GEOPHYSICAL RESEARCH LETTERS, VOL. 35, L18808, doi:10.

1029/2008GL034031, 2008

Emissions of CH4 and N2O over the United States and Canada based
on a receptor-oriented modeling framework and COBRA-NA
atmospheric observations
Eric A. Kort,1 Janusz Eluszkiewicz,2 Britton B. Stephens,3 John B. Miller,4,5
Christoph Gerbig,6 Thomas Nehrkorn,2 Bruce C. Daube,1 Jed O. Kaplan,7
Sander Houweling,8 and Steven C. Wofsy1
Received 17 March 2008; revised 9 July 2008; accepted 22 August 2008; published 26 September 2008.

[1] We present top-down emission constraints for two non- 2006]. Of non-CO2 anthropogenic greenhouse gases, CH4
CO2 greenhouse gases in large areas of the U.S. and southern has the largest RF, while N2O has the third-largest RF
Canada during early summer. Collocated airborne [Hofmann et al., 2006].
measurements of methane and nitrous oxide acquired during [3] Recently, increased attention has been focused on
the COBRA-NA campaign in May–June 2003, analyzed assessing CH 4 sources. Global atmospheric inversions
using a receptor-oriented Lagrangian particle dispersion [Bergamaschi et al., 2007; Chen and Prinn, 2006] and global
model, provide robust validation of independent bottom-up emissions sensitivity tests [Houweling et al., 2006] have
emission estimates from the EDGAR and GEIA inventories. begun to tighten atmospheric constraints on CH4 sources,
We find that the EDGAR CH4 emission rates are slightly low though capabilities are limited on a regional (104 km2 – 106
by a factor of 1.08 ± 0.15 (2s), while both EDGAR and GEIA km2) scale due reliance on data from remote sites and coarse
N2O emissions are significantly too low, by factors of 2.62 ± meteorological resolution (typically 1°  1° or larger). These
0.50 and 3.05 ± 0.61, respectively, for this region. Potential shortcomings can be overcome using a regionally focused
footprint bias may expand the statistically retrieved approach, as taken in the Amazon by Miller et al. [2007].
uncertainties. Seasonality of agricultural N2O emissions may Global studies of N2O have focused on agricultural and
help explain the discrepancy. Total anthropogenic U.S. and industrial emissions from seasonal to interannual time scales
Canadian emissions would be 49 Tg CH4 and 4.3 Tg N2O [Nevison et al., 2007; Hirsch et al., 2006; Prinn et al.,
annually, if these inventory scaling factors applied to all of 1990], but have not attempted to place strong constraints on
North America. Citation: Kort, E. A., J. Eluszkiewicz, B. B. regional budgets.
Stephens, J. B. Miller, C. Gerbig, T. Nehrkorn, B. C. Daube, J. O. [4] In this paper, we constrain emissions of CH4 and N2O
Kaplan, S. Houweling, and S. C. Wofsy (2008), Emissions of CH4 over large areas of the U.S. and southern Canada using
and N2O over the United States and Canada based on a receptor- models and data specifically designed to address regional
oriented modeling framework and COBRA-NA atmospheric scales at high spatial and temporal resolution. We employ a
observations, Geophys. Res. Lett., 35, L18808, doi:10.1029/ Lagrangian particle diffusion model (LPDM) to directly
2008GL034031. connect atmospheric measurements with surface fluxes in
a receptor-oriented framework. Our measurements consist
1. Introduction of flask samples collected during the CO2 Boundary Layer
Regional Airborne – North America (COBRA-NA) cam-
[2] Knowledge of greenhouse gas emissions is central to paign in May and June 2003. A direct comparison of
understanding the Earth’s climate system. Though CO2 is simulations with atmospheric measurements on a fine,
the single most important anthropogenic greenhouse gas, regional scale enables us to avoid many of the problems
non-CO2 anthropogenic greenhouse gases play a significant that hamper global inversions, extending methods and
role in the Earth’s energy balance, contributing nearly as applications of LPDMs to regional studies of surface
much to radiative forcing (RF) as carbon dioxide (0.977 W emissions [e.g., Stohl et al., 2003; Gerbig et al., 2003;
m 2 compared to 1.626 W m 2 in 2004) [Hofmann et al., Matross et al., 2006; Gimson and Uliasz, 2003].

1
School of Engineering and Applied Sciences and Department of Earth 2. Study and Measurements
and Planetary Sciences, Harvard University, Cambridge, Massachusetts, USA.
2
Atmospheric and Environmental Research, Inc., Lexington, Massachusetts, [5] The COBRA-NA campaign consisted of 38 flight
USA.
3
legs < 4 hours, from 23 May 2003 through 28 June 2003
National Center for Atmospheric Research, Boulder, Colorado, USA. (for details see Hurst et al. [2006]). Extensive vertical
4
NOAA Earth System Research Laboratory, Boulder, Colorado, USA.
5
Also at Cooperative Institute for Research in Environmental Sciences, profiling in well-mixed conditions (avoiding frontal and
University of Colorado, Boulder, Colorado, USA. convective activity) was targeted to collect an appropriate
6
Max-Planck-Insitut für Biogeochemie, Jena, Germany. data set for assessing surface emissions of measured gases.
7
Swiss Federal Institute for Forest, Snow, and Landscape Research, Locations of the 300 flasks collected and used in this study,
Lausanne, Switzerland.
8
Netherlands Institute for Space Research and Institute for Marine and
and their collective footprint, can be found in Figure 1.
Atmospheric Research Utrecht, Utrecht, Netherlands. These flasks were analyzed for CO2, CO, CH4, H2, SF6, and
N2O concentrations, and for O2/N2, Ar/N2, 13C in CO2, and
18
Copyright 2008 by the American Geophysical Union. O in CO2 ratios. The CH4 and N2O measurements used
0094-8276/08/2008GL034031

L18808 1 of 5
L18808 KORT ET AL.: CH4 AND N2O EMISSIONS IN NORTH AMERICA L18808

Figure 1. (a) Plus markings indicate locations of airborne flask collection. The average footprint from all the flasks is
shown, indicating regions to which our measurements are most sensitive. (b) CH4 emissions in North America in June
(EDGAR and JK). (c) N2O emissions in North America (EDGAR).

here are calibrated on the NOAA2004 [Dlugokencky et al., tion of particles to reach the boundary, but going beyond
2005] and NOAA2006 [Hall et al., 2007] scales, respec- 6 days has no effect on the footprint, and an insignificant
tively. Further details of flask sampling can be found in the effect on boundary values of the gases considered here. For
auxiliary materials.1 each receptor, we calculate the footprint, which represents
the sensitivity of the receptor point to surface sources, in
3. Methodology units of ppb/(mmol m 2 s 1). The footprint is calculated by
counting the number of particles in a surface-influenced
[6] The LPDM employed in our work is the Stochastic region (defined as 1=2 of the estimated PBL height [Gerbig et
Time Inverted Lagrangian Transport (STILT) model, run in al., 2003]) and the time spent in the region (for details, see
the time-reversed (receptor-oriented) mode. STILT has been Lin et al. [2003]). The footprint is multiplied by an a priori
developed and applied to regional emission of CO2 previ- emission field to compute the associated contribution to the
ously, as described by Lin et al. [2003], Gerbig et al. mixing ratio at the receptor.
[2003], and Matross et al. [2006].
3.2. Emissions Inventories
3.1. Meteorological Input and Footprint Estimation [9] We need to represent sources and sinks that affect
[7] We drove STILT with meteorological fields from atmospheric concentrations on regional spatial and temporal
the Weather Research and Forecasting (WRF) model scales. For CH4, we include natural wetland and anthropo-
[Skamarock et al., 2005]. We modified WRF to output genic sources in our a priori emission estimates. Wetland
time-averaged mass fluxes (rather than instantaneous ad- sources are given by the ‘JK’ wetland inventory described
vective velocities) to drive STILT, which results in very by Bergamaschi et al. [2007], indexed on a 1=2 °  1=2 ° grid
good mass conservation (a critical consideration for surface and changing monthly. Anthropogenic sources are from the
flux estimates), and we also obtained WRF convective mass EDGAR 32FT2000 inventory [Olivier et al., 2005] (avail-
fluxes that are used directly in STILT. We employed 40-km able on a 1°  1° grid, constant in time). See Figure 1b for
resolution in WRF version 2.2 that includes analysis nudg- spatial distribution of emissions.
ing to improve meteorological realism. Effects of turbulence [10] For N2O, two different inventories are used, EDGAR
in the Planetary Boundary Layer are represented as a 32FT2000 and GEIA [Bouwman et al., 1995]. EDGAR
Markov chain process in STILT [Lin et al., 2003]. includes all anthropogenic sources of N2O, but does not
[8] Given input meteorological data, the STILT model include natural sources. GEIA includes both natural and
transports ensembles of 100 particles (air parcels) back- anthropogenic sources. Both N2O inventories are held
wards in time 6 days for the set of 300 receptor points; constant in time and are available on 1°  1° grids. See
advecting particles further back in time increases the frac- Figure 1c for spatial distribution of EDGAR emissions.
1
Auxiliary materials are available in the HTML. doi:10.1029/ [11] Emissions categories for GEIA include soils (natural
2008GL034031. and agricultural), animal excreta, post-forest clearing en-

2 of 5
L18808 KORT ET AL.: CH4 AND N2O EMISSIONS IN NORTH AMERICA L18808

Figure 2. (a) Model vs. measurement for CH4, solid line is 1:1 line, dashed line is fit. Grayed points are outliers not
included in fit (>3s from the mean). (b) Bin-averaged vertical distribution showing scaled model in blue, and measurements
in black, whiskers illustrate 95% confidence intervals. (c) Total mission footprint multiplied by CH4 emissions, showing
surface contribution to receptor points. (d) Surface contribution of N2O to receptor points.

hanced soil emissions, agricultural waste burning, biofuels, [13] Neither data nor global model results are available to
fossil fuels, industrial sources, and ocean emissions. ED- set a boundary condition for N2O. We plotted the observed
GAR categories for both CH4 and N2O include fossil fuels, N2O vs. CH4 above 3km, where surface influence is small,
biofuels, industrial sources, agricultural sources, and waste to determine the ratio of N2O variations vs. CH4 at the
handling. Seasonal factors for the different source categories boundary (0.010). We then multiplied the CH4 boundary
in EDGAR exist for testing purposes, but the majority of the values by this factor and added the relevant intercept, 299.8
sources considered here do not exhibit strong temporal nmol/mol. Using our prior fluxes, variation in background is
variation, and large uncertainty exists in these estimates, responsible for 57% of the variation for methane, and 50%
so they are not included. The exception is agricultural of the variation for nitrous oxide.
emissions of N2O, and seasonal factors are considered here
as discussed in section 4.2. 3.4. Assessment/Optimization Approach
[14] We fit model results to observations including un-
3.3. Upwind Boundary Condition certainty in both x and y, using the c2 (fitexy) method of
[12] To compare with measurements, the mixing ratio Press et al. [1992], to ensure unbiased fit parameters [Miller
enhancements due to surface fluxes must be added to a and Tans, 2003]. We assume that errors in the prior sources
background mixing ratio advected from the domain bound- are constant in space and time, and find one scaling factor
ary. We retrieve background mixing ratios from a two- (inverse of the fit slope) for the entire emissions field,
dimensional (latitude, height), temporally-varying array providing a simple and straightforward approach to assess
placed at 145°W longitude. Mixing ratios are attributed to model performance and determine the North American
particles as they cross this boundary. If particles do not budgets. We do not have sufficient data to decouple
reach the boundary, they are given a background mixing different source-types or to examine inventory errors in
ratio taken from the 145°W boundary at the same latitude smaller spatial regions. The intercept of the fit, representing
and altitude as the particle location 6 days traveled back in a correction to the boundary condition, has little or no effect
time. For CH4, TM3 ‘‘scenario one’’ model output as on the scaling factor. Further information on the effect of the
described by Houweling et al. [2006] is used for the bound- boundary, chi-square optimization, and uncertainties (in-
ary, shifted down by 27 nmol/mol. This bias is found by cluding footprint bias), is given in the auxiliary materials.
comparing model with COBRA-NA measurements above
3km where surface emissions are found to have minimal 4. Results and Discussion
influence on weekly timescales (on average less than 5 nmol/
4.1. Methane
mol, for methane). It is also consistent with the known offset
of 20 nmol/mol found by Houweling et al. [2006]. [15] Results for CH4 are remarkable in the point-by-point
agreement between simulated and measured mixing ratios

3 of 5
L18808 KORT ET AL.: CH4 AND N2O EMISSIONS IN NORTH AMERICA L18808

Figure 3. (a) Model vs. measurement for N2O, EDGAR case, solid line is 1:1 line, dashed line is fit. Grayed points are
outliers not included in fit (>3s from the mean) (b) GEIA case. (c) EDGAR case scaled by found multiplier (2.62). (d)
Binned vertical distribution showing un-scaled model (EDGAR case) in red, scaled model in blue, scaled model with upper
and lower bounds on scaling factor in gray, and measurements in black, whiskers illustrate 95% confidence intervals.

(Figures 2a and S1b), with a fitted slope remarkably close to similar, with both simulations underestimating atmospheric
one (0.924 ± 0.13), implying a scaling factor for the measurements. Fitting model vs. EDGAR yields a slope of
inventory of 1.08 ± 0.15 (note potential unaccounted 0.381 ± 0.072 and vs. GEIA a slope of 0.328 ± 0.065.
footprint bias may expand the uncertainty). Applying the Increasing emissions by factors of 2.62 ± 0.50 and 3.05 ±
scaling factor to the emissions gives the vertical distribution 0.61 for EDGAR and GEIA, respectively, improves model-
illustrated in Figure 2b; the very good agreement suggests measurement agreement in the shape of the vertical profile
that the WRF-driven STILT model is able to accurately (see Figures 3c and 3d for the EDGAR inventory). The
connect measurement locations with source regions up- offset above 3km is likely caused by the upwind boundary
stream, and thus to faithfully model the vertical distribution condition. Though the slope from the 1km to 3km bin
of CH4 in the atmosphere—a critical indicator of surface appears somewhat steeper in the measurement than the
emissions. The agreement above 3km indicates that the model, the upper error bound on the scaling, seen in gray,
advected boundary condition is sufficiently accurate for nearly captures the same trend. There is a potential that
our purposes. The fidelity of the model to the observations vertical mixing errors could bias the analysis, though the
also suggests accuracy in the spatial distribution of the prior methane results suggest otherwise. A more likely cause is
emissions inventory, though potentially cancelling errors or the advected boundary condition, though consistency be-
footprint bias cannot be ruled out (discussed further in the tween model and measurement, when considering error
auxiliary materials). Anthropogenic sources dominate the bounds, suggests the effect on our analysis is small. Spa-
inventory in the region that we sampled: fossil fuels, waste tially, we have coverage of the strong emissions regions
handling, and agriculture produce 97% of the total modeled through the Midwest and the Northeast (Figure 2d).
enhancement. Natural wetlands are hardly represented [17] These results indicate that in early summer current
(< 3% total modeled enhancement) in our measurements, inventories give N2O emissions too low by a factor of 2 – 3
and so our assessment of the inventory is restricted to for much of the USA and southern Canada. Since agricul-
anthropogenic emissions. Spatially, we are sensitive to the tural emissions dominate the source signal, seasonality may
high emissions regions in the northern Midwest and North- explain some of the under-representation. But applying the
east, but not to the large emissions from the gulf coast region published seasonal factors from EDGAR would actually
(Figure 2c). worsen agreement, as the scaling factor for May and June is
0.85. However, the peak scaling factor, for March, is 2.35. If
4.2. Nitrous Oxide agricultural emissions peaked in May – June instead of
[16] Results for N2O are shown in Figure 3. Simulations March, then our results would be consistent with the
based on the GEIA and EDGAR inventories are very EDGAR inventory scaled by this earlier seasonal peak,

4 of 5
L18808 KORT ET AL.: CH4 AND N2O EMISSIONS IN NORTH AMERICA L18808

and the overall annual emissions might not be higher by the Hall, B. D., G. S. Dutton, and J. W. Elkins (2007), The NOAA nitrous
oxide standard scale for atmospheric observations, J. Geophys. Res., 112,
indicated factor. D09305, doi:10.1029/2006JD007954.
Hirsch, A. I., A. M. Michalak, L. M. Bruhwiler, W. Peters, E. J. Dlugo-
kencky, and P. P. Tans (2006), Inverse modeling estimates of the global
5. Conclusion nitrous oxide surface flux from 1998 – 2001, Global Biogeochem. Cycles,
20, GB1008, doi:10.1029/2004GB002443.
[18] This study simultaneously estimated continental- Hofmann, D. J., J. H. Butler, E. J. Dlugokencky, J. W. Elkins, K. Masarie,
scale surface fluxes of two important greenhouse gases, S. A. Montzka, and P. Tans (2006), The role of carbon dioxide in climate
CH4 and N2O, using a top-down approach built upon the forcing from 1979 – 2004: Introduction of the annual greenhouse gas
STILT LPDM, constrained by collocated atmospheric meas- index, Tellus, Ser. B, 58, 614 – 619.
Houweling, S., T. Röckmann, I. Aben, F. Keppler, M. Krol, J. F. Meirink,
urements from aircraft and optimized with a simple regres- E. J. Dlugokencky, and C. Frankenberg (2006), Atmospheric constraints
sion approach. Results suggest the EDGAR inventory for on global emissions of methane from plants, Geophys. Res. Lett., 33,
CH4 accurately accounts for emissions over much of the L15821, doi:10.1029/2006GL026162.
Hurst, D. F., J. C. Lin, P. A. Romashkin, B. C. Daube, C. Gerbig, D. M.
U.S. and southern Canada (accurate to 8 ± 14%). In Matross, S. C. Wofsy, B. D. Hall, and J. W. Elkins (2006), Continuing
contrast, N2O emissions in both EDGAR and GEIA are global significance of emissions of Montreal Protocol – restricted halocar-
underestimated (by factors of 2.62 ± 0.50 and 3.05 ± 0.61, bons in the United States and Canada, J. Geophys. Res., 111, D15302,
doi:10.1029/2005JD006785.
respectively). Potential footprint bias may expand uncer- Lin, J. C., C. Gerbig, S. C. Wofsy, A. E. Andrews, B. C. Daube, K. J. Davis,
tainty but we believe this effect to be small (see auxiliary and C. A. Grainger (2003), A near-field tool for simulating the upstream
materials). Seasonality in agricultural emissions could be influence of atmospheric observations: The Stochastic Time-Inverted La-
responsible for the discrepancy. However, if seasonality grangian Transport (STILT) model, J. Geophys. Res., 108(D16), 4493,
doi:10.1029/2002JD003161.
does not explain the under-representation in future studies, Matross, D. M., et al. (2006), Estimating regional carbon exchange in
this underestimate would have important consequences for New England and Quebec by combining atmospheric, ground-based
predictions of climate change, given the strong radiative and satellite data, Tellus, Ser. B, 58, 344 – 358.
Miller, J. B., and P. P. Tans (2003), Calculating isotopic fractionation from
forcing by N2O. If these scaling factors applied uniformly atmospheric measurements at various scales, Tellus, Ser. B, 55, 207 – 214.
over the USA and Canada, then the yearly anthropogenic Miller, J. B., L. V. Gatti, M. T. S. d’Amelio, A. M. Crotwell, E. J. Dlugo-
contribution of these countries to global emissions would be kencky, P. Bakwin, P. Artaxo, and P. P. Tans (2007), Airborne measure-
ments indicate large methane emissions from the eastern Amazon basin,
49 Tg CH4 and 4.3 Tg N2O. Extending this approach with Geophys. Res. Lett., 34, L10809, doi:10.1029/2006GL029213.
measurements covering at least one full year would provide Nevison, C. D., N. M. Mahowald, R. F. Weiss, and R. G. Prinn (2007),
valuable information on the seasonality of emissions. The Interannual and seasonal variability in atmospheric N2O, Global Biogeo-
two main potential sources for bias in this approach, the chem. Cycles, 21, GB3017, doi:10.1029/2006GB002755.
Olivier, J. G. J., J. A. Van Aardenne, F. Dentener, L. Ganzeveld, and J. A.
upwind boundary condition and transport error, could be H. W. Peters (2005), Recent trends in global greenhouse gas emissions:
further quantified and reduced with measured vertical Regional trends and spatial distribution of key sources, in Non-CO2
profiles along the boundary, and analysis with gases such Greenhouse Gases (NCGG 4), edited by A. van Amstel, pp. 325 –
330, Millpress, Rotterdam, Netherlands.
as SF6, respectively. Press, W. H., S. A. Teukolsky, W. T. Vetterling, and B. P. Flannery (1992),
Numerical Recipes in FORTRAN: The Art of Scientific Computing, 2nd
[19] Acknowledgments. Work at AER has been supported by the ed., Cambridge Univ. Press, New York.
NASA Terrestrial Ecology Program. E.A.K. acknowledges financial support Prinn, R., D. Cunnold, R. Rasmussen, P. Simmonds, F. Alyea, A. Crawford,
from the Department of Defense through the NDSEG fellowship program. P. Fraser, and R. Rosen (1990), Atmospheric emissions and trends of
COBRA-2003 was funded by the NASA Large Scale Biosphere-Atmo- nitrous oxide deduced from 10 years of ALE-GAGE data, J. Geophys.
sphere Experiment in Amazonia (NASA NCC5-590), by NASA support to Res., 95(D11), 18,369 – 18,385.
J.B.M. and B.B.S. (NAG5-11430), and through the NOAA Office of Skamarock, W. C., J. B. Klemp, J. Dudhia, D. O. Gill, D. M. Barker,
Oceanic and Atmospheric Research. We also thank Ed Dlugokencky and W. Wang, and J. G. Powers (2005), A description of the advanced
Patricia Lang for analysis and quality control of all CH4 and N2O data. The research WRF version 2, Tech. Note 468+STR, 88 pp., Natl. Cent. for
National Center for Atmospheric Research is sponsored by the National Atmos. Res., Boulder, Colo.
Science Foundation. Stohl, A., C. Forster, S. Eckhardt, N. Spichtinger, H. Huntrieser, J. Heland,
H. Schlager, S. Wilhelm, F. Arnold, and O. Cooper (2003), A backward
modeling study of intercontinental pollution transport using aircraft mea-
References surements, J. Geophys. Res., 108(D12), 4370, doi:10.1029/
Bergamaschi, P., et al. (2007), Satellite chartography of atmospheric 2002JD002862.
methane from SCIAMACHY on board ENVISAT: 2. Evaluation based
on inverse model simulations, J. Geophys. Res., 112, D02304,
doi:10.1029/2006JD007268. B. C. Daube, Department of Earth and Planetary Sciences, Harvard
Bouwman, A. F., K. W. van der Hoek, and J. G. J. Olivier (1995), Un- University, 20 Oxford Street, Cambridge, MA 02138, USA.
certainties in the global source distribution of N2O, J. Geophys. Res., J. Eluszkiewicz and T. Nehrkorn, Atmospheric and Environmental
100(D2), 2785 – 2800. Research, Inc., 131 Hartwell Avenue, Lexington, MA 02421-3126, USA.
Chen, Y.-H., and R. G. Prinn (2006), Estimation of atmospheric methane C. Gerbig, Max-Planck-Insitut für Biogeochemie, P.O. Box 10 01 64,
emissions between 1996 and 2001 using a three-dimensional global che- D-07701 Jena, Germany.
mical transport model, J. Geophys. Res., 111, D10307, doi:10.1029/ S. Houweling, Netherlands Institute for Space Research, Sorbonnelaan 2,
2005JD006058. NL-3584 CA, Utrecht, Netherlands.
Dlugokencky, E. J., R. C. Myers, P. M. Lang, K. A. Masarie, A. M. J. O. Kaplan, Swiss Federal Institute for Forest, Snow, and Landscape
Crotwell, K. W. Thoning, B. D. Hall, J. W. Elkins, and L. P. Steele Research, Station 2, CH-1015 Lausanne, Switzerland.
(2005), Conversion of NOAA atmospheric dry air CH4 mole fractions E. A. Kort, School of Engineering and Applied Sciences, Harvard
to a gravimetrically prepared standard scale, J. Geophys. Res., 110, University, Geomuseum 411, 24 Oxford Street, Cambridge, MA 02138,
D18306, doi:10.1029/2005JD006035. USA. ([email protected])
Gerbig, C., J. C. Lin, S. C. Wofsy, B. C. Daube, A. E. Andrews, B. B. J. B. Miller, NOAA Earth System Research Laboratory, 325 Broadway,
Stephens, P. S. Bakwin, and C. A. Grainger (2003), Toward constraining Boulder, CO 80305, USA.
regional-scale fluxes of CO2 with atmospheric observations over a con- B. B. Stephens, National Center for Atmospheric Research, 1850 Table
tinent: 2. Analysis of COBRA data using a receptor-oriented framework, Mesa Drive, Boulder, CO 80305, USA.
J. Geophys. Res., 108(D24), 4757, doi:10.1029/2003JD003770. S. C. Wofsy, School of Engineering and Applied Sciences, Harvard
Gimson, N. R., and M. Uliasz (2003), The determination of agricultural University, Pierce Hall 110-D, 20 Oxford Street, Cambridge, MA 02138,
methane emissions in New Zealand using inverse modeling techniques, USA.
Atmos. Environ., 37, 3903 – 3912.

5 of 5