acp-16-14003-2016.pdf

Atmos. Chem. Phys., 16, 14003–14024, 2016

www.atmos-chem-phys.net/16/14003/2016/

doi:10.5194/acp-16-14003-2016

Seasonal variability of stratospheric methane: implications for

constraining tropospheric methane budgets using total column

observations

Katherine M. Saad

, Debra Wunch

1,2

, Nicholas M. Deutscher

3,4

, David W. T. Griffith

, Frank Hase

Martine De Mazière

, Justus Notholt

, David F. Pollard

, Coleen M. Roehl

, Matthias Schneider

, Ralf Sussmann

Thorsten Warneke

, and Paul O. Wennberg

Environmental Science and Engineering, California Institute of Technology, Pasadena, California, USA

Department of Physics, University of Toronto, Toronto, Ontario, Canada

Center for Atmospheric Chemistry, School of Chemistry, University of Wollongong, Wollongong, NSW, Australia

Institute of Environmental Physics, University of Bremen, Bremen, Germany

Institute for Meteorology and Climate Research, Karlsruhe Institute of Technology, IMK-ASF, Karlsruhe, Germany

Royal Belgian Institute for Space Aeronomy, Brussels, Belgium

National Institute of Water and Atmospheric Research, Omakau, New Zealand

Institute for Meteorology and Climate Research, Karlsruhe Institute of Technology, IMK-IFU,

Garmisch-Partenkirchen, Germany

Correspondence to:

Katherine M. Saad (katsaad@caltech.edu)

Received: 7 April 2016 – Published in Atmos. Chem. Phys. Discuss.: 10 May 2016

Revised: 24 October 2016 – Accepted: 25 October 2016 – Published: 11 November 2016

Abstract.

Global and regional methane budgets are

markedly uncertain. Conventionally, estimates of methane

sources are derived by bridging emissions inventories with

atmospheric observations employing chemical transport

models. The accuracy of this approach requires correctly

simulating advection and chemical loss such that modeled

methane concentrations scale with surface fluxes. When total

column measurements are assimilated into this framework,

modeled stratospheric methane introduces additional poten-

tial for error. To evaluate the impact of such errors, we com-

pare Total Carbon Column Observing Network (TCCON)

and GEOS-Chem total and tropospheric column-averaged

dry-air mole fractions of methane. We find that the model’s

stratospheric contribution to the total column is insensitive

to perturbations to the seasonality or distribution of

tropo-

spheric

emissions or loss. In the Northern Hemisphere, we

identify disagreement between the measured and modeled

stratospheric contribution, which increases as the tropopause

altitude decreases, and a temporal phase lag in the model’s

tropospheric seasonality driven by transport errors. Within

the context of GEOS-Chem, we find that the errors in tropo-

spheric advection partially compensate for the stratospheric

methane errors, masking inconsistencies between the mod-

eled and measured tropospheric methane. These seasonally

varying errors alias into source attributions resulting from

model inversions. In particular, we suggest that the tropo-

spheric phase lag error leads to large misdiagnoses of wet-

land emissions in the high latitudes of the Northern Hemi-

sphere.

1 Introduction

Identifying the processes that have driven changes in at-

mospheric methane (CH

), a potent radiative forcing agent

and major driver of tropospheric oxidant budgets, is criti-

cal for understanding future impacts on the climate system.

Methane’s growth rate, which had been decreasing through

the 1990s from about 10 to 0 ppb per year, began to in-

crease again in 2006 and over the past decade has averaged

5 ppb per year (Dlugokencky et al., 2011). Developing ro-

bust constraints on the global CH

budget is integral to un-

derstanding which processes produced these decadal trends

Published by Copernicus Publications on behalf of the European Geosciences Union.

14004

K. M. Saad et al.: Assimilation of total column methane into models

(e.g., Bergamaschi et al., 2013; Wecht et al., 2014a, b; Turner

et al., 2015).

One common approach to quantifying changes in the spa-

tial distribution of sources are atmospheric inversions, which

incorporate surface fluxes estimated by bottom-up invento-

ries as boundary conditions for a chemical transport model

(CTM). The modeled CH

concentrations are compared to

observations within associated grid boxes, and prior emis-

sions are scaled to minimize differences with measured dry-

air mole fractions (DMFs), producing posterior estimates.

The accuracy of these optimized emissions depends on how

well the CTM simulates atmospheric transport and CH

sinks, which are generally prescribed.

Pressure-weighted total column-averaged DMFs (

gas

)

provide a relatively new constraint and have previously been

shown to improve estimates of regional and interhemispheric

gradients in trace gases (Yang et al., 2007). Infrared spec-

trometers can measure CH

DMFs (

) from ground-

based sites, such as those in the Total Carbon Column Ob-

serving Network (TCCON) and Network for the Detection

of Atmospheric Composition Change (NDACC), and satel-

lites, including SCanning Imaging Absorption spectroMe-

ter for Atmospheric CartograpHY (SCIAMACHY) (Berga-

maschi et al., 2007), Greenhouse gases Observing SATel-

lite (GOSAT) (Parker et al., 2011), and the upcoming TRO-

POspheric Monitoring Instrument (TROPOMI) (Butz et al.,

2012). These observations complement surface measure-

ments because they add information about the vertically av-

eraged profile and are sensitive in the free troposphere (Yang

et al., 2007). Additionally, they complement aircraft obser-

vations by measuring trace gases at higher temporal fre-

quency, although they share the limitation of not measur-

ing in inclement weather. Satellite measurements add global

coverage that can fill in gaps where in situ observations are

sparse. Fraser et al. (2013) found that assimilating GOSAT

columns into the GEOS-Chem CTM with an ensem-

ble Kalman filter reduced posterior emissions uncertainties

by 9–48 % for individual source categories and by more than

three times those of inversions that only assimilated surface

data for most regions. Wecht et al. (2014b) determined from

their analysis of observing system simulation experiments

(OSSEs) that TROPOMI’s daily frequency and global cover-

age performs similarly to aircraft campaigns on sub-regional

scales, and could provide a constraint on California’s CH

emissions similar to CalNex aircraft observations (Santoni

et al., 2014; Gentner et al., 2014).

Incorporating total columns into modeling assessments

can also be used to diagnose systematic issues with model

transport. For example, comparing carbon dioxide (CO

)

from TCCON and TransCom (Baker et al., 2006), Yang et al.

(2007) found that most models included in the comparison

lack sufficiently strong vertical exchange between the plane-

tary boundary layer (PBL) and the free troposphere, thereby

dampening the seasonal cycle amplitude of

. The lim-

itations of models to accurately represent vertical transport

can lead to radically different spatial distributions of fluxes;

Stephens et al. (2007) found, for example, that the north-

ern terrestrial carbon land sink and tropical emissions were

overestimated by 0.9 and 1.7 PgC year

−

, respectively, when

comparing models to aircraft CO

profiles. More recent stud-

ies attribute to model transport errors the tendency of simu-

lated CH

in the Southern Hemisphere to be higher at the

surface than the free troposphere, in contrast with measure-

ments (Fraser et al., 2011; Patra et al., 2011).

Tropospheric CH

typically does not vary radically with

height above the PBL; above the tropopause, however, the

vertical profile of CH

exhibits a rapid decline with altitude

as a result of its oxidation and the lack of any source beyond

advection from the troposphere. Fluctuations in stratospheric

dynamics, including the height of the tropopause, change the

contribution of the stratosphere to the total column. CH

pro-

files with similar tropospheric values can thus have signif-

icant differences in

(Saad et al., 2014; Washenfelder

et al., 2003; Wang et al., 2014).

Provided that simulations replicate seasonal and zonal

variability of stratospheric CH

loss, tropopause heights, and

vertical exchange across the upper troposphere and lower

stratosphere (UTLS), posterior flux estimates from inver-

sions incorporating

measurements would not be sen-

sitive to stratospheric processes. However, most models do

not accurately represent stratospheric transport, producing

low age-of-air values and zonal gradients in the subtropi-

cal lower stratosphere that are less steep than observations

(Waugh and Hall, 2002). The TransCom-CH

CTM inter-

comparison assessment of transport using sulfur hexafluoride

(SF

) showed a strong correlation between the stratosphere–

troposphere exchange (STE) rate and the model’s CH

bud-

get, and a weaker correlation between the CH

growth rate

and vertical gradient in the model’s equatorial lower strato-

sphere (Patra et al., 2011). These forward model dependen-

cies of CH

concentrations on vertical transport, both within

the troposphere and across the tropopause, have the potential

to introduce substantial errors in atmospheric inversions. As

temporal and spatial biases in a model’s vertical profile will

alias into posterior emissions, inversions that incorporate to-

tal column measurements must ensure that the stratosphere

is sufficiently well described so as to not introduce spurious

seasonal, zonal, and interhemispheric trends in CH

concen-

trations and consequently emissions.

In this analysis, we identify systematic model errors in

the seasonal cycle and spatial distribution of CH

DMFs by

comparing TCCON total and tropospheric columns (Saad

et al., 2014) to vertically integrated profiles derived from

the GEOS-Chem CTM (Bey et al., 2001; Wang et al., 2004;

Wecht et al., 2014a). We assess the impact of errors in the

characterization of stratospheric processes on the assimila-

tion of

and resulting posterior emissions estimates. In

Sect. 2 we describe the TCCON column measurements and

GEOS-Chem setup and characteristics. In Sect. 3 we present

the results of the measurement–model comparison. In Sect. 4

Atmos. Chem. Phys., 16, 14003–14024, 2016

www.atmos-chem-phys.net/16/14003/2016/

K. M. Saad et al.: Assimilation of total column methane into models

14005

Table 1.

TCCON sites, coordinates, altitudes, start date of measurements and locations used in this analysis.

Site

Latitude

Longitude

Elevation

Start date

Location

Data reference

(

◦

)

(

◦

)

(km)

Bialystok

53.2

23.0

0.18

Mar 2009

Bialystok, Poland

Deutscher et al. (2014)

Bremen

53.1

8.9

0.03

Jan 2007

Bremen, Germany

Notholt et al. (2014)

Karlsruhe

49.1

8.4

0.11

Apr 2010

Karlsruhe, Germany

Hase et al. (2014)

Orleans

48.0

2.1

0.13

Aug 2009

Orleans, France

Warneke et al. (2014)

Garmisch

47.5

11.1

0.75

Jul 2007

Garmisch, Germany

Sussmann and Rettinger (2014)

Park Falls

45.9

−

90.3

0.47

Jan 2005

Park Falls, WI, USA

Wennberg et al. (2014b)

Lamont

36.6

−

97.5

0.32

Jul 2008

Lamont, OK, USA

Wennberg et al. (2014c)

JPL

34.2

−

118.2

0.39

Jul 2007

Pasadena, CA, USA

Wennberg et al. (2014d, a)

Saga

33.2

130.3

0.01

Jul 2011

Saga, Japan

Kawakami et al. (2014)

Izaña

28.3

−

16.5

2.37

May 2007

Tenerife, Canary Islands

Blumenstock et al. (2014)

Darwin

−

12.4

130.9

0.03

Aug 2005

Darwin, Australia

Griffith et al. (2014a)

Réunion Island

−

20.9

55.5

0.09

Sep 2011

Saint-Denis, Réunion

De Mazière et al. (2014)

Wollongong

−

34.4

150.9

0.03

Jun 2008

Wollongong, Australia

Griffith et al. (2014b)

Lauder

−

45.0

169.7

0.37

Jan 2005

Lauder, New Zealand

Sherlock et al. (2014a, b)

we compare the base case simulation to one in which emis-

sions do not vary within each year and quantify the sensitivity

of source attribution of the biggest seasonal emissions sector,

wetlands, to the tropospheric seasonal delay.

2 Methods

2.1 Tropospheric methane columns

TCCON has provided precise measurements of

and

other atmospheric trace gases for over ten years (Wunch

et al., 2011a, 2015). Developed to address open questions

in carbon cycle science, the earliest sites are located in Park

Falls, Wisconsin, United States and Lauder, New Zealand

at about 45

◦

north and south, respectively. Since 2004, the

ground-based network of Fourier transform spectrometers

has expanded greatly.

are processed with the current

version of the TCCON software, GGG2014, to be consistent,

and thereby comparable, across sites. Total column retrievals

are generated with the GFIT nonlinear least-squares fitting

algorithm, which calculates the best spectral fit of the solar

absorption signal to an a priori vertical profile and outputs

a scaling factor. The pressure-weighted integration of the

scaled a priori profile produces column abundances, which

are then divided by the dry air column, calculated using con-

currently retrieved oxygen (O

) columns (Wunch et al., 2010,

2011a, 2015). Trace gas a priori profiles are derived with

empirical models, which are generated incorporating aircraft

and balloon in situ and satellite measurements (see Wunch

et al., 2015, for a complete list), and for CH

include a sec-

ular increase of 0.3 % per year and an interhemispheric gra-

dient in the altitude dependence of the vertical profiles (Toon

and Wunch, 2014). These models are fit to daily noontime

National Centers for Environmental Protection and National

Center for Atmospheric Research (NCEP/NCAR) reanalysis

pressure grids (Kalnay et al., 1996), interpolated to the sur-

face pressure measured real-time on site. Because the profile

of CH

drops off rapidly in the stratosphere, the accuracy of

the a priori shape, and thus the retrieved column, depends on

correctly determining the tropopause.

Tropospheric columns have been shown to represent the

magnitude and seasonality of in situ measurements (Saad

et al., 2014; Washenfelder et al., 2003; Wang et al., 2014).

The tropospheric CH

column-averaged DMFs (

) are

derived by the hydrogen fluoride (HF) proxy method de-

scribed in Saad et al. (2014), which uses the relationship be-

tween CH

and HF in the stratosphere, derived from ACE-

FTS satellite measurements (Bernath, 2005; De Mazière

et al., 2008; Mahieu et al., 2008; Waymark et al., 2014), to

calculate and remove the stratospheric contribution to

The

used in this analysis have been processed consis-

tently with the GGG2014 TCCON products, with air-mass

dependence and calibration factors calculated for and ap-

plied to

(Wunch et al., 2010, 2015). Additional details

about the tropospheric CH

measurements can be found in

Appendix A.

With the exception of Eureka and Sodankylä, which are

highly influenced by the stratospheric polar vortex, all TC-

CON sites that provide measurements before December 2011

are included in this analysis (Fig. 1). Table 1 lists locations

and data collection start dates for each of the sites.

2.2 GEOS-Chem model

Model comparisons use the offline CH

GEOS-Chem ver-

sion 9.02 at 4

◦

horizontal resolution on a reduced

vertical grid (47L). CH

loss is calculated on 60 min in-

tervals and is set by annually invariable monthly 3-D

fields: hydroxyl radical (OH) concentrations in the tropo-

sphere (Park et al., 2004) and parameterized CH

loss rates

www.atmos-chem-phys.net/16/14003/2016/

Atmos. Chem. Phys., 16, 14003–14024, 2016

14006

K. M. Saad et al.: Assimilation of total column methane into models

Figure 1.

Map of TCCON sites used in this analysis. Site colors are

on a spectral color scale in order of latitude, with Northern Hemi-

sphere sites designated by cool colors and Southern Hemisphere

sites designated by warm colors.

per unit volume in the stratosphere (Considine et al., 2008;

Allen et al., 2010; Murray et al., 2012). Emissions are re-

leased at 60 min time steps and are provided by the GEOS-

Chem development team for 10 sectors: (i) gas and oil, (ii)

coal, (iii) livestock, (iv) waste, (v) biofuel and (vi) other an-

thropogenic annual emissions from EDGAR v4.2 (European

Commission Joint Research Centre, Netherlands Environ-

mental Assessment Agency, 2011; Wecht et al., 2014a), (vii)

other natural annual emissions from Fung et al. (1991), (viii)

rice agriculture (European Commission Joint Research Cen-

tre, Netherlands Environmental Assessment Agency, 2011)

and (ix) wetland (Pickett-Heaps et al., 2011) monthly emis-

sions, which incorporate GEOS5 annual and monthly mean

soil moisture values, and (x) biomass burning daily emission

from GFED3 estimates (Mu et al., 2011; van der Werf et al.,

2010). Loss via soil absorption (Fung et al., 1991), set annu-

ally, is subtracted from the total emissions at each time step.

2.2.1 Model setup

We initialized zonal CH

distributions with GGG2014 data

version a priori profiles (Toon and Wunch, 2014) pro-

duced at horizontal grid centers, which we adjusted ver-

tically to match the zonally averaged daily mean model’s

tropopause, derived from the National Aeronautics and

Space Administration Global Modeling and Assimilation

Office (NASA/GMAO) Goddard Earth Observing System

Model, Version 5 (GEOS5). The model was run from De-

cember 2003, the first month in which GEOS5 meteoro-

logical data were available, to June 2004, the beginning of

the TCCON time series; we then ran the model repeatedly

over the June 2004–May 2005 time frame, which allowed

us to make comparisons with the TCCON data at Park Falls

and Lauder, until CH

concentrations reached equilibrium. A

number of perturbation experiments were run in this way to

quantify the sensitivity of CH

distribution and seasonality to

the offline OH fields, prescribed emissions, and tropopause

levels (Table 2). These model experiments are described in

greater detail in Appendix B1.

∆

(ppb)

-15

-10

-5

Tropospheric column

∆

(ppb)

-15

-10

-5

Total column

∆

(ppb)

-15

-10

-5

Stratospheric contribution

Figure 2.

Seasonality of the difference between base and aseasonal

for tropospheric, total and stratospheric contribution to total

columns. Site colors are as in Fig. 1.

Using CH

fields for 1 January 2005 from the equilibrium

simulation as initial conditions, model daily mean CH

mole

fractions were computed through 2011. These were con-

verted to dry mole fractions, as described in Appendix B2. In

addition to the default emissions scheme, an aseasonal sim-

ulation setup, in which rice, wetland, and biomass burning

emissions were disabled and aseasonal emissions scaled up

such that total annual zonal fluxes approximate those in the

base simulation, was similarly run to equilibrium and used

as initial conditions for the 2005–2011 run. The model in-

frastructure posed difficulties for setting the seasonally vary-

ing fluxes constant throughout each year; thus we implement

this scaling technique as an alternative to assess first-order

impacts of emission seasonality. The resulting changes to the

spatial distribution of CH

emissions are shown in Fig. B1.

For comparisons with column measurements, model ver-

tical profiles were smoothed with corresponding TCCON

averaging kernels, interpolated for the daily mean so-

lar zenith angles, and prior profiles, scaled with daily median

scaling factors, following the methodology in Rodgers and

Connor (2003) and Wunch et al. (2010). Averaging kernels

and prior profiles were interpolated to the model’s pressure

grid, and all terms in the smoothing equation were interpo-

lated to daily mean surface pressures measured at each site.

Tropospheric columns were integrated in the same manner

as the total columns up to the grid level completely below

Atmos. Chem. Phys., 16, 14003–14024, 2016

www.atmos-chem-phys.net/16/14003/2016/

K. M. Saad et al.: Assimilation of total column methane into models

14007

Table 2.

Sensitivity experiments.

Simulation name

Description

lifetime

Final CH

burden

(years)

(Tg)

Base

Default OH and emissions

9.55

4825

Aseasonal

Constant monthly emission rates

9.57

4872

Updated OH

Monthly OH fields from standard chemistry

biogenic VOCs

8.53

4828

Figure 3.

Smoothed daily mean

and stratospheric contribution to

at Park Falls (blue) and Lauder (red) for

(a)

base equilibrium

simulation and the difference between the base and

(b)

aseasonal and

(c)

updated OH simulations.

the daily mean tropopause, consistent with how GEOS-Chem

partitions the atmosphere in the offline CH

simulation. To

test the dependence of our results on the chosen vertical inte-

gration level, tropospheric columns were also calculated as-

suming the tropopause was one and two grid cells above this

level. While

changed slightly, by a median of about

1 and 5 ppb for a one and two-level increase respectively,

shifting the tropopause did not alter the findings discussed in

this paper. A description of the model smoothing methodol-

ogy and assumptions is provided in Appendix B3. The strato-

spheric contribution to the total column, which is calculated

as the residual between the

and

, is the amount by

which the stratosphere attenuates

via stratospheric loss

and transport (see Appendix C for the derivation).

2.2.2 Model features

The seasonal amplitude of the differences between base and

aseasonal simulations are small – within

4 ppb – for all ver-

tical levels in the Southern Hemisphere (Fig. 2). In the North-

ern Hemisphere, however, the difference is much larger and

primarily impacts the troposphere, where it varies between

−

10 and

13 ppb. The insensitivity of the stratosphere to

the seasonality of emissions is due to the common source

of stratospheric air in the tropics (Boering et al., 1995) and

the loss of seasonal information as the age of air increases

(Mote et al., 1996).

Due to the relatively short photochemical lifetime of CH

in the stratosphere, about 22 months in the base simula-

tion, stratospheric CH

concentrations stabilize much more

quickly than in the troposphere (Fig. 3a). This rapid re-

sponse time of the stratosphere occurs regardless of perturba-

tions to the troposphere, such as the seasonality of emissions

(Fig. 3b) or tropospheric OH fields (Fig. 3c). In both hemi-

spheres the differences between the base and experimental

simulations asymptotically approach steady state with sea-

sonal variability over a decade in the troposphere, but oscil-

late seasonally around a constant mean in the stratosphere.

Stratospheric differences between simulations are consider-

ably smaller than the seasonal amplitude of the base run:

within 6 and 1 ppb, respectively, vs. a seasonal range of

30 ppb at Park Falls. By contrast,

have differences

within 30 and 10 ppb, respectively, vs. a seasonal range of

20 ppb at Park Falls. The stratosphere at Lauder is even less

sensitive to tropospheric perturbations.

www.atmos-chem-phys.net/16/14003/2016/

Atmos. Chem. Phys., 16, 14003–14024, 2016

14008

K. M. Saad et al.: Assimilation of total column methane into models

Figure 4.

Daily median TCCON and smoothed daily mean GEOS-Chem base (top) and aseasonal (bottom) DMFs for

(a)

(b)

and

(c)

stratospheric contribution. Site colors are as in Fig. 1. Northern Hemisphere least squares regression equations are in the top left, and

Southern Hemisphere least squares regression equations are in the bottom right of each plot. Dashed lines mark the one-to-one lines.

3 Measurement–model comparison

The TCCON daily median and GEOS-Chem daily mean

column-averaged DMFs demonstrate a strong inter-

hemispheric difference for

and

in both the base

and aseasonal simulations (Fig. 4). The Northern Hemi-

sphere

slope deviates from the one-to-one line more

than the

slope (0

02 vs. 0

03), and the

correlation coefficients are equivalent (

41), which

indicates that the poorer agreement between measurements

and models in the troposphere drives the scatter in the total

column.

The stratospheric contribution comparison between TC-

CON and the base simulation for the Northern Hemisphere

sites has an equivalent slope (0

1) and higher corre-

lation coefficient (

68) compared to

(Fig. 4c).

GEOS-Chem’s larger stratospheric contribution to the total

column, coupled with lower tropospheric values, depresses

. Because this effect on

occurs more at higher

latitudes, zonal errors in the model’s stratosphere balances

those in the troposphere. The result is better measurement–

model agreement in the total columns.

The aseasonal simulation produces lower slopes and cor-

relation coefficients for,

(slope

0.42

0.02,

32),

(slope

03,

26), and the

stratospheric contribution (slope

01,

66)

in the Northern Hemisphere. Removing the seasonality of

emissions increases both measurement–model differences

and scatter, as we would expect given the seasonality of

Northern Hemisphere emissions noted in bottom-up stud-

ies (Kirschke et al., 2013). The aseasonal simulation also

reduces the offset between TCCON and GEOS-Chem,

whereby modeled

and

are systematically low.

TransCom-CH

showed that GEOS-Chem CH

concentra-

tions tend to be lower than the model median, and much

lower than the range of other models when using the same

OH fields (Patra et al., 2011). The aseasonal emissions used

in this analysis likely reduce this documented imbalance with

the model’s tropospheric OH fields.

The

and

regression equations across Southern

Hemisphere sites are nearly equivalent, which suggests that

the Southern Hemisphere is not as impacted by the STE er-

rors as the Northern Hemisphere. This consistency between

and

could also be a function of the zonal depen-

dence of the stratospheric error: whereas more than half of

the Northern Hemisphere sites are north of 45

◦

N, the most

poleward site in the Southern Hemisphere is located at 45

◦

The increased scatter associated with the slightly lower

value of 0.63, compared to the

value of 0.88,

does indicate that the Southern Hemisphere is not exempt

from model errors associated with emissions, the OH distri-

bution, or transport. The lower

slope of the aseasonal

simulation (1.1 vs. 1.3) illustrates the influence of emissions:

removing their seasonality leads to better measurement–

model agreement, evidenced by a slope closer to both the

one-to-one line and the zero-intercept. We hypothesize that

either the seasonality of Southern Hemispheric emissions

is too strong or, more likely, errors in the Northern Hemi-

spheric seasonality of emissions drive measurement–model

mismatch in the Southern Hemisphere via interhemispheric

Atmos. Chem. Phys., 16, 14003–14024, 2016

www.atmos-chem-phys.net/16/14003/2016/

K. M. Saad et al.: Assimilation of total column methane into models

14009

Latitude

-60

-40

-20

Pressure (hPa)

100

150

200

250

300

350

400

450

Spring

Latitude

-60

-40

-20

100

150

200

250

300

350

400

450

Fall

-150

-100

-50

100

150

∆

(ppb)

Figure 5.

Zonally averaged ACE minus GEOS-Chem climatological CH

mole fractions for boreal spring and fall. Black line represents the

mean zonal tropopause level. Site colors of squares on the

axis are as in Fig. 1.

transport. If this effect was solely due to a changed emis-

sions distribution, we would expect the

slope to also

change for the Southern Hemisphere sites, if only slightly;

instead the slope is equivalent to the base simulation

and

slopes, and

87, only marginally less than

the base simulation

correlation coefficient.

The stratospheric contribution regression equations dif-

fer only slightly between the base and aseasonal simula-

tions:

(

14,

68, vs.

(

15,

67. The insensitivity of both the stratospheric

contribution and the total columns in the Southern Hemi-

sphere to perturbations in the seasonality of tropospheric

emissions could be driven by the smaller vertical gradient

across the UTLS that results from the influence of Northern

Hemispheric air both in the free troposphere (Fraser et al.,

2011) and the stratosphere (Boering et al., 1995). This ef-

fect would also support the interpretation of Northern Hemi-

spheric emission errors driving disagreement between obser-

vations and the model in the Southern Hemisphere.

In the troposphere, CH

increases from south to north;

the stratospheric contribution of CH

, however, increases

from the Equator to the poles due to the zonal gradient in

tropopause height. In the Northern Hemisphere total col-

umn, the zonal gradient largely disappears: at high lati-

tudes, the larger tropospheric emissions balances the larger

stratospheric contribution. By contrast, zonal gradients in the

Southern Hemisphere troposphere and stratosphere are addi-

tive, and greater south to north differences are apparent in the

total column.

Figure 5 illustrates how the model differs from ACE-

FTS CH

measurements in the stratosphere over boreal

spring (March–April–May) and fall (September–October–

November). Except above the tropical tropopause, CH

considerably lower in the ACE-FTS climatology (v. 2.2,

Jones et al., 2012) compared to GEOS-Chem. The difference

varies both with altitude and latitude, especially in the North-

ern spring poleward of 40

◦

N. The vertical gradient is the

least pronounced in Lauder, where the stratospheric contri-

butions of TCCON and GEOS-Chem fall most closely to the

one-to-one line (Fig. 4). The low CH

in the tropical mid and

upper stratosphere in GEOS-Chem could be a result of too-

weak vertical ascent to the stratosphere; however, the ACE-

FTS data gaps in the tropical troposphere make this hypoth-

esis difficult to test.

3.1 Dependence on tropopause height

In the Northern Hemisphere, the measurement–model mis-

match of the stratospheric contribution increases as the

tropopause altitude shifts downward (Fig. 6). As the model’s

stratospheric portion of the pressure-weighted total column

increases, the error in stratospheric CH

is amplified, caus-

ing a larger disagreement with measurements. Because the

tropopause height decreases with latitude, and this gradi-

ent increases during winter and spring, this introduces both

zonal and seasonal biases. The disagreement exhibits a large

spread for relatively few tropopause pressure heights because

the model’s effective tropopause, that is, the pressure level

at which the model divides the troposphere from the strato-

sphere in GEOS-Chem, is defined at discrete grid level pres-

sure boundaries.

The tropospheric mismatch (

), by contrast, de-

creases with tropopause height for the majority of days and

exhibits a much weaker correlation to tropopause height,

099 vs. 0

22 for the stratospheric contribution. Thus, as ex-

pected, the tropopause height explains less of the variance in

the measurement–model mismatch in

: the upper tro-

posphere is generally well-mixed, and chemical loss does

not vary with altitude as much as in the lower stratosphere.

This weaker relationship also demonstrates that the choice of

tropopause used in the tropospheric profile integration does

not strongly impact

www.atmos-chem-phys.net/16/14003/2016/

Atmos. Chem. Phys., 16, 14003–14024, 2016