data set and for that these differences, examination of the frequency distribu-
subset of the data with nonzero concentration changes typical tions of flux estimates from both models confirmed that

Table 1. Total number of chamber deployments with each of the N2O accumulation curve types defined in Fig. 1, number of deployments
of each type for which both concentration changes exceeded the minimum detectable concentration difference (MDCD), mean N2O
fluxes for each group computed from Eq. [1] and linear regression (LR), and the number of linear regression flux estimates that were
significantly different from zero (P s 0.05).
All cases Cases > MDCD
Mean N2O flux Mean N2O flux
type Cases Eq. [1] LR P s 0.05 Cases Eq. [1] LR P < 0.05
g N ha d - ' S N ha" 1 d -i
1 1172 11.7 6.5 234 449 20.4 13.1 145
2 450 2.7 4.5 151 128 6.8 9.2 76
3 213 undefined! 0.7 0 21 undefined 0.9 0
4 61 -7.6 -1.7 0 2 -3.3 -1.7 0
S 35 -0.7 -2.7 0 6 -1.8 -3.4 0
6 293 undefined -0.2 0 48 undefined 0.1 0
Total 2224 654
t For curve types 3 and 6, the logarithmic term in Eq. [1] is undefined.
1310 SOIL SCI. SOC. AM. J., VOL. 59, SEPTEMBER-OCTOBER 1995

Eq. [1] yielded substantially fewer near-zero flux esti- those events exhibit nonlinear behavior, employing a
mates, and that their mean, median, and range were all linear model to describe the exchange process may seri-
greater than for linear model results. Note also that Table ously underestimate the flux, thereby introducing down-
1 data for curve Types 2 through 6 support our contention ward bias in the summary statistics for the experiment,
that their apparent concentration differences largely rep- as well as regional- and global-scale budgets based par-
resent measurement error. For example, for the 602 tially or wholly on those data. Limitations of Eq. [1]
cases of Types 3,4,5, and 6, none of the linear regression include that it is applicable to only a subset of chamber
flux estimates were significantly different from zero, the deployments, does not account for measurement variabil-
mean absolute flux was only about 0.7 g N ha"1 d~', ity, and is highly sensitive to that variability when the
and both concentration changes were greater than the flux is small. In addition, because the equation utilizes
MDCD in only 77 cases (13%). Type 2 exceptions to only three chamber concentration measurements, neither
this generalization represent situations where the flux its goodness of fit to those observations nor statistical
was large compared with measurement variability, but so significance of the resulting flux estimates can be tested.
nearly linear that even relatively small errors sometimes Development of a nonlinear regression approach based
caused the second concentration change to exceed the on gas transport theory is needed to estimate the soil-
first, which violates one of the assumptions on which atmosphere exchange rates of N2O and other trace gases,
Eq. [1] is based. particularly in highly porous soils and other situations
Because the unperturbed NzO flux was unknown, it where the rate of change in non-steady-state chamber
was impossible to test which of the two flux models best headspace concentrations is not constant with time.
represented the "true" exchange process in each case.
However, because Eq. [1] is based on diffusion theory,
we adopted it for the 449 Type 1 cases with both N2O
concentration changes greater than the MDCD, estimat-
ing the flux for all other cases by linear regression. This
approach yielded a mean for the 2224 deployments of
5.74 g N ha"1 d"1, 34% greater than the mean that
resulted from applying linear regression to all cases
(4.28 g N ha"1 d"1). The large change in mean flux that
resulted from changing the model for only a small fraction
of the deployments occurred because the selected group
included most of the data with the greatest influence on
the magnitude and dynamics of N2O exchange at our
field site. For example, of the 263 fluxes greater than
10 g N ha"1 d"1, 192 (or 73%) were included among
the approximately 20% of total deployments that met
the criteria established for using Eq. [1]. Most of the
remaining 80% were included among the 1570 cases
with one or both concentration changes smaller than the
MDCD, a population for which the mean flux was only
1.55 g N ha"1 d"1.
It is typical that trace gas exchange has episodic charac-
ter, and that the nature and dynamics of source and sink
ecosystems are defined principally by the large fluxes
that characterize a few short-term exchange events. If
non-steady-state chamber concentration changes during

Вам также может понравиться