Global Budget of Methanol: Constraints from Atmospheric Observations

Abstract

We use a global three-dimensional model simulation of atmospheric methanol to examine the consistency between observed atmospheric concentrations and current understanding of sources and sinks. Global sources in the model include 128 Tg yr−1 from plant growth, 38 Tg yr−1 from atmospheric reactions of CH3O2 with itself and other organic peroxy radicals, 23 Tg yr−1 from plant decay, 13 Tg yr−1 from biomass burning and biofuels, and 4 Tg yr−1 from vehicles and industry. The plant growth source is a factor of 3 higher for young than from mature leaves. The atmospheric lifetime of methanol in the model is 7 days; gas-phase oxidation by OH accounts for 63% of the global sink, dry deposition to land 26%, wet deposition 6%, uptake by the ocean 5%, and aqueous-phase oxidation in clouds less than 1%. The resulting simulation of atmospheric concentrations is generally unbiased in the Northern Hemisphere and reproduces the observed correlations of methanol with acetone, HCN, and CO in Asian outflow. Accounting for decreasing emission from leaves as they age is necessary to reproduce the observed seasonal variation of methanol concentrations at northern midlatitudes. The main model discrepancy is over the South Pacific, where simulated concentrations are a factor of 2 too low. Atmospheric production from the CH3O2 self-reaction is the dominant model source in this region. A factor of 2 increase in this source (to 50–100 Tg yr−1) would largely correct the discrepancy and appears consistent with independent constraints on CH3O2 concentrations. Our resulting best estimate of the global source of methanol is 240 Tg yr−1. More observations of methanol concentrations and fluxes are needed over tropical continents. Better knowledge is needed of CH3O2 concentrations in the remote troposphere and of the underlying organic chemistry. Earth and Planetary Sciences Version of Record

Full text

Page 1

Globalbudgetofmethanol:Constraintsfromatmosphericobservations
Daniel J. Jacob,1Brendan D. Field,1Qinbin Li,1,2Donald R. Blake,3Joost de Gouw,4
Carsten Warneke,4Armin Hansel,5Armin Wisthaler,5Hanwant B. Singh,6
and A. Guenther7
Received 28 June 2004; revised 5 January 2005; accepted 2 February 2005; published 26 April 2005.
[1] We use a global three-dimensional model simulation of atmospheric methanol to
examine the consistency between observed atmospheric concentrations and current
understanding of sources and sinks. Global sources in the model include 128 Tg yr?1from
plant growth, 38 Tg yr?1from atmospheric reactions of CH3O2with itself and other
organic peroxy radicals, 23 Tg yr?1from plant decay, 13 Tg yr?1from biomass burning
and biofuels, and 4 Tg yr?1from vehicles and industry. The plant growth source is a
factor of 3 higher for young than from mature leaves. The atmospheric lifetime of
methanol in the model is 7 days; gas-phase oxidation by OH accounts for 63% of the
global sink, dry deposition to land 26%, wet deposition 6%, uptake by the ocean 5%, and
aqueous-phase oxidation in clouds less than 1%. The resulting simulation of atmospheric
concentrations is generally unbiased in the Northern Hemisphere and reproduces the
observed correlations of methanol with acetone, HCN, and CO in Asian outflow.
Accounting for decreasing emission from leaves as they age is necessary to reproduce
the observed seasonal variation of methanol concentrations at northern midlatitudes.
The main model discrepancy is over the South Pacific, where simulated concentrations are
a factor of 2 too low. Atmospheric production from the CH3O2self-reaction is the
dominant model source in this region. A factor of 2 increase in this source (to 50–100 Tg
yr?1) would largely correct the discrepancy and appears consistent with independent
constraints on CH3O2concentrations. Our resulting best estimate of the global source of
methanol is 240 Tg yr?1. More observations of methanol concentrations and fluxes are
needed over tropical continents. Better knowledge is needed of CH3O2concentrations in
the remote troposphere and of the underlying organic chemistry.
Citation: Jacob, D. J., B. D. Field, Q. Li, D. R. Blake, J. de Gouw, C. Warneke, A. Hansel, A. Wisthaler, H. B. Singh, and
A. Guenther (2005), Global budget of methanol: Constraints from atmospheric observations, J. Geophys. Res., 110, D08303,
doi:10.1029/2004JD005172.
1.Introduction
[2] Methanol is the second most abundant organic gas in
the atmosphere after methane. It is present at typical con-
centrations of 1–10 ppbv in the continental boundary layer
and 0.1–1 ppbv in the remote troposphere [Singh et al.,
1995; Heikes et al., 2002]. It is a significant atmospheric
source of formaldehyde [Riemer et al., 1998; Palmer et al.,
2003a] and CO (B. N. Duncan et al., Global model study of
the interannual variability and trends of carbon monoxide
(1988–1997): 1. Model formulation, evaluation, and sensi-
tivity, submitted to Journal of Geophysical Research, 2004,
hereinafter referred to as Duncan et al., submitted manu-
script, 2004), as well as a minor term in the carbon cycle
[Heikes et al., 2002] and in the global budgets of tropo-
spheric ozone and OH [Tie et al., 2003]. Most of the
observations of atmospheric methanol concentrations con-
sist of short-term records in surface air [Heikes et al., 2002].
Recent aircraft missions have added a new dimension to our
knowledge of methanol concentrations in the global tropo-
sphere [Singh et al., 2000, 2001, 2003a, 2004; Lelieveld et
al., 2002]. We use here a global 3-D chemical transport
model (CTM) to examine the constraints that these aircraft
observations provide on current understanding of methanol
sources and sinks.
[3] Global budgets of atmospheric methanol have been
presented previously by Singh et al. [2000], Galbally and
Kristine [2002], Heikes et al. [2002], Tie et al. [2003], and
von Kuhlmann et al. [2003a, 2003b]. They are summarized
in Table 1. Plant growth is the principal source. Additional
JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 110, D08303, doi:10.1029/2004JD005172, 2005
1Division of Engineering and Applied Science, Harvard University,
Cambridge, Massachusetts, USA.
2Now at Jet Propulsion Laboratory, Pasadena, California, USA.
3Department of Chemistry, University of California, Irvine, California,
USA.
4NOAA Aeronomy Laboratory, Boulder, Colorado, USA.
5Institute of Ion Physics, University of Innsbruck, Innsbruck, Austria.
6NASA Ames Research Center, Moffett Field, California, USA.
7Atmospheric Chemistry Division, National Center for Atmospheric
Research, Boulder, Colorado, USA.
Copyright 2005 by the American Geophysical Union.
0148-0227/05/2004JD005172$09.00
D08303
1 of 17

Page 2

sources include plant decay, biomass burning, atmospheric
oxidation of methane and other volatile organic com-
pounds (VOCs), vehicles, and industrial activities. Chemi-
cal loss by oxidation by OH results in an atmospheric
lifetime for methanol of about 10 days. Deposition,
exchange with the ocean, and heterogeneous reactions in
aerosols and clouds are additional sinks but are poorly
constrained.
[4] Large discrepancies are apparent between the different
budgets of Table 1. Singh et al. [2000] found a factor of
three imbalance between their independently estimated
sources and sinks, implying a missing sink. Heikes et al.
[2002] achieved closure between sources and sinks in their
budget but with large uncertainties. Their estimated plant
growth source is four times larger than that of Singh et al.
[2000] and is partly compensated by an assumed large
deposition sink. Galbally and Kirstine [2002] enforced
closure by simulating methanol in a six-box global model
of the atmosphere-ocean system. Tie et al. [2003] conducted
global 3-D model simulations of methanol with two esti-
mates for the plant growth source, a high value of 312 Tg
yr?1from their bottom-up source inventory and a low value
reducedbyafactorof3tomatchthatofGalballyandKirstine
[2002]. They presented order-of-magnitude comparisons to
observationsbutdidnotconcludeastotheimpliedconstraints
on the methanol budget. Von Kuhlmann et al. [2003b]
evaluated a global 3-D model simulation of methanol with
observations from two aircraft campaigns (SONEX and
PEM-Tropics B). We present here a more detailed evaluation
with a large ensemble of measurements from aircraft and
ships (Figure 1).
2.
2.1. General Description
[5] We conducted a 1-year simulation of atmospheric
methanol for 2001 with the GEOS-CHEM CTM v4.32
(http://www-as.harvard.edu/chemistry/trop/geos) [Bey et
al., 2001]. The model is driven by GEOS-3 assimilated
meteorological observations (including convective mass
fluxes) from the NASA Global Modeling and Assimilation
Office. The GEOS-3 data have a temporal resolution of
6 hours (3 hours for mixing depths and surface quantities),
a horizontal resolution of 1? ? 1?, and 48 layers in the
vertical. We average them here over a 4? latitude ?
5? longitude grid for input to GEOS-CHEM.
[6] The simulation is conducted for 18 months starting
from low concentrations in July 2000. The first six months
are used for initialization and we focus on the 1-year results
for 2001. The methanol budget in the model is given in
Table 1 and details are given below. Methanol originating
from each source in Table 1 is transported as a separate
species, for a total of 5 species transported in the model. The
model losses of methanol are linear, so that the sum of
transported species adds up to the total methanol concen-
tration within a few percent (nonlinearity in the semi-
Lagrangian advection algorithm prevents perfect closure).
We use the model results to compare to observations not just
Model Description
Table 1. Global Atmospheric Budgets of Methanol Reported in the Literature
Reference
Singh et al.
[2000]a
Heikes et al.
[2002]a
Galbally and Kirstine
[2002]a
Sources (Tg yr?1)
100 (37–212)
13 (5–31)
13 (6–19)
4 (3–5)
19 (14–24)
149 (83–260)
Tie et al.
[2003]b
von Kuhlmann et al.
[2003a, 2003b]This Workc
Plant growth
Plant decay
Biomass burningd
Urbane
Atmospheric productionf
Total source
75 (50–125)
20 (10–40)
6 (3–17)
3 (2–4)
18 (12–24)
122 (75–210)
280 (50–>280)
20 (10–40)
12 (2–32)
8 (5–11)
30 (18–30)
350 (90–490)
104–312 77 128 (100–160)
23 (5–40)
13 (10–20)
4 (1–10)
38g(50–100)
206h(170–330)
15
2
28
123
31
135–343
Sinks (Tg yr?1)
109 (60–203)
5 (2–15)
24 (11–43)
0.3 (0.2–0.6)
11 (5–20)
149 (82–273)
3.4
8
Gas-phase oxidation by OH
In-cloud oxidation by OH(aq)
Dry deposition (land)
Ocean uptakek
Wet deposition
Total sinks
Atmospheric inventory (Tg)
Atmospheric lifetime (days)
100 (25–150)
10 (5–20)
70 (35–210)
50 (?20–150)
10 (4–36)
270 (160–570)
3.9
9 (5)m
59–14977i
129j
<1
55
10
12
206
4.0
7 (5–10)
32–85 37
16–50
107–284
1.9–4.7
12
9
123
40–50l
aBest estimates and ranges in parentheses.
bThe authors present two budgets, one with their best estimate of the plant growth source (312 Tg yr?1), and one with that source reduced by a factor of 3
to match the estimate of Galbally and Kirstine [2002]. The numbers given here are the range defined by these two budgets.
cValues used in the GEOS-CHEM CTM simulation presented in the text, with best estimates of ranges in parentheses. Singh et al. [2004] reported a
preliminary version of this budget in their study comparing GEOS-CHEM model results to TRACE-P aircraft observations (see text).
dIncluding biofuel use.
eIncluding vehicles, solvent use, and manufacturing.
fFrom reactions of CH3O2with organic peroxy radicals. The self-reaction CH3O2+ CH3O2accounts for 85% of this source on a global basis according to
our work.
gAlthough a value of 38 Tg yr?1is used in our simulation, comparison to methanol observations over the remote oceans suggests that this source is too
low by a factor of two (see text). Our resulting best estimate for the range is 50–100 Tg yr?1.
hAfter doubling of the source from atmospheric production (see footnote g), our best estimate for the total source is 240 Tg yr?1.
iVon Kuhlmann et al. [2003b] give relative contributions from each sink, which we convert here to absolue values on the basis of their global source.
jIncluding 3 Tg yr?1in the stratosphere.
kNet ocean uptake; the ocean is both a source and a sink of methanol [Heikes et al., 2002].
lFrom combined effects of gas-phase oxidation by OH and dry deposition (deposition velocity of 0.1 cm s?1).
mHeikes et al. [2002] give a lifetime of 9 days in the text; however their global budget table implies a lifetime of 5 days.
D08303
JACOB ET AL.: GLOBAL METHANOL BUDGET
2 of 17
D08303

Page 3

from 2001 but from other years as well, assuming that
interannual variability is a relatively small source of error.
[7] Correlations of the methanol simulation with consis-
tent simulations of CO, HCN, and acetone for 2001 are also
presented for comparison with ship and aircraft observa-
tions. The CO and HCN simulations are as described by Li
et al. [2003]. The acetone simulation is as described by
Jacob et al. [2002] but without an ocean source (for reasons
to be discussed in section 4.2).
2.2. Sources of Methanol
2.2.1.Plant Growth
[8] We use the plant physiology model of Galbally and
Kirstine [2002] in which methanol emission scales as net
primary productivity (NPP) with emission factors per unit
carbon of 0.020% for grasses and 0.011% for other plants.
This offers a process-based parameterization for global
mapping of the methanol source from plant growth. We
apply the emission factors to a monthly NPP database from
the CASA 2 biosphere model with 1? ? 1? spatial resolution
[Potter et al., 1993; Randerson et al., 1997]. The global
NPP in that database is 17 Pg C yr?1for grasslands and
41 Pg C yr?1for other plants. The resulting methanol
emission is 128 Tg yr?1. Comparison to previous studies
in Table 1 would suggest at least a factor of 2 uncertainty on
this global source but our simulation of atmospheric obser-
vations implies a narrower range, as discussed later.
[9] The Galbally and Kirstine [2002] relationship of
methanol emission to NPP has yet to be tested with field
observations, and NPP estimates are themselves subject to
substantial uncertainty [Karl et al., 2004]. Our initial
simulations used monthly mean NPP values to distribute
methanol emissions seasonally, but there resulted large
model overestimates of observed methanol concentrations
at northern midlatitudes in late summer and fall. Laboratory
and field data indicate in fact that methanol emissions from
young leaves are a factor of 2–3 higher than from mature
leaves [McDonald and Fall, 1993; Nemecek-Marshall et al.,
1995; Karl et al., 2003]. We fit these results by scaling our
monthly mean NPP-based emissions with the following
parameterization adapted from A. Guenther:
ai¼ b 1 þ 2max
Li? Li?1
Li
;0
????
ð1Þ
where Liis the local leaf area index for month i, aiis the
monthly scaling factor, and b is a normalizing factor such
12
ai= 12 for each grid square. The normalization
thatP
ensures consistency with the Galbally and Kirstine [2002]
NPP-based algorithm on a yearly basis. Leaf area indices in
the model are computed monthly as a function of ecosystem
type, NPP, and global vegetation index (GVI), following the
algorithm of Guenther et al. [1995] as implemented by
Wang et al. [1998]. The resulting methanol emissions at
midlatitudes peak in spring, when they may be as much as a
factor of three larger than in summer. Within a given month
we distribute the methanol source evenly over the daytime
hours, assuming zero emission from green plants at night
[Nemecek-Marshall et al., 1995; Schade and Goldstein,
2001; Warneke et al., 2002].
i¼1
Figure 1. Atmospheric observations of methanol used for comparison with model results. Ship cruises
indicated by lines include INDOEX over the Indian Ocean in March 1999 [Wisthaler et al., 2002] and
AOE-2001 over the Arctic Ocean in July August 2001 (A. Hansel and A. Wisthaler, unpublished data,
2001). Aircraft missions indicated by lines include TRACE-P over the North Pacific in March–April
2001 [Singh et al., 2003a, 2004] and TOPSE over the North American Arctic in February–May 2000
(D. R. Blake, unpublished data, 2000). Additional aircraft missions indicated by boxes include SONEX
over the North Atlantic in October–November 1997 (regions 1–2) [Singh et al., 2000], MINOS over the
eastern Mediterranean in August 2001 (region 3) [Lelieveld et al., 2002], ITCT 2K2 over the northeast
Pacific in April–May 2002 (region 4) [Nowak et al., 2004], and PEM-Tropics B over the South Pacific in
February–March 1999 (regions 5–9) [Singh et al., 2001]. The symbol labeled ‘‘10’’ indicates the
locations of Innsbruck (Austria) [Holzinger et al., 2001] and Zugspitze (Germany, 2650 m ASL)
(A. Hansel and A. Wisthaler, unpublished data, 2003), both at (47N, 11E).
D08303
JACOB ET AL.: GLOBAL METHANOL BUDGET
3 of 17
D08303

Page 4

2.2.2.
[10] Warneke et al. [1999] reported the abiotic emission
of methanol from decaying plant matter with an emission
factor of 3–5 ? 10?4g per g of C oxidized. We apply this
emission factor to monthly mean heterotrophic respiration
rates with 1? ? 1? resolution from the CASA 2 model. The
global heterotrophic respiration rate is 58 Pg C yr?1and the
resulting methanol source is 17–29 Tg yr?1(best estimate
23 Tg yr?1). Galbally and Kirstine [2002] point out that
part of the methanol produced by plant decay may be
consumed within the litter, but they also point out that
additional biotic processes contribute to the methanol
source from plant decay. Their best estimate for the total
source from plant decay is 13 Tg yr?1(range 5–31).
2.2.3. Biomass Burning and Biofuels
[11] We use a methanol/CO emission factor of 0.018 mol
mol?1for combustion of different types of biomass, based
on compilations of literature data [Yokelson et al., 1999;
Andreae and Merlet, 2001]. Yokelson et al. [1999] find little
variability in the emission factor between different types of
fires (range 0.006–0.031 mol mol?1). Singh et al. [2004]
find a mean emission factor of 0.016 ± 0.002 mol mol?1for
fire plumes from Southeast Asia sampled over the NW
Pacific. Christian et al. [2003] find a mean value of
0.024 mol mol?1for Indonesian fuels. Holzinger et al.
[2004] find a mean value of 0.038 mol mol?1for aged
biomass burning plumes sampled over the Mediterranean
Sea, which is relatively high and which they attribute to
secondary production; but the fire plumes sampled by Singh
et al. [2004] were also aged.
[12] We apply the 0.018 mol mol?1emission factor to
gridded CO emission inventories for biomass burning
(climatological, monthly) [Duncan et al., 2003] and bio-
fuels (aseasonal) [Yevich and Logan, 2003]. Global emis-
sions in these inventories are 440 Tg CO yr?1for biomass
burning and 161 Tg CO yr?1for biofuels, and the resulting
methanol sources are 9 and 4 Tg yr?1respectively. Inverse
modeling estimates of the global biomass burning source of
CO constrained with surface air observations fall in the
range 600–740 Tg CO yr?1[Bergamaschi et al., 2000;
Pe ´tron et al., 2002], about 50% higher than used here.
2.2.4.Urban
[13] We refer to ‘‘urban’’ as the ensemble of methanol
sources from fossil fuel combustion, other vehicular
emissions, solvents, and industrial activity [Galbally and
Kirstine, 2002]. We use the aseasonal gridded (1? ? 1?)
EDGAR V2.0 global anthropogenic emission inventory
for 1990, which gives a total urban alkanol emission of
8.2 Tg yr?1[Olivier et al., 1994], and assume that methanol
accounts for half of this total or 4.1 Tg yr?1. Goldan et al.
[1995] reported a concentration ratio of methanol to nitro-
gen oxides (NOx) of 0.17 mol mol?1in urban air in
Colorado in winter; scaling of this source to a global fossil
fuel combustion NOxsource of 23 Tg N yr?1would yield
an anthropogenic source of methanol of 9.1 Tg yr?1.
Aircraft measurements by J. deGouw (unpublished) indi-
cate methanol/CO enhancement ratios of 0.050 mol mol?1
in Denver but 0.011–0.014 mol mol?1in other U.S. cities.
An emission ratio of 0.013 mol mol?1, combined with a
global fossil fuel source of CO of 480 Tgyr?1(Duncan etal.,
submittedmanuscript, 2004),wouldimplyaglobalmethanol
source of 7.1 Tg yr?1. Holzinger et al. [2001] report a
Plant Decay
methanol:benzene enhancement ratio of 0.8 mol mol?1for
urban air in Innsbruck, Austria. The EDGARV2.0 inventory
gives aglobal benzene emission of 1.2Tg yr?1from vehicles
and industry, which would imply a methanol source of only
0.4 Tg yr?1. The urban source of methanol is thus highly
uncertain but is clearly small on a global scale.
2.2.5. Atmospheric Production
[14] Methanol is produced in the atmosphere by reactions
of the methylperoxy (CH3O2) radical with itself and with
higher organic peroxy (RO2) radicals [Madronich and
Calvert, 1990; Tyndall et al., 2001]:
CH3O2þ CH3O2! CH3OH þ CH2O þ O2
ðR1Þ
CH3O2þ RO2! CH3OH þ R0CHO þ O2
ðR2Þ
Alternate branches for these reactions, not producing
methanol, are
CH3O2þ CH3O2! CH3O þ CH3O þ O2
ðR10Þ
CH3O2þ RO2! CH3O þ RO þ O2
ðR20Þ
The CH3O2 and RO2 radicals are produced in the
atmosphere by oxidation of VOCs. The sum of reactions
(R1), (R10), (R2), and (R20) typically accounts for less than
10% of the CH3O2sink in current chemical mechanisms.
The dominant atmospheric sinks are the reactions with HO2
and NO, which do not produce methanol:
CH3O2þ HO2! CH3OOH þ O2
ðR3Þ
CH3O2þ NO ! CH3O þ NO2
ðR4Þ
[15] We calculate the atmospheric source of methanol
from (R1) and (R2) with a GEOS-CHEM simulation of
tropospheric ozone-NOx-VOC chemistry [Fiore et al.,
2003]. Primary VOCs in that simulation include methane,
ethane, propane, higher alkanes, >C2 alkenes, isoprene,
acetone, and methanol. The simulation uses recommended
data from Tyndall et al. [2001] for the kinetics and methanol
yieldsof(R1),andalsofor(R3)involvingtheacetonylperoxy
radical (CH3C(O)CH3O2) produced by oxidation of acetone.
Themethanol yields at298 K forthese two reactions are0.63
and 0.5, respectively. We assume a 0.5 yield for all other
CH3O2+ RO2reactions, following Madronich and Calvert
[1990].Theresultingglobalsourceofmethanolis38Tgyr?1.
Reaction (R1) contributes 85% of that global total. The
remaining 15%, contributed by (R2), mainly involves RO2
radicals produced from biogenic isoprene and is largely
confined to the continental boundary layer, where it is
much smaller than the primary emission from plant
growth. In the remote atmosphere, where the atmospheric
production of methanol is of most interest, the CH3O2
radicals driving (R1) originate mainly from the oxidation
of methane.
[16] Previous literature estimates of the atmospheric
source of methanol from (R1) and (R2), similarly obtained
with global tropospheric chemistry models, are in the range
D08303
JACOB ET AL.: GLOBAL METHANOL BUDGET
4 of 17
D08303

Page 5

18–31 Tg yr?1(Table 1), lower than our estimate. These
include 30 Tg yr?1from Heikes et al. [2002] computed with
an earlier version of GEOS-CHEM. Discrepancies between
estimates reflect differences in the abundances of NO and
HO2, the CH3O2 reaction rate constants, the yield of
methanol from (R1) and (R2), and the importance of
(R2). Uncertainties in the (R1) rate constant and in the
corresponding methanol yield are each about 30% at 298K
[Tyndall et al., 2001] and higher at colder temperatures. As
we will see, observations from the PEM-Tropics B aircraft
mission over the South Pacific suggest an even larger
atmospheric source of methanol than is used here.
2.3. Sinks of Methanol
2.3.1.Gas-Phase Oxidation by OH
[17] Weuse the rate constant k= 3.6 ?10?12exp[?415/T]
cm3molecule?1s?1recommended by Jimenez et al. [2003]
with an uncertainty of 20%. We apply this rate constant to a
3-D archive of monthly mean OH concentrations from the
Fiore et al. [2003] GEOS-CHEM simulation of tropospheric
ozone-NOx-VOC chemistry. The lifetime of methylchloro-
form against tropospheric oxidation by OH (proxy for the
global mean tropospheric OH concentration) is 5.6 years in
that simulation [Martin et al., 2003]. This is within the range
constrained by the methylchloroform observations, which
imply a 25% uncertainty in global mean OH concentrations
[Intergovernmental Panel on Climate Change (IPCC),
2001]. Stratospheric loss of methanol is computed using
OH concentrations archived from a global 2-D stratospheric
chemistry model [Schneider et al., 2000] and amounts to
only 2% of the loss in the troposphere. Our computed
lifetime of methanol against oxidation by OH is 11 days.
Addition of errors in quadrature implies an uncertainty of
30% on this value.
2.3.2. Aqueous-Phase Oxidation by OH(aq)
[18] Methanol dissolves in aqueous aerosols (Henry’s law
constant H = 1.6 ? 10?5exp [4900/T] M atm?1) and can
then be oxidized by OH(aq) (k = 1.0 ? 109exp [?590/T]
M?1s?1; Elliot and McCracken [1989]). The corresponding
first-order loss rate constant (s?1) for atmospheric methanol
is k0= HLRTk[OH(aq)] where L is the dimensionless liquid
water content, R is the gas constant, and [OH(aq)] is the
OH(aq) concentration in units of M (moles per liter of
water). The reaction takes place mainly in clouds, where L
is more than 3 orders of magnitude higher than in a clear-
sky atmosphere. The GEOS-3 meteorological archive pro-
vides 3-D cloud optical depths t from which we estimate L=
4tr/3QDZ where ris the effective cloud droplet radius (taken
tobe 10 mm),Q ? 2is thecloud droplet extinctionefficiency,
and DZ is the vertical thickness of the gridbox. The OH(aq)
concentration in cloud droplets is largely determined by
aqueous-phase cycling with HO2(aq)/O2
complicated way on cloud composition [Jacob, 1986]. We
assume here a simple parameterization [OH(aq)] = d[OH(g)]
where [OH(g)] is the gas-phase concentration calculated in
GEOS-CHEMwithoutconsiderationofaqueous-phasecloud
chemistry, and d = 1 ?10?19M cm3molecule?1is chosen to
fit the cloud chemistry model results of Jacob [1986]. The
resulting lifetime of methanol against in-cloud oxidation by
OH(aq) is longer than 1 year. Heikes et al. [2002] and
Galbally and Kirstine [2002] found a larger role for in-
cloud oxidation by OH(aq) in their global methanol
?and depends in a
budgets, amounting to 5–10% of the gas-phase loss
(Table 1). Their assumed OH(aq) concentrations and
aqueous-phase CH3OH + OH rate constants are higher
than ours.
[19] A few studies have raised the possibility of a
missing heterogeneous sink for methanol in aerosols or
clouds. Singh et al. [2000] suggested that such a sink
might explain their observed decrease of methanol con-
centrations from the middle to upper troposphere at north-
ern midlatitudes (SONEX aircraft mission). Jaegle ´ et al.
[2000] found that unexpectedly high formaldehyde con-
centrations measured in the upper troposphere during
SONEX correlated with methanol, and speculated that
reactive uptake of methanol in cirrus clouds with a reaction
probabilityg=0.01couldprovideanexplanation.Yokelsonet
al. [2003] observed the depletion of methanol in a cloud
polluted by biomass burning smoke, and Tabazadeh et al.
[2004] proposed that surface reactions of methanol on cloud
droplets could be responsible. However, statistical compari-
son of in-cloud versus clear-sky methanol concentrations in
the large data set from the TRACE-P aircraft mission indi-
cates no significant differences at any altitude [Singh et al.,
2004]. Upper tropospheric observations from the SONEX
mission indicate no depletion in air processed by deep
convection or cirrus clouds [Jaegle ´ et al., 2000]. Laboratory
studies indicate no significant reactive uptake ofmethanol by
sulfuric acid aerosols [Iraci et al., 2002], ice [Hudson et
al., 2002; Winkler et al., 2002], or mineral oxide surfaces
[Carlos-Cuellar et al., 2003]. Kane and Leu [2001] report
fast reaction of methanol with sulfuric acid in concen-
trated solutions but Iraci et al. [2002] attribute this result
to an experimental artifact. At present, the weight of
evidence does not support a fast heterogeneous sink for
methanol.
2.3.3. Deposition to Land
[20] Microbial and foliar uptake of methanol by vegeta-
tion and soils is difficult to separate from the plant growth
and decay sources, either observationally or for model
purposes. Investigators have placed soil and leaf litter
[Schade and Goldstein, 2001] and foliage [Nemecek-
Marshall et al., 1995] in enclosures and observed a net
emission of methanol. However, these studies used enclo-
sures that were flushed with methanol-free air. Recent
measurements by A. Guenther (unpublished) indicate that
there can be a net uptake of methanol when enclosures are
flushed with air containing ambient levels of methanol.
[21] Measurement of the diurnal cycle of methanol con-
centrations at continental surface sites provides some sepa-
ration of microbial uptake from plant growth emission,
which is restricted to daytime. Kesselmeier et al. [2002]
observed no significant diurnal variation at a site in the
Amazon forest, but other observations at sites in the tropics
and northern midlatitudes indicate decreases over the course
of the night ranging from about 30% [Goldan et al., 1995] to
several-fold [Riemer et al., 1998; Holzinger et al., 2001;
Karl et al., 2004], suggesting surface uptake. A 30%
decrease over the course of an 8-hour summer night, and
for a typical nighttime mixing depth of 100 m, would imply
a methanol deposition velocity of 0.12 cm s?1. Karl et al.
[2004] measured methanol deposition to a tropical forest
ecosystem by using a combination of eddy covariance and
vertical gradients, and found a mean methanol deposition
D08303
JACOB ET AL.: GLOBAL METHANOL BUDGET
5 of 17
D08303

Page 6

velocity of 0.27 ± 0.14 cm s?1. On the basis of the above
information, and acknowledging the large uncertainty, we
assume here a constant dry deposition velocity of 0.2 cm s?1
to land. We assume that this deposition velocity also applies
to ice-covered surfaces since Boudries et al. [2002] found
that Arctic snow is a net sink for methanol. The resulting
atmospheric lifetime of methanol against dry deposition is
26 days. Including a methanol sink from dry deposition
improves significantly the ability of the model to reproduce
observations at northern midlatitudes in fall and winter.
2.3.4. Wet Deposition
[22] We simulate wet deposition of methanol with the
GEOS-CHEM wet deposition scheme described by Liu et
al. [2001]. This scheme accounts for scavenging of water-
soluble species by convective updrafts, convective anvils,
and large-scale precipitation. The scavenging coefficients
used by Liu et al. [2001] are scaled here to the dissolved
fraction of methanol inferred from the local liquid water
content and Henry’s law constant. Scavenging is assumed to
take place in warm clouds only (T > 268 K). The retention
efficiencyofmethanoluponclouddropletfreezingistakento
be 0.02, by analogy with previous assumptions for CH3OOH
and CH2O [Mari et al., 2000], so that scavenging is ineffi-
cient in mixed liquid-ice clouds where the precipitation
involves riming. Laboratory studies indicate that methanol
coverageoficesurfacesis low,less than10?4ofamonolayer
at equilibrium [Hudson et al., 2002; Winkler et al., 2002],
supporting the assumption of a low retention efficiency. The
resulting lifetime of methanol against wet deposition is
120 days. With such a long lifetime, wet deposition does
not significantly affect the vertical profile of methanol in
the troposphere [Crutzen and Lawrence, 2000].
2.3.5.Uptake by the Ocean
[23] Simple solubility considerations imply that the ocean
mixed layer must be a large reservoir for methanol [Galbally
and Kirstine, 2002; Singh et al., 2003a]. Heikes et al. [2002]
pointoutthatmethanolisbothproducedandconsumedinthe
ocean. Singh et al. [2003a] observed a gradient of increasing
methanol concentrations from the marine boundary layer to
the free troposphere over the remote North Pacific during
TRACE-P, implying an ocean sink with a mean deposition
velocity of 0.08 cm s?1. Measurements taken at the Mace
Head coastal site in Ireland also show evidence of methanol
uptake by the ocean, with a similar deposition velocity
[Carpenteretal.,2004].OurGEOS-CHEMresultspresented
in the work of Singh et al. [2003a] show that an ocean
saturation ratio of 90% for methanol, combined with a
standard two-layer parameterization of ocean-atmosphere
exchange, gives a good simulation of the observed
TRACE-P vertical gradients. In the absence of better infor-
mation we assume that the 90% saturation ratio holds
globally. The resulting lifetime of methanol against net
uptake by the ocean is 130 days.
3.
Distribution of Methanol
Global Model Budget and Atmospheric
[24] Table 1 summarizes our global model budget of
methanol. The global source of 206 Tg yr?1includes
contributions from plant growth (62%), atmospheric oxida-
tion of VOCs (18%), plant decay (11%), biomass burning
and biofuels (6%), and vehicles and industrial activities
(2%). The global sink balancing that source is dominated by
gas-phase oxidation by OH (63%) with minor contributions
from dry deposition to land (26%), wet deposition (6%), net
uptake by the ocean (5%), and aqueous-phase oxidation in
clouds (<1%). The resulting atmospheric lifetime of meth-
anol is 7 days and the global atmospheric burden is 4.0 Tg.
A preliminary version of our model budget was presented
by Singh et al. [2004] in a study comparing GEOS-CHEM
results to their TRACE-P aircraft observations (reference to
our budget is given there as B. D. Field et al., manuscript in
preparation, 2003). That preliminary version did not include
deposition of methanol to land or the dependence of
methanol emissions on leaf age.
[25] Our global source of methanol is in the range of
previous literature cited in Table 1 (122–350 Tg yr?1).
Differences reflect principally the magnitude of the plant
growth source. Our value for that source (128 Tg yr?1) is
close to that of Galbally and Kirstine [2002] (100 Tg yr?1),
as would be expected since we followed their algorithm.
Our atmospheric source of methanol from CH3O2reactions
(38 Tg yr?1) is larger than previous estimates (18–31 Tg
yr?1).
[26] Gas-phase oxidation by OH is a major methanol sink
in all the inventories of Table 1. However, the principal
methanol sink in the Heikes et al. [2002] budget is dry
deposition to land and oceans, with an assumed deposition
velocity of 0.4 cm s?1. As mentioned above, the TRACE-P
aircraft data of Singh et al. [2003a] imply a much lower
deposition velocity to the ocean. Fast deposition to land in
the Heikes et al. [2002] budget (90 Tg yr?1) partly com-
pensates for their high terrestrial biogenic source (280 Tg
yr?1). The relative contributions to the methanol sink from
gas-phase oxidation by OH, dry deposition, and wet depo-
sition in our model are essentially the same as in the
previous global 3-D model study by von Kuhlmann et al.
[2003a] (63%, 30%, and 7% respectively). That study did
not report separate deposition terms to land and ocean.
[27] Figure 2 shows the mean concentrations of methanol
simulated by the model in surface air and at 500 hPa, for
January and July. Surface air concentrations over land
exceed 5–10 ppbv in the tropics and at northern midlatitudes
in summer. The particularly high concentrations over Siberia
in July are due to the long continental fetch and the late
emergence of leaves. Surface air concentrations at northern
midlatitudes in winter drop to 0.5–2 ppbv. Concentrations
over land decrease typically by a factor of 2–5 from
the surface to 500 hPa, reflecting the surface source and
the 7-day lifetime. There is by contrast little vertical gradient
over the oceans. Concentrations over the northern hemi-
sphere oceans (0.5–1 ppbv) show little seasonal variation,
because faster chemical loss in summer balances the effect of
the larger continental source. Concentrations over the south-
ern hemisphere oceans are higher in winter (0.5–1 ppbv)
than in summer (0.2–0.5 ppbv) since the continental source
in the southern hemisphere is mainly in the tropics and has
little seasonal variation. Concentrations are lowest (below
0.2 ppbv) in surface air over Antarctica in summer.
4.Evaluation With Observations
[28] We examine in this section how the above estimates
of methanol sources and sinks, when implemented in a
D08303
JACOB ET AL.: GLOBAL METHANOL BUDGET
6 of 17
D08303

Page 7

global CTM, succeed in reproducing the observed atmo-
spheric concentrations of methanol. Our primary focus is on
observations from aircraft and ships, which integrate source
information over large regions. Observations from surface
continental sites, reviewed by Heikes et al. [2002], are few
and show high spatial variability that we cannot expect to
reproduce given our crude model representation of sources.
We use these surface observations mainly to examine large-
scale features, including seasonal variations and tropical
concentrations, that are of particular interest for testing the
model.
4.1. Surface Concentrations
[29] The review of Heikes et al. [2002] gives representa-
tive surface air concentrations of 20 (range 0.03–47) ppbv
in urban air, 10 (1–37) ppbv over forests, 6 (4–9) ppbv
over grasslands, 2 (1–4) ppbv for continental background,
and 0.9 (0.3–1.4) ppbv over the northern hemispheric
oceans. Our model values are roughly consistent with these
ranges, as shown in Figure 2. Urban air is not resolved on
the scale of the model.
[30] The Heikes et al. [2002] compilation includes no
surface data for tropical land ecosystems, where the model
predicts high concentrations year-round. Measurements by
Kesselmeier et al. [2002] in the Rondo ˆnia tropical forest of
Brazil (10S, 63W) for a 7-day period in October 1999 (end
of dry season) indicate a range of 1 to 6 ppbv. Our mean
simulated concentration for that site and month is much
higher, 10 ppbv, with dominant contributions from plant
growth (6 ppbv), plant decay (2 ppbv), and biomass burning
(2 ppbv). Measurements by Karl et al. [2004] at a tropical
forest site in Costa Rica (10N, 84W) for a 20-day period in
April–May 2003 (peak of dry season) indicate a mean
concentration of 2.2 ppbv. Our mean simulated concentra-
tion for that site and 2-month period is 2.1 pppbv, in
agreement with observations, with a major contribution
from plant growth (1.2 ppbv) and minor contributions of
about 0.3 ppbv each from plant decay, biomass burning, and
atmospheric production. The lower model value in Costa
Rica than in Rondo ˆnia reflects the shorter continental fetch.
The model shows little seasonal variation at the Rondo ˆnia
site (monthly means range from 8 to 10 ppbv) but more at
the Costa Rica site (1.2 to 2.7 ppbv). At the latter site, the
seasonal maximum is in the wet season (June–October)
when the plant growth source is high, and the seasonal
minimum is in the early dry season (January–February)
before biomass burning. Karl et al. [2004] also made
methanol flux measurements at the Costa Rica site, which
indicate a value 50% higher than the parameterization of
Galbally and Kristine [2002] when normalized to the
estimated tropical forest NPP. On the other hand, the data
from Kesselmeier et al. [2002] would suggest that the NPP-
based parameterization of Galbally and Kristine [2002] is
too high for the Amazon forest. More observations of
Figure 2. Simulated monthly mean concentrations of methanol (ppbv) in surface air and at 500 hPa for
January and July 2001.
D08303
JACOB ET AL.: GLOBAL METHANOL BUDGET
7 of 17
D08303

Page 8

methanol concentrations and fluxes are clearly needed for
tropical land ecosystems.
[31] To our knowledge, the only methanol observations in
the literature extending over a full yearly cycle are those of
Holzinger et al. [2001] taken in the outskirts of Innsbruck,
Austria (47N, 11E). These authors used concurrent mea-
surements of benzene (emitted by vehicles) to separate the
local urban from the regional nonurban components of
methanol. Figure 3 shows the observed ranges for the
nonurban component. There is a spring-summer maximum
and fall-winter minimum, reflecting biogenic emission.
Concentrations drop by more than a factor of 2 from June
to September. Also shown in Figure 3 are the simulated
methanol concentrations and the contributions from the
individual model sources. Plant growth is the main model
source except in winter, when plant decay and urban
sources become relatively important. The model reproduces
the seasonal variation in the observations, including the rise
from winter to spring and the decrease from spring to fall.
The latter reflects in the model the weaker emission from
mature leaves.
[32] Ship measurements from Wisthaler et al. [2002] over
the Indian Ocean during INDOEX 1999 (March 1999)
provide another important data set for methanol in surface
air. The cruise track extended from 20N to 12 S, starting
from the west coast of India and extending south of the
intertropical convergence zone (Figure 1). Observed meth-
anol concentrations exceed 1 ppbv in Indian outflow and
drop to 0.5–0.6 ppbv in southern hemispheric air. Wisthaler
et al. [2002] reported a strong correlation of methanol with
CO, which we compare in Figure 4 to our mean monthly
model results for 2001 sampled along the cruise track. The
agreement is remarkably good. The model captures the
Figure 3. Seasonal variation of methanol concentrations at
Innsbruck, Austria. The ranges of observations reported by
Holzinger et al. [2001] for different times of year in 1996–
1997 are shown as vertical lines with symbols. These
observations are for the non-urban component of methanol,
after subtraction of the urban component based on
correlation with benzene. Model results are shown as the
solid line, with additional lines identifying contributions
from individual sources in the model: plant growth (short
dashes), plant decay (dots), urban (thin solid), biomass
burning and biofuels (long dashes), and atmospheric
production (dash-dot).
Figure 4. Methanol-CO correlations along the INDOEX 1999 cruise track over the Indian Ocean
(Figure 1). Observations from Wisthaler et al. [2002] are shown as small dots, with linear regression as
solid line. Monthly mean GEOS-CHEM results along the cruise track are shown as large open circles,
with linear regression as dashed line. Coefficients of determination (r2) and slopes of the linear
regressions (S) are shown inset.
D08303
JACOB ET AL.: GLOBAL METHANOL BUDGET
8 of 17
D08303

Page 9

clean 0.5–0.6 ppbv background and the observed relation-
ship with CO, although it does not capture the population of
observations with relatively high CO but low methanol
(hence the stronger methanol-CO correlation in the model,
r2= 0.76 versus r2= 0.47). Over India in the model the
methanol is mostly from plant growth, while CO is mostly
from combustion; the methanol-CO correlation in Indian
outflow reflects varying degrees of dilution with the marine
background rather than a commonality of methanol and CO
source processes.
[33] Additional unpublished methanol measurements
have been made by A. Wisthaler and A. Hansel on a ship
cruise in the Arctic in July August 2001 (AOE-2001), and at
Zugspitze in southern Germany (47N, 11E, 2650 m altitude)
for four months in October 2002 to January 2003. These are
shown in Figure 5 together with the corresponding model
results. The model reproduces the low Arctic cruise obser-
vations north of 85N (0.30–0.35 ppbv), reflecting deposi-
tion to the ocean, but not the even lower concentrations
(0.1–0.2 ppbv) observed further south; these low concen-
trations are from only a few measurements and suggest a
strong local ocean sink. Observations at Zugspitze show a
decrease from 0.7 ppbv in October to 0.3 ppbv in January.
The model also shows a decrease, mostly from the deposi-
tion sink, but not as strong as observed. This would suggest
a nonphotochemical sink of methanol missing from the
model. However, the TOPSE aircraft observations over
the North American Arctic in winter, to be discussed in
section 4.3, show higher concentrations than at Zugspitze
and do not suggest such a missing sink.
4.2. Asian Outflow (TRACE-P Mission)
[34] The TRACE-P aircraft mission in March–April 2001
provided extensive data for methanol and other species in
Asian outflow over the North Pacific [Jacob et al., 2003].
The outflow included a major contribution from seasonal
biomass burning in Southeast Asia [Heald et al., 2003a].
Flight tracks are shown in Figure 1. Our model simulation
is for the same meteorological year as TRACE-P. We
sample the model along the flight tracks and for the flight
days for comparison to observations. GEOS-CHEM simu-
lations for the TRACE-P period have been evaluated
previously with observations for a number of species
including in particular CO [Kiley et al., 2003; Palmer et
al., 2003b] and HCN [Li et al., 2003], which we discuss
below in the context of their correlations with methanol.
The model gives a good simulation of Asian outflow
pathways with no evident transport bias [Liu et al., 2003;
I. Bey et al., Characterization of transport errors in chemical
forecasts from a global tropospheric chemical transport
model, submitted to Journal of Geophysical Research,
2005]. Simulation of the CO2 observations indicates a
45% underestimate of the net biospheric carbon emission
for China in the CASA 2 model during the TRACE-P
period [Suntharalingam et al., 2004]. However, as we will
see below, there is no evident bias in our methanol
simulation.
4.2.1. Vertical Profiles
[35] Figure 6 compares the mean simulated and observed
vertical profiles of methanol concentrations for the four
TRACE-P quadrants separated at 30N, 150E. The four
quadrants contrast fresh Asian outflow west of 150E to
more background North Pacific air to the east.
[36] The observed methanol concentrations in the NW
quadrant decrease from 1.5–2 ppbv in the boundary layer to
0.9–1.3 ppbv in the free troposphere, rise again to a
maximum in excess of 2 ppbv in the upper troposphere,
and then decline to 0.1 ppbv above the tropopause at 12 km.
Figure 5. Mean methanol concentrations from the AOE-2001 Arctic cruise in July August 2001 as a
functionoflatitudeandfromZugspitze(southernGermany,2650mASL)inOctober2002toJanuary2003
as a function of month. Unpublished observations from A. Wisthaler and A. Hansel (closed circles) are
compared tomodel results (open circles). Standarddeviations ontheobservations areshown forZugspitze.
D08303
JACOB ET AL.: GLOBAL METHANOL BUDGET
9 of 17
D08303

Page 10

The model reproduces these features although it is up to a
factor 2 too low in the middle troposphere. Stratospheric
concentrations cannot be compared because of excessive
stratosphere-troposphere exchange in the GEOS assimilat-
ed meteorological data used to drive the model [Liu et al.,
2001; Tan et al., 2004]. The plant growth source in the
model contributes a relatively featureless background of
0.3 ppbv throughout the troposphere in this NW quadrant.
The boundary layer enhancement reflects a mix of Chi-
nese emissions from plant decay, biofuels, and urban
sources. The upper tropospheric enhancement is due to
outflow from deep convection in Southeast Asia and is
contributed mainly by the plant growth and biomass
burning components.
[37] The observed mean profile for the SW quadrant in
Figure 6 shows greater influence from Southeast Asia than
the NW quadrant. Concentrations decrease gradually with
altitude, from 1.5 ppbv in the boundary layer to 0.5–1 ppbv
in the middle and upper troposphere but with a convective
enhancement apparent at 8–9 km. The model reproduces
this structure with no evident bias.
[38] Observations in the NE quadrant show a marked
increase with altitude, in contrast to the western quadrants.
Asian influence is principally in the middle and upper
troposphere. The model underestimates this Asian influ-
ence, by about the same factor as for the free tropospheric
Asian outflow in the NW quadrant. The relatively low
methanol concentrations in the lower troposphere, both in
the model and in the observations, reflect gas-phase oxida-
tion and ocean uptake of methanol as the Asian air masses
subside [Singh et al., 2003a].
[39] Asian influence during TRACE-P was weakest in the
SE quadrant, which lies south of the dominant outflow
track. The observed layer of high methanol concentrations
at 2–4 km altitude is from one single Hawaii-Guam flight
where the aircraft sampled repeatedly an Asian outflow
plume that had traveled southward and subsided [Crawford
et al., 2004]. High CO concentrations (200 ppbv) were
observed in that layer. Previous analysis of the GEOS-
CHEM CO simulation shows that this layer is displaced
upward and diluted in the model relative to the observations
[Heald et al., 2003b]. A similar bias is found for methanol,
as the model enhancement is at 4–5 km altitude and peaks
at only 0.7 ppbv. The model is also low in the marine
boundary layer (0.4 ppbv, versus 0.6 ppbv in the observa-
tions); atmospheric production is the most important model
source there and is probably too weak, as discussed later in
the context of the PEM-Tropics B observations over the
South Pacific.
[40] Singh et al. [2004] previously compared results from
our preliminary model version (not including dry deposition
to land or the decrease in the plant growth source as leaves
age) to their mean observed background vertical profile of
methanol concentrations for the ensemble of the mission.
This showed agreement within 10% up to 6 km altitude, but
a growing model overestimate at higher altitudes (up to a
factor of 3 at 11 km) that they speculated could reflect
heterogeneous chemical loss not captured by the model. In
Figure 6. Vertical profiles of methanol concentrations from the TRACE-P aircraft mission over the
North Pacific (March–April 2001) averaged over four quadrants separated at 30N, 150E (Figure 1).
Means and standard deviations of observations are shown as symbols and horizontal lines; the number of
observations contributing to each average is indicated. Model results are shown as solid lines, with
additional lines identifying contributions from individual model sources: plant growth (short dashes),
plant decay (dots), urban (thin solid), biomass burning and biofuels (long dashes), and atmospheric
production (dash-dot).
D08303
JACOB ET AL.: GLOBAL METHANOL BUDGET
10 of 17
D08303

Page 11

the Singh et al. [2004] comparison, observed methanol
concentrations decrease from 0.8 ppbv at 6 km to 0.35 ppbv
at 11 km (filtered against stratospheric influence); whereas
model values increase from 0.9 ppbv at 6 km to 1.1 ppbv at
11 km. Our comparisons in Figure 6 do not show such a
model bias in the upper troposphere, but instead consider-
able regional difference in simulated and observed vertical
profiles for the different quadrants, as discussed above. The
decreasing trend from 6 to 11 km in the observations is
found only for the southern quadrants. In addition, the
model values presented here are lower than in the prelim-
inary simulation reported by Singh et al. [2004] due to our
subsequent inclusion of a land deposition sink in the model.
This affects inparticularthesimulation ofuppertropospheric
concentrations, which include a major contribution from
convective outflow of tropical continental air.
4.2.2.Correlations With Other Species
[41] Further insight can be gained by examining the
observed correlations of methanol with other species. In
the ensemble of TRACE-P observations we find that
methanol correlates most strongly with acetone (r2= 0.83,
slope = 0.38 mol mol?1), HCN (r2= 0.78, slope = 0.12 mol
mol?1), and CO (r2= 0.63, slope = 64 mol mol?1). These
correlations are shown in Figure 7. The slopes are given
with methanol as denominator. We generated corresponding
correlations from the model results sampled along the flight
tracks. These are also shown in Figure 7. The CO and HCN
simulations are as described in the work of Li et al. [2003].
The acetone simulation presented here is that of Jacob et al.
[2002] applied to the TRACE-P period, but without an
ocean source since the TRACE-P observations indicate that
the ocean was in fact a net sink for acetone [Singh et al.,
2003a]. We find that an ocean source for acetone in the
model would destroy the correlation with methanol since
the ocean is a sink for methanol.
[42] The model reproduces the correlation between meth-
anol and acetone found in the observations (r2= 0.49,
slope = 0.48 mol mol?1in the model). The biogenic acetone
source in the model is scaled to isoprene emission [Jacob et
al., 2002], while that of methanol is scaled to NPP. The
stronger correlation in the observations suggests that the
sources of acetone and methanol are governed by more
similar processes than is assumed in the model. Measure-
ments at rural U.S. sites in summer have previously shown a
strong correlation between acetone and methanol with an
acetone/methanol slope of 0.21–0.27 mol mol?1[Goldan et
al., 1995; Riemer et al., 1998]. The higher slope observed
here is due to anthropogenic sources of acetone.
Figure 7. Correlations of methanol with CO, acetone, and HCN concentrations in the ensemble of
TRACE-P data. Observations (top panels) are compared to model results (bottom panels). Coefficients of
determination (r2), regression lines, and corresponding slopes (S) are indicated.
D08303
JACOB ET AL.: GLOBAL METHANOL BUDGET
11 of 17
D08303

Page 12

[43] The observed correlation between methanol and
HCN is well simulated by the model (r2= 0.43, slope =
0.10 mol mol?1). The dominant source of HCN variability
in the TRACE-P data is biomass burning in Southeast Asia
[Li et al., 2003; Singh et al., 2003b]. The observed HCN/
CO molar emission ratio from biomass burning varies in the
literature over a large range from 0.03% to 1.1%, with a best
fit for the TRACE-P conditions of 0.27% [Li et al., 2003].
Combined with our estimated methanol/CO molar emission
ratio from biomass burning of 1.8% mol mol?1this yields
an HCN/methanol emission ratio of 0.15 mol mol?1. The
weaker slope in the model and in the observations likely
reflects biogenic methanol emissions in Southeast Asia that
are collocated with the biomass burning emissions.
[44] Good agreement between model and observations is
also found in the correlation of methanol with CO (r2=
0.44, slope = 69 mol mol?1in the model). CO here serves
as a general tracer of Asian outflow and the correlation
mainly provides support for the overall magnitude of
methanol export from the Asian continent. Although the
model underestimates the TRACE-P methanol observations
in the middle troposphere (Figure 6), the methanol-CO
correlation is principally driven by strong outflow events
in the boundary layer [Liu et al., 2003].
4.3. Other Aircraft Observations
[45] We now compare model results to methanol obser-
vations from other aircraft missions including TOPSE at
North American high latitudes in February–May 2000
(D.R. Blake, unpublished data), SONEX over the North
Atlantic in October–November 1997 [Singh et al., 2000],
MINOS over the eastern Mediterranean in August 2001
[Lelieveld et al., 2002], ITCT 2K2 over the Northeast
Pacific in April–May 2002 [Nowak et al., 2004], and
PEM-Tropics B over the South Pacific in February–March
1999 [Singh et al., 2001]. Aside from MINOS, these
missions were conducted for years other than the 2001
model year. We use monthly mean vertical profiles in the
model over the flight regions (Figure 1) to compare to the
mean observations. The measurements in TRACE-P,
SONEX, and PEM-Tropics B were made by real-time gas
chromatography (GC). The measurements in ITCT 2K2 and
MINOS were made by proton transfer mass spectrometry
(PTR-MS). A ship-based intercomparison of online PTR-
MS and real-time GC-MS methods indicates high correla-
tion between the two and agreement within a few percent
[de Gouw et al., 2003]. Calibration differences of up to 20%
may be expected between the GC measurements of Singh et
al. from different missions. The TOPSE measurements were
made by GC analysis from collected air canisters, all with
the same relative humidity and the same lapse of time
between collection and analysis. The accuracy is estimated
to be 30% and the precision is much better.
[46] Additional aircraft methanol data are available from
the PEM-West B mission over the NW Pacific in February–
March 1994 [Singh et al., 1995] and from the LBA-
CLAIRE mission over the rain forest in Surinam in March
1998 [Williams et al., 2001]. The PEM-West B data are
from the same region and season as TRACE-P, and show
similar concentrations [Singh et al., 2004], but represent a
much sparser data set. The LBA-CLAIRE data indicate low
methanol concentrations, averaging 1.1 ppbv in the bound-
ary layer and 0.6 ppbv in the free troposphere. These would
suggest, consistent with Kesselmeier et al. [2002], that
methanol emission from tropical forests of South America
is much lower than predicted from the Galbally and Kirstine
[2002] parameterization. However, quantitative comparison
to the model is difficult because the observations were taken
Figure 8. Latitudinal profiles of methanol concentrations at North American high latitudes during the
TOPSE mission (Figure 1), for February and April and for 0–2 and 4–6 km altitude. Mean observations
(D. R. Blake, unpublished data, 2000) are shown as solid circles; the number of observations used to
compute the mean is also shown. Model results are shown as open circles.
D08303
JACOB ET AL.: GLOBAL METHANOL BUDGET
12 of 17
D08303

Page 13

under conditions of marine inflow and only about 12 hours
downwind of the coast.
[47] The TOPSE measurements are of particular interest
as they characterize the latitudinal gradient at high northern
latitudes during the transition from winter to spring [Atlas et
al., 2003]. We show in Figure 8 the latitudinal profiles
measured at 0–2 and 4–6 km altitude in February and April
(only a few observations are available in May). Also shown
are the corresponding model profiles, sampled as monthly
means along the TOPSE flight tracks. Boundary layer
concentrations in February show a latitudinal decrease from
1 ppbv at northern midlatitudes to 0.4 ppbv in the Arctic,
both in the model and the observations, due to the shutting
down of the biogenic source and the effect of surface
deposition. The same latitudinal gradient is found in April
but concentrations are about 0.2 ppbv higher, similarly in
the model and in the observations, reflecting the springtime
source. Concentrations at 4–6 km altitude show by contrast
little latitudinal gradient, which the model explains as
reflecting a weaker influence of the dry deposition sink.
The February observations at 4–6 km (0.5 ppbv) are higher
than the wintertime Zugspitze observations discussed in
section 4.1 and are more consistent with the model results.
The model misses the observed April enhancement at 4–
6 km south of 60N although it captures it in the 0–2 km
data; this could be due to model error in the springtime
onset of the source. We have no explanation for the high
concentrations (above 1 ppbv) observed north of 80N at 4–
6 km in April.
[48] Aircraft observations at northern midlatitudes are
available from the SONEX, MINOS, and ITCT 2K2 mis-
sions. These are shown in Figure 9 together with the
corresponding model results. Observed concentrations in
the middle and upper troposphere in SONEX are low,
0.4 ppbv on average, and this is reproduced in the model
with only a slight positive bias. Reduced methanol emission
from mature leaves is critical for simulation of the SONEX
observations; without this reduction the simulated methanol
concentrations would be much higher in fall because of the
weak photochemical sink. The global 3-D model study of
von Kuhlmann et al. [2003b] also found good agreement
with the SONEX observations using a global methanol
source 40% lower than ours (Table 1). They did not account
for the decrease of methanol emission with leaf age.
[49] The model reproduces the free tropospheric concen-
trations observed in MINOS (0.6–1 ppbv) over the Medi-
Figure 9. Vertical profiles of methanol concentrations from the SONEX, MINOS, ITCT 2K2, and
PEM-Tropics B aircraft missions averaged over the regions of Figure 2. Symbols and lines are as for
Figure 6. The MINOS observations are taken from Lelieveld et al. [2002] without information on the
number of observations.
D08303
JACOB ET AL.: GLOBAL METHANOL BUDGET
13 of 17
D08303

Page 14

terranean in July, and also the high values in surface air. It is
too low in the lower free troposphere (2–4 km), possibly
reflecting difficulties in simulating vertical mixing and
transport over this highly heterogeneous region. Salisbury
et al. [2003] observed a mean concentration of 3.3 ppbv at a
site in Crete during MINOS, and as shown in Figure 9 this
is consistent with model results. Comparisons to the von
Kuhlmann et al. [2003a, 2003b] model presented in that
paper found it to be too low by a factor of four on average,
possibly reflecting its low plant growth source.
[50] Observations from the ITCT 2K2 campaign off the
California coast in April–May average 0.9–1.1 ppbv in the
1–7 km column with lower values near the surface,
consistent with an ocean sink. The model is 0.2–0.6 ppbv
too low. The comparison near the surface is subject to
uncertainty because of land-ocean contrast (the model has
more continental influence than the observations, but this
may be due to numerical diffusion). The free tropospheric
observations in ITCT 2K2 are similar in magnitude to
the TRACE-P observations for the northeast quadrant
(Figure 6), but the model shows lower values in ITCT
2K2 because of the longer distance from the Asian conti-
nent. Atmospheric production makes a relatively large
contribution (0.2 ppbv) to the simulated concentrations in
ITCT 2K2, reflecting the strong radiation and low NOx
concentrations over the subsiding northeast Pacific. A
possibility, discussed further below, is that this atmospheric
source in the model is too weak.
[51] Observations from the PEM-Tropics B campaign
over the South Pacific are typically 0.6–1.2 ppbv with
little horizontal or vertical structure (Figure 9). These
values are remarkably high considering the remoteness
from land. The model is about a factor of 2 too low,
averaging about 0.4–0.5 ppbv with little structure. A
unique feature of model results for this region is that
atmospheric production is the dominant source, contribut-
ing 0.2–0.4 ppbv. This source is favored by the high UV
radiation (stimulating CH3O2production), low NOxcon-
centrations (suppressing competition from the CH3O2+ NO
reaction), and low CO/CH4 ratio (resulting in a high
CH3O2/HO2ratio). The model underestimate in this region
suggests that atmospheric production in the model is too
low. A doubling of this source would largely correct the
discrepancy and would be consistent with the featureless
character of the observations.
[52] We previously discussed in section 2 the uncertain-
ties in computing the atmospheric source of methanol from
CH3O2 reactions. Examination of model results for the
PEM-Tropics (B) region indicates that self-reaction (R1)
accounts for 5–10% of the CH3O2sink, reaction (R4) with
NO for 20%, and reaction (R3) with HO2 for the rest;
reaction (R2) is negligible. CH3OOH produced by (R3) has
an atmospheric lifetime of about a day against losses by
reaction with OH and photolysis:
CH3OOH þ OH ! CH3O2þ H2O
ðR5aÞ
CH3OOH þ OH ! CH2O þ OH þ H2O
ðR5bÞ
CH3OOH þ hn ! CH3O þ OH
ðR6Þ
[53] Reaction (R5a) recycles CH3O2 while reactions
(R5b) and (R6) do not. For typical atmospheric conditions,
(R5a) accounts for about 50% of CH3OOH loss and (R5b)
and (R6) each for about 25%. Assuming chemical steady
state for CH3OOH, the effective sink of CH3O2from (R3) is
determined by k3(1 ? f), where the CH3O2 recycling
efficiency f = k5a/(k5a + k5b + k6) is about 50%. Jet
Propulsion Laboratory (JPL) [2003] gives uncertainty esti-
mates of 30% for k3, 40% for k5, and 50% for k6(it gives no
uncertainty estimate for the recommended branching ratio
70/30 for k5a/k5b). Thus the overall uncertainty on the
effective CH3O2sink from (R3) could be a factor of two.
In addition, Martin et al. [2002] showed that the NO
concentrations simulated by GEOS-CHEM over the South
Pacific are about 50% higher than the PEM-Tropics (B)
observations. Combination of these two factors could easily
allow a doubling of the source of methanol from (R1),
considering that this source is quadratic in the CH3O2
concentration. Photochemical model calculations by Olson
et al. [2001] along the PEM-Tropics (B) flight tracks
indicated a 50% overestimate of observed CH3OOH, which
would be consistent with an overestimate of (R3). They also
indicated a factor of 2 underestimate of CH2O (common
product of CH3O2degradation pathways) [Heikes et al.,
2001]. As discussed by Heikes et al. [2001], the latter
discrepancy suggests a major contribution from VOCs other
than methane to the supply of CH3O2.This would provide
an alternate explanation for increasing the methanol source
from the CH3O2self-reaction.
5.
Budget From Atmospheric Observations
Discussion: Constraints on the Methanol
[54] Observed atmospheric concentrations of methanol
are consistent with the view that plant growth provides
the principal global source. Laboratory and field studies
indicate a factor of 2–3 decrease in this source from young
to mature leaves [McDonald and Fall, 1993; Nemecek-
Marshall et al., 1995; Karl et al., 2003]. We find that this
is consistent with the observed seasonal variation of meth-
anol concentrations at northern midlatitudes. A global plant
growth source of 128 Tg yr?1, computed in our model using
the NPP-based algorithm of Galbally and Kirstine [2002],
provides an overall unbiased simulation at northern midlat-
itudes. The tropics are the largest global contributor to this
source but observations there are few. Data for tropical
South America [Williams et al., 2001; Kesselmeier et al.,
2002] suggest that our model source is too high, while data
for Costa Rica [Karl et al., 2004] are consistent with the
model. More work is clearly needed to improve understand-
ing of methanol emission from tropical land ecosystems.
The model also predicts high surface air concentrations
(>10 ppbv) over Siberia in July, reflecting a large seasonal
plant growth source as well as a long continental fetch.
Measurements in this region would be of great value.
[55] The net flux of methanol from the terrestrial bio-
sphere in the model is 96 Tg yr?1including 128 Tg yr?1
from plant growth and 23 Tg yr?1from plant decay,
balanced by 55 Tg yr?1dry deposition to land (Table 1).
The atmospheric observations offer limited constraints for
separating these three different terms. The plant growth
source appears to be responsible for the observed decline of
D08303
JACOB ET AL.: GLOBAL METHANOL BUDGET
14 of 17
D08303

Page 15

concentrations from spring to fall, while the deposition sink
appears to be responsible for the observed diurnal variation
of concentrations as well as the decline at northern latitudes
over the course of the winter.
[56] Observed concentrations of methanol over the South
Pacific during the PEM-Tropics B aircraft mission (0.6–
1.2 ppbv) are much higher than would be expected from the
terrestrial vegetation source and show little vertical or
longitudinal structure. Diffuse atmospheric production of
methanol from the CH3O2self-reaction (R1), where CH3O2
originates mainly from methane oxidation, is the principal
model source in the region and contributes a 0.2–0.4 ppbv
model background. Our computed global magnitude of this
source (38 Tg yr?1) is larger than previous estimates (18–
31 Tg yr?1) but still would need to be doubled to approach
the South Pacific observations. Methanol observations over
the remote north tropical and subtropical Pacific (TRACE-P
and ITCT 2K2 aircraft missions) also suggest that atmo-
spheric production of methanol from (R1) in the model is
too low.
[57] Computation of the methanol source from (R1) in
global tropospheric chemistry models is affected by uncer-
tainties in the concentration of CH3O2and in the branching
ratio of CH3O2reactions. A doubling to 50–100 Tg yr?1
does not appear to be inconsistent with independent con-
straints, and is specifically within the constraints offered by
ancillary PEM-Tropics B observations of CH3OOH and
CH2O. It could reflect errors in kinetic rate constants, or
the presence of VOCs other than methane in the remote
marine atmosphere to provide sources of CH3O2. Some
observations of aldehydes in remote marine air [Heikes et
al., 2001; Singh et al., 2001, 2004] suggest the latter, and
imply a more active organic photochemistry than is cur-
rently described in models, while other observations are
more consistent with current understanding [Fried et al.,
2003]. Measurements of HO2 and total peroxy (HO2 +
CH3O2 + RO2) radical concentrations by the chemical
amplifier method during TRACE-P indicated concentration
ratios of organic peroxy radicals to HO2that are consistent
with current models [Cantrell et al., 2003], but there is
substantial uncertainty in these measurements. Direct mea-
surements of CH3O2concentrations in the remote atmo-
sphere are needed.
[58] Our model yields a global mean atmospheric lifetime
for methanol of 7 days, with losses from gas-phase reaction
with OH (63% of total loss), dry deposition to land (26%),
wet deposition (6%), and uptake by the ocean (5%).
Aqueous-phase oxidation by OH(aq) in clouds is negligible.
The general ability of the model to reproduce the observed
variability of methanol concentrations (cf. Figure 7) sug-
gests a relatively narrow range of uncertainty in the lifetime,
5–10 days. Chemical loss of methanol by reaction with OH
is uncertain by only about 30% [Jimenez et al., 2003] and
most likely represents the dominant global sink for metha-
nol. The low methanol concentrations observed at high
northern latitudes in winter imply an additional nonphoto-
chemical sink, which we attribute in the model to dry
deposition to land. This attribution is supported by obser-
vations of methanol fluxes and of the diurnal variation of
methanol concentrations at vegetated sites, although there is
considerable variability from site to site. Heterogeneous loss
in aerosols and clouds could provide an alternate explana-
tion for the observed wintertime depletion of methanol but
is not supported by independent evidence. Aircraft obser-
vations of vertical profiles over the oceans suggest that
ocean uptake is only a weak sink, although ship data from
the AOE-2001 summertime Arctic campaign suggest the
possibility of rapid local uptake. Wet deposition, con-
strained by the Henry’s law solubility for methanol, is only
a weak sink.
[59] Our best estimate of the global source of methanol
(after allowing for doubling of atmospheric production to
76 Tg yr?1) is 240 Tg yr?1. Singh et al. [2004] estimated a
much lower global source, 110 Tg yr?1, by scaling the
global acetone source of 95 Tg yr?1from Jacob et al.
[2002] to the mean acetone/methanol concentration ratio of
2.1 g g?1that they measured in TRACE-P and to the ratio
of lifetimes of acetone (15 days, from Jacob et al. [2002])
and methanol (9 days, from Heikes et al. [2002]). However,
the concentration ratio measured in TRACE-P is lower than
the expected global mean because of the large regional
anthropogenic contribution to acetone [Jacob et al., 2002]
and the relatively weak biogenic methanol emission from
East Asia at that time of year. A methanol source of 110 Tg
yr?1in our model would result in a low bias relative to
observations. Singh et al. [2004] acknowledge that their
emission estimate is subject to large uncertainty.
6.Implications for Atmospheric Chemistry
[60] Our budget implies a minor but nonnegligible role
for methanol in global tropospheric chemistry. Oxidation of
methanol by OH produces CH2OH (major pathway) and
CH3O (minor), which both go on to produce CH2O and
HO2[Jimenez et al., 2003]. The global tropospheric rate of
methanol oxidation calculated in our model on a per-carbon
basis, 47 Tg C yr?1, amounts to 12% of the corresponding
value of 380 Tg C yr?1for methane oxidation [IPCC,
2001]. Methanol provides a significant source of CH2O
both in the continental boundary layer [Riemer et al., 1998;
Palmer et al., 2003a] and in the remote troposphere [Heikes
et al., 2001]. The production of CO from methanol oxida-
tion amounts to 4–6% of the global CO source of 1800–
2700 Tg yr?1(Duncan et al., submitted manuscript, 2004).
Inclusion of methanol in GEOS-CHEM at the levels simu-
lated here decreases the global tropospheric OH concentra-
tion by 2%, an effect similar to that reported previously by
Tie et al. [2003]. Regional variations in this effect on OH
are discussed by Tie et al. [2003] and are relatively small.
[61] Acknowledgments.
Atmospheric Chemistry Program of the U.S. National Science Founda-
tion. H. B. Singh acknowledges support from the NASA IDS program.
A. Wisthaler and A. Hansel thank Caroline Leck (MISU), the coordi-
nator of the Atmospheric Program of AOE-2001, and Ulrich Poeschl
(TU-Munich), the coordinator of the Zugspitze SCAVEX Program. The
GEOS-CHEM model is managed at Harvard University with support
from the NASA Atmospheric Chemistry Modeling and Analysis Pro-
gram. We thank the reviewers for their very useful comments.
This work was funded at Harvard by the
References
Andreae, M. O., and P. Merlet (2001), Emission of trace gases and aerosols
from biomass burning, Global Biogeochem. Cycles, 15, 955–966.
Atlas, E. L., B. A. Ridley, and C. A. Cantrell (2003), The Tropospheric
Ozone Production about the Spring Equinox (TOPSE) Experiment:
Introduction, J. Geophys. Res., 108(D4), 8353, doi:10.1029/
2002JD003172.
D08303
JACOB ET AL.: GLOBAL METHANOL BUDGET
15 of 17
D08303