global modeling of atmospheric methane sources and sinksinez/msri-ncar_carbonda/...methane is the...

Global modeling of atmosphericmethane sources and sinks

Sander Houweling

De omslag toont de bewolkingverdeling over de aarde op een willekeurig moment, afgeleiduit berekeningen met het ECMWF model. Deze verdeling illustreert de horizontale mengingin de troposfeer (met dank aan het KNMI).

Global modeling of atmosphericmethane sources and sinks

Mondiale modellering van bronnen en verwijderingprocessenvan atmosferisch methaan

(met een samenvatting in het Nederlands)

PROEFSCHRIFT

TER VERKRIJGING VAN DE GRAAD VAN DOCTOR AANDE UNIVERSITEIT UTRECHT OP GEZAG VAN DE RECTORMAGNIFICUS, PROF. DR. H. O. VOORMA, INGEVOLGE HETBESLUIT VAN HET COLLEGE VOOR PROMOTIES IN HET

OPENBAAR TE VERDEDIGEN OP 7 FEBRUARI 2000DES NAMIDDAGS OM 16:15 UUR

door

Sander Houweling

geboren op 18 maart 1970, te Hilversum.

Promotor: Prof. Dr. J. Lelieveldfaculteit Natuur- en Sterrenkunde, Universiteit Utrecht

Co-promotor: Dr. F. J. Dentenerfaculteit Natuur- en Sterrenkunde, Universiteit Utrecht

Paranimfen: A. M. Bleeker en J. Bosman

Dit proefschrift is mede moglijk gemaak door financiele steun van het Nederlands OnderzoekProgramma (NOP-II).

ISBN: 90-393-2296-1Universal Press, 1999

He deals the cards to find the answer,the sacred geometry of chance,the hidden law of a probable outcome,the numbers lead a dance.

Sting

Contents

1 Introduction 31.1 The role of methane in the atmosphere . . . . . . . . . . . . . . . . .. . . . 31.2 Sources of methane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51.3 Atmospheric chemistry and methane removal . . . . . . . . . . .. . . . . . 71.4 Chemistry transport models . . . . . . . . . . . . . . . . . . . . . . . .. . . 81.5 Inverse modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

1.5.1 Mathematical concepts . . . . . . . . . . . . . . . . . . . . . . . . . 101.5.2 Application to CH4 . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

1.6 Research aims and thesis outline . . . . . . . . . . . . . . . . . . . .. . . . 14

2 The impact of nonmethane hydrocarbon compounds on tropospheric photochem-istry 172.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182.2 Model description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .19

2.2.1 NMHC chemistry representation . . . . . . . . . . . . . . . . . . .. 192.2.2 TM3 description . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

2.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272.3.1 Ozone . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272.3.2 Carbon monoxide . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332.3.3 The hydroxyl radical . . . . . . . . . . . . . . . . . . . . . . . . . . 342.3.4 Nitrogen containing NMHC . . . . . . . . . . . . . . . . . . . . . . 362.3.5 Nonmethane hydrocarbons . . . . . . . . . . . . . . . . . . . . . . . 41

2.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 432.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

3 Inverse modeling of methane sources and sinks using the adjoint of a globaltransport model 493.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 503.2 Inversion method and unknowns . . . . . . . . . . . . . . . . . . . . . .. . 513.3 Model description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .543.4 Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 553.5 A priori assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .563.6 Chemical methane loss . . . . . . . . . . . . . . . . . . . . . . . . . . . . .583.7 Methane inversion results . . . . . . . . . . . . . . . . . . . . . . . . .. . . 603.8 Sensitivity tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 713.9 Case study: Emissions from southeast Asia . . . . . . . . . . . .. . . . . . 75

1

2 CONTENTS

3.10 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 773.11 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

4 The modeling of tropospheric methane; how well can point measurements bereproduced by a global model? 814.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 824.2 Model description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .83

4.2.1 Chemistry transport model . . . . . . . . . . . . . . . . . . . . . . .834.2.2 Methane simulations . . . . . . . . . . . . . . . . . . . . . . . . . . 84

4.3 Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 864.4 Interhemispheric exchange rate . . . . . . . . . . . . . . . . . . . .. . . . . 874.5 Results and discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 87

4.5.1 In situ measurements . . . . . . . . . . . . . . . . . . . . . . . . . . 904.5.2 Flask measurements . . . . . . . . . . . . . . . . . . . . . . . . . . 934.5.3 Wind sector selection . . . . . . . . . . . . . . . . . . . . . . . . . . 964.5.4 Non-local air selection . . . . . . . . . . . . . . . . . . . . . . . . .974.5.5 Influence of wetland emissions . . . . . . . . . . . . . . . . . . . .. 100

4.6 Summary and conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . .105

5 Simulation of pre-industrial atmospheric methane to constrain the global sourcestrength of natural wetlands 1115.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1125.2 Model description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1135.3 Pre-industrial sources . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . 1145.4 Pre-industrial sinks . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . 1175.5 Isotopic ratios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1195.6 Results and discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 121

5.6.1 Pre-industrial methane . . . . . . . . . . . . . . . . . . . . . . . . .1215.6.2 Methane increase during industrialization . . . . . . . .. . . . . . . 1245.6.3 δ13C-CH4 during industrialization . . . . . . . . . . . . . . . . . . . 125

5.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

6 General discussion and future perspectives 129

Appendix A Gas-phase chemistry mechanism 133

Appendix B NOAA measurement stations 139

Bibliography 159

Summary 161

Samenvatting 165

Nawoord 169

Curriculum Vitae 171

Chapter 1

Introduction

Methane is the most abundant hydrocarbon in the Earths atmosphere. The first section ofthis introduction highlights the importance of atmospheric methane (CH4), and the influenceof human activities on this trace gas. Subsequently, a briefoverview is given of the mainprocesses that introduce methane in the atmosphere, and theatmospheric photochemistrythat leads to its removal. Further, a short introduction is given to inverse modeling and howthis technique has been applied to estimate sources and sinks of CH4. Finally, the scientificaim of this thesis is formulated and an overview is given of the research activities that arepresented.

1.1 The role of methane in the atmosphere

Polar ice caps provide convincing evidence of a drastic increase of the atmospheric methaneconcentration over the last centuries. Measurements of airtrapped in the ice reveal thatbefore industrialization the globally averaged methane concentration was∼700 nmol mol−1,being about a factor of 2.5 lower than today [Etheridge et al., 1998]. Moreover, from theAntarctic ice core record we learn that presently CH4 concentrations are the highest of thepast 420,000 yr [Petit et al., 1999]. Since 1983, CH4 concentrations are monitored on aroutine basis at a number of observatories distributed overthe globe [Dlugokencky et al.,1994b]. These measurements show that the growth rate of methane slowly decreased overthe last two decades. The CH4 trend also shows substantial variability on a time scale ofa few years. Particularly, in 1992 the CH4 growth rate abruptly decreased, followed by apartial recovery in subsequent years (1993,1994) [Dlugokencky et al., 1996]. This growth rateanomaly is generally attributed to the eruption of Mt. Pinatubo on 15 June 1991, althoughthe mechanisms associated with this event remain uncertain.

The increased level of methane has important implications for the energy balance andthe chemical composition of the atmosphere. These effects can largely be explained by thechemical and physical properties of the methane molecule. As an important characteristic,methane absorbs and emits longwave radiation at the wavelengthsλ=3.31µm and 7.66µm.The energies of photons of these wavelengths correspond to energy differences between dif-ferent vibrational states of the CH4 molecule [Herzberg, 1945]. Hence, these photons can beabsorbed by a CH4 molecule, which leads to a transition of the CH4 bend-vibrational stateto one of a higher energy. Moreover, the vibrational transitions are associated with rotational

3

4 Introduction

100

80

60

40

20

0

Ab

so

rba

nce

(%

)R

ad

iative

flu

x (

no

rma

lize

d)

(a)

(b) 2 3 4 5 6 7 8 910 15 20 30 50 100

288 K

CH4 CH4

wavelength (µm)

Figure 1.1: (a) The normalized blackbody emission spectrum for the Earth as a function ofwavelength. (b) The fraction of radiation absorbed while passing from the surface to the topof the atmosphere as a function of wavelength. Absorption peaks of CH4 are indicated (theabsorption atλ=7.7µm partly overlaps that of N2O) [Goody and Yung, 1989].

transitions. This process is reversed by the emission of a photon or by the transfer of energyto other molecules by collision.

The Earth loses energy to space by long wave radiation, emitted at wavelengths as de-termined by the local temperature. This energy transfer is most efficient at wavelengths thatare intensively emitted by the Earth, and at wavelengths that are in a relatively transparentpart of the absorption spectrum of the atmosphere. As can be seen in Figure 1.1 both criteriaare met betweenλ=8µm and 12µm, known as the atmospheric infrared window. Because thelongwave absorption of CH4 at λ= 7.66µm occurs within the atmospheric window region,CH4 is an important greenhouse gas [Herzberg, 1945].

From a chemical viewpoint methane can be regarded as a reduced form of carbon. Sincethe oxidation potential of the atmosphere is relatively high, reduced carbon is oxidized insequences of chemical reactions, ultimately leading to carbon dioxide. Intermediate productsof such sequences influence the concentrations of other gases such as O3, OH, CO and H2O.For example, the increase of CH4 has contributed to the increase of tropospheric O3. This isimportant, because O3 is also a greenhouse gas, and because elevated levels of thistoxic gasnear the surface harm the biosphere. The chemical removal ofmany atmospheric constituentsis initiated by the reaction with OH radicals. As a result, this radical largely determines the at-mospheric residence time of many compounds that are emittedfrom the surface. Increases ofCH4 tend to decrease OH, and, since this also influences the CH4 lifetime, it introduces a pos-itive feedback. In the stratosphere, methane oxidation is an important source of stratospheric

1.2 Sources of methane 5

water vapor, which is important for the chemistry and the radiation of the ozone layer.The atmospheric abundance of CH4 is about a factor 200 smaller than that of CO2. Still it

contributes significantly to the enhanced greenhouse effect owing to a relatively high warmingefficiency. This efficiency is quantified by the Greenhouse Warming Potential (GWP), definedas the induced radiative forcing relative to CO2 (W kg−1/W kgCO−1

2 ) integrated over a certaintime period. If the direct (radiative) and indirect (chemical) contributions of CH4 are added,a GWP of 21 is calculated for a 100 year integration time. The contribution of CH4 to theenhanced greenhouse effect is estimated at 0.57 W m−2, or 22% of all greenhouse gases(36% of CO2) for the period 1850–1992 [Lelieveld et al., 1998]. Since the recent decline ofthe CH4 growth rate is poorly understood, future contributions aredifficult to predict.

In summary, methane plays an important role in chemical and radiative processes in theatmosphere. Therefore, to understand and quantify these processes, a better understanding ofthe global cycle of methane is needed. From a broader perspective, methane plays a key rolein important environmental issues, and an improved understanding of its sources and sinks iscrucial.

1.2 Sources of methane

Methane is formed by various different processes that can beclassified as thermogenic, py-rogenic, and microbial. Their combined global source strength is estimated at∼525–625Tg(CH4) yr−1 (Tg= teragrams or 1012 g) (see Chapter 4), as constrained largely by the at-mospheric CH4 budget. In comparison, the source strengths of individual processes are moreuncertain (see Chapter 3, Table 3.1). The contribution of anthropogenic activities varies persource category, being dominant for thermogenic (fossil energy) and pyrogenic sources. Theanthropogenic contribution to the microbial source is less(∼50%), although still importantin absolute terms, since this category makes up a large fraction of the total emission (∼75%).

Thermogenic methane is formed under high pressures and temperatures at several kilome-ters depths within the Earth, as a product of the thermal degradation and cracking of organicmaterial [Tissot and Welte, 1984]. Geological reservoirs of coal, gas, and oil are increasinglyexploited to meet our demand for energy. At each step along the line of mining, transmission,distribution, and processing of fossil fuels, fractions ofCH4 are released to the atmosphere.In particular, coal mining and the transmission of natural gas are important CH4 sources onthe global scale. For example, in underground coal mines large volumes of air are ventilatedfor safety reasons. In some countries, CH4 in the vented air is flared or burned for energy,but in most cases the CH4 is released to the atmosphere. In surface mines, CH4 escapesfrom exposed coal layers, especially, those that are disturbed by blast and mechanical oper-ations [Khalil, 1993]. The pipelines used to transmit and distribute natural gas constitute animportant source of methane to the atmosphere. Continuous emissions result from leakingcomponents such as valves and seals. Occasionally large amounts are released by mainte-nance procedures or accidents. Remaining industrial sources include fossil fuel combustion,production of cokes, processing of iron and steel, and petroleum refining. Although, individ-ually, these sources are of relatively minor importance, collectively they may contribute 15Tg(CH4) yr−1 [Khalil, 1993]. All anthropogenic emissions, and natural sources such as vol-canic activity and crustal movements, are associated with significant uncertainties, however,their sum is relatively well constrained by the global14C budget. Atmospheric measure-ments of14CH4 indicate that the total fossil, or radioactively “dead” methane source amounts

6 Introduction

90± 66 Tg(CH4) yr−1 [Quay et al., 1999]. For a further discussion on isotopes see Chapter 5.

Pyrogenic methane is produced by the incomplete combustionof organic (non-fossil) ma-terial, by, for example, the burning of forests, savannas, agricultural wastes, and biofuels. Inparticular, during the dry season in the tropics large areasare burned to facilitate the cultiva-tion of crops. This so called shifting cultivation contributes about 40% to the global biomassburning emission of CH4. In total, 40±30 Tg(CH4) is released by biomass burning peryear, of which approximately 90% is confined to the tropics [Lelieveld et al., 1998;Hao andWard, 1993]. Temperate and boreal forest fires account for the remainder. The CH4 yield offires is difficult to estimate since the fraction of carbon released as CH4 is dependent on thefuel type, burning phase (open fire or smoldering), and season. For example,Hao and Ward[1993] estimated that tropical forest fires produce about 4 times more CH4 than savannahfires, although a larger amount of savannah biomass is burned. In addition, the interannualvariability of these sources is relatively strong, closelyrelated to climatic variabilities. Forexample, in 1997 large areas were burned over Indonesia caused by an unusual drought inconnection with El Nino.

Methane is produced by methanothropic bacteria that use acetate or CO2 and H2 as elec-tron acceptors in their metabolism. By this process, natural wetlands such as swamps, mires,and bogs produce large quantities of CH4 (∼145 Tg(CH4) yr−1 [Lelieveld et al., 1998]).CH4 emissions from rice paddies (cultivated wetlands) contribute about 60 Tg(CH4) yr−1

[Olivier et al., 1999], and are explained by the same microbial mechanism. Rice fields areparticularly efficient CH4 sources, owing to the tropical climate of important rice producingcountries and the regulation of water by irrigation. Water management, rice variety and theamount and type of manure that is applied, are important factors that determine the sourcestrength of rice fields [Denier van der Gon, 2000a, b]. The total CH4 emission from riceagriculture is difficult to estimate since these factors areall highly variable.

Methanothropic bacteria that are active in the interface between oxic and anoxic layersconsume methane. As a result, the net amount of methane that is emitted largely dependson the balance between microbial production and consumption, which is determined by theavailability of oxygen and the efficiency of different transport pathways. Besides wetlands,the same processes play an important role in the decomposition of waste, where the degreeof landfill coverage determines to what extent anaerobic fermentation takes place and theproduced CH4 can reach the atmosphere [Bergamaschi et al., 1998].

Important amounts of CH4 are produced by ruminants. The digestive tracts of these an-imals are developed to ferment cellulose and hemicelluloses by a symbiosis with bacteria.These bacteria produce fatty acids that are utilized by the host, and H2 and CO2 that areconverted to CH4 by methanogenic bacteria [Khalil, 1993]. The CH4 production per animaldepends on many factors, such as nutrition level and the digestability of the food [Crutzenet al., 1986]. Further, CH4 is produced by anaerobic fermentation of animal excreta, depen-dent on its use and the method of storage. The global CH4 emission of domestic ruminantsis estimated at 110±25 Tg(CH4) yr−1, including a contribution of animal waste of 30±5Tg(CH4) yr−1 [Lelieveld et al., 1998]. The emission from wild animals, such as deer, moose,gazelle, and wildebeest is estimated at∼5 Tg(CH4) yr−1 [Crutzen et al., 1986].

Further, a significant amount of CH4 is produced by termites (20±10 Tg(CH4) yr−1

[Lelieveld et al., 1998]), largely related to the tropics where they are most abundant. Termitesplay an important role in the degradation of organic matter and humus formation. Theirability to digest polysaccharidic plant compounds is largely due to symbiotic relationships

1.3 Atmospheric chemistry and methane removal 7

with anaerobic bacteria, which also explains the emission of CH4 [Khalil, 1993].

1.3 Atmospheric chemistry and methane removal

The high abundance of molecular oxygen in the atmosphere (21%) and its relatively highredox potential suggest that this gas is important for the oxidative removal of other gases.Nevertheless, direct reactions of molecular oxygen with gases such as CO, CH4, CH2O, andmore reactive hydrocarbons are unimportant under atmospheric conditions, owing to the largeactivation energies required. Instead, most reactions aredriven by highly reactive radicalssuch as OH and HO2, although these radicals are still closely linked to molecular oxygen.Because of the importance of these radicals, the chemical reaction mechanisms occurring inthe atmosphere can be characterized as radical chain reactions, including radical initiation,propagation, and termination.

An important process of radical initiation in the troposphere is the photolysis of O3 andthe reaction of the resulting excited oxygen radical O(1D) with H2O,

O3hν−→ O2 +O(1D) (R1)

O(1D)+H2O −→ 2 OH (R2)

Methane is oxidized by a series of propagation reactions, resulting in the production offormaldehyde (CH2O) as the first stable product. Under high NOx conditions this chain isrepresented by,

CH4 +OH −→ CH3 +H2O (R3)CH3 +O2 −→ CH3O2 (R4)CH3O2 +NO −→ CH3O+NO2 (R5)CH3O+O2 −→ CH2O+HO2 (R6)HO2 +NO −→ OH+NO2 (R7)

CH4 +2 O2 +2 NO −→ CH2O+2 NO2 +H2O (R8)

CH2O is further oxidized by photolysis and reaction with OH yielding CO, and, after subse-quent oxidation by OH, CO2. The photolysis of CH2O is also an important source of radicals.As a result, under high NOx conditions the oxidation of CH4 to CO2 leads to a net productionof radicals. Thus, under these conditions an increase of CH4 leads to an increase of OH.

The level of NOx is crucial since it determines how effectively HO2 and CH3O2 arerecycled to OH, i.e. how many times reactions (R3) to (R7) arerepeated until the chainreaction is terminated. Under low NOx conditions reaction (R5) is significantly reduced bycompetition with the termination reactions

CH3O2 +HO2 −→ CH3O2H +O2 (R9)CH3O2 +CH3O2 −→ CH2O+CH3OH +O2, (R10)

which effectively shortens the reaction chain length. The relative importance of the pathways(R5) and (R9, R10) is determined by the ratio of reactive C andN. As a result, the chemicalresponse to a change of the CH4 concentration is highly non-linear, and will be different fordifferent parts of the atmosphere. In the troposphere, termination reactions such as (R9) and(R10) are so important that, on average, increases of CH4 lead to a reduction of OH.

8 Introduction

It should be realized that the oxidation of CH4 to CH2O, as outlined above, representsonly a small fraction of the reactions occurring in the troposphere. For example, hydrocar-bons with longer carbon chain lengths (non-methane hydrocarbon compounds or NMHC),introduce an enormous amount of complexity. Although the oxidation of single carbon atomsof NMHC often largely resembles the oxidation of methane, several alternative pathways playa role. As an important example, olefinic carbon bonds (C=C) in molecules such as ethene,propene and isoprene, react with O3 and NO3 besides OH. The O3 and NO3 pathways areparticularly important during nighttime, as opposed to theOH pathway. O3 has a sufficientlylong lifetime to “survive” through the night, while OH must be sustained by photolysis re-actions during daytime. As opposed to OH, NO3 is destroyed by photolysis reactions, and,therefore, in absence of these sinks during nighttime its concentration increases. As a result,the oxidation of olefinic NMHC continues during the night, while the oxidation of aliphaticcompounds such as CH4 is restricted to day time. It should be noted that the radicals pro-duced by these O3 and NO3 reactions lead to some OH production during the night, that,according to recent measurements, may become quite important under polluted conditions[Apel et al., 1999].

Another important NMHC reaction pathway leads to the production of organic nitrates;compounds that can be regarded as chemical reservoirs of NOx. An important example isperoxyacetyl nitrate (PAN), produced by

CH3C(O)O2 +NO2 ⇀↽ CH3C(O)O2NO2 (“PAN”). (R11)

The decomposition of PAN is strongly temperature dependent, and, as a result, its lifetimevaries between a few days in the boundary layer and a few months in the upper troposphere.Once transported to the free troposphere, or produced at higher altitudes, PAN can travellong distances. In this way, NOx emitted over polluted centers of the world can reach remoteregions at larger distances than its own lifetime would permit. This process has implicationsfor methane also, because it influences the global distribution of the OH radical.

Until now, we specifically focussed our attention on tropospheric photochemistry. Thisis justified by the fact that most of the atmospheric methane oxidation takes place in the tro-posphere (∼90%). The remaining part is oxidized in the stratosphere (10–50 km altitude),where, except oxidation by OH, reactions of methane with Cl and O(1D) radicals are impor-tant. Although these reactions occur in the troposphere as well, abundances of Cl and O(1D)are too low to make significant contributions. In the stratosphere, CFC oxidation significantlyenhances Cl and more intense UV-radiation and relatively high O3 mixing ratios (O3-layer)enhance O(1D). The photolysis of CH4 increases with altitude too, although this processremains relatively unimportant.

1.4 Chemistry transport models

Most of the results presented in this thesis are derived froman off-line chemistry transportmodel (CTM). The use of such models aims to realistically simulate processes that occur inthe real atmosphere, with the goal to quantify the global distribution and time evolution ofcertain atmospheric constituents. Since computational resources are limited, this can onlybe done if the real world is rigorously simplified by the model. How realistic these stronglysimplified models are depends on the careful representations of the most important processes,and of discretizations that resolve crucial temporal and spatial scales. Note that “off-line”refers to the fact that dynamic transport equations are not solved within the model. Instead,

1.4 Chemistry transport models 9

horizontal and vertical air mass fluxes are prescribed, usually based on the output of GeneralCirculation Models (GCMs).

To validate the performance of a CTM, observed concentrations of various trace gases areused. Generally, these measurements are taken at certain points in space and time. The spatialand temporal scales that are represented by the measured concentrations are determined bythe variability of the examined trace gas. Since the horizontal resolution of global 3-D modelsis generally quite coarse (several degrees) the applicability of certain measurements is limited,for example, over source regions, where this variability may occur on smaller scales thanresolved by the model.

Tracers such as CFC-11, SF6, and85Kr are of particular interest in the context of modelvalidation since they can be used to test an important model property; the rate of interhemi-spheric exchange. These gases are well suited for this purpose since their sources are solelyanthropogenic and relatively well quantified. In addition,their atmospheric residence times,ranging fromτ=15.5 yr for 85Kr and >800 yr for SF6 [Levin and Hesshaimer, 1996], arelong compared with the e-folding time of cross-hemispherictransport (∼1 yr). Atmosphericconcentrations of these gases are monitored on a regular basis at stations that are distributedover different latitudes.

To validate transport properties on smaller time scales222Rn is used (τ=5 days). Thisradio-isotope is formed by the nuclear decay of226Ra in soils, and, as a consequence, itssources are limited to the continents only. Measurements of222Rn at marine stations clearlyidentify air-masses of continental origin (so called radonstorms), which can be used to testthe advective transport and the vertical mixing in the model. Further, measurements of222Rnhelp to identify source regions of other simultaneously analyzed tracers (see Chapter 4 andDentener et al.[1999]).

To validate the model chemistry, measurements of various gases can be used, such asO3, CO, NOx, NMHC, and PAN. In this thesis, the validation of OH receivesspecial at-tention, since this compound is of great importance to our understanding of CH4. Directmeasurements of this radical are, however, very complicated owing to its short lifetime (or-der seconds) and high variability. Recently, spectroscopic techniques for measuring OH havebecome operational, but no measurements on a routine basis have been reported yet.

Large scale averaged OH concentrations can be derived indirectly using measurements ofmethyl chloroform (CH3CCl3) [Prinn et al., 1995;Krol et al., 1998]. CH3CCl3 has almostexclusively industrial sources, except for a relatively unimportant contribution of biomassburning. The main sink of CH3CCl3 is by reaction with OH radicals in the troposphere,resulting in a tropospheric turnover time of 4.7 yr. Becauseof its relatively long lifetime,CH3CCl3 has so far mainly provided a constraint on the globally averaged level of OH. Atpresent, the CH3CCl3 concentrations in the atmosphere decrease drastically, caused by rig-orous emission reductions by the implementation of the Montreal Protocol of substancesthat deplete the ozone layer. It can be shown that if the sources become relatively unimpor-tant, the CH3CCl3 concentrations become increasingly sensitive to the OH distribution. Asa consequence, in the near future CH3CCl3 may provide a better constraint on, for example,the north-south distribution of OH. Besides CH3CCl3, other tracers as CHF2Cl (HCFC-22),CH2Cl2, and14CO could potentially provide constraints on OH, although, at present, this islimited by the amount of measurements [Spivakovski et al., 1999].

10 Introduction

1.5 Inverse modeling

Sources and sinks of methane can be quantified using two fundamentally different approaches;up-scaling and inverse modeling. The first aims at integrating all existing information aboutsource and sink processes, including results of local field experiments and statistics on re-gional or national levels. To obtain a global coverage data-gaps are filled by inter- and extrap-olation of the available information. The inverse modelingapproach is to use measurementsof atmospheric methane mixing ratios, and determine sourceand sink distributions that leadto optimal agreements between model simulations and these observations. As we will seelater, in spite of the different sources of information thatare utilized by each method, theresulting source estimates are not necessarily independent.

This subsection briefly describes some fundamental concepts that are used in this thesis tosolve inverse problems. A full description would require a discussion of Bayesian theory, thatprovides the mathematical foundation for our method. Sincethe resulting equations are stillrelatively transparent and illustrative, instead, I focusthe discussion on their interpretation.Instructive introductions to Bayesian theory are given by,for example,Menke[1989] andTarantola[1987]. The second part of this section deals with the application of inverse theoryto the determination of CH4 sources and sinks.

1.5.1 Mathematical concepts

Suppose, we want to improve the model simulated concentrationCmod by utilizing the ob-served concentrationCobs. Generally, both model results and measurements are known witha limited certainty. To optimally benefit from both pieces ofinformation, the updated a pos-teriori estimateCpost is defined as a weighted average ofCmod andCobs,

Cpost =a

a+bCmod+

ba+b

Cobs (1.1)

wherea andb express our confidence in the estimatesCmod andCobs, respectively. Differentmethods can be used to quantify the weightsa andb. From Bayesian theory it follows thatif the probability distributions of the estimates ofC (Cmod andCobs) are Gaussian, thenCpost

is also Gaussian shaped. If we assume Gaussian distributions, the mean ofCpost (Cpost) is atthe minimum of the cost functionJ , defined as,

J (C ) =12(

1

σ2obs

(C−Cobs)2 +

1

σ2mod

(C−Cmod)2) (1.2)

whereσ2obs andσ2

mod represent the variances ofCobs andCmod, respectively. SinceCpost isat the minimum of equation 1.2 this requires minimal squareddeviations ofCpost from Cobs

andCmod in units of their correspondingσ-uncertainties. Therefore, this method, which isgenerally used in optimal interpolation, is known as the least square criterion.

In inverse modeling we follow a similar approach, with the exception that instead ofconcentrationsC we update fluxesf (sources and/or sinks). There is a causal relationshipbetween fluxes and concentrations of any atmospheric constituent, defined by its transportthrough the atmosphere (T), expressed by

C = T f (1.3)

1.5 Inverse modeling 11

Note that equation 1.3 assumes a linear relationship between concentrations and the fluxes,which is satisfied to a good approximation for many sources, but, in case of methane, is nottrue for the photochemical sink. Here, for simplicity we will assume linearity (see Chapter3 for the treatment of non-linearities). After substitution of equation 1.3 in equation 1.2 weobtain

J ( f ) =12(σ−2

obs(T f −Cobs)2 + σ−2

apr( f − fapr)2) (1.4)

where “apr” refers to a first-guess or a priori estimate of fluxf, and fpost is at the minimum ofJ ( f ). For linear problemsJ ( f ) is simply a quadratic function off. At the minimum ofJ ( f ),its first derivative (J ′( f )) is zero. Since theJ ′( f ) is a linear function off, this minimum caneasily be derived from the Newton formula, which yields

fpost = fapr− J ′( fapr)/J ′′. (1.5)

The first and second derivative ofJ follow readily from equation 1.4. Substitution in equation1.5 leads to

fpost = fapr +(Tσ−2obsT + σ−2

apr)−1Tσ−2

obs(Cobs−T fapr). (1.6)

From Bayesian theory it follows that the uncertainty of the aposteriori (post) flux equals1/J ′′, or

1

σ2post

= T1

σ2obs

T +1

σ2apr

. (1.7)

Equation 1.7 implies that the uncertainty offpost is large for a cost functionJ with a broadminimum, and small for a (well defined) sharp minimum. Note that equation 1.7 resemblesthe well known expression for the sum of 2 parallel resistances. The transport operators onth right hand side relate aσ change of concentration to aσ change of source as determinedby atmospheric transport. With parallel resistances, the total resistance is determined mainlyby the smallest resistance, and the sum of 2 resistances is always smaller than the individ-ual resistances. Similarly, by combining two sets of information - measured concentrationsand first-guess knowledge - we always reduce the uncertaintyof the estimated (a posteriori)flux, and the a posteriori uncertainty is mostly determined by the source of information thatconstrains the a posteriori source most efficiently.

So far, we only discussed a problem of a single flux and a singlemeasured concentra-tion. These concepts can be generalized, however, to problems with many fluxes and mea-surements (see Chapter 3). The vector/matrix equations that apply to the multi-dimensionalproblem are highly similar to those just discussed.

1.5.2 Application to CH4

In the past, a number of studies have been devoted to CH4 source and sink quantification byinverse modeling [Brown, 1993, 1995;Kandlikar, 1997;Hein et al., 1997;Saeki et al., 1998].These studies, including the one presented in this thesis, all apply fairly similar optimiza-tion procedures (see Chapter 3). Also the applied measurement stations are rather similar,however, this thesis puts a relatively strong emphasis on sites near continents. These mea-surements have been performed by the Cooperative Air Sampling Network of the National

12 Introduction

180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E

60S

30S

EQ

30N

60N

MLO

KUMMID

SHM

BRW

CBA

ALTMBC

SMO

P01

P02P03

P04

CGO

CMO UTA

NWR

ITN

KEY

BME

BMW

RPB

PSA

SPO

ICE

ZEPSTM

IZO

ASCSEY

SYO

BAL

GOZ

GMI

TAPQPC

UUM

SC1

SC7

MHT

Figure 1.2: Geographical locations of NOAA observational stations used in this thesis toestimate sources and sinks of CH4 by inverse modeling.

Oceanic and Atmospheric Administration (NOAA) Climate Monitoring and Diagnostics Lab-oratory (CMDL). At present, this network of observatories distributed over the globe consistsof ∼50 stations. Figure 1.2 shows the locations of the stations that were used in this thesis.In addition,Hein et al.[1997] used measurements of isotopic ratios (13CH4/13CH4). Here wedo not consider isotopes, since previous studies indicatedthat, given the number of availablemeasurements, isotopes introduce relatively weak additional constraints [Hein et al., 1997;Brown, 1995].

Important differences between the published inverse modeling studies of CH4 are thetype of model and the definition of the fluxes that are estimated. Brown[1993] applied a 2-Dmodel to improve the estimated net flux of CH4 per month and per latitudinal zone.Hein et al.[1997] used a 3-D model and updated the annual CH4 flux per process, such as ruminants andrice paddies, keeping their spatial and temporal distributions fixed. Alternatively,Hartley andPrinn [1993] used parts of continents in an investigation of sources of CFCs. These studiesdo not indicate advantages of using a certain definition overothers, and it remains unclearif/how specific definitions influence results.

Since measurements are applied of a limited set of, mainly remote, stations, it is expectedthat these inversions mainly resolve large scale fluxes. This supports the definition of fluxesthat comprise relatively large regions, also leading to a number of fluxes that is of the sameorder as the number of measurements. A larger number of independent measurements thanfluxes allows to uniquely determine the fluxes without a priori assumptions, as used byBrown[1993]. The limitations of this approach appeared from a study on seismic tomography byTrampert and Snieder[1996] showing that such a “low resolution” inversion in combinationwith an inhomogeneous measurement network may lead to severely biased estimates. A the-oretical analysis of this bias is outside the scope of this thesis, therefore, I will restrict theexplanation to a simple example. Suppose we want to improve the estimated emission overEurope by measuring the CH4 concentration over Utrecht when the wind is from the east. We

1.5 Inverse modeling 13

Forward mode:

x x x x x

x x x x x

x x

x x

x x

x x x x x

x x x x x

x x x x x

Reverse mode:

[ x x x ]

[ x x x ]

x x x

x x x

x x x

x x x

x x x

x x x

x x

x x

x x

x x x x x

x x x x x

x x x x x

x x x x x

[ x x x ]

[ x x x ]

[ x x x x x ]

[ x x x x x ]

[ x x x ]

[ x x ]

x x

x x

x x

x x x x x

x x x x x

x x x x x

x x x x x

x x x

x x x

x x x

=

=

=

=

=

=

Figure 1.3: Example of forward and reverse modes illustrating the differences in the storagerequirements and in the number of operations. In the forwardmode the matrix product isevaluated from right to left, and in the reverse mode from left to right. Matrices containingthe intermediate results are printed in bold (taken fromKaminski et al.[1999a]).

scale the European emissions by a single factor such that agreement is obtained between themodel simulated and measured concentration. As a result, weadjust the sources of, for exam-ple, the south of France, while these sources have no relationship with our measurements. If,however, France was defined as a separate region, this sourcewould not have been adjustedsince this would not change the simulated concentration over Utrecht.

As in the example, the definition of fluxes on smaller spatial and temporal scales leadsto a reduction of the bias. Alternatively, if regions are chosen that have accurately knowndistributions of sources and sinks this bias is reduced also. For example, the open oceans areexpected to be a relatively uniform source of methane, although the integrated source strengthis quite uncertain. Unfortunately, this prerequisite is not satisfied for most of the continentalsources of CH4. A recent inverse modeling study of CO2 sources and sinks byKaminskiet al. [1999c] shows that the sizes of these biases are comparable with the fluxes themselvesfor resolutions similar to those used byBrown [1993, 1995];Kandlikar [1997]; Hein et al.[1997]. For methane such an analysis has not been performed yet, however, similar resultsare expected. In this study, to minimize the contribution ofbiases, fluxes are defined on thesmallest possible (model grid) scale.

The transport operators (T), equivalent to the sensitivities of concentrations to flux changes

14 Introduction

(δC/δ f ), can efficiently be computed by a code that represents the first derivative of the trans-port model. For this purpose the Tangent Linear and Adjoint Model Compiler (TAMC) byGiering [1996] is used. TAMC is an automatic tool that converts the code of the regularmodel to its corresponding derivative. Actually, the first derivative is taken of every individ-ual operation of the transport model. The overall derivative is related to the derivatives peroperation by the chain rule. The chainrule can either be evaluated in forward or backward di-rection. If, to minimize biases, fluxes are defined per grid, the resulting number of unknowns(sources and sinks) is much larger than the amount of measurement data. As illustrated byFigure 1.3, in this case the backward mode is more efficient. To evaluate the chain rule inreverse direction, adjoints of the derivatives of operations are needed, which, in case of ma-trices, are just their corresponding transposed matrices.Because of this, the model used toevaluate derivatives in reverse mode is called an adjoint model.

1.6 Research aims and thesis outline

The central theme of this thesis is the quantification of sources and sinks of methane froman atmospheric perspective, i.e. by using atmospheric measurements and interpret them witha model. The aim of this work is to improve the methods and boundary conditions thatwere used previously, to provide more firm and realistic constraints on sources and sinks ofmethane. The original objective was to focus on southeast Asia to improve our understand-ing of CH4 emissions from rice agriculture. This turned out to be very ambitious, both withrespect to the modeling and the observational requirementsposed by this problem. Unfor-tunately, the latter could not be improved within the framework of this project. Therefore,the focus was shifted to larger scales, although improved estimates of rice field emissionsremained a major goal.

As discussed in section 1.3, accurate knowledge of OH is needed to constrain CH4removal. To realistically simulate OH it is important to represent chemical reactions ofNMHC. In Chapter 2, a condensed photochemistry parameterization is presented accountingfor NMHC chemistry. Its performance is compared with other and more extensive schemes.Further, the scheme is implemented in a global 3-D model, andsimulated concentrations ofimportant compounds, such as O3, OH, CO, HNO3, and PAN, are compared with measure-ments. Also, the influence of NMHC chemistry on the simulatedconcentrations of thesegases, in particular of OH, is quantified and discussed.

In Chapter 3, a global inverse modeling study of CH4 is presented, using the adjointtechnique. Measurements and a priori constraints on sources and sinks are described, anddifferences between our method and previously applied methods are discussed. The averagedannual emissions and seasonal cycle are estimated for a three year target period, assuming thesame sources and sinks each year (quasi-stationary state).Differences between a priori anda posteriori derived emissions are presented, and the potential impact of model errors on theestimated emissions is quantified. Further, the sensitivity of the a posteriori derived emissionsto the a priori assumed uncertainties is tested, and alternative methods for quantifying the apriori uncertainties are proposed.

In Chapter 4, direct or “forward model” simulations of CH4 are presented. Here, in-stead of multi-year averaged measurements, direct comparisons between model calculatedand measured concentrations are presented. For this purpose, multi-year reanalyzed mete-orological input fields are used, representative of the simulated period. The ability of themodel to reproduce short-term variability is tested at remote stations, and stations that are

1.6 Research aims and thesis outline 15

increasingly influenced by continental sources. From theseresults, methods are derived toimprove the comparability of model and measurements. Further, regions are identified thatshow the largest discrepancies between measured and model simulated CH4 concentrations.The most probable causes of these discrepancies are discussed.

In Chapter 5, pre-industrial simulations of CH4 are presented, with the aim to reducethe uncertainty of natural wetland emissions, based on the assumption that these have notchanged much during the past centuries. A pre-industrial source and sink scenario is de-fined using reported historical emission estimates, and results of pre-industrial photochem-istry simulations. Except for the pre-industrial global mean concentration, also the interpolardifference and isotopic composition of CH4 has been derived from ice cores. The relative im-portance of each of these constraints on wetland emissions is studied. A best guess estimateand uncertainty range of natural wetland emissions is derived.

Chapter 6 summarizes the main conclusions of this work, and recommendations are givenfor future investigations.

Chapter 2

The impact of nonmethanehydrocarbon compounds ontropospheric photochemistry

Nonmethane hydrocarbons (NMHC) play an important role in global scale troposphericphotochemistry. The representation of NMHC chemistry in three-dimensional troposphericchemistry transport models requires a highly parameterized description of only the most im-portant processes, as the number of reactions and compoundsinvolved is very large. In thispaper a chemical scheme is presented, based on the Carbon Bond Mechanism 4 (CBM-4),modified for use in the global Tracer Model 3 (TM3). The original scheme has been extendedwith reactions important under background conditions, andreaction rates and product dis-tributions have been updated. Box model tests show that the modified CBM-4 and a detailedreference mechanism agree well for a broad range of chemicalconditions. Results of TM3runs with and without NMHC chemistry are compared with observations, illustrating the ef-fects of NMHC on key compounds of tropospheric photochemistry, such as ozone, the hydroxylradical, carbon monoxide, and NOx. In particular, the simulation of ozone over polluted re-gions improves when NMHC chemistry is accounted for. Globally, the contribution of NMHCto net photochemical ozone production is estimated at about40%, leading to a 17% increaseof the tropospheric ozone column. OH is depleted over the continents owing to reactions withNMHC, which is most evident in regions with strong biogenic emissions. Although NMHCsignificantly influence the global OH distribution, their effect on its total tropospheric contentappears to be marginal. Results of sensitivity runs with andwithout organic peroxy nitratesshow that this N reservoir may significantly change the global NOx distribution, leading toa NOx increase over the oceans of the order of 50%. This increase improves the agreementbetween simulated and observed nitrate levels at remote stations, although large discrepan-cies remain. In general, the present treatment of NMHC reproduces the main features oftropospheric chemistry.

1Published inJournal of Geophysical Research,103,10,673-10,696, with F. J. Dentener and J. Lelieveld as co-authors.

17

18 NMHC and tropospheric photochemistry

2.1 Introduction

The trace gas composition of the troposphere is to a large extent regulated by photochemicalprocesses. Therefore to understand the present tropospheric composition and predict possiblefuture trends of trace gases, such as methane and ozone, it isessential to identify and studythe key processes that control photochemistry. The chemistry of nonmethane hydrocarbons(NMHC) is one particular example, being the subject of a large number of studies duringthe past decade. Laboratory studies increased the knowledge about hydrocarbon oxidationpathways, while models have been applied to determine regional and global budgets.

In the early years of tropospheric chemistry modelingChameides and Cicerone[1978]andBrewer et al.[1983] concluded that the influence of NMHC on both ozone and OH wouldprobably be less than a few percent, based one-dimensional model studies with strongly sim-plified representations of NMHC. Subsequently,Liu et al.[1987],Trainer et al.[1987],Kast-ings and Singh[1986], andIsaksen et al.[1985] showed that, on a regional scale and underpolluted conditions, photochemistry is significantly influenced by NMHC and that the globalscale effects of NMHC might have been underestimated previously. Since then, several two-dimensional and three-dimensional model studies have beenperformed, using different repre-sentations of NMHC chemistry, for example, those ofHough[1991],Strand and Hov[1994],Kanakidou et al.[1991], andMuller [1995]. Few of these studies, however, have quantifiedthe effect of NMHC on tropospheric ozone and hydroxyl radicals.

Hough[1991] calculated a contribution of organic peroxy radicals (excluding methyl per-oxide) to photochemical ozone production of less then 10%, but the contribution of NMHC-derived HO2 to this production was not specified.Strand and Hov[1994] estimated a 10%reduction of tropospheric ozone production in the northernhemisphere during summer, as aresult of a 50% reduction of anthropogenic NMHC emissions. This result roughly indicatesthe magnitude of the global NMHC effect on tropospheric ozone, as the largest effect of therelatively short lived hydrocarbons on ozone production isexpected in industrialized regions.Other studies, however, suggest that biogenic hydrocarbonemissions may also contributesignificantly to ozone production [Trainer et al., 1987;Roselle, 1991].

Besides ozone, the abundance of the hydroxyl radical, the principal oxidizing agent of thetroposphere, may be influenced by NMHC chemistry as well. In fact, since the global andannual amount of oxidized carbon derived from NMHC is estimated to be larger than thatfrom methane, NMHC have the potential to significantly affect OH concentrations. Nitrogen-containing organics, like peroxyacetyl nitrate (PAN), which are products of NMHC oxidation,may also be important because of their influence on the globaldistribution of NOx, and thusindirectly on ozone.Kanakidou et al.[1991] studied the effect of PAN derived from propaneand ethane oxidation in a two-dimensional model. They calculated that the decompositionof PAN increases NOx mixing ratios in the lower troposphere in the tropics by only5–10%.The simulated PAN concentrations, however, were low in comparison with measurements,and it was concluded that the chemistry of other hydrocarboncompounds should also havebeen accounted for. In addition, the use of three-dimensional models was recommended be-cause of the relatively small spatial scales and the nonlinear nature of hydrocarbon oxidationprocesses.

The main objective of this paper is to study the impact of NMHCon tropospheric pho-tochemistry. In particular, we focus on the global budgets and distributions of ozone, thehydroxyl radical, and nitrates. We used the global three-dimensional chemistry transportmodel Tracer Model 3 (TM3), an updated version of TM2 [Heimann, 1995], supplementedwith a newly developed NMHC chemistry module. Incorporation of an NMHC gas phase

2.2 Model description 19

chemistry scheme in a three-dimensional chemistry transport model (CTM) implies that alarge number of chemical compounds and reactions need to be accounted for. In the real at-mosphere, thousands of reactions and species are involved in the chemistry of NMHC. Thusany representation of this complicated chemistry in globalscale three-dimensional CTMs re-quires a rigorous selection of the most important reactionsand species. Even if only the mostabundant species are represented, the computational load of including all the reaction prod-ucts and transport of the relatively long lived intermediates is very high. The amount of detailneeded to represent NMHC chemistry depends on two importantquestions. First, how wellcan the species of interest (ozone, CO, OH and NOx in his study) still be reproduced usinga simplified NMHC representation. Second, how does the addedvalue of including extra ormore complicated NMHC reaction pathways relate to other model uncertainties, for exam-ple, of the emission estimates and the ozone exchange between stratosphere and troposphere,which may exceed 50%. However, these considerations may be useful in defining selectioncriteria; in practice, they will not be very strict, as the model uncertainties are difficult toquantify.

The number of reactions in the NMHC representations can be reduced by the definition ofsurrogate species. Each of these surrogates represents an important characteristic of NMHCchemistry, often corresponding to certain molecular structures, as, for example, paraffinic car-bon atoms or olefinic carbon bounds. Widely used chemical mechanisms, based on this prin-ciple, are the Carbon Bond Mechanism 4 (CBM-4) [Gery et al., 1989] and the Regional AcidDeposition Model (RADM) chemistry [Stockwell et al., 1990]. In this study, we developeda condensed chemical mechanism, based on CBM-4. As CBM-4 wasoriginally designed tosimulate regional scale photochemistry under polluted conditions, we added some reactionsimportant under background, low NOx conditions. Moreover, some product yields have beenupdated using information of the more recent and extensive Regional Atmospheric Chem-istry Model (RACM) [Stockwell et al., 1997]. A description of our chemistry scheme is givenin section 2.2, which also presents a brief outline of the TM3model, focusing on processesrelevant for tropospheric chemistry. In section 2.3, modelresults for different trace gasesare compared with measurements, and the effects of running the model with and withoutNMHC chemistry are demonstrated. In addition, the results of some sensitivity experimentsare presented, illustrating the importance of some NMHC-related processes. In section 2.4the results are discussed, and section 2.5 presents the conclusions.

2.2 Model description

2.2.1 NMHC chemistry representation

A main reason for selecting the CBM-4 mechanism as a startingpoint for our scheme is itshighly parameterized representation of NMHC chemistry. The scheme is very CPU efficientas a result of the small number of reactions and compounds compared with other chemicalmechanisms. In addition, it has been extensively tested against smog chamber experimentsand more comprehensive schemes [Gery et al., 1988;Derwent, 1990;Paulson and Seinfeld,1992;Tonnesen and Jeffries, 1994]. In the original CBM-4 scheme, a number of reactioncoefficients have been adjusted to the results of smog chamber experiments, for example,the representation of isoprene. Other parts, for example, the oxidation of “PAR,” whichrepresents all paraffinic carbon atoms, have been based merely on mechanistic knowledge.

Since the publication of the scheme in 1989, several studieshave improved the under-


Table 2.1.: Abbreviated Compound Names

Name Description Name Description

ALD2 acetaldehyde and higher aldehydes ISOP isopreneC2O3 peroxyacetyl radical MGLY methylglyoxalPAN peroxyacetyl nitrate and higher PANs ORGNIT lumped alkyl nitratesPAR paraffinic carbon atoms XO2 NO to NO2 operatorOLE olefinic carbon bonds XO2N NO to alkyl nitrate operatorETH ethene RXPAR PAR budget correctorROOH lumped organic peroxides (>C1)

standing of NMHC chemistry, in particular, regarding the oxidation of isoprene and reactionsof ozone with olefines. We have maintained the mechanistic parts of CBM-4, where possible,and added reactions needed to represent background chemistry. The remaining part has beenreadjusted, using RACM [Stockwell et al., 1997] as a reference mechanism. We evaluated allreactions and compounds concerning their importance for global scale tropospheric photo-chemistry. As a result, some CBM-4 compounds, for example, HONO and O(3P), have beeneliminated. The reaction rates have been updated accordingto Atkinson[1994] andDeMoreet al. [1994]. The present version consists of 30 species and 68 reactions, slightly less thanthe original CBM-4 (33 species and 81 reactions). The modifications can be summarized asfollows: (1) Extension of the methane oxidation chemistry;(2) Introduction of organic perox-ide and organic nitrate reactions; (3) Elimination of the species HONO, toluene and xylene,and explicit treatment of O(3P) and O(1D); (4) Update of the product distribution of isopreneand ozone to olefine reactions; and (5) Update of the reactionrates.

In CBM-4, the chemistry of methane is treated in a very simplified way, which is justifiedfor the temporal and spatial scales of regional models. Compared with urban polluted condi-tions, in the relatively clean background atmosphere, methane chemistry is important relativeto higher hydrocarbons. Therefore a more precise treatmentof its chemistry is required in aglobal model. We replaced the implicit representation of methane in CBM-4 by the explicitreaction sequence (R26)–(R31) (see Appendix 6). Organic peroxy radicals are importantshort lived reactants, produced in the oxidation of hydrocarbons. Under polluted NOx-richconditions, the major reaction of these radicals is with NO,yielding alkoxy radicals and NO2.As a result of this transformation of NO to NO2, net ozone production takes place. UnderNOx-poor conditions the competing reaction with HO2 becomes important, which leads toperoxy radical removal without the formation of NO2. Since CBM-4 only accounts for theorganic peroxy radical reactions with NO and reactions between peroxy radicals, we extendedthe scheme with the HO2 reactions ((R66) and (R67)). These reactions yield organicperox-ides, represented in the new scheme by the species “ROOH,” (see Tabel 2.1) of which thechemistry is similar to that of methyl peroxide (CH3O2H) ((R59) and (R60)).

Organic nitrates may be important, similar to organic peroxy nitrates, as reservoir speciesof NOx. Although CBM-4 accounts for NOx removal due to organic nitrate production, nosubsequent reactions of these products are accounted for. This assumption is not tenablefor regions at some distance of NMHC and NOx sources, where the net effect of organicnitrate chemistry changes from NOx removal to NOx production. Therefore we introduced


the “ORGNIT” component, representing the lumped organic nitrates, and its most importantchemical sinks, i.e., photolysis and the OH radical reaction ((R61) and (R62)). The lumpingof organic nitrates is complicated by the fact that the OH reaction rate increases with carbonchain length. For a limited number of organic nitrates, kinetic data for the OH reaction andphotolysis cross sections and quantum yields have been published [Atkinson, 1994]. Weadopted an average turnover time for the OH reaction and photolysis, which is representativefor a C4 mononitrate. This time corresponds to a lifetime of about 6 days under summertimeconditions at midlatitudes.

The size of the modified CBM-4 scheme has been reduced by the elimination of thespecies O(3P), O(1D), HONO, and aromatic compounds. O(3P) reactions with hydrocarbonscan become important under smog chamber conditions. Under tropospheric conditions, how-ever, the OH, O3, and, to a lesser extent, NO3 reactions with hydrocarbons dominate. Theinorganic reactions of O(3P) are very fast and can be considered to proceed instantaneously.The same is valid for O(1D), which can react with H2O and O2/N2 to yield OH and O(3P),respectively. The OH yield is calculated directly from O3 photolysis and the rates of thesubsequent reactions.

HONO may be produced in significant amounts in heterogeneousreactions during night-time, under highly polluted conditions. The exact mechanism and reaction rates remain, how-ever, uncertain. The HONO production, based on gas phase chemistry only, appears to be tooslow to yield significant amounts of HONO. Since an adequate description of heterogeneousproduction mechanisms is not available, we have neglected HONO.

Aromatics contribute about 10% to the total global anthropogenic NMHC emissions[Olivier et al., 1996], with car exhaust being the principal source. In regions with relativelystrong traffic emissions, the ratio of aromatics to total NMHC emissions may be substantiallyhigher than 10%. For example, in the United Kingdom, the total contribution of aromatics toozone production is estimated at 30% [Derwent et al., 1996]. Although the neglect of tolueneand xylene may thus lead to a significant underprediction of ozone production at some spe-cific locations, it can be expected that the effects on large-scale ozone budgets are relativelysmall. In view of the CPU costs of calculating the CBM-4 aromatics chemistry (10 reactions,6 compounds) and the large uncertainties related to its representation, these species have beenomitted. As a simple approximation, the first stable products of xylene oxidation in CBM-4,methyl glyoxal and PAR, are emitted as surrogates of xylene.

The product yields of the reactions of ethene and lumped olefines (OLE) with ozone havebeen updated, using the products of the corresponding RACM reactions ((R50) and (R53)).The carbon chain length of the RACM surrogate for terminal olefines (OLT) is larger thanthe CBM-4 compound OLE, which has been compensated for by theremoval of PAR in ourscheme (by producing the PAR budget corrector RXPAR). In a similar way, the product dis-tributions of the isoprene reactions have been revised ((R56)–(R58)). The product yield of theisoprene reaction with OH (R56) has been based on the RACM formulation of the isopreneperoxy radical reaction with NO, being the major reaction pathway. This reaction yieldsmethacrolein, which is not represented by CBM-4. The product yield of the methacrolein+ OH reaction has been approximated by a combination of OLE and methyl glyoxal, whichhas the advantage that no new species is introduced. A similar procedure has been appliedto the isoprene + ozone and isoprene + NO3 reactions. The production of ketones has beenneglected, since these compounds are not very reactive. This practice is consistent with theCBM-4 representation of PAR, in which ketones are also neglected.

The modified CBM-4 has been extensively tested in a box model against RACM, usingthe Facsimile Gear method for the numerical integration of the chemical equations. Five-


0 10 20 30 40 50 60NMHC/NOx

0.0

0.5

1.0

1.5

2.0

mod

.CB

M-4

/RA

CM

0 100 200 300 400 500NMHC/NOx

0.0

0.5

1.0

1.5

2.0

mod

.CB

M-4

/RA

CM

Figure 2.1: Box model comparison of modified CBM-4 and RACM for the case of industry(left), and tropical rainforest (right). Diamonds, mean OH; triangles, O3 production; squares,mean organic nitrogen.

day simulations have been performed, accounting for emissions and constant removal ofozone and HNO3 to maintain the initial chemical regime throughout each simulation. TheNMHC carbon to NOx ratio has been varied between 1 and 500 by adjusting the NMHCinitialization and emission. Different chemical conditions have been simulated using fourtest cases: industry, biomass burning, tropical rainforest, and remote oceans.

Figure 2.1 shows the generally good agreement between our scheme and RACM, for theindustry and tropical rainforest cases. In these figures, results of the modified CBM-4 havebeen divided by corresponding RACM results for ozone production, mean hydroxyl radicaland mean organic nitrogen (PAN + organic nitrate) concentrations over the total model run.

The largest relative deviations are found for OH, which is underestimated and overesti-mated by the scheme dependent on the chemical conditions. Inthe industry case, PAR isthe dominant hydrocarbon compound, in contrast to the rainforest case in which isoprenedominates. Paraffins consume more hydroxyl radicals than are produced by the reactions ofits products, and this effect increases at higher NMHC to NOx ratios. The modified CBM-4 appears to overestimate this effect. The NMHC to NOx ratio under polluted conditionsin Europe, to which this case applies, is typically of the order of 10–30. Therefore underindustry affected conditions, OH deviations by more than 25%, compared with RACM, arenot likely to occur frequently. The tropical case shows thatthe modified CBM-4 overpre-dicts OH compared with RACM, which has its origin in the representation of isoprene. AtNMHC-NOx ratios higher than 50, the absolute amount of OH is, however,very low (<105

molecules/cm3), which means that the impact of the large relative differences between RACMand CBM-4 is only small. It should be emphasized that the chemistry of isoprene is not wellquantified under NOx-poor conditions, so that the overall uncertainty is relatively large.

The agreement in ozone production between both schemes is remarkably good (within10%), which is also true for organic nitrogen, especially inthe tropical case. The somewhathigher alkyl nitrate yield in our PAR representation, relative to RACM, partly explains thedifference in organic nitrogen for the industry case. The effects on ozone are limited, as thelargest alkyl nitrate differences occur in the NOx-rich regime, where ozone formation is notNOx limited.

Figures 2.2 and 2.3 present the results of two cases of the Intergovernmental Panel of


0 1 2 3 4 5day

2022

24

26

28

30

3234

O3

ppbv

0 1 2 3 4 5day

1

10

100

1000

NO

x pp

tv

Figure 2.2: PhotoComp IPCC box model intercomparison for the case of land-bio. Squares,mean concentrations at noon of the schemes involved in the IPCC intercompar ison for O3(upper panel) and NOx (lower panel); dashed lines, 1 rms error of the IPCC mean concentra-tions; asterisks, noon concentrations for modified CBM-4.

Climate Change (IPCC) Photochemical Model Intercomparison (PhotoComp) [Olsen et al.,1997] together with results of our modified CBM-4 calculations. The land-bio and plume-HCcase, representing conditions over “remote” continental regions and middle tropospheric pol-luted plumes, respectively, have been chosen here because in the remaining cases NMHC arenot represented. Generally, our results are well within 1 rms error of the schemes participat-ing in the IPCC intercomparison. Similar agreement is foundbetween results obtained withour scheme and reported contributions to the EUROTRAC Chemical Mechanism WorkingGroup intercomparison [Poppe et al., 1996].

2.2.2 TM3 description

The Tracer Model 3 is an updated version of TM2 as described byHeimann[1995]. Thespatial resolution of the version used in this study is 5◦ in the longitudional and 3.8◦ inthe latitudional direction with 19 vertical levels. Near the surface these levels are defined


0 1 2 3 4 5day

4060

80

100

120

140

160180

O3

ppbv

0 1 2 3 4 5day

1

10

100

1000

10000

NO

x pp

tv

Figure 2.3: Same as Figure 2.2 but for the plume-HC case.

as terrain following sigma coordinates, in the upper stratosphere by pressure levels, and inbetween by a hybrid of the two. The horizontal and vertical transport of tracers is based onsix hourly mean meteorological fields, including wind, surface pressure, temperature, andhumidity. For this purpose, preprocessed data can be used from the European Centre forMedium-Range Weather Forecasts (ECMWF) model or other general circulation models. Inthis study, reanalyzed ECMWF fields are used for the year 1993. The advective transport iscalculated using the “slopes scheme” ofRussell and Lerner[1981]. Using our new NMHCchemistry scheme, 16 chemical tracers need to be transported (see Table 2.2). This number ofchemical tracers may be compared with our most simplified CH4, CO, HOx, NOx backgroundchemistry scheme with seven transported tracers. The subgrid scale convective fluxes areevaluated using the cloud scheme ofTiedke[1989], including entrainment and detrainment inupdrafts and downdrafts. Turbulent vertical transport is based on stability dependent verticaldiffusion [Louis, 1979]. The tracer transport in TM3 has been tested by comparing radonobservations at different locations to model simulations.The results of these tests will bepresented in the near future [Dentener et al., 1999]. From this study it appears that measuredand simulated radon concentrations agree quite well; generally, deviations are<50%.


Table 2.2.: Model Transported Tracers

Tracer Tracer

1. O3 9. PAR2. NOx 10. ETH3. HNO3 11. OLE4. H2O2 12. ISOP5. CH4 13. ALD26. CH3O2 14. ROOH7. CH2O 15. PAN8. CO 16. ORGNIT

The chemical equations are integrated using Eulerian backward iterative (EBI), as formu-lated byHertel et al.[1993]. In a standard TM3 run a fixed time step of 2400 s is used forthe chemistry. The number of iterations applied depends on the chemical lifetime of a partic-ular compound, with a maximum of eight. Increasing the number of iterations only slightlychanges the model results. Photolysis rates are derived from the radiation code ofBruhl andCrutzen[1988], and daytime average values are updated twice a month. As a first-order ap-proximation of the diurnal variation of photolysis rates, asingle period of a normalized (sin)2

function is used during daytime. A more sophisticated scheme will be implemented in thenear future.

O3 transport from the stratosphere into the upper level of the TM3 model domain is ac-counted for by fixing the ozone concentration in the model toplayer, based on 10 hPa ozoneconcentrations derived from UARS data. Stratospheric HNO3 is treated similarly, based onUARS derived O3/HNO3 ratios at 10 hPa (B. Bregman, personal communication, 1997). Thestratospheric destruction of methane by photolysis and reaction with OH, Cl, and O(1D) isderived from two-dimensional photochemical model calculationsBruhl and Crutzen[1993],scaled to a global loss of 40 Tg/yr in agreement withCrutzen[1995].

The total emissions of NOx, CO, CH4, and anthropogenic and biogenic NMHC in TM3are listed in Table 2.3. The methane source is represented bykeeping the surface layer con-centration fixed, using monthly and zonally averaged valuesbased on the work ofHein et al.[1997]. Lightning NOx emissions are scaled to the distribution of deep convectivecloud topheights and associated mass fluxes. Lightning over the oceans is assumed to be a factor of 10less effective in producing NOx than over land, and the cloud-to-ground flashes are assumedto be a factor of 3 more effective than cloud-to-cloud flashes. These values are in accord withthe parameterizations ofLevy[1996] andGalliardo-Klenner[1996]. The global amount ofNOx produced by lightning of 5 TgN/yr is in agreement withGalliardo-Klenner and Cooray[1996].

The annual totals and spatial distributions of the emissions of anthropogenic NMHC, an-thropogenic CO, isoprene, soil, and industrial NOx are based on 1◦ x 1◦ GEIA and EDGAR-V2.0 emission inventories [Olivier et al., 1996;Guenther et al., 1995;Yienger and Levy,1995; Benkovitz et al., 1996]. The anthropogenic hydrocarbon emission database uses acompound classification into 25 categories. Each category has been translated into the corre-sponding CBM-4 compound representation according toGery et al.[1989]. As a result, 78%


Table 2.3.: Total Emissions and Model Calculated CO Production (Teragrams per Year)

Industry Biomass Burning Vegetation Other Total

NOx (Tg N/yr) 22 5 11a 38CO 478 496 75 40b 1089NMHC (Tg C/yr) 90 18 400 508CO from CH4c 886CO from NMHCc 408CH4 517

aValues of 5 Tg N from lightning and 5.5 Tg N from soils.bOcean emissions.cTotal model domain.

of the anthropogenic emissions are represented as PAR, 13% as OLE, and 9% as ETH, ALD2,MGLY, and CH2O. In total, 87% of the EDGAR NMHC-carbon is represented by CBM-4compounds. Ocean emissions of CO have been distributed in 15◦ latitude bands, based onthe work ofBates et al.[1995]. The annual total CO emission from oceans is a compromisebetween different estimates [Bates et al., 1995;Erickson, 1989]. CO emissions from vegeta-tion have been scaled to the global distribution of net primary production (NPP) as derivedfrom the climate assessment model IMAGE [Minnen et al., 1996;Kreileman, 1996]; the totalamount is based on the work ofBauer et al.[1979]. Isoprene emissions have been reducedby 100 TgC/yr compared with the GEIA recommendations (see section 2.4).

The seasonal pattern of isoprene emissions is accounted forin the GEIA isoprene emis-sion database, which has a temporal resolution of 1 month. Biomass burning emissions aredistributed over the year according to the seasonal dependence estimate ofHao et al.[1991].CO emissions are expected to be higher during winter in the northern hemisphere, associ-ated with energy use. Therefore a seasonality derived from SO2 emissions is applied to COsources north of 45◦. No temporal variations are applied to anthropogenic NMHC emissionsthat are not related to biomass burning. All emissions are kept constant during day and night,with the exception of biogenic NMHC emissions. The rate at which isoprene is emitted byvegetation is known to be a function of both temperature and light intensity. In TM3 it isapproximated by scaling of the isoprene emissions to the daily variation of photolysis rates.

Dry deposition is accounted for by using a resistance analogy-based parameterization[Ganzeveld et al., 1997]. In this scheme the deposition velocity is calculated from the aero-dynamic resistance, the quasi-laminar boundary layer resistance, and the surface resistance.The computation of the surface resistances is based on the work of Wesely[1989]. In addi-tion, the surface resistances of some compounds have been scaled to structure analogs: thenitrate radical (NO3) surface resistance as NO2, organic nitrates as PAN, and higher organicperoxides (ROOH) as methyl peroxide. CO dry deposition overland has been accounted forby using a surface resistance which leads to an average deposition velocity of about 0.2 mm/s.

The wet deposition of HNO3 is based on a parameterization ofJunge and Gustafson[1957]. The wet removal of less soluble species is scaled to HNO3 according to the Henry’s

2.3 Results 27

Table 2.4.: Wet Removal Coefficients

Species H298 −∆H/R Referencea

HNO3 2.1x105 8700 SWH2O2 7.4x104 6621 LKCH3O2H 2.2x102 5607 LKCH2O 6.3x103 6425 LBROOH 2.2x102 5607 as CH3O2H

Henry constants in mol l−1 atm−1; temperature dependency,H(T) =H298e−∆H/R(1/T−1/298).

aSW,Schwartz and White[1981]; LB,Ledbury and Blair[1925]; LK,Lind and Kok[1986].

law coefficients [Dentener, 1993]. Higher organic peroxides are assumed to be removed atthe same rate as methyl peroxide. Since monoalkyl nitrates and PANs are not very soluble,their wet deposition has been neglected [Kames and Schurath, 1992;Roberts, 1990]. TheHenry’s law coefficients applied are summarized in Table 2.4. The heterogeneous removalof N2O5 on sulfate aerosols has been accounted for by using a parameterization byDentenerand Crutzen[1993].

2.3 Results

2.3.1 Ozone

In this section, results of two model simulations are compared with measurements, one runwith and one without the representation of NMHC chemistry. The first will be referred to as“NMHC” and the latter as “CH4-only.” Both runs have been initialized with the same con-centration fields, except for longer-lived hydrocarbons (PAR) and nitrogen-containing hy-drocarbons (PAN and alkyl nitrates), which have been set to zero in the “CH4-only” case.In Figures 2.4 and 2.5, simulated monthly mean surface ozonemixing ratios and verticalozone profiles are compared with observations [Oltmans et al., 1989;Oltmans and Levy II,1992, 1994;Komhyr et al., 1989]. The measurement data represent monthly and seasonalmeans, averaged over the total observational record at eachstation, which covers between 1and 20 years. Although it would have been desirable to compare our results to 1993 measure-ments, those data were not available. The precision of the ozone measurements is claimed tobe within 5%.

Six stations have been selected for comparison of surface ozone mixing ratios (see Figure2.4), such that a range of different chemical conditions is covered. Payerne (Switzerland) wasselected because the ozone maximum during summer represents about the mean of severalEuropean stations, at some distance of large cities. As indicated by Figure 2.4, simulatedozone concentrations appear to be significantly higher whenwe account for NMHC. ThisNMHC effect is most pronounced for the stations Wallops Island, Payerne, and Tateno, which


Mace Head( 53N 10W 30m)

J F M A M J J A S O N D 0

10

20

30

40

50

60

70

O3

(ppb

)

Barbados( 13N 59W 59m)


10

20

30

40

50

60

70

O3

(ppb

)

Samoa( 14S 171W 82m)


10

20

30

40

50

60

70

O3

(ppb

)

Tateno( 36N 140E 31m)


10

20

30

40

50

60

70

O3

(ppb

)

Wallops Island( 38N 75W 13m)


10

20

30

40

50

60

70

O3

(ppb

)

Payerne( 47N 7E 491m)


10

20

30

40

50

60

70

O3

(ppb

)

Figure 2.4: Observed versus calculated surface ozone concentrations. Squares, observations;solid lines, NMHC; dashed lines, CH4-only.

2.3 Results 29

Marambio 64S 57W

0 40 80 120 160 200O3 (ppb)

1000

900

800

700

600

500

400

300

200

pres

s (h

Pa)

Natal 6S 35W

0 40 80 120 160 200O3 (ppb)

1000

900

800

700

600

500

400

300

200

pres

s (h

Pa)

Hohenpeissenberg 48N 11E

0 40 80 120 160 200O3 (ppb)

1000

900

800

700

600

500

400

300

200

pres

s (h

Pa)

Resolute 75N 95W

0 40 80 120 160 200O3 (ppb)

1000

900

800

700

600

500

400

300

200

pres

s (h

Pa)

Figure 2.5: Ozone sonde observations and calculated vertical ozone profiles for June to Au-gust. Squares, sondes; solid lines, NMHC; dashed lines, CH4-only.

are located in industrialized regions. The maximum differences occur during summer, whenNMHC chemistry approximately doubles surface ozone concentrations. In the case of Wal-lops Island and Payerne, NMHC chemistry helps representingthe photochemically inducedozone maximum during summer, although the maximum at Wallops Island is overestimatedby about 10 ppbv. The model predicts large ozone gradients atthe east coast of the UnitedStates during summer, which are not well resolved by the relatively coarse model grid. Thisfeature may partly explain the overestimation of summertime ozone mixing ratios at WallopsIsland. At the remote stations Samoa and Barbados the differences between the NMHC andCH4-only runs are smaller, up to 30% during July. In these cases, the seasonal cycle is hardlyaffected by NMHC chemistry, and the ozone increase due to NMHC is similar during all


Figure 2.6: The ratio of surface ozone, “NMHC” divided by “CH4-only,” for July. The blackdashed lines indicate the 1.0 contours.

seasons. Samoa is at too large a distance from the continent to be significantly influenced bylocal NMHC chemistry, and therefore the ozone differences are likely related to long-rangetransport. The same is true for Barbados, where easterly winds from the Atlantic Ocean pre-vail, indicating that the ozone concentration over part of the Atlantic has been overpredictedby the model.

Figure 2.6 illustrates the global effect of NMHC chemistry on surface ozone mixing ra-tios for July. The surface ozone concentration field derivedfrom the NMHC run has beendivided by the CH4-only run, yielding the relative differences between both runs. Again, itappears that NMHC have the largest effect in industrializedareas, becoming less significantin relatively clean environments. A decrease in ozone due toNMHC chemistry is found overthe tropical Amazon rainforest, where net photochemical ozone destruction prevails underconditions of low NOx and high biogenic hydrocarbon concentrations. The vertical profilesin Figure 2.5 indicate that the effect of NMHC is not confined to the boundary layer but af-fects the whole model domain. The effect is largest in the northern hemisphere, is smallerin Natal, and is negligible in winter in the southern hemisphere. Stations are selected on ahistory of a relatively large number of sonde measurements compared with other stations rep-resentative of the same latitudinal band. The agreement between the selected measurementsand the TM3 NMHC simulation turns out to be somewhat fortunate. In general, deviationsare within 30% with a tendency of the model to overestimate ozone as compared with thesonde measurements. An ozone maximum near the surface is calculated in the NMHC modelrun at Hohenpeissenberg, Germany, due to photochemical ozone production in the pollutedboundary layer. Unfortunately, the vertical resolution ofthe available sonde data is too lowto confirm this.

In Table 2.5 the annual tropospheric ozone budget has been summarized. The amountof ozone which is removed by dry deposition is relatively lowcompared with the results ofMuller [1995] (1100 Tg/yr), and the ozone budget compilation listed byWorld Meteorologi-cal Organization[1995] (953–1178 Tg/yr). However, our simulated deposition of 681 Tg/yragrees well with the 740 Tg/yr calculated byRoelofs and Lelieveld[1995]. The stratosphere-troposphere exchange (STE) of ozone is in fair agreement with the range of 528–846 Tg/yr

2.3 Results 31

Table 2.5.: Annual Tropospheric Ozone Budget (Teragrams per Year)

North. Hemisphere South. Hemisphere Global

NMHC CH4-only NMHC CH4-only NMHC CH4-only

Chem. production 2448 1824 1531 1070 3979 2894Chem. destruction 2366 1834 1699 1315 4065 3149Dry deposition 441 337 240 195 681 533Strat/trop exchangea 768 740

Burden 174 149 137 118 311 266

The tropopause has been defined at 100 hPa in the tropics and 200 hPa at latitudes polewardof 30◦. The ozone inbalance for “CH4-only” is caused by a decreasedannual mean ozoneburden compared with the initialization.

aIncludes two-way exchange across the extratropical tropopause.

reported byWorld Meteorological Organization[1995]. It should be noted that the calcu-lated STE of 768 Tg/yr also accounts for some two-way exchange across the extratropicaltropopause and is thus somewhat larger than the STE from the “overworld” across the 100hPa level. Our annual mean tropospheric ozone column of 311 Tg exceeds the 236 Tg re-ported byRoelofs and Lelieveld[1995], who did not account for NMHC chemistry; however,surrogate carbon monoxide emissions were used to compensate for it. Ozone burdens byHough[1991] andMuller [1995] cannot be used for comparisons, since ozone was integratedover their total model domain, which also includes part of the stratosphere. The chemical pro-duction of ozone, here defined as the NO2 photolysis minus the reaction of ozone with NO,of 3979 Tg/yr is in fair agreement with the total model domainintegrated value of 4550 Tg/yrby Muller [1995].

Based on our calculations, the contribution of NMHC chemistry to the photochemicalproduction of ozone is 38%, resulting in a 17% increase in thetropospheric ozone columnwhen NMHC chemistry is accounted for. Test runs have been performed to determine thesensitivity of two important processes associated with NMHC chemistry: the reactions ofreservoir-N compounds and biogenic emissions. When all chemical interactions of PAN andalkyl nitrates are turned off in the model, the troposphere-integrated photochemical ozoneproduction due to NMHC reduces by 22%. Significant differences between this run and thestandard NMHC run are also found in surface ozone distribution (see section 2.3.4). Whenthe biogenic emissions are turned off, the NMHC-induced photochemical ozone productiondecreases by as much as 70%. The difference in ozone column between this run and “CH4-only,” which is thus caused by anthropogenic NMHC emissionsonly, is just 6%, illustratingthe strong influence of isoprene in our model.


Hungary( 47N 16E 240m)


100

200

300

400

500

600

700

CO

(pp

bv)

Grifton( 35N 77W 505m)


100

200

300

400

CO

(pp

bv)

Mace Head( 53N 10W 25m)


50

100

150

200

250

300

CO

(pp

bv)

Qinghai Province( 36N 101E 3810m)


50

100

150

200

250

CO

(pp

bv)

Barbados( 13N 59W 3m)


50

100

150

200

CO

(pp

bv)

Samoa( 14S 170W 42m)


50

100

150

200

CO

(pp

bv)

Figure 2.7: Simulated versus observed surface CO concentrations. Pluses, observations (totalrecord) and standard deviation; asterisks, observations (1993); solid lines, NMHC; dashedlines, CH4-only.

2.3 Results 33

2.3.2 Carbon monoxide

Figure 2.7 shows a comparison between surface measurementsof CO [Novelli et al., 1992](P. C. Novelli, personal communication, 1996) and the TM3 simulations. The uncertainty,resulting from both sampling and measurement errors, was estimated to be less than 8%.The observational record lengths, from which the monthly averaged data have been derived,vary between 1 and 7 years. Also monthly averaged observations have been included for1993, consistent with the 1993 meteorological model input used. As expected, the NMHCrun yields higher CO concentrations than “CH4-only,” as a result of CO production fromNMHC oxidation. The relative differences in CO between bothruns depend on the ratio ofNMHC and CO emissions and the amount of NMHC and CO which is oxidized near thesources, which again depends on the reactivity of the NMHC mix and the OH level. Overall,the largest relative effects of NMHC chemistry on CO are found in regions with relativelyhigh biogenic emissions. At most stations, NMHC chemistry appears to have little effect onthe seasonality of the carbon monoxide surface mixing ratio. Note that the close agreementbetween both runs at the beginning of the year is caused by using the same carbon monoxideinitialization. The high carbon monoxide levels calculated for Hungary in winter may be re-lated to a stagnant episode over central Europe during the first three weeks of February 1993.Also the 1993 averaged observations for March are distinctly higher in comparison with theaverages derived from the total record, ranging from March 1993 until 1996. Indeed, a 1992meteorology-driven TM3 run yields much lower CO levels in Hungary during this part of theyear (not shown). Generally, however, the use of 1993 data instead of total record-derivedmonthly means does not significantly improve the agreement between model and observa-tions. At the polluted stations Grifton and Hungary, the agreement between observationsand simulations does not significantly improve when NMHC chemistry is accounted for. Atthe remote stations Barbados and Samoa the underestimationof CO in the CH4-only run isreduced by hydrocarbon chemistry in the NMHC run, although differences up to 30% remain.

In October 1984 the Space Shuttle measured CO within the framework of the the NASALangley Measurement of Air Pollution from Satellite (MAPS)experiment. The MAPS ob-servations yield mean free tropospheric CO concentrations, weighted for the sensitivity of theinstrument [Reichle et al., 1990]. This sensitivity is a function of altitude, with a maximum atabout 400 hPa. We have derived a sensitivity factor for each model layer based on a reportedrelationship between instrument sensitivity and pressure[Reichle et al., 1990]. This factorand the air mass per grid box have been used to define weightingfactors used to calculateaverage CO concentrations. Since the first publication of the data in 1990, the instrumentcalibration has changed. The data presented in this study have been retrieved from the NASALangley URL “eosdis.nasa.gov” in March 1997.

Figure 2.8 shows a comparison of 1984 MAPS data collected from October 6–13, andOctober mean NMHC run results. The light parts of the MAPS plot indicate that no data areavailable, which applies to all latitudes poleward of 60◦S and 60◦N. The regions with highCO mixing ratios in the tropics coincide with the biomass burning regions during October.This feature is also represented by the model, although the observed mixing ratios are higherand the plumes appear to cover larger areas. The exact location of the 1984 biomass burningplume extending from Africa to the Indian Ocean is, however,difficult to reproduce by themodel, since it is determined by the meteorology and fire locations at the time the MAPS datawere collected. The average column-integrated mixing ratio over the northern hemisphere asdeduced from the modeled CO field agrees well with the MAPS observations. The lowest COmixing ratios of 30–50 ppbv were measured over the Pacific Ocean, which is also reproduced


MAPS CO Oct.1984

180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E

60S

30S

EQ

30N

60N

isolines are (>): 30.0 50.0 70.0 90.0 110.0 ppbv

TM3 CO Oct.1993

180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E

60S

30S

EQ

30N

60N

isolines are (>): 30.0 50.0 70.0 90.0 110.0 ppbv

Figure 2.8: Comparison of observed and simulated column-integrated CO.

by our model calculations, although in the model this regionstretches out over most of thesouthern hemisphere, which is not supported by the observations.

2.3.3 The hydroxyl radical

Methyl chloroform (MCF) has been used as a tracer to validatethe modeled OH fields. Inthe troposphere the reaction with OH is the most important MCF removal process. Otherremoval processes are ocean uptake and loss to the stratosphere, which both contribute about5%. Methyl chloroform is emitted by industry in amounts quantified with a relatively smalluncertainty of about 10% [Midgley, 1989;Midgley and McCulloch, 1995]. Surface concen-trations have been measured at stations of the ALE/GAGE network since 1978 [Prinn et al.,1992, 1994].

The MCF simulation method applied to test the TM3 OH field has been described byKanakidou et al.[1995]. To simulate the ocean uptake of MCF, a two-dimensional distri-bution of thermocline depth and sea surface temperature have been used, derived from the

2.3 Results 35

Mace Head(53N,10W)

1975 1980 1985 1990 1995year

50

100

150

200

MC

F (

pptv

)

Barbados(13N,59W)

1975 1980 1985 1990 1995year

40

60

80

100

120

140

160

MC

F (

pptv

)

Tasmania(41S,145E)

1975 1980 1985 1990 1995year

40

60

80

100

120

140

160

MC

F (

pptv

)

Figure 2.9: ALE/GAGE MCF observations and model results. Dots, observations; solidlines, NMHC; dashed lines, CH4-only. Note that the NMHC and CH4-only simulationsdeviate only marginally.


global ocean circulation model HAMOCC3 [Maier-Reimer, 1993] (K. D. Six, personal com-munication, 1996). The calculated turnover time of MCF due to ocean uptake amounts to 87years, which agrees well to the estimate ofButler et al.[1991]. Model runs start at 1970,using initial MCF fields based on a linear extrapolation of ALE/GAGE MCF records, beingabout 20 and 50 ppt over the southern and northern hemisphere, respectively. Since we useda model spin-up time of 8 years, which is longer than the tropospheric lifetime of methylchloroform, the initialization has little influence on the results.

Monthly averaged OH fields derived from a TM3 model run with and without NMHCchemistry have been used to calculate MCF oxidation. Figure2.9 indicates that the calculatedmean OH is represented in both cases within a few percent. Forthe NMHC run we calculatea turnover time due to the hydroxyl reaction in the troposphere of 5.5 years. Accounting forhydroxyl radical, ocean, and stratospheric losses (τstrat = 55 years) an overall MCF lifetimeof 4.7 years is derived for the total model domain, which agrees well with the MCF lifetimeestimate byPrinn et al.[1995] of 4.8± 0.3 years.

In Figure 2.10 the monthly mean hydroxyl radical concentration field of the NMHC runis presented, divided by that of the CH4-only run. Concentration ratios are shown for the“boundary layer,” i.e., the lowest three model layers (±1 km), as well as the “free tropo-sphere” (from boundary layer top to tropopause). Both figures indicate that OH is signif-icantly depleted over the continents, particularly in the lowest model layers. This effect iscompensated for over the oceans and in the upper layers of themodel. The largest dif-ferences are found over the Amazon rainforest due to the large biogenic emissions, whichstrongly reduce OH through isoprene chemistry. A sensitivity run without isoprene emis-sions shows that OH minima over the continents remain, but the effect is much smaller. Fromthese results it appears that, although the OH distributions of the NMHC and CH4-only runare quite different, these differences hardly affect the methyl chloroform test results. Thisfinding suggests that further OH distribution validation requires simulation of a tracer with amuch shorter lifetime. However, such a validation puts strong constrains on the accuracy ofemission estimates.

2.3.4 Nitrogen containing NMHC

Figures 2.11 to 2.12 show comparisons of calculated and simulated PAN mixing ratios at thesurface as well as vertical profiles. The Kejimkujik, Nova Scotia, data cover a time periodof about 5 years [Bottenheim et al., 1994]. The overall uncertainty of sampling and analysisis estimated at 30%. The Kolummerwaard, The Netherlands, PAN data have been derivedfrom a measurement record of 1991–1994 [Roemer, 1996]. It should be kept in mind thatall measurements represent peroxyacetyl nitrate, whereasour simulated PAN represents thesum of all PAN-like compounds. Therefore it is to be expectedthat the model overestimates“real” PAN measurements to some extent. Measurements of other peroxy nitrates indicate,however, that peroxyacetyl nitrate is the most abundant of all organic peroxy nitrates. Forexample, Shepson et al. measured peroxypropionyl nitrate (PPN) to peroxy acetyl nitrateratios in Ontario of 0.1 [Shepson et al., 1992]. At more remote stations, even lower PPN/PANratios have been observed [Singh et al., 1986].

The thermal decomposition of PAN is very sensitive to temperature changes, leading tolonger PAN lifetimes at relatively lower temperatures. As aresult, at Nova Scotia, maximumPAN mixing ratios are observed during February and March. This feature is reproduced bythe model; however, the PAN mixing ratios are overestimatedthroughout the year by at least afactor of 2. As will be illustrated in section 2.3.5, mixing ratios of NMHC are overestimated

2.3 Results 37

Figure 2.10: The ratio of hydroxyl radical concentrations, “NMHC” divided by “CH4-only”for July; boundary layer (top), and free troposphere (bottom). Ratios south of 60◦S are notgiven (because of the polar night). Black dashed lines indicate 1.0 contours.

by the model in regions with relatively strong biogenic emissions. PANs produced in iso-prene oxidation show an opposite seasonal dependence, withmaxima during summer whenisoprene emissions are strongest. Therefore an overestimated PAN production from isopreneoxidation is not likely to be the only explanation for the discrepancy between measured andsimulated PAN levels at Nova Scotia. At Kolummerwaard, simulated PAN levels agree wellwith the observations.

At Izana, Tenerife, at 2300 m altitude, the model significantly overestimates PAN duringwintertime. The monthly mean observations are derived frommeasurements between 1991and 1993 performed within the framework of EUROTRAC Tropospheric Ozone Research(TOR) (URL “www.tor.rivm.nl”). Similar differences are found at Mauna Loa (not shown),which indicates that the model may overestimate free tropospheric PAN mixing ratios. Thereaction of OH with PAN, which has not been accounted for, cannot explain the discrepancy,



100

200

300

400

PA

N(p

ptv)


200

400

600

800

PA

N(p

ptv)


500

1000

1500

2000

PA

N(p

ptv)

Figure 2.11: Observed versus modeled monthly surface concentrations of PAN. (top) Tener-ife (16◦W, 28◦N, 2368 m); squares, observations +10/90 percentile. (middle) Nova Scotia(65◦W, 44◦N); dots, mean of observations; dashed lines, 10/90 percentile of observations.(bottom) Kolummerwaard (53◦N, 6◦E); squares, observations + mean of cleanest and mostpolluted wind sectors. Solid line in each plot, TM3.

2.3 Results 39

0 100 200 300 400 500PAN(pptv)

1000900

800

700

600

500

400

300

200100

pres

s (h

Pa)

0 800 1600 2400 3200 4000NOY(pptv)

1000900

800

700

600

500

400

300

200100

pres

s (h

Pa)

Figure 2.12: Observed versus modeled vertical profiles. (left) PAN at Barrow (156◦W,71◦N, June), (right) NOy at North Bay (67◦W, 55◦N, June). Dots, measurements; solid lines,NMHC; dashed lines, CH4-only.

since the turnover time of PAN due to this reaction exceeds 0.5 year. The OH reaction withhigher PANs, of which no kinetic data are available, is expected to be faster. It cannot,however, explain the differences at Izana, as the contribution of higher PANs is only small.

The vertical PAN profile for June at Barrow [Singh et al., 1992] shows increased mix-ing ratios with altitude, as a result of a lifetime increase with decreasing temperature, whichis well represented by the model. In the upper model layers, PAN photolysis becomes in-creasingly important, and, as a result, a PAN maximum is predicted by the model at about300 hPa. A comparison of aircraft measurements of NOy over North Bay, Canada, for June[Singh et al., 1994] with model calculations indicates that NMHC chemistry in the modelconsiderably influences NOy under these conditions (see Figure 2.12). This fact does not,however, lead to a significant improvement of the agreement with the observations.

In Table 2.6 the observed average annual wet and total deposition of nitrate is comparedwith model calculations. The measurement compilation is taken fromDentener and Crutzen[1993], where further details can be found. The selected stations are classified as polluted,background, or tropical. In the NMHC run, more NOx is transported away from the sources,which leads to a lower nitrate deposition over polluted regions and higher values at back-ground stations as compared with the CH4-only run. In general, this changed nitrate deposi-tion distribution does not improve the agreement between model and observations. Both runsdeviate from the observations at some stations by more than afactor of 2, indicating that thewet deposition of nitrate is not well reproduced by the model. Subgrid scale variation in bothprecipitation patterns and NOx emissions may partly explain this.

Figure 2.13 shows measurements of gaseous HNO3 and particulate nitrate at remote sta-tions at different latitudes compiled byGalliardo-Klenner[1996] compared with model cal-culations. Both model runs underestimate nitrate at these locations; however, accounting forNMHC improves the agreement. The heterogeneous removal of organic nitrates on aerosols,which has not been accounted for, may contribute to this difference, although our presentknowledge about this process is insufficient.

Figure 2.14 shows the results of a TM3 run without PAN on surface level mixing ratios


Table 2.6.: Observed and Calculated Nitrate Wet Deposition

Site Coordinates Measured NMHC CH4-only

Polluted

Norwaya 69◦N 25◦E 45 73 88Sweden 57◦N 18◦E 345 184 218Baltic a 58◦N 22◦E 283 274 316Ireland 52◦N 10◦W 117 76 83Portugal 42◦N 7◦W 25 50 59Florida 25◦N 80◦W 162 70 69Massachusetts 42◦N 70◦W 271 165 208Nova Scotia 45◦N 63◦W 226 145 174New Brunswick 46◦N 67◦W 281 123 152Quebec 50◦N 67◦W 229 163 207Beijing 40◦N 116◦E 126 105 126Guizhou 27◦N 106◦E 164 77 129Hong Kong 22◦N 114◦E 226 63 79Linan 30◦N 120◦E 398 92 123Allabahad 25◦N 82◦E 144 104 132Ryori 39◦N 141◦E 174 116 143Congo south 4◦S 12◦E 181 131 134Congo north 5◦N 17◦E 407 87 105Nigeria 8◦N 8◦E 200 87 105

Background

Amsterdam Island 38◦S 78◦E 8 9 6Samoa 14◦S 170◦W 16 28 20Hawaii 20◦N 156◦W 17 21 15Torres de Paine 52◦S 73◦W 5 6 4Poker Flat 67◦N 147◦W 11 6 6Cape Grim 41◦S 144◦E 30 11 10Cape Point 34◦S 18◦E 28 7 7Dye 3 65◦N 43◦W 56 19 23South Pole 85◦S 1 4 3

Tropical

Amazon Basin 5◦S 55◦W 110 60 129San Carlos 2◦N 67◦W 175 35 106Lake Calado 10◦S 50◦W 100 59 112Costa Rica 10◦N 85◦W 50 75 71C. Amazonia 3◦S 60◦W 70 54 118

Nitrate wet deposition is in mg N m−2 yr−1

aTotal nitrate deposition (wet + dry).

2.3 Results 41

-40 -20 0 20 40 60lat(degrees)

0

50

100

150

HN

O3(

pptv

)

Figure 2.13: Comparison of a compilation of observed total HNO3 + nitrate at remote sitesat different latitudes and model results. Squares, measurements; solid line, NMHC; dashedline, CH4-only.

of NOx and O3 in July. These results confirm that a significant fraction of NOx is transportedfrom the continents to the oceans through PAN. NOx is particularly enhanced over tropicaloceans in the vicinity of the continents, in accord with the relative short PAN lifetime athigh temperatures. Ratios as large as a factor of 5 are found in these regions, which agreeswell with model calculations byMoxim et al.[1996]. The effect on ozone is in agreementwith the expected effect of the changed NOx distribution; similar patterns can be seen inthe distribution of OH (Figure 2.10). The annual wet deposition of nitrate over the tropicalcontinents for the NMHC run (Table 2.6), however, suggests that this transport of NOx fromthe continent to the ocean is overestimated by the model.

2.3.5 Nonmethane hydrocarbons

The CBM-4 compound PAR is compared with the sum of C2-C6 paraffins as measured fromAugust 1993 until July 1994 in New England, United States [Goldstein et al., 1995]. Theoverall accuracy of the analysis method is estimated to be better than 18%. The error bars inFigure 2.15 represent the 0.1 to 0.9 quantile of the measuredlognormal concentration distri-bution per month. The agreement between model results and measurements is reasonable, inparticular during summer. A small overestimation by the model is to be expected, since PARalso represents>C6 compounds. However, it cannot fully explain the differencebetweenmodel and measurements during winter.

In Table 2.7 a number of published hydrocarbon measurementshave been compared withmodel results. A selection has been made such that regions with different chemical conditionsare covered. The agreement between measurements ofShepson et al.[1993] and simulatedPAR in Ontario is quite comparable to the differences found for New England. At Birkenes,Norway, the differences may well be explained by the contribution of >C5 components inPAR. The rate of the reaction between PAR and OH, (R46), has anaverage value represtative


Figure 2.14: (top) Ratio of surface NOx mixing ratios for July comparing NMHC runs with-out/with PAN. (bottom) A comparison as in the top panel for surface O3 mixing ratios forJuly.

of hydrocarbon mixtures in the vicinity of sources. As a consequence, the PAR removalat larger distances from the sources is too fast, leading to underestimation of PAR over thePacific Ocean and, to a lesser extent, over the Atlantic Ocean.

The calculated isoprene mixing ratios are considerably higher than indicated by mea-surements. The spatial concentration distribution of thiscompound is characterized by largegradients in the vicinity of sources due to its short chemical lifetime. Therefore many subgridscale factors may obscure the comparison of modeled and measured isoprene mixing ratios.Nevertheless, isoprene mixing ratios have often been measured close to the sources for ex-ample in forests, so that it is to be expected that measured values are higher than the modelpredictions. Since the opposite is found in regions with relatively high biogenic emissionslike the Amazon basin and the southeastern the United States, we conclude that isoprene issignificantly overestimated by the model. The simulated total NMHC carbon in Alabama

2.4 Discussion 43


10

20

30

40

5060

par(

ppbv

)

Figure 2.15: Observed and modeled surface total paraffinic carbon in NewEngland (72◦W,43◦N). Squares, observations + 10/90 percentile; solid line, TM3 model.

appears to be mainly determined by isoprene and its oxidation products, which also leads toa significant overestimation as compared with measurementsof Goldstein et al.[1995].

At Tsukuba, Japan,Yokouchi[1994] observed little difference in the isoprene concentra-tion during summer and autumn. It has been suggested that smaller emissions during autumnare balanced by a longer chemical lifetime. This balance is not confirmed by our results, sincethe model predicts a tenfold decrease of isoprene during autumn compared with summer. Itindicates that in this region the estimated isoprene emission seasonality may not be correct.

2.4 Discussion

The comparisons between model results and measurements show that the model satisfactorilyreproduces observed mixing ratios of several key compoundsimportant in NMHC chemistry.When interpreting these results, an important point of concern is to what extent these modelresults and measurements can be expected to agree. In general, differences in spatial andtemporal scales limit the comparability. In polluted regions the relatively high variability of-ten requires a larger number of observations than are available to determine the concentrationdistribution at a particular site. One way to deal with this problem is to test global models atremote sites only, where local influences are relatively small. However, the largest NMHCeffects are expected in source regions, so that we necessarily maintain this ill-conditionedcomparison, failing a better method to test our model results. In all cases we tried to selectmeasurements including the largest amounts of data to minimize influences of local vari-ability. Therefore for species like ozone and carbon monoxide, multiyear averages have beencompared with model results, although the latter are based on one particular year of ECMWF-derived meteorology (1993). TM3 results obtained with 1992meteorology generally indicatea relatively small interannual variability, and the general picture, as appearing from the com-parisons presented, changes little. A more detailed study focusing on interannual variability


Table 2.7.: Hydrocarbon Measurements and Model Calculations

Compound Location Coordinates ModelaMeasureda Ref.

Σ NMHC carbon Alabama (U.S.) 32◦N, 88◦W 201 85 GolΣ paraffinic carbon Ontario (Canada) 44◦N, 80◦W 42 32 SheC2-C8

b Atlantic Ocean 55◦N, 20◦W 0.9 2.5 PenC2-C8

c Atlantic Ocean 55◦N, 20◦W 15 12 PenC2-C4 Pacific Ocean 0◦N, 150◦W 0.2 2.4 AtlC2-C5

b Norway 59◦N, 8◦E 11 9 HovC2-C5

c Norway 59◦N, 8◦E 27 22 Hov

Isoprene Amazon 60◦W, 3◦S 71 2.7 ZimIsoprene Pennsylvania (U.S.) 42◦N, 78◦W 0.7 5.8 MarIsoprene Ontario (Canada) 45◦N, 79◦W 0.8 3.8 LinIsoprene Alabama (U.S.)d 32◦N, 88◦W 21 3.8 GolIsoprenee Tsukuba (Japan) 36◦N, 140◦E 0.2 0.3 YokIsoprenef Tsukuba (Japan) 36◦N, 140◦E 0.02 0.2 Yok

Gol, Goldstein et al.[1995]; She,Shepson et al.[1993]; Pen,Penkett[1993]; ATl, Atlaset al. [1993]; Hov, Hov et al. [1991]; Zim, Zimmerman et al.[1988]; Mar, Martin et al.[1991]; Lin, Lin et al. [1992]; Yok,Yokouchi[1994].

aConcentrations are in ppbv.bDuring summer.cDuring winter.dDaytime measurements versus diurnal averaged model.eDuring July.fDuring November.

will be performed in the future. Another criterion used in selecting measurements has beenthe representation of different chemical regimes. It is considered a principal requirement ofa global model that accurate results are obtained under all conditions.

In spite of the insufficient model resolution in urban regions, the modeled and measuredozone levels agree reasonably well, i.e., within 30% at mostlocations. Considering the mag-nitude of subgrid scale variations at the polluted sites, better agreement can hardly be ex-pected. The representation of NMHC chemistry clearly improves the ability of the model toreproduce summertime ozone levels in these regions. This improvement has, for example,been demonstrated for Payerne, Switzerland, and Tateno, Japan. For European stations inparticular, the poor treatment of aromatics in our scheme may influence ozone predictions.Considering the large variation in observed ozone at different stations close to Payerne, how-ever, it is expected that the coarse model resolution puts a much larger constrain on the abilityto reproduce ozone over this continent. At Wallops Island, our NMHC chemistry represen-tation leads to an overestimation of ozone during summer, although a satisfactory agreementwith measurements is found during other seasons. From comparisons of simulated hydro-

2.4 Discussion 45

carbon levels to observations, it appears that in regions with large biogenic emissions ourpresent NMHC treatment is hardly capable of reproducing measured concentrations. There-fore in some parts of the United States, the oxidation of NMHCis suppressed and, as a result,NMHC mixing ratios are strongly overestimated. At the east coast of the United States NOx

emissions are sufficiently high to effectively recycle OH. This fact, in combination with anoverestimated transport of NMHC from the adjacent gridboxes, may explain the high simu-lated ozone levels at Wallops Island during July and August.

It has been shown that carbon monoxide measurements are quite well reproduced by themodel. The introduction of the EDGARV2.0 CO emission inventory in our model clearlyimproved the predicted CO levels over the continents. A comparison of the column-averagedCO mixing ratio to MAPS measurements indicates that the applied biomass burning emis-sions may be too low. To some extent, it may also explain the underestimated CO levels atthe remote stations Samoa and Barbados. Since the MCF test indicates that the global hy-droxyl radical level is reproduced by the model within a few percent, this underprediction islikely related to a source of CO rather than the sink of OH. We calculated that by omittingthe methanol production from methane oxidation, at most 30 Tg/yr CO per year is neglected,which corresponds to 1.3% of the total CO emission only.

It has been demonstrated that the representation of NMHC chemistry reduces OH in thecontinental boundary layer. Since the MCF simulation appears to be quite insensitive to con-tinental scale differences in OH distribution, this OH testdoes not indicate whether the OHdistribution improves by the representation of NMHC. The comparison of simulated hydro-carbon concentrations to measurements suggests that the OHlevel may be underestimatedover the continents when NMHC chemistry is used, especiallyin regions where biogenicemissions dominate. Accumulation of isoprene to levels of 50 ppbv or even higher occursin the model, in particular, in regions where high isoprene emissions coincide with low NOxabundances. The chain length of radical propagation reduces under low NOx conditions,since radical termination due to peroxide formation increases. In addition, isoprene oxidationyields organic nitrates and PANs, which further reduces theavailability of NOx. As the OHlevel decreases, the isoprene concentration further increases, and consequently the isoprene–O3 reaction becomes an increasingly important ozone sink. These oxidant-depleting chemicalconditions occur in the model over the Amazon basin, which results in local OH and ozoneminima. In contrast,Jacob and Wofsy[1990] find that the observed ozone concentrations inthe Amazon boundary layer are largely regulated by dry deposition and influx from the freetroposphere, which implies that the net effect of boundary layer photochemistry on ozonemust be relatively small. Measurements of isoprene (<5 ppbv) and ozone (10 ppbv) in theAmazon region [Zimmerman et al., 1988] also suggest that the simulated isoprene concentra-tions are to a large extent an artifact of the model. To account for the likely overestimationof isoprene emissions, the emitted amount has been reduced by 100 TgC/yr compared withthe 500 TgC/yr recommended byGuenther et al.[1995]. Given the uncertainty in this esti-mate, which may exceed a factor of 3 according toGuenther et al.[1995], we consider thisreduction to be justified. Since the modified CBM-4 overestimates OH relative to a moredetailed scheme under these conditions, it is not likely that a more detailed chemical schemewould lead to much improvement. In addition, the coarse model resolution may be an impor-tant factor. Because of the relatively short lifetimes of NMHC compared with methane andCO, spatial gradients of photochemical NMHC reactants are relatively large. Therefore theconcentration gradients of the short lived NMHC compounds are not well resolved. Also thespatial resolution at which other processes are described,for example, photolysis rates, may,as a result, become increasingly critical.


A fundamental difference between the modified CBM-4 and moredetailed mechanismsis that the first applies uniform reaction rates to compoundsbelonging to the same lumpedgroup. As a result, nearly all NMHC are chemically removed within a few thousand kilome-ters from the sources. The underestimation of hydrocarbon concentrations at larger distancesis not a point of concern, since at these locations its effecton photochemistry is marginal.Overestimated NMHC oxidation near the sources, however, may lead to significant errors.It is inherent to parameterizations of NMHC chemistry that many reactions directly yieldend products, since intermediate compounds of reaction pathways have been eliminated. Asa consequence, the scheme tends to oxidize carbon atoms too fast, which also leads to anoverestimation of NMHC oxidation in the vicinity of sourcesin the model.

Further, heterogeneous removal of NMHC has been a largely unexplored field. For ex-ample, the neglect of carbonaceous particle formation may partly explain the overestimationof NMHC and, consequently, of CO. In fact, the “bookkeeping”of carbon atoms in the PARrepresentation of CBM-4 is very strict. As a result, paraffins are almost completely oxidizedto CO and ultimately to CO2, whereas a significant fraction of the reaction intermediates arenonvolatile or become soluble. Therefore it is to be expected that heterogeneous removal ofhydrocarbons becomes important at some stages during theiroxidation.

2.5 Conclusions

We have developed an NMHC oxidation mechanism for global scale modeling of tropo-spheric photochemistry, with the aim to improve the simulation of key compounds such asO3, CO, OH, and NOx, using the least possible amount of chemical detail. Box model testsindicate that, over a range of chemical conditions, the agreement between our scheme and anextensive reference scheme (RACM) is well within 25% for most compounds. Occasionally,somewhat larger differences are found for OH; however, within the range of chemical condi-tions characteristic for global scale photochemistry, this difference does not exceed 50%. Byaccounting for NMHC chemistry, the agreement between modeled and observed ozone overindustrialized regions improves significantly. During summer, at some stations the monthlymean ozone mixing ratios increase by a factor of 2 owing to NMHC chemistry. Integratedover the troposphere, we calculate an increase of about 40% in photochemical ozone produc-tion, which leads to a 17% increase of the tropospheric ozonecolumn. Biogenic emissionscontribute about 70% to this increased ozone production. Considering the discrepancies be-tween model results and observations, which are notably large in regions with high biogenicemissions, these numbers are quite uncertain.

The agreement between modeled and observed CO does not significantly improve whenNMHC chemistry is accounted for, although a better agreement with the observations is foundat the remote stations. Since the CO concentrations are largely determined by the surfaceemissions applied, which are quite uncertain, the differences at these stations may well be ex-plained by an underestimation of CO emissions. The model is able to reproduce MCF at theALE/GAGE stations within a few percent, slightly overestimating MCF in the northern hemi-sphere. Since the MCF sinks to the ocean and stratosphere arerelatively small, deviations ofthe tropospheric OH content remain within a few percent. Therepresentation of NMHC hasa marginal effect on the overall tropospheric OH burden; however, a drastic change in OHdistribution is found. Near the surface over the continents, OH is depleted owing to reactionswith hydrocarbons. This depletion is particularly evidentunder conditions of low NOx andhigh biogenic NMHC. A comparison of observed and simulated levels of NMHC suggests

2.5 Conclusions 47

that the model underestimates OH near the surface.Deviations between calculated and observed concentrations of NMHC, such as paraffins

and isoprene, at least partly result from the crude representation of NMHC in TM3. Weconclude that the modified CBM-4 is not well suited for studying the global distribution andbudget of particular hydrocarbon compounds. The chemical shortcomings do not, however,seem to strongly influence the ability to reproduce the overall oxidant budget, as indicatedby the simulated levels of ozone and OH. Therefore the schemefulfills the primary require-ment that the simulation of global scale photochemistry is improved. The simulated effectsof NMHC on photochemistry confirm the need of representing NMHC chemistry in globalmodels. Since detailed chemical mechanisms are too computationally expensive for globalthree-dimensional models, we conclude that our modified CBM-4 offers an acceptable com-promise.

Acknowledgments

We like to thank Peter van Velthoven and Ad Jeuken of the KNMI for their efforts to developTM3 and to preprocess ECMWF data. We are also grateful to PaulNovelli for providing COobservations at polluted locations. Further, we like to thank Maria Kanakidou and NathaliePoisson for their help with box model chemistry intercomparisons, and useful discussions.This work has been supported by the Dutch Global Change program, NOP project 951202and the EC supported SINDICATE project.

Chapter 3

Inverse modeling of methanesources and sinks using the adjointof a global transport model

An inverse modeling method is presented to evaluate the sources and sinks of atmosphericmethane. An adjoint version of a global transport model has been used to estimate thesefluxes at a relatively high spatial and temporal resolution.Measurements from 34 monitor-ing stations and 11 locations along two ship cruises by the National Oceanographic andAtmospheric Administration have been used as input. Recentestimates of methane sources,including a number of minor ones, have been used as a priori constraints. For the targetperiod 1993–1995 our inversion reduces the a priori assumedglobal methane emissions of528 to 505 Tg(CH4) yr−1 a posteriori. Further, the relative contribution of the NorthernHemispheric sources decreases from 77% a priori to 67% a posteriori. In addition to makingthe emission estimate more consistent with the measurements, the inversion helps to reducethe uncertainties in the sources. Uncertainty reductions vary from 75% on the global scale to∼1% on the grid-scale (8◦x10◦), indicating that the grid scale variability is not resolved bythe measurements. Large scale features such as the interhemispheric methane concentrationgradient are relatively well resolved and therefore imposestrong constraints on the estimatedfluxes. The capability of the model to reproduce this gradient is critically dependent on theaccuracy at which the interhemispheric tracer exchange andthe large-scale hydroxyl radicaldistribution are represented. As a consequence, the inversion-derived emission estimates aresensitive to errors in the transport model and the calculated hydroxyl radical distribution. Infact, a considerable contribution of these model errors cannot be ignored. This underscoresthat source quantification by inverse modeling is limited bythe extent to which the rate ofinterhemispheric transport and the hydroxyl radical distribution can be validated. We showthat the use of temporal and spatial correlations of emissions may significantly improve ourresults; however, at present the experimental support for such correlations is lacking. Ourresults further indicate that uncertainty reductions reported in previous inverse studies ofmethane have been overestimated.

1Accepted for publication inJournal of Geophysical Research, with T. Kaminski, F. J. Dentener, J. Lelieveld,and M. Heimann as co-authors.

49

50 Inverse modeling of methane

3.1 Introduction

Our understanding of atmospheric methane is to a large extent limited by a poor quantificationof its sources. Although the dominant methane-producing processes seem to be identified,their global distributions as well as globally integrated source strengths are still highly un-certain. Some sources for which relatively reliable statistics exist, for instance, emissions bydomestic ruminants, are relatively well constrained by integrating or scaling-up the sources.For sources that have a large natural variability, for example, emissions from rice paddiesand natural wetlands, the opposite is true. Despite considerable effort to constrain these im-portant sources, progress is slow. For example, the amount of methane released from ricepaddies has recently been estimated at 80±50 Tg(CH4) yr−1 [Lelieveld et al., 1998], in-dicating an uncertainty range similar to theIPCC [1990] estimate of nearly a decade ago(20–150 Tg(CH4)/yr). Recent estimates of global methane emissions from natural wetlands,the largest single source of methane, byLelieveld et al.[1998] andHein et al.[1997] disagreestrangely (145±30 and 232±27 Tg(CH4) yr−1, respectively).

Measurements over the past decade show a significant variability of the globally aver-aged methane trend. The fluctuations in the early 1990s have been merely attributed to theMount Pinatubo eruption, but the underlying processes remain unclear [Dlugokencky et al.,1994a, 1996]. Recent measurements indicate that the methane trend continues to decline[Dlugokencky et al., 1998]. For the changes observed so far the sources and sinksof methaneare not well enough constrained to provide a reliable explanation. As a consequence, thefoundation for any future scenario of methane in the atmosphere is weak. It is important toimprove our understanding of the atmospheric budget of methane since it plays an importantrole in radiative and chemical balances in the atmosphere.

In addition to upscaling techniques, inverse modeling can be used to estimate sources.This method makes use of measurements of trace gas concentrations, which are translatedto constraints on the sources by means of an atmospheric transport model. Generally, thenumber of available measurements is a limiting factor in this approach. For such an inverseproblem to have a single solution either the number of emission parameters to be estimatedmust be adjusted according to the available measurements, or different constraints must beintroduced. Often it is required that the solution be close to existing knowledge by intro-ducing first guess or a priori information of the sources. This can be done in a consistentway by adopting a so-called Bayesian approach, in which all parameters are expressed asstatistical probability distributions. A solution in the form of a superposition of all statisticaldistributions involved can be computed, from which means and covariances can be derived.

Until now a number of studies have quantified global-scale sources using inverse meth-ods aiming at different trace gases, such as CFCs [Brown, 1993;Hartley and Prinn, 1993;Plumb and Zheng, 1996;Mulquiney and Norton, 1998;Mulquiney et al., 1998], CH4 [Brown,1993, 1995;Hein et al., 1997], CO2 [Enting et al., 1995;Rayner et al., 1996;Law and Sim-monds, 1996;Kaminski et al., 1999b]. In all these studies except the last, fluxes are aggre-gated into a few large regions. This “big region” approach has the disadvantage that emissiondistributions over predefined regions are assumed to be perfectly well known, which, in prac-tice, is justified by computational limitations rather thanby knowledge about the sources. Infact, such an approach may lead to significant biases in the estimated emissions caused bynonhomogeneous sampling by the measurement network as shown byTrampert and Snieder[1996]. Presently, it is unknown at which resolution fluxes should be represented in order toreduce this bias to an insignificant size. Sensitivity studies, however, show that significantbiases occur when continental-scale regions are applied (T. Kaminski, personal communica-

3.2 Inversion method and unknowns 51

tion, 1998).The aim of this paper is to study the sources and sinks of methane on a global scale and

to assess the usefulness of the existing measurement data asconstraints. In this study, as op-posed to previous studies, biases are minimized using the method ofKaminski et al.[1999a].To compute efficiently the large number of sensitivities needed to perform an inversion in thisapproach, an adjoint model is used. Except for the definitionof the fluxes and the methodto derive the sensitivities the method we adopt resembles the method byHein et al.[1997].Compared with this study, the measurements and first-guess assumptions have been extendedand updated.

Using our method, the inverse problem is strongly underdetermined because of the largeincrease in the number of unknowns compared withHein et al.[1997], whereas the numberof constraints (measurements) is about the same. This can beinterpreted as a reduction of apriori information since temporal and spatial distributions of the sources and sinks within theregions are no longer assumed to be well known. In fact, we choose the opposite starting pointby assuming that the uncertainties in the fluxes are fully uncorrelated. From this viewpointthe big region approach can be considered as a case in which all fluxes over a region areassumed to be fully correlated. Hence our confidence in the a priori emission distributionscan be expressed in the spatial and temporal correlation of source and sink uncertainties. Ourmethod allows different scenarios for treating these correlations.

In the next section the inverse modeling technique will be explained. We describe themodel in section 3.3, a priori knowledge on sources and sinksand measurements in sections3.4 and 3.5, and the relevant chemistry in section 3.6. In section 3.7 the results of the methaneinversion are presented, focusing on differences between apriori and a posteriori estimates.The impact of correlated uncertainties is studied in section 3.8, where we also present resultsusing different scenarios. Sensitivity tests have been performed to investigate the potentialinfluence of the major assumptions on the estimated sources and sinks. Finally, the ability ofthis method to solve questions related to different emissions scenarios has been studied withrice paddies as a test case in section 3.9.

3.2 Inversion method and unknowns

To deduce information about surface emissions from measured concentrations, we use aglobal atmospheric chemistry transport model (CTM) that relates these emissions to con-centrations. In mathematical terminology the CTM defines a mapping (or function) of aparameter set (including the emissions) on the simulated concentrations. For any measuredmethane concentrationdi,t at locationi and montht this function can be written in the form

di,t = di,t [f,Q(f,sOH,sstra,c0),R(f,soh,sstra,c0),c0], (3.1)

where any componentf j ,m of the vectorf represents the integrated surface emission over re-gion j and monthm. Sincef j ,m may either represent a surface source or sink, in the followingthe term “flux” will be used to denote exchange between the surface and the atmosphere. ThesOH parameterizes the chemical removal of methaneQ due to the reaction with the hydroxylradical in each tropospheric grid box. Similarly,sstra andR refer to stratospheric methaneloss as a result of the reactions with O(1D), Cl, Br, OH radicals, and methane photolysis.Finally, c0 is defined as the global mean methane concentration at the start of a simulation.The aim of the inversion is to find the set of most probable values for the parametersf, sOH,


sstra, andc0. For this purpose we iteratively apply a linear inversion procedure that combinesthe set of observationsd and a set of a priori estimatesf, sOH, sstra, andc0 using (3.1).

We need to iterate because the chemistry introduces a nonlinearity as explained below.The amount of methaneQk,l oxidized by hydroxyl radicals in a particular troposphericgridboxk and monthl can be represented as

Qk,l = sOHkOH(T)[OH]k,l [CH4]k,l , (3.2)

wherekOH is the temperature-dependent rate constant for the reaction between the hydroxylradical and methane. The dependency of the amount of methaneoxidized on its concen-tration [CH4]k,l causes a nonlinear response to the fluxes at the observational sites. Sincetropospheric methane oxidation influences the hydroxyl radical concentration, there is also afeedback mechanism through changes of[OH]k,l . Similar nonlinearities occur because of thestratospheric sink.

Different strategies can be adopted to define the fluxes represented by the parametersf j ,m. One approach is to classify fluxes geographically, for example, per continent or country[see, e.g.,Hartley and Prinn[1993];Mulquiney et al.[1998]]. Alternatively, a classificationdistinguishing different flux-producing processes can be used, as is done, for instance, byHein et al. [1997]. Unless both the emission distributions over the specified regions andthe model are perfect, non homogeneous sampling by the sparse network will lead to biasedemission estimates [Trampert and Snieder, 1996]. This bias can be reduced by decreasing thesize of the regions or by defining the regions such that the uncertainty in the flux distributionover each region is minimal. Because of the large spatial variability of CH4 emissions, welack such well-defined regions. Therefore, over the continents we choose to represent thesurface flux in every continental surface grid box of our CTM by a separate flux parameter. Incontrast, ocean grid boxes without linkages to continents or continental shelfs are aggregatedinto one single region. For all fluxes, including those from the oceans, a temporal resolutionof 1 month is used; hence seasonal cycles are also subject to optimization.

The open ocean source is mainly driven by a small methane supersaturation in the surfacelayer, ranging from 0.95 to 1.17 over the oceans [Bates et al., 1996]. Although the methaneconcentration in ocean water has been determined at a limited number of sites only, the avail-able data suggest that the net emission is only small and thatthe variability is not large enoughto influence significantly the atmospheric methane concentration distribution. The north andsouth polar ice shields are treated similarly to the oceans,which is justified by the absenceof any known significant surface source or sink in these regions. As a result of both fluxaggregations, the total number of surface flux parameters isreduced by about a factor of 2.

In (3.2) we scale the amount of methane oxidized in every tropospheric grid box by asingle parametersOH, which implies that the entire spatiotemporal distribution of the OHradical field is kept constant. Unlike the ocean and polar sources, such a treatment of theOH sink is not sustained by our knowledge of the troposphericOH distribution. Treatingthe OH sink this way is equivalent to assuming that the OH distribution is perfectly known,which is not the case. In fact, we lack the tools to validate model-simulated OH distributionsfor aspects other than large-scale averages. The fact that other reactive trace gases, such ascarbon monoxide and ozone, are simulated realistically by our model [Houweling et al., 1998]indicates, however, that the large-scale chemistry and transport are adequately represented.Therefore we fix the OH distribution, assuring that the optimized methane flux fields areconsistent with our knowledge of global tropospheric chemistry. We return to this discussionin section 3.6.

3.2 Inversion method and unknowns 53

For the stratospheric sink we used a fixed spatiotemporal CH4 loss distribution derivedfrom the two-dimensional (2-D) chemistry transport model of Bruhl and Crutzen[1993] sinceour CTM does not have an accurate representation of the stratosphere. The error introducedby this method is expected to be small as it appears that surface methane concentrations arerelatively insensitive to changes in this distribution because of the relatively long exchangetime between the troposphere and stratosphere. Formally, the stratospheric sink also intro-duces a nonlinearity. Since the CH4 loss distribution has been fixed, the feedback of themethane concentration on stratospheric oxidation is neglected, which eliminates this nonlin-earity.

The linearized form of (3.1), including the simplificationsas discussed above, is repre-sented by

di,t = ∆c0 + ∑j ,m

∂di,t

∂ f j ,m∆ f j ,m−

∂di,t

∂sstra∆sstra

− ∑k,l

∂di,t

∂Qk,l(∑

k,l

∂Qk,l

∂[CH4]k,l∆[CH4]k,l +

∂Qk,l

∂sOH∆sOH)

+ di,t [f,Q( ˜[CH4], sOH), sstra, c0], (3.3)

where the tilde designates a first-guess estimate. In general, the relative changes in the CH4

concentration resulting from the inversion-derived changes in the fluxes are small (<5%), so

that the term∂Qk,l

∂[CH4]k,l∆[CH4]k,l in (3.3) can be neglected. In a second iteration of the inversion

procedure the changes in[CH4]k,l are taken into account by linearizing around the improvedvalues forf,sOH,sstra andc0. Generally, after the second iteration the changes in the resultshave become insignificant (<1%).

In vector notation, (3.3) can be formulated as

∆d = T∆f −S∆s+ ∆c0, (3.4)

wheres includes bothsOH andsstra. The Jacobian matrixT contains concentration responsesto the surface fluxes at the measurement sites as a result of atmospheric transport only. Thismatrix is efficiently computed using the adjoint version of our CTM since the number ofmeasurement stations is small compared with the number of surface flux parameters [Kamin-ski et al., 1999a]. Since the atmospheric transport acts linearly on the concentration and thesources are assumed to be independent of the atmospheric methane mixing ratio, the corre-sponding terms in (3.4) are linear.

The matrixS, representing the concentration response to methane sinksin the atmosphere,is calculated using the standard “forward” model version. First, the tropospheric OH distri-bution is computed using the full chemistry version [Houweling et al., 1998]. The OH fieldsobtained are used in a methyl chloroform (CH3CCl3) simulation and scaled to give an opti-mal fit to observed CH3CCl3 concentrations (see section 3.6). In a second model run these

scaled OH fields are used to compute the∂Qk,l∂sOH

using ˜[CH4]. The stratospheric sink responseis determined in a separate run of the forward model using theprescribed sink distribution.

A model spin-up time of 3 years has been applied to derive bothmatrices in (3.4), withthe responses being stored in the fourth year. In each year ofmodel simulation the samesources and sinks are used. According toHein et al.[1997] andTans[1997] a period of 3years is sufficiently long for the atmosphere to be well mixed. As a result, the global-scaleconcentration gradients, caused by the constant fluxes, change little after this period, and theinitial concentrations contribute to the global mean only.


The solution to the inverse flux optimization is defined as theset of parameter valuesthat optimally satisfy two requirements. First, the optimized or a posteriori fluxes shouldbe as close as possible to the first-guess (a priori) fluxes. Second, the inversion-derived (aposteriori) fluxes should lead to an as close as possible agreement between modeled andmeasured concentrations. In both instances, misfits are quantified as sums of squares in unitsof observational and a priori standard deviations, respectively. Mathematically, this solutioncorresponds to the minimum of a cost functionJ defined as

J (x) =< d(x) − dobs,C−1d (d(x) − dobs) > + < x − xapr,C−1

x,apr(x − xapr) >, (3.5)

where<> denotes an inner product, the subscriptsobsandapr denote observed and a priorivalues, andCd andCx,apr are the covariance matrices for the corresponding vectorsdobs andxapr (x contains bothf ands). It can be shown that at the minimum ofJ the fluxes satisfy

x = xapr +(ATC−1d A +C−1

x,apr)−1ATC−1

d (dobs−Axapr), (3.6)

where matrixA contains bothT and S [Tarantola, 1987]. A second, equally important,result of the inversion is the uncertainty of the inferred parameters, which is derived fromthe uncertainties of the observations and the uncertainties of the a priori estimates. The aposteriori covariance matrix of the uncertainty in the inferred parameters is [Tarantola, 1987]:

Cx = (ATC−1d A +C−1

x,apr)−1. (3.7)

Technically, (3.6) and (3.7) are solved by means of a singular vector decomposition.


In this study we use the “off-line” global 3-D atmospheric Transport Model 2 (TM2) de-veloped byHeimann[1995]. The spatial resolution of the model is 10◦ in the longitudinaldirection and 7.8◦ in the latitudinal direction with nine vertical sigma levels from the surfaceup to 10 hPa (∼30 km altitude). The horizontal and vertical transport of tracers is basedon 12 hourly mean air mass fluxes and sub-grid scale transportdata. For this purpose, ana-lyzed meteorological fields from the European Centre for Medium-Range Weather Forecast(ECMWF) model have been preprocessed as described byHeimann and Keeling[1989] andHeimann[1995]. In the present study, ECMWF-analyzed data for the year 1987 have beenused for each year of model simulation.

The advective transport is calculated using the “slopes scheme” of Russell and Lerner[1981]. The sub-grid scale convective mass fluxes are evaluated using the cloud schemeof Tiedke[1989], which includes entrainment and detrainment in updrafts. Turbulent verticaltransport is based on stability-dependent vertical diffusion [Louis, 1979]. The tracer transporthas been validated by comparing85Kr, SF6, 222Rn, and CFC-11 measurements to simulations.The results indicate that the boundary layer mixing over thecontinents is reasonably wellrepresented. The cross-equatorial transport, however, issomewhat underestimated in themodel (see section 3.6).

To compute relationships between fluxes and concentrationsefficiently, an adjoint versionof the model has been used. The adjoint model was developed byKaminski et al.[1996]using the tangent-linear and adjoint model compiler (TAMC)by Giering [1996]. This modelhas previously been applied in a study of the global sources and sinks distribution of CO2[Kaminski et al., 1999b].

3.4 Measurements 55

To simulate the interactions between methane and tropospheric photochemistry, the for-ward code has been extended with the photochemistry routineof Houweling et al.[1998].Chemical equations are integrated using a Eulerian backward iterative (EBI) scheme as for-mulated byHertel et al. [1993]. Emissions of photochemical tracers other than CH4 arebased on Global Emission Inventory Activity (GEIA) and Emission Database for Global At-mospheric Research (EDGARV2.0) inventories [Olivier et al., 1996;Guenther et al., 1995;Yienger and Levy, 1995;Benkovitz et al., 1996]. The treatment of wet and dry depositionof tracers is comparable toHouweling et al.[1998]. Photolysis rates are derived from theradiation code ofBruhl and Crutzen[1988], and daytime average values are updated twicea month. As a first-order approximation of the diurnal variation of photolysis rates, a singleperiod of a normalized (sin)2 function is used during daytime.

3.4 Measurements

Observational data have been derived from measurements at 34 monitoring stations and 11positions along two cruise tracks of the National Oceanic and Atmospheric Administration(NOAA)/ Climate Monitoring and Diagnostics Laboratory (CMDL) cooperative air samplingnetwork [Dlugokencky et al., 1994b] (see Appendix 6). In general, a duplicate sample at eachstation is taken about once a week. The precision of the gas chromatographic analysis is es-timated at 0.2% [Dlugokencky et al., 1994b]. Measurements “flagged” by NOAA, indicatingthat the sampled air may have been influenced by local sources, have not been used.

The NOAA methane concentration records show significant interannual variability causedby year to year differences in methane fluxes and atmosphericcirculations. These variabilitiescannot be reproduced by the model since the same fluxes and winds are used for each year ofmodel simulation. The assumption of constant fluxes impliesthat a constant methane trendis also computed. To satisfy this assumption as much as possible for recent years, we havechosen January 1993 until December 1995 as a target period. Stations with a regular datarecord over this period have been selected. At each station amultiyear averaged seasonalcycle and a spline fit to the trend have been computed, on the basis of data sampled over thistarget period.

A global mean surface CH4 trend has been derived from a linear fit to the spline trends at21 selected background locations (see Appendix 6). For thispurpose, stations were selectedfrom which a reliable trend over this relatively short target period could be derived, favoringremote stations with little data gaps. In this way we computed a trend of 6.1 ppbv yr−1 witha 1σ uncertainty interval of 1.2 ppbv yr−1 over the period 1993–1995. For the 5.06x1018 kgatmospheric load in our model this corresponds to 17.1±3.5 Tg(CH4) yr−1. From the globalmean trendT , the multi-year averaged seasonal cycleS , and the interceptI per station at themidpoint of the target period (tmid) the observations are derived as

dobs(i,t) = Ii +(t − tmid)T + Si,t . (3.8)

The uncertainty in the derived monthly mean concentration is taken to be the root meansquare of the residual between the averaged duplicate samples and the fitted data as derivedfrom (3.8). The contribution of the analysis error to the uncertainty is small and has thereforebeen neglected. This data-fitting procedure is described indetail byHein et al.[1997].

The ship-based measurements over the pacific (P01–P05) and South China Sea (SC1–SC7) have been sampled at 5◦ and 3◦ latitude intervals, respectively. Because of variationsin the cruise tracks, the longitudinal coordinates of the cruise measurements are not constant.


From the cruise-specific coordinates (E. J. Dlugokencky, personal communication, 1997) av-erage longitudes have been computed. Data from cruises thatfollowed alternative tracks havebeen rejected. Variations in the longitudinal coordinatesare generally within 10◦ over the Pa-cific and 2◦ over the South China Sea. These differences are considered to be reasonablysmall given the horizontal resolution of our model and the relatively smooth methane con-centration gradients over the Pacific. The Pacific locationsP01–P05 are only a minor subsetof all Pacific cruise locations at which samples are taken by NOAA. For the remaining loca-tions, however, too little data were available over the target period to derive reliable seasonalcycles.

3.5 A priori assumptions

Recent estimates of the global distribution and annual total for each surface source and sinkprocess have been used as a priori inputs. Table 3.1 lists allthe surface fluxes accounted for,the associated annual totals, and the assumed uncertainties per process. It should be notedthat the applied inverse modeling formalism (as outlined insection 3.2) requires that the first-guess information be independent of the measurements. The applied emission distributions toa large degree satisfy this requirement, being merely basedon bottom-up derived statistics of,for example, population density, industrially produced quantities, etc. The global budget maywell have been taken into consideration, however, in establishing the first-guess annual totaland corresponding uncertainties. Although, formally, it is incorrect to use these numbers,they are being used in this study for lack of alternative estimates. In this section, only thosesources and sinks are discussed that are not fully explainedby the references listed in Table3.1. Section 3.6 discusses the a priori treatment of the tropospheric sink.

On the basis of a compilation of methane measurements in seawater [Lambert and Schmidt,1993], methane fluxes from open oceans and coastal zones havebeen estimated as 3.6 and6.1 Tg(CH4) yr−1 respectively. In addition, about 5 Tg(CH4) yr−1 of methane is assumedto be emitted from seepages through the sediments of continental shelfs. Open ocean andcontinental shelf emissions are assumed to be uniformly distributed, where the continentalshelfs have been defined as the coastal zone with sea depths<200 m.

The global distribution of methane emissions from wild ruminants have been approx-imated using the method described byBouwman et al.[1997]. As proposed byWarneck[1988] 3–6% of the net primary produced (NPP) vegetation is consumed by wild animalsof which a constant fraction is assumed to be emitted as methane. We used an annual NPPdistribution as derived from the Integrated Model to Assessthe Greenhouse Effect (IMAGE)[Minnen et al., 1996;Kreileman, 1996] discarding cultivated regions on the basis of land usedatabase byMatthews[1983]. Similar toBouwman et al.[1997], we assume that in forestedecosystems only 20% of NPP consists of consumable grass or leaves. Vegetation types havebeen assigned on the basis of the landcover database byOlson et al.[1983].

Sulphur emissions are used as a proxy for methane emissions from volcanic degassing us-ing the time-averaged distribution of volcanic sulphur emissions from continuously emittingvolcanoes on the basis ofAndres and Kasgnoc[1998] scaled to 3.5 Tg(CH4) yr−1 [Lacroix,1993]. Biomass burning emission estimates byHao et al.[1991] andOlivier et al. [1996]do not account for the contribution by the Australian continent. In this study these emissionshave been derived from estimated forest and savanna burningover Australia (L. Bouwman,personal communication, 1998). We compute that the Australian biomass burning sourcecontributes 6% to the global total, assuming the same methane release per unit biomass for

3.5 A priori assumptions 57

Table 3.1.: A Priori CH4 Fluxes and Uncertainties

Process Flux Uncertaintya Seasonality Referenceb,c

Anthropogenic Surface Fluxes

Oil/gas production 51 ± 30 . . . OL(1,3)Coal mining 38 ± 15 . . . OL(1,3)Landfills 48 ± 20 . . . OL(1,3)Domestic ruminants 93 ± 35 . . . OL(1,3)Biomass burning 40 ± 30 part of the dry LE(1), OL(3),

season H91(4)Rice agriculture 80 ± 50 dependent on the LE(1), M91(3,4)

growing seasonOther sourcesd 20 ± 16 . . . OL(1,3)

Natural Surface Fluxes

Tropical wetlands 91 ± 26 . . . LE(1), M87(3)Boreal/arctic wetlands 54 ± 15 temperature LE(1), M87(3),

dependent RH(4)Termites 20 ± 20 . . . LE(1), SA(3)Oceans 15 ± 10 . . . LE(1), PS(3)Volcanoes 3.5 ± 3 . . . LA(1), PS(3)Wild animals 5 ± 5 . . . LE(1), PS(3)Soil oxidation -30 ± 15 . . . LE(1), FU(3)

Total 528 ± 90

Fluxes and uncertainties in Tg(CH4) yr−1.aA 95% uncertainty interval (± 2σ) is used.bReferences are 1 annual and global flux total; 2 uncertainty in 1 (Lelieveld et al.[1998] if

not stated explicitly); 3 spatial distribution; and 4 seasonality.cReferences are OL,Olivier et al. [1996]; CR,Crutzen[1995]; LE,Lelieveld et al.[1998];

H91,Hao et al.[1991]; M91,Matthews et al.[1991]; M87,Matthews and Fung[1987]; RH,Hein et al.[1997]; SA,Sanderson[1996]; LA, Lacroix [1993]; FU,Fung et al.[1991]; andPS, present study (see text).

dOther sources are the sum of fossil fuel and domestic biofuelcombustion, and industrialproduction of iron, steel, and chemicals.


each continent. The seasonal cycles of Australian biomass burning emissions have beenestimated using monthly averaged precipitation based on 5 years of ECMWF output (1992–1996). The length and the distribution of the emissions overthe burning season are taken fromHao et al.[1991], with the start of burning season in the second month of the dry period.

Uncertainty estimates of methane sources and sinks are mainly available for the globallyand annually integrated fluxes. Since the a priori uncertainties per grid cell are needed wescale these integrated values down to “local” uncertaintiesσi, j for the flux by processi in eachsurface grid cell and month (indexj for both) in which the process is active. Subsequently,for each surface grid cell we compute the combined uncertainty of the flux due to all activeprocesses. It has been assumed for each processi that (1) the relative uncertaintieski ofemissionsfi, j from each active grid cell in each month are constant (σi, j = ki fi, j ) and that (2)the local uncertaintiesσi, j are uncorrelated (σ2

i = ∑ j σ2i, j ). Assuming that in a particular grid

cell and month the local uncertainty of the fluxes due to all active processes are independent,the local uncertainties for the sum of these fluxesσ j are given by:

σ j =√

∑i

(ki fi, j )2. (3.9)

As a consequence, the local relative uncertainties are muchlarger than uncertainties in theglobally integrated emissions (see also section 3.7). Thisprocedure for our standard inversionis modified in testing the effect of spatial and temporal correlations in section 3.8.

3.6 Chemical methane loss

About 90% of the methane removal from the atmosphere is due toreaction with the hydroxylradical in the troposphere. This means that to simulate the methane cycle accurately, a realis-tic representation of OH is of critical importance. Methyl Chloroform (1,1,1 trichloroethane,called CH3CCl3 hereafter) can be used to constrain OH, because its sources are relatively ac-curately known, and the hydroxyl radical reaction constitutes the most important sink. To testand optimize the model-simulated hydroxyl radical fields, asimulation of CH3CCl3 has beencarried out. Simulated CH3CCl3 concentrations have been compared to measurements fromfive stations of the Atmospheric Lifetime Experiment (ALE)/Global Atmospheric Gases Ex-periment (GAGE) network [Prinn et al., 1992, 1994] from 1978 to 1994.

To simulate CH3CCl3, surface emissions are applied on the basis ofMidgley [1989] andMidgley and McCulloch[1995]. Ocean uptake and stratospheric and tropospheric loss havebeen represented according toKanakidou et al.[1995]. The calculated turnover time ofCH3CCl3 due to ocean uptake amounts to 92 years, well within the rangeof 59–128 years es-timated byButler et al.[1991]. The stratospheric turnover time has been scaled to∼50 years,in agreement withKanakidou et al.[1995]. A scaling factorαOH for the global OH field hasbeen computed minimizing the r m s differences between measured and simulated CH3CCl3,weighted by the reciprocal standard deviations of the monthly mean measurements, similar toHein et al.[1997]. These weighting factors also take into account a 5% systematic uncertaintyin the absolute calibration of the CH3CCl3 measurements [Prinn et al., 1995]. The minimumis evaluated using the method ofBrent [1973], applying inverse parabolic interpolation. Thisprocedure yields an optimized scaling factorαOH = 0.95, which has subsequently been usedto scale the OH fields. The corresponding atmospheric lifetime of CH3CCl3 amounts to 5.0years, which is on the high end of the range of 4.8±0.3 years calculated byPrinn et al.[1995]. Krol et al. [1998] derived an even lower lifetime of 4.5–4.7 years over the period

3.6 Chemical methane loss 59

Mace Head(53N,10W)

1977.0 1978.6 1980.3 1981.9 1983.5 1985.2 1986.8 1988.5 1990.1 1991.7 1993.4 1995.0year

60

80

100

120

140

160

CH

3CC

l3 (

pptv

)

Oregon(45N,124W)

1977.0 1978.6 1980.3 1981.9 1983.5 1985.2 1986.8 1988.5 1990.1 1991.7 1993.4 1995.0year

60

80

100

120

140

160

CH

3CC

l3 (

pptv

)

Barbados(13N,59W)

1977.0 1978.6 1980.3 1981.9 1983.5 1985.2 1986.8 1988.5 1990.1 1991.7 1993.4 1995.0year

60

80

100

120

140

CH

3CC

l3 (

pptv

)

Samoa(14S,171W)

1977.0 1978.6 1980.3 1981.9 1983.5 1985.2 1986.8 1988.5 1990.1 1991.7 1993.4 1995.0year

40

60

80

100

120

CH

3CC

l3 (

pptv

)

Tasmania(41S,145E)

1977.0 1978.6 1980.3 1981.9 1983.5 1985.2 1986.8 1988.5 1990.1 1991.7 1993.4 1995.0year

40

60

80

100

120

CH

3CC

l3 (

pptv

)

Figure 3.1: Comparison of modeled 1,1,1-trichloroethane (CH3CCl3) concentrations to mea-surements at the Atmospheric Lifetime Experiment (ALE)/ Global Atmospheric Gases Ex-periment (GAGE) stations. Solid lines are optimized OH fields; dashed lines are original OHfields; and squares are monthly mean measurements. Error bars represent the 1σ interval ofthe measurements.


1978–1993, taking into account a possible trend in the hydroxyl radical concentration. Theoptimized hydroxyl radical level leads to a troposphere-integrated methane turnover time of9.0 years. This corresponds to the oxidation of 485 Tg of methane per year for a concentra-tion level representative of 1994. The uncertainty in the tropospheric OH content is estimatedat∼10% byKrol et al. [1998]. We apply a 2σ uncertainty, which is a factor of 2 smaller thanthis estimate (25 Tg(CH4); ∼5%), to keep the a posteriori derived sink within a reasonablerange (see section 3.10).

Figure 3.1 shows a comparison between simulated and measured CH3CCl3 concentra-tions at the 5 sites of the ALE/GAGE network. From Figure 3.1 it appears that the latitudinalgradient of CH3CCl3 is overestimated by the model. This indicates that tracer transport inthe model is too slow and/or that the simulated ratio betweenthe Northern and SouthernHemispheric (NH/SH) OH content is too low. Indeed, on the basis of 85Kr, SF6, and CFC-11 simulations we conclude that the interhemispheric exchange time is underestimated by∼20%, in line with earlier findings [Hein, 1994;Heimann and Keeling, 1989]. From simu-lations carried out with the horizontal and vertical diffusion coefficients tuned to reproducethese tracers optimally, we conclude that transport explains about half of the underestimationof the CH3CCl3 gradient. We aim to improve this as part of future model developments.

3.7 Methane inversion results

Figures 3.2–3.4 show comparisons of first-guess and optimized surface fluxes of methane.The per grid differences appear to be significant, with values of the same order as the assumedfirst-guess fluxes. As a result, flux parameters that are a priori assigned to be net sources ofmethane may represent a net sink a posteriori and vice versa.The changes are mainly limitedto the continents, which is explained by the relatively small a priori uncertainty in the oceanicflux. Over the continents, regions with relatively strong emission increases are typicallyadjacent to regions with strong decreases, for example, over southeast Asia during summer.Clearly, the simulated concentrations are most sensitive to flux changes in those grid boxes, inwhich measurement sites are located. Consequently, in these grid boxes relatively small fluxadjustments can compensate for misfits between simulated and modeled concentrations. Thisis also illustrated by maps of this sensitivity for CO2 as given byKaminski et al.[1999b]. Tosatisfy large-scale budget constraints, these corrections are then compensated for elsewhere.

Concentrations as computed using a priori and a posteriori flux fields have been comparedto observations (Figure 3.5). As expected, the optimized flux field improves the agreementbetween modeled and measured concentrations. In general, the a priori flux field appears tooverestimate the global north–south methane concentration gradient significantly. Figure 3.6shows that this feature, particularly, has been corrected in the optimization. A decrease of thelatitudinal concentration gradient can be achieved in two different ways: first, by a decreasein the ratio between the emissions in the Northern and Southern Hemispheres and, second, bya decrease of the globally integrated sources and sinks. Thelatter will not affect the globallyintegrated methane burden provided that the difference between the integrated sources andsinks matches the trend in methane. In fact, many combinations of sources and sinks satisfythis global budget constraint. It can be verified, however, that the latitudinal concentrationgradient is related to this combination such that larger sources and sinks lead to an increasedgradient. From the differences in the a priori and a posteriori integrated flux totals (Table 3.2)it appears that both mechanisms contribute to the decreaseda posteriori latitudinal gradient.

3.7 Methane inversion results 61

a

b

c

Figure 3.2: Estimated surface fluxes of methane for January: (a) a priori, (b) a posteriori, and(c) difference a posteriori minus a priori.


a

b

c

Figure 3.3: As Figure 3.2 for July.


a

b

c

Figure 3.4: As Figure 3.2 for annual means.


ALT

J F M A M J J A S O N D

1.75

1.80

1.85

CH

4(pp

mv)

CMO


1.72

1.74

1.76

1.78

1.80

1.82

1.84

CH

4(pp

mv)

KUM


1.68

1.70

1.72

1.74

1.76

1.78

1.80

1.82

CH

4(pp

mv)

MID

J F M A M J J A S O N D 1.68

1.70

1.72

1.74

1.76

1.78

1.80

1.82

CH

4(pp

mv)

P24


1.62

1.64

1.66

1.68

1.70

1.72

1.74

CH

4(pp

mv)

SPO


1.62

1.64

1.66

1.68

1.70

1.72

CH

4(pp

mv)

Figure 3.5: Comparison of simulated concentrations to measurements.Dashed lines repre-sent simulations based on a priori fluxes; solid lines represent simulations based on a posteri-ori fluxes; and squares represent monthly means of observations. Error bars represent the 1σinterval of the measurements. The offset of the a priori datahas been adjusted according tothe offset at the South Pole.


-90 -45 0 45 90Latitude (degrees)

1.65

1.70

1.75

1.80

1.85

1.90

CH

4(pp

mv)

Figure 3.6: Observed and simulated latitudinal methane concentration gradients (represent-ing annual means). Squares represent simulations based a priori fluxes; triangles representsimulations based on a posteriori fluxes; and solid circles represent measurements. The offsetin the a priori data has been adjusted according to the offsetat the South Pole.

Regional-scale flux changes can often be explained by examining the fits at the neareststations. For example, the decreased emissions over central China in Figure 3.4 are largelyimposed by the South China Sea cruise measurements. In autumn, when predominantlycontinental air is sampled at these sites, measured concentrations are significantly lower thanpredicted by a model simulation using a priori flux fields. Therefore a better fit at thesesites is obtained when fluxes over central China are reduced.Indeed, excluding the SouthChina Sea cruises from the inverse optimization almost completely eliminates the region ofdecreased emissions over China in Figure 3.4. Also, measurements at the Mongolian stationUlaan Uul and Qinghai Province in central China support a decrease in methane emissionsover central China relative to the a priori assumptions. Thesimulated concentrations at thesehigher-altitude stations are, however, less sensitive to emission adjustments in this region andtherefore impose weaker constraints.


Table 3.2.: Standard Scenario Integrated CH4 Fluxes and Uncertainties

A priori A posteriori

Mean Uncertaintya Mean Uncertaintya

Surface fluxGlobe 528 ± 90 505 ± 24Northern Hemisphere (NH) 405 ± 81 340 ± 19Southern Hemisphere (SH) 123 ± 40 165 ± 18

Tropospheric OH 485 ± 25 451 ± 22Stratospheric loss 40 ± 10 37 ± 9.7

Trend 3 ± 94 18 ± 7c0 1.696 . . . b 1.694 ± 0.009

Fluxes and uncertainties are in Tg(CH4) yr−1.aA 95% uncertainty interval (± 2σ) is used.bInitial methane concentration is in ppmv (see equation (3.3)). Uncertainty inc0 is set to a

high value.

A decrease in the a posteriori estimated emissions over southeast Australia is a persistentfeature over all seasons. This is related to the measurements at Cape Grim and is mostprobably caused by the fact that these measurements and the model simulated concentrationsat this station represent something different. At Cape Grimair is sampled only if the winddirection is between west and southwest to avoid contamination by local sources. As a resultof this baseline selection, monthly mean measurements are expected to be lower than thosecalculated by the model since in the model such a sampling protocol is not used. As shownby Ramonet and Monfray[1996] for CO2, the seasonal cycle at this station is also influencedby the sampling procedure.

The emission decrease over Europe, extending over northwest Russia, appears not tobe imposed by the measurements of the European stations at the Baltic Sea, Mace Head,Gozo, and Ocean Station M. Removing these stations from the inverse optimization onlyslightly influences the a posteriori emission patterns overthese regions. Since the a posterioritropospheric OH sink has decreased as compared with the firstguess, decreased fluxes helpbalance the concentration level. In other words, the a posteriori flux reductions in theseregions are more strongly imposed by the global budget than by the regional observations.

The South American continent is poorly sampled by the observational network. Ascen-sion Island, which is at a relatively small distance from South America, predominantly sam-ples air transported from the African continent [Kaminski et al., 1999b]. Other stations areat large distances where plumes of continental air have largely been dispersed. As a result,the flux changes relative to the a priori assumptions over South America are not caused byparticular misfits at a few stations but rather contribute tothe average level and seasonal cy-


cles at a large number of stations. Since most of these stations are surrounded by oceanicgrid boxes, which are all part of a larger flux region, local flux adjustments are suppressed.Flux adjustments at the continents need to be larger to compensate for the lower sensitivi-ties at the oceanic stations. The standard deviations of themeasurements at remote SouthernHemispheric stations like the South Pole, Palmer Station, and Syowa are, however, relativelysmall, and therefore these measurements receive a relatively large weight in the inversion.Therefore, despite their relatively low sensitivity thesemeasurements can still constrain theestimated continental source strengths.

Figure 3.7 shows the reduction in the surface flux uncertainty as a result of flux constraintsimposed by observational data in the inversion. Evidently,the uncertainty reduction is mostpronounced in the direct vicinity of continental measurement stations. From (3.7) it followsthat the uncertainty reduction is determined by two factors: (1) the a priori uncertainty rel-ative to the measurement uncertainty and (2) the sensitivity of the simulated concentrationat a station toward the fluxes. If both factors are relativelylarge, as is the case for thesecontinental fluxes, the uncertainty reduction is relatively strong as well. At some locations asignificant uncertainty reduction is computed at larger distance from the measurement sites.For example, the spot with relatively strong uncertainty reduction in Angola (Figure 3.7) isrelated to constraints from the measurements at Ascension Island. At oceanic stations, nosharp sensitivity maxima are found because of to the aggregation of oceanic fluxes. Since,in addition, the a priori uncertainties assigned to the oceanic emissions are relatively low, thecomputed uncertainty reduction over the ocean is low.

The difference in the integrated uncertainties in Table 3.2and the uncertainty reductionper grid in Figure 3.7 indicates that the relative uncertainty reduction increases toward largerscales. Such strong uncertainty reductions can only be explained if the a posteriori uncer-tainties in the contributing grids are predominantly anticorrelated. Figure 3.8 shows the co-variances of the a posteriori uncertainties of all surface fluxes for July and two flux elementsin central China and in Brazil. Indeed, the uncertainties inthe a posteriori fluxes are pre-dominantly anticorrelated with respect to the single elements. Such anticorrelations indicatethat the grid box fluxes are not resolved by the measurements.In other words, the sum of acluster of parameters is constrained by the measurements rather than by the contributions ofindividual elements. The covariance plot for the Chinese flux element shows a dipole struc-ture similar to the flux differences in Figures 3.2–3.4. Thisconfirms that such structures areimposed by regional budget constraints.

The validation of the a posteriori flux fields is difficult since it would require a source ofinformation independent of those already used in the inversion. Alternatively, we can per-form a series of inversions in each of which we omit the information of one of the stations.Such a test shows to what extent the inversion-derived emission fields improve the simulatedmethane concentrations. Figure 3.9 shows results of these tests at different sites. In particular,at the remote stations over the oceans (GMI and ASC) the agreement between simulated andmeasured concentrations has improved. At continental sites, such as Grifton, North Carolina(ITN), concentrations are determined by the source and sinkcomposition in the direct sur-roundings of a station. Therefore the agreement is expectedto be worse there. As illustratedin previous studies [Kaminski et al., 1999b;Plumb and Zheng, 1996], the singular vectorscan be used to determine in which direction in the flux space the measurement informationis most efficiently mapped. Although these studies focus on different trace gases (CO2 andCFCs, respectively), similar features are found for methane. Such an analysis indicates thata few directions are relatively well resolved, i.e., the global mean mixing ratio, the latitudi-nal gradient, and the seasonal cycle. From Figure 3.9 it can be seen that indeed, the main


a

b

c

Figure 3.7: Reduction in surface flux uncertainty gained by the inversion (1−

σaposteriori/σapriori): (a) January, (b) July, (c) annual mean.


a

b

Figure 3.8: (a) A posteriori uncertainty covariances between the Julyflux for a grid in Braziland all remaining fluxes for July and (b) the same as Figure 7a but for a grid in China.Covariances are expressed in units of the corresponding (Brazilian or Chinese) variances.(Consequently, these Chinese and Brazilian grids both havea value of 1.)

improvements in the a posteriori flux derived concentrations are related to these features.To quantify the amount of information in the observations ofparticular stations, inver-

sions have been carried out in which only a single station is used. Table 3.3 shows a rankingof stations according to their globally and locally integrated uncertainty reduction. Here localhas been defined as the grid box in which a station is located plus all grid boxes adjacent tothis grid box (nine in total), which means all emissions within a range of∼1500 km. Theuncertainty reduction has been divided by the uncertainty reduction in a full (45 station) in-version. In other words, this uncertainty reduction represents the percentage of uncertaintyreduction in a full inversion, which can already be achievedby introducing only a singlestation.

From the size of the globally integrated uncertainty reductions in Table 3.3 it is clearthat in a multistation inversion the uncertainty reductionper station is less than in the singlestation inversion. The reason for this follows directly from (3.7). On the global scale, remote


CBA


1.72

1.74

1.76

1.78

1.80

1.82

1.84

1.86

CH

4(pp

mv)

MHT


1.74

1.76

1.78

1.80

1.82

1.84

CH

4(pp

mv)

ITN


1.75

1.80

1.85

1.90

CH

4(pp

mv)

MLO


1.68

1.70

1.72

1.74

1.76

1.78

1.80

CH

4(pp

mv)

GMI


1.66

1.68

1.70

1.72

1.74

1.76

1.78

CH

4(pp

mv)

SC5


1.70

1.75

1.80

CH

4(pp

mv)

ASC


1.64

1.66

1.68

1.70

1.72

1.74

CH

4(pp

mv)

CGO


1.62

1.64

1.66

1.68

1.70

1.72

CH

4(pp

mv)

SPO


1.62

1.64

1.66

1.68

1.70

1.72

CH

4(pp

mv)

Figure 3.9: Comparison between simulated and observed concentrations. Measurements atthese locations have been excluded from the inversion. Dashed lines represent a priori; solidlines represent a posteriori; and squares represent measurements. Error bars represent 1σinterval of the measurements.

stations as, such as SPO and SYO, have the largest impact on the uncertainties. This can partlybe explained by the relatively small uncertainties of thesemeasurement data owing to therelatively small variability at these stations. Furthermore, the uncertainty in the local fluxesis relatively small so that fluxes at a greater distance receive relatively more weight. Lowestin rank on the global scale are stations for which the opposite argument holds. Generally,stations with a high score in reducing global-scale uncertainties have a low score on the localscale and vice versa. This analysis is dependent on the a priori assigned uncertainties and theJacobian matrixA. Therefore the exact order may be quite different when usingother a prioriuncertainties or when studying a different trace gas.

3.8 Sensitivity tests 71

Table 3.3.: Uncertainty Reduction per Station

Global Local

Station Uncertainty Reduction,a% Station Uncertainty Reduction, %

1. SPO 83 QPC 452. SYO 81 TAP 393. P24 75 GMI 394. PSA 74 SC7 325. P25 71 SC5 206. P22 63 SC6 187. CGO 58 SEY 168. ASC 48 UUM 169. SMO 46 SC4 1510. MLO 42 SC3 1411. P20 37 GOZ 1212. GMI 37 MHT 1213. QPC 34 IZO 1114. MID 34 STM 1015. SHM 34 UTA 9

......

41. TAP 13 CMO 242. UTA 13 P25 143. GOZ 13 P24 144. ITN 5 ALT <0.545. BAL 4 SPO <0.5

aUncertainty reductions are relative to the uncertainty reductions in a standard inversion(see text).

3.8 Sensitivity tests

In the previous section it was assumed that all errors in the measurements and fluxes areuncorrelated. One can think, however, of realistic conditions that violate this assumption.In fact, correlations may even serve as additional constraints on the fluxes. Consider, forexample, the European continent, with the most prominent methane sources being intensivefarming and industrial processes. None of these sources areexpected to change substantiallyover the seasons. Hence it is likely that these sources and related errors are positively corre-lated in time. Another example is the use of emission factorsthat are, for instance, derivedfrom statistics of fossil fuel use or food production. A biasin such emission factors may leadto correlated errors both in space and time. Emission factors are often relatively uncertainsince they are generally based on only a few case studies, which have been extrapolated tolarge regions. Provided that it is reasonable to assume thatemission factors are constant, suchbiases may indeed occur. It can be argued, however, that the largest uncertainty introduced


by using emission factors is in the assumption that they are constant. At smaller distancesbetween sources, however, the assumption of constant emission factors is probably less vio-lated since source processes are expected to be more similar. In this case the correlation isexpected to increase with decreasing distance.

Regarding the measurements, for instance, systematic errors in the sampling analysis leadto positively correlated errors. For the NOAA glass flask analyses, however, such errors areexpected to be small since the same calibration standard is used for all stations. Probably, therepresentativity of point samples for the time- and spaceaveraged concentrations computedby the model is a much larger source of errors. Although the measurements are screened for“pollution events,” the remaining samples may still be influenced by conditions that the modelis not able to reproduce. Only in some special cases may such representation errors may alsobe correlated. For example, systematic differences in the origin of air masses sampled inthe model and measurements, as is the case, for example, at Cape Grim, introduce positivelycorrelated errors. In general, however, representation errors are expected to be dominatedby random influences of “coincidental” events. Therefore itseems reasonable to assume thatmeasurement errors are uncorrelated.

The information needed to quantify spatial and temporal correlations in the fluxes is prac-tically absent. In general, uncertainty analyses of emission inventories are scarce, and ifpresent, they come in a wide range of different forms that cannot readily be combined forpurposes such as the present study. To investigate the importance of a priori assumed cor-relations, two sensitivity tests have been carried out, onefor temporal and one for spatialcorrelations.

Figure 3.10 shows estimated seasonal cycles of the fluxes in some selected grid boxes.In the time-correlated inversions a correlation coefficient of 0.9 has been applied to monthlyaveraged surface fluxes at the same location representing different months. In addition, forsources that are known to be strongly dependent on the season, for example, biomass burning,rice cultivation, and natural wetlands, a temporal correlation length of 2 months has beentaken into account. In these cases the correlation is assumed to decrease exponentially intime. The correlation length is defined as the time in which the correlation decreases by afactor 1/e. Such time dependent correlations lead to a smoothing of the estimated seasonalcycles. The timing and length of the season, which are generally quite uncertain for thesesources, can still be adjusted by the measurements.

Both for the Pacific Ocean and central Europe (see Figure 3.10) the a posteriori seasonalcycles are influenced substantially by the assumed time correlation. Assuming uncorrelatedsources, the computed seasonal variations cannot be explained by the most probable pro-cesses involved. The time-correlated seasonal cycles are considerably closer to their firstguesses. Also, at the Chinese and African sites the time-correlated cycles are substantiallycloser to the first guess, despite the larger freedom for flux adjustments, because of to the 2month correlation length assigned to biomass burning and rice paddy emissions. In Zambiathe biomass burning peak during the dry season is suppressedin the time-correlated inversion.Since biomass burning emissions may well peak in one particular month, the smoothing ofthe emissions in the time correlated inversion probably does not improve the results for thislocation.

In Figure 3.11, differences between first-guess and optimized fluxes are presented usingspatial correlations. A spatial correlation coefficient of1.0 with a correlation length of 2000km (∼3 grid cells) has been applied to all sources. Probably, for some sources the spatialdistribution is better known than for others, which is, for example, the case for ruminantscompared with biomass burning. This would justify the use ofsource-specific correlations.

3.8 Sensitivity tests 73

Figure 3.10: Seasonal cycles in the estimated fluxes for four regions. Solid lines are uncor-related fluxes; dashed lines are time-orrelated fluxes (see text); and dotted lines are a priorifluxes.

Since we lack the information to quantify these differences, all processes are treated the same.Note that inversions that differ in the assumed correlations also have different local a prioriflux uncertainties since additional correlation terms appear in equation (3.9). Since no anti-correlations are used and the same global uncertainties areassumed, the local uncertaintieswill, as a result, become smaller. As a consequence, the a priori information on the fluxesreceive more weight. A comparison of Figures 3.11 and 3.2–3.4 indicates that spatial cor-relations significantly change the estimated fluxes. As expected, emission changes extendover larger regions. Compared with the time-correlated inversion, it is difficult to determinewhether or not the use of space correlations improves the results. Little is learned from a com-parison as presented in Figure 3.9 since the differences between the concentrations derivedfrom the correlated and uncorrelated inversions appear to be small.

Two additional sensitivity tests have been carried out to test the influence of the assigneda priori uncertainties (also listed in Table 3.4) that determine the weights of misfits in thecost function. One may question how fair this weighting is since these uncertainties are, ingeneral, poorly quantified. The method to derive grid-scaleuncertainties from global-scaleestimates has been selected for its mathematical consistency but, in fact, does not give veryrealistic values. The assumed global uncertainty estimates, as listed in Table 3.1, lead to


a

b

c

Figure 3.11: Spatial correlation: difference between a posteriori anda priori (a po sterioriminus a priori) surface flux estimates for (a) January , (b) July, and (c) annual mean.

3.9 Case study: Emissions from southeast Asia 75

relative uncertainties per source per grid per month of∼600%. Since Gaussian probabilitydistributions are assumed that do not conserve sign, grids apriori dominated by sourcescould a posteriori easily be changed into sinks. In regions where a change in sign of the apriori assumed flux is unrealistic it is not appropriate to assume a Gaussian distribution. Toinvestigate the sensitivity of estimated emissions to a priori prescribed flux uncertainties, twosensitivity simulations have been carried out. In the first simulation it is assumed that we donot know whether one source is more uncertain than another. Therefore each surface flux isassigned the same relative uncertainty, taken to be 600%, except for the oceans. In the secondinversion, similar to the first, every surface flux is considered to be equally well (or poorly)known; an uncertainty level of 100% is chosen so that changesin signs are always outsidethe 1σ interval.

Finally, an inversion is performed in which the ratio in the OH radical abundance of theNorthern and Southern Hemispheres has been changed. A ratioof 2 (NH/SH) has been cho-sen, which, using a priori fluxes only, leads to a simulated latitudinal methane concentrationthat is almost correct. The OH field, as derived from the tropospheric chemistry version ofour model, has a NH/SH ratio of 1.06. Thus, in this sensitivity test the overestimated methanegradient is compensated by adjusting the OH distribution.

To quantify and illustrate the results of sensitivity tests, fluxes have been integrated overhemispheres and regions. Obviously, the sensitivity of theoptimized fluxes to a certain as-sumption is dependent on the weight the tested assumptions receive in the inversion. Sinceonly a priori assumptions are tested, these weights are approximately determined by the rel-ative weights of a priori and measurement constraints. Therefore two regions have been de-fined: region 1 over southeast Asia (75◦–135◦E, 10◦–40◦N), which is relatively well resolvedby the measurements and Region 2 over central Africa (5◦W–35◦E, 15◦S–25◦N), which isrelatively poorly resolved by the measurements. Table 3.4 summarizes these integrated emis-sions and the associated uncertainties for all sensitivitytests performed.

Overall, the a posteriori estimates are rather similar compared with the differences be-tween a priori and a posteriori estimates. Even for the African region, which is relativelypoorly sampled, the estimated totals are quite robust. The apriori and a posteriori uncertain-ties are so high, however, that none of the differences in Table 3.2 are significant. Lookingat the global totals, surprisingly, the spatially correlated and 100% relative uncertainty inver-sions, which both have relatively low a priori uncertainties compared to the standard simu-lation, have the highest adjustments in the globally integrated fluxes. The Northern versusSouthern hemispheric emission ratio, however, is closer tothe a priori value. A different com-promise between globally integrated flux adjustment and north–south emission adjustment tocorrect for the simulated latitudinal concentration gradient is favored in these inversions. Theinversion in which the OH ratio has been adjusted shows that both the north–south emissionratio and the globally integrated source and sink are close to their a priori values, which isagain explained by the corresponding effects on the latitudinal gradient in methane.

3.9 Case study: Emissions from southeast Asia

Rice agriculture constitutes an important methane source in southeast Asia. Its relatively highuncertainty results from the high temporal and spatial variability in the source strength of ricepaddies and the relatively poor statistics on rice management in developing countries. Recentstudies [Huang et al., 1997;Denier van der Gon, 2000a] point to a global source strengthin the lower part of the range of the IPCC estimate of 60±40 Tg(CH4) yr−1 IPCC [1994]


Table 3.4.: Sensitivity Tests, Integrated CH4 Fluxes, and Uncertainties

Globe NH SH Region 1a Region 2b

A Priori

All inversions 528 (±90)c 405 (±81) 123 (±40) 111 (±56) 51 (±28)

A Posteriori

Standard 505 (±24) 340 (±19) 165 (±18) 77 (±23) 65 (±20)Time correlation 500 (±23) 338 (±17) 162 (±16) 80 (±20) 65 (±18)Space correlation 481 (±22) 333 (±15) 148 (±13) 78 (±15) 63 (±14)Rel. uncertainty (600%) 511 (±25) 343 (±21) 169 (±20) 76 (±27) 68 (±25)Rel. uncertainty (100%) 479 (±15) 338 (±9) 142 (±8) 75 (±8) 54 (±6)OH ratio (NH/SH=2) 516 (±24) 395 (±20) 121 (±17) 103 (±23) 69 (±20)

Fluxes and uncertainties in Tg(CH4) yr−1.aRegion 1 is 75◦–135◦E, 10◦–40◦N (southeast Asia).bRegion 2 is 5◦W–35◦E, 15◦S–25◦N (central Africa).cA 95% uncertainty interval (± 2σ) is used. The a priori uncertainties correspond to the

standard scenario only.

and of the 80±50 Tg(CH4) yr−1 [Lelieveld et al., 1998] used in this study. On the basis ofthese recent studies and several estimates of the rice agriculture-related methane emissionsin China [Kern et al., 1995;Cao et al., 1995;Dong et al., 1996;Yao et al., 1996;Kern et al.,1997], H. A. C. Denier van der Gon (personal communication, 1998) derived a best guessestimate of 30±15 Tg(CH4) yr−1.

To test whether the inversion results are sensitive to the a priori emission scenario as-sumed, two inversions called ”standard” and ”low rice” havebeen compared in which eachestimate has been used as a priori input. The low rice scenario equals the standard scenarioexcept for a reduction in rice paddy emissions, for which tropical wetland emissions havebeen substituted. The global source estimates of these two processes are expected to beparticularly influenced by global budget information; thatis, neither are independent of themeasurements. It is questionable, however, how well the measurements constrain the rela-tive magnitudes of these sources. Table 3.5 lists the emissions derived from both inversionsintegrated over hemispheres, a zonal band (10◦–40◦N), and a region (75◦–135◦W part of thezone). TheMatthews et al.[1991] rice emission map indicates that∼80% of the rice paddiesare located in the selected zonal band, and∼75% are confined to the region.

The a posteriori integrated emissions appear to be quite insensitive to the a priori scenarioapplied. Globally, both scenarios show decreased a posteriori totals compared with the firstguesses. In the standard scenario the decrease over the region of intensive rice cultivation(31%) is relatively large compared with the Northern Hemispheric emission change (16%),which can be interpreted as a regional decrease superimposed on the global decrease. For thelow rice scenario the opposite is found, with a regional decrease (8%) slightly smaller than the

3.10 Discussion 77

Table 3.5.: Standard and Low Rice Scenarios, Integrated CH4 Fluxes, and Uncertainties

Scenario Globe NH SH Zonea Regionb

A Priori

Standard 528 (±90) 405 (±81) 123 (±40) 212 (±66) 111 (±56)Low rice 528 (±77) 384 (±66) 143 (±38) 185 (±47) 74 (±31)

A Posteriori

Standard 505 (±24) 340 (±19) 165 (±18) 169 (±25) 77 (±23)Low rice 508 (±24) 342 (±18) 166 (±17) 164 (±23) 68 (±18)

Fluxes are in Tg(CH4) yr−1.aThe zonal band is from 10◦–40◦N.bThe region is 75◦–135◦W, 10◦–40◦N.cA 95% uncertainty interval (± 2σ) is used.

Northern Hemispheric emission decrease (11%). In summary,on the basis of the standard andlow rice simulations, relatively low emissions over southeast Asia are defensible, however,not as low as those indicated above. On the basis of these results neither scenario can beruled out. It should be emphasized that the methane optimization is dominated by global-scale changes in the emissions, which largely obscure regional-scale changes.

3.10 Discussion

Inverse modeling is a useful tool to complement bottom up estimates of global methanesources and sinks. An important question is to what extent inverse modeling can contribute.Since many different procedures can be used to perform such an inversion, the answer maybe different depending on the technique, the model, and the assumptions used. Since themeasurement data set is limited, the use of a priori knowledge is necessary to constrain theinverse problem, i.e. to find the most likely emission scenario consistent with the measure-ments. Compared to previous methane studies [Hein et al., 1997;Brown, 1993] that forcomputational reasons have to prescribe a few fixed flux patterns and optimize the corre-sponding scaling coefficients, we greatly release this rigid a priori constraint. In our case thesystem has a much higher degree of freedom to adjust fluxes. Because a priori knowledge isused as input, our inversion does not provide a fully independent emission estimate. Instead,a different type of result is obtained addressing three important topics. First, this techniquehelps to establish the consistency between estimated emissions and measured concentrations.Second, the most likely emission changes can be derived to reduce inconsistencies. Third, bycombining constraints imposed by measurements and a prioriknowledge, uncertainties arereduced.

Uncertainty reductions appear to be a function of scale, with larger reductions going to


larger scales. This reflects a trade-off between resolutionand uncertainty reduction, which iscommon in inverse problems. This means that at small scales the individual sources and sinksare not resolved by the measurements, except those close to measurement stations. Closer tothe stations, however, the uncertainty reduction becomes increasingly affected by the resolu-tion of the model. Generally, in inverse modeling, uncertainties are a measure of the rangeover which parameters may change given the constraints. This range is dependent on thedimension of the flux space. The higher the number of possiblesolutions, the more optionsthere are to compensate for ”outlying” fluxes while still keeping the constraints satisfied.Therefore, in general, modeled a posteriori uncertaintiesare expected to be underestimatesof the real uncertainty since the real world is always more complex than the model. Indeed,compared to big region approaches, we find relatively high a posteriori uncertainties. Sincein our inversion the resolution is less limiting, our uncertainty reductions are expected to bemore realistic.

The case study presented in section 3.9 illustrates the consequences this may have. Theresults ofHein et al.[1997] suggest that inverse modeling puts a strong enough constrainton rice paddy emissions to rule out one of the scenarios sincethe standard scenario is on thehigh side and the low rice scenario is significantly lower than their inverse modeling-derived95% confidence interval for rice paddy emissions. In this study, however, both scenarios arestill within this uncertainty range.

In addition to uncertainty estimates, a posteriori derivedmethane concentration distribu-tions indicate how much we learn in the inversion. Such a comparison should be interpretedwith care, however, since it is an indirect test of fluxes. It is a direct test of the methane con-centration field, however, and as such, it leads to the conclusion that the a posteriori derivedmethane concentration distribution has improved. Obviously, this is useful for applicationswhere methane concentration fields are needed rather than flux fields. Although the concen-tration field has improved, the flux field may give improved results for the wrong reason. Asmentioned in section 3.7, this is the case when flux adjustments compensate for model errors.

In fact, all inconsistencies between measurements and modeled concentrations are at-tributed to the fluxes and the measurements since the model isassumed to be perfect. A use-ful diagnostic of model error is the interhemispheric exchange time. As discussed in section3.6, simulation of85Kr, SF6, and CFC-11 indicates that the exchange time is underestimatedby ∼20%. As a test, the exchange rate was increased by horizontaldiffusion [Prather et al.,1987;Keeling et al., 1989], which showed that transport explains at most half ofthe differ-ence between the measured and the simulated methane gradient if the a priori values for themethane fluxes are used. As a consequence, the adjustments inthe ratio between Southernand Northern Hemispheric emissions and globally integrated sources and sinks by the inver-sion, as shown in section 3.7, can partly be attributed to this model error. In addition, an errorin the relative OH abundances of both hemispheres may also contribute. Within the uncer-tainty ranges of the CH3CCl3 test the OH ratio between both hemispheres may vary betweenabout 0.5 and 2 [Spivakovski et al., 1999]. This indicates that the CH3CCl3 test is not ideallysuited for validating this ratio, at least for the period of CH3CCl3 measurements that we usedto test our OH fields. A north–south ratio of 2, in addition to acorrected interhemisphericexchange time in the model, can overcompensate the underestimated CH4 gradient. From anatmospheric chemistry point of view such a high interhemispheric OH ratio is highly unlikelyand cannot be explained by the models. Quantification of suchchemical constraints is, how-ever, very difficult. Our results indicate that to estimate sources and sinks of methane on aglobal scale accurately a more accurate validation tool forthe global OH radical distributionis needed.

3.11 Conclusions 79

The number of available measurements as compared with the number of unknowns con-tinues to be an important limitation on inversion studies. Unless considerable progress inmeasuring methane from satellites is made, this is not expected to change much in the nextdecade. Alternatively, measurements of isotope ratios, which provide process-specific infor-mation, may be used. Studies byBrown[1995] andHein et al.[1997], however, indicate thatthe isotopic data set available at present does not provide an important constraint. Here someprogress may be expected in the future since more data will become available.

Alternatively, the source information contained in the measurements may be used moreefficiently. For example, it is likely that by using multiyear averaged seasonal cycles, muchinformation is lost. When, instead of using averaged data, the actual measured data are used,the methane inversion should be improved in a number of ways.Meteorological input repre-sentative of each year of the target period should then be used, as opposed to repeatedly usingone year, as is done in this study. Furthermore, quantitative information about sources thathave a high interannual variability, such as dependencies on precipitation and temperature,should be provided. Representation errors between measured and modeled concentrationthen become increasingly critical since coincidental disturbances are averaged out to a lesserdegree. These topics will be the focus of future work.

3.11 Conclusions

We presented an inverse modeling method to study the global-scale sources and sinks ofmethane. The measurements have been treated in accordance with the quasi-stationary stateassumption, as byHein et al. [1997], assuming a constant trend and a multiyear averagedseasonal cycle. The inversion has been performed at a relatively high spatial and temporalresolution (per grid and per month) over the continents using an adjoint version of the globaltransport model byKaminski et al.[1999b]. Recent estimates of the sources and sinks ofmethane have been used as first-guess input, including minorsources such as oceans, volca-noes, termites, and wild animals.

The inversion-derived net global surface source of methaneamounts to 505 Tg(CH4)yr−1, using a first guess of 528 Tg(CH4) yr−1 for the target period 1993–1995. The relativecontribution of the Northern Hemispheric sources decreases from 77 to 67%. The chemicalsinks in the troposphere and stratosphere decrease from 485to 451 Tg(CH4) yr−1 and 40to 37 Tg(CH4) yr−1, respectively. The computed a posteriori methane trend of 18 Tg(CH4)yr−1 is well within the 17.1±3.5 Tg(CH4) yr−1 as derived from the measurements over thetarget period 1993–1995. Simulations of85Kr, CFC-11, and SF6 indicate that our modelunderestimates the interhemispheric exchange rate by∼20%, which explains up to 50% ofthe differences between the a priori and a posteriori emission estimates. In addition, errorsin the simulated north–south distribution of the hydroxyl radical may contribute to thesedifferences. Within the ranges of uncertainty the combinedeffect of model errors couldpotentially explain the differences in the observed and simulated methane gradient, whichunderscores the importance of reliable tools to validate the interhemispheric transport rateand the hydroxyl radical distribution.

In addition to providing new “measurement”-consistent emission estimates, inverse mod-eling helps to reduce the uncertainties. These uncertaintyreductions are strongly related toscale. Small (<1%) reductions are computed at the grid scale, with some exceptions close tomeasurement stations. At the scale of hemispheres, uncertainty reductions as large as 75%are obtained, for example, from±80 a priori to±20 Tg(CH4) yr−1 a posteriori for the North-


ern Hemisphere. The sharp uncertainty decrease with scale is reflected by the predominantlynegative correlation of the a posteriori flux uncertainties, indicating that at the grid scale,fluxes are not resolved by the measurements.

The sensitivity of the inversion-derived emission estimates to a number of a priori as-sumptions, such as the assumed a priori uncertainties, the OH radical distribution, and un-certainty correlations, has been tested. The conditions used in each sensitivity test are withinthe ranges of uncertainty. Estimated emissions integratedover larger regions appear to bequite robust to these assumptions, except for the large-scale OH radical distribution, whichcould compensate for the emission adjustments derived using the standard scenario. Definingmore accurately the a priori state in terms of spatial and temporal correlations among un-certainties helps to reduce the a posteriori uncertainties. Since the observational informationfor justifying the use of any strong positive or negative correlations in space is essentiallylacking, the improvement is limited in this work. It may, however, be a useful constraint ifmore information were available. The use of temporal correlations leads to more significantimprovements and helps, in particular, constrain seasonalcycles in the fluxes. We emphasizethat source quantification by inverse modeling would benefitfrom bottom up estimates ofsource correlations, which could, for example, be derived from source process modeling.

As a case study, we have focused on emissions from rice fields in southeast Asia. Thecontinental NOAA stations Qinghai Province, Tae-ahn Peninsula, and Ulaan Uul and theSouth China Sea cruises helped to constrain the emission estimates for this region. The aposteriori derived fluxes point to lower emissions as compared to the a priori assumptions. Acase study illustrates, however, that the uncertainty related to the a posteriori derived emis-sions is still too large to reduce the uncertainty of rice paddy emissions significantly. In fact,the present inversion method allows a larger range of fluxes from rice paddies than is reportedby Hein et al.[1997]. Since the use of a high-resolution inversion is expected to improve theuncertainty estimates, our larger uncertainty range is expected to be more realistic.

Acknowledgments

We thank E. J. Dlugokencky (NOAA) for providing all kinds of information about the NOAAsampling network and for the hospitality in letting us have aglimpse in the kitchen. Also, weare grateful to R. Hein (DLR) for a helpful introduction to inverse modeling and for kindlyproviding us data. Further, we would like to acknowledge a useful cooperation with H. A. C.Denier van der Gon (Department of Soil Science and Geology, Wageningen University). Thiswork has been supported by the Dutch Global Change program, NOP project 951202.

Chapter 4

The modeling of troposphericmethane; how well can pointmeasurements be reproduced by aglobal model?

Global model simulations of tropospheric methane are presented, using state of the art rep-resentations of its terrestrial sources. Parameters critical for its tropospheric sink and trans-port have been evaluated using CH3CCl3 and SF6. We assess how well available methanemeasurements can be reproduced by the model, and how model and measurements can mostefficiently be compared. Using ECMWF re-analyzed meteorological fields, direct compar-isons between model results and flask or in situ measurementsare presented, as opposed tocomparing multi-annual averaged seasonal cycles and trends as was done in previous stud-ies. When comparing monthly means derived from weekly flask sampling and the model, theagreement at stations as Bermuda East and Mace Head is improved if, instead of samplingthe model at each model time step, samples are taken at the same times as the measurementswere taken. A method is presented to estimate the potential influence of sub-grid variabilityusing a marked tracer that is emitted in the vicinity of observational stations only. Fromthe contribution of this tracer to the computed methane concentration at a particular sta-tion the potential contribution of sub-grid sources can be estimated.222Rn is used to selectbaseline conditions in the model to improve the comparability of model and measurementswhen a clean air sector is selected for sampling. Comparisons of model results and mea-surements, screened for local influences and artifacts of wind sector selection, indicate thatthe model has in particular difficulty reproducing seasonalcycles at higher latitude stationsof the Northern Hemisphere. Sensitivity simulations show that the simulated annual varia-tion at these stations is sensitive to the parameterizationof wetland emissions. Also at theSouth China Sea, model simulations point to errors in the representation of methane sources.Marked tracer simulations indicate that this is most likelyrelated to emissions from naturalwetlands and rice paddies, in line with recent inverse modeling and up-scaling estimates.

1Accepted for publication inJournal of Geophysical Research, with F. J. Dentener, J. Lelieveld, B. Walter, andE. J. Dlugokencky as co-authors.

81

82 Forward modeling of methane

4.1 Introduction

Methane is an important trace gas in the earth’s atmosphere.It is the second most importantincreasing greenhouse gas, with an estimated contributionto the present climate forcing of35% of that by increasing CO2 [Lelieveld et al., 1998]. This is associated with absorption oflong-wave radiation (mainly atλ=7.6µm), and indirect warming by tropospheric ozone andstratospheric water vapor, that are products of its oxidation. Further, methane influences thelifetimes of various chemically reactive trace gases owingto a photochemical feedback onthe hydroxyl radical [Crutzen and Zimmerman, 1991;Krol et al., 1998].

Tropospheric methane concentrations are monitored by several measurement networksstarting about two decades ago [Dlugokencky et al., 1994b;WMO, 1997;GLOBALVIEW-CH4, 1999]. Since methane is relatively evenly distributed over the globe, owing to a tro-pospheric residence time of about 8 years [Lelieveld et al., 1998], these measurements quitewell constrain its tropospheric abundance. On the other hand, sources and sinks of methaneare relatively weakly constrained. As a consequence, it is difficult to understand the timeevolution of methane [Dlugokencky et al., 1994a, 1996, 1998] and anticipate future changes.

To improve our understanding of the present methane budget,modelers have tried tointegrate all available information on sources, sinks and observed concentrations, using eitherforward [Fung et al., 1991;Lelieveld et al., 1998] or inverse techniques [Brown, 1993;Heinet al., 1997;Kandlikar, 1997;Saeki et al., 1998;Houweling et al., 1999a]. Inverse modelingis a useful tool to relate accurate methane concentration measurements to relatively uncertainsources and sinks. Previous inversions of methane and of other long lived trace gases suchas CO2 indicate, however, that a large number of measurements is required to substantiallyconstrain fluxes since their smooth concentration distributions renders the inverse problemill-conditioned. In addition, model errors may seriously affect emission estimates derived byinverse modeling [Houweling et al., 1999a]. Although, for example, the Bayesian approach toinverse modeling allows model errors to be taken into account, in practice the computationalcosts become very high.

In this study, forward modeling is used to re-assess the agreement between measurementsand model results given a state of the art description of sources and sinks, to identify the ma-jor shortcomings of our understanding of the methane cycle.Secondly, we investigate howsources and sinks of methane in the model can most efficientlybe constrained by measure-ments. In previous studies [Fung et al., 1991;Hein et al., 1997;Houweling et al., 1999a]multi-annual data records were used to derive averages or polynomial fits representative ofclimatological means, trends and seasonal cycles at observational stations. Preferably stationswere used at large enough distance from the sources, such that measurements represented thescales resolved by the models. Unfortunately, at such stations the fingerprint of sources andsinks has largely been dispersed by atmospheric mixing. In addition, measurements are oftensmoothed which further reduces the information content of the measurements on sources andsinks. Current global models become increasingly advancedin reproducing smaller tempo-ral and spatial scales. To optimally profit from this model development, more sophisticatedmethods of using measurements to test these models are needed. To investigate this, first westudy which scales can be reproduced by our model, using measurements without any pre-treatment by interpolation or smoothing techniques. Different methods of model sampling aretested at background oceanic stations as well as relativelypolluted coastal and inland stations.For all model simulations meteorological input data are used which are representative of thesimulated period, to reproduce as accurately as possible the atmospheric conditions at themoment of sampling. When testing the model using limited numbers of samples at relatively


polluted sites representation errors may potentially growto significant or even dominatingproportions. Methods are proposed to minimize these errorsor to quantify their potentialinfluence.

This paper is organized as follows: Section 4.2 describes our chemistry transport model,focusing on the representation of the methane cycle. Section 4.3 describes in situ and flaskmeasurements of methane, used to test the model. To validatelarge scale transport prop-erties of the model, simulated and measured latitudinal concentration gradients of SF6 arecompared (section 4.4). In section 4.5, the results of methane simulations are presented andmethods to reduce representation errors are demonstrated.Further, we focus on the moststriking discrepancies between model and measurements, using these methods. Finally, ourconclusions are presented in section 4.6.


4.2.1 Chemistry transport model

Model simulations presented in this study have been carriedout using the global three di-mensional (3-D) Tracer Model 3 (TM3) [Houweling et al., 1998;Dentener et al., 1999]. Thegeographical resolution applied is 5◦ in the longitudinal and 3.75◦ in the latitudinal directionwith 19 vertical levels (unless noted). The vertical levelshave been defined as terrain fol-lowing coordinates near the surface, pressure levels in thestratosphere, and a hybrid of thetwo in between. The horizontal and vertical transport of tracers is based on six hourly meanmeteorological fields, including wind, surface pressure, temperature, and humidity, derivedfrom European Centre for Medium-Range Weather Forecasts (ECMWF) re-analyses for theyears 1980–1993 and analysis for the years 1994–1996. Each year of model simulation thecorresponding meteorological input fields have been used, which applies to all model resultspresented in this work. The advective transport is calculated using the “slopes scheme” ofRussell and Lerner[1981]. The sub-grid scale convective airmass fluxes are evaluated us-ing the cloud scheme ofTiedke[1989], including entrainment and detrainment in updraftsand downdrafts. Turbulent vertical transport is based on stability dependent vertical diffu-sion [Louis, 1979]. As a test of boundary layer mixing and regional scaletransport,222Rnsimulations at various model resolutions have been compared with observations at continen-tal and remote locations [Dentener et al., 1999]. From this study it follows that measuredand simulated radon concentrations agree quite well; generally, deviations are<50%. In thepresent study, results of SF6 simulation are compared with measurements as a test of largescale transport (see section 4.4).

Tropospheric chemistry is represented using a modified version of the Carbon BondMechanism 4 [Houweling et al., 1998;Gery et al., 1989], accounting for CH4/CO and non-methane hydrocarbon (NMHC) chemistry, including isoprene. Chemical equations are in-tegrated using a Eulerian Backward Iterative (EBI) scheme,as formulated byHertel et al.[1993]. Emissions of photochemical tracers other than CH4 are based on the GEIA andEDGAR emission inventories [Olivier et al., 1996;Guenther et al., 1995;Yienger and Levy,1995;Benkovitz et al., 1996]. Wet and dry deposition of soluble and reactive tracers has beendescribed byHouweling et al.[1998] andGanzeveld et al.[1997]. Photolysis rates are basedon a highly efficient parameterization of the DISSORT radiative transfer code, and a parame-terization proposed for wave-length dependent cross sections and quantum yields accountingfor multiple scattering by clouds [Krol and Van Weele, 1997;Landgraf and Crutzen, 1998].


O3 transport from the stratosphere into the upper level of the TM3 model domain is accountedfor by constraining the ozone concentration in the upper three model layers, based on ozoneconcentrations from the climatology byFortuin and Kelder[1998]. Stratospheric HNO3 istreated similarly, based on UARS-derived O3/HNO3 ratios [Kumer et al., 1997;Bailey et al.,1997].

The CTM-derived hydroxyl radical fields have been validatedusing 1,1,1 trichloroethane(CH3CCl3). CH3CCl3 can be used to constrain OH, because its sources are relatively accu-rately known, and the hydroxyl radical reaction constitutes the most important sink. Surfaceemissions of CH3CCl3 are based onMidgley [1989] andMidgley and McCulloch[1995].Stratospheric and oceanic losses of CH3CCl3, both small compared with the hydroxyl rad-ical sink, are represented as described byHouweling et al.[1998] andKanakidou et al.[1995]. CH3CCl3 measurements of the Atmospheric Lifetime Experiment (ALE)/ GlobalAtmospheric Gases Experiment (GAGE) network [Prinn et al., 1992, 1994] for 1980–1992have been compared with model results. More recent data havebeen omitted since the sourcesof CH3CCl3 have become less certain after implementation of the Montreal Protocol on Sub-stances that Deplete the Ozone Layer. The 1980–1992 ECMWF re-analysis meteorologicaldata were used to simulate the transport of CH3CCl3. In addition, for the same period we tookvariations and trends of emissions and ozone columns into account to compute hydroxyl rad-ical distributions [Aardenne et al., 1999;Fortuin and Kelder, 1998;Lelieveld and Dentener,1999]. From the CH3CCl3 simulation a scaling factor is derived, optimizing the agreementbetween measured and model simulated CH3CCl3 as described byHouweling et al.[1999a].The computed atmospheric lifetime of CH3CCl3, after optimization, ranges from 5.0 yr in1980 to 4.7 yr in 1990, within the 4.8± 0.3 years calculated byPrinn et al.[1995] (unscaledhydroxyl radical fields yield 11% longer lifetimes).

4.2.2 Methane simulations

To simulate methane, four-year model simulations have beenperformed, of which methaneconcentrations computed for the final year are compared withmeasurements only. Accordingto Hein et al.[1997] andTans[1997], a spin-up time of three years is sufficiently long to avoida strong influence of the initial CH4 distribution on the results. To initialize methane, a uni-form mixing ratio is prescribed for the whole model domain, derived from results of inversemodeling [Houweling et al., 1999a]. When model results are compared with measurementsthe contribution of the initial methane field to the global mean concentration is eliminatedby adjusting the offset in the computed concentrations to background stations (details can befound in Figure captions). The CH3CCl3 optimized hydroxyl radical fields are prescribed asthree-dimensional monthly means. This method yields a tropospheric methane turnover timeof 7.8 yr, and a tropospheric removal of 509 Tg(CH4) for 1993.

The sources of methane (roughly representative of 1990) arerepeated each year, exceptfor natural wetlands, because dependencies on rainfall andtemperature are accounted for ac-cording toWalter[1998]. The interannual variability of biomass burning andrice paddy emis-sions are difficult to quantify, and have therefore been neglected. For the remaining sourcesthe interannual variability is expected to be small. The observed trend in the global meanmethane concentration decreased from 0.6% to 0.2% growth per year Dlugokencky et al.[1998] over the simulated period (1989–1996), suggesting amore or less stationary globalsource strength. In absence of indications that a specific source has significantly changedover this relatively short period, we assume that no trend inany of the sources is present. Thestratospheric destruction of methane by photolysis and reaction with OH, Cl, and O(1D) is


Table 4.1.: Prescribed CH4 Surface Fluxes

Process Annual Total Referencea,b

Anthropogenic Surface Fluxes

Oil/Gas production 51 OL(1,2)Coal mining 38 OL(1,2)Waste treatmentc 73 LE(1),OL(1,2)Domestic ruminantsd 93 OL(1,2)Biomass burning 40 LE(1), OL(2),

H91(3), SH(2,3)Rice agriculture 80 LE(1), M91(2,3)Other sourcese 20 OL(1,2)

Natural Surface Fluxes

Natural wetlands 145 LE(1), BW(2,3)Termites 20 LE(1), SA(2)Oceans 15 LE(1), SH(2)Volcanos 3.5 LA(1), SH(2)Wild animals 5 LE(1), SH(2)Soil oxidation -30 LE(1), AR(2,3)

Total 554

All fluxes in Tg(CH4) yr−1.aReferences for (1) annual and global total, (2) spatial distribution, (3) seasonality

(if not specified a constant source is assumed).bOL, Olivier et al.[1996]; LE,Lelieveld et al.[1998]; H91,Hao et al.[1991]; M91,

Matthews et al.[1991]; SA,Sanderson[1996]; LA, Lacroix [1993]; FU,Fung et al.[1991]; BW,Walter[1998]; AR,Ridgwell et al.[1999]; SH,Houweling et al.[1999a].

cIncludes landfills and waste water.dIncludes digestion (OL) and waste (LE).eSum of fossil fuel and domestic biofuel combustion, industrial production of iron,

steel, and chemicals.


derived from two-dimensional photochemical model calculations [Bruhl and Crutzen, 1993],scaled to a global loss of 40 Tg(CH4) yr−1 in agreement withCrutzen[1995].

Sources of methane are represented in the model as specified in Table 4.1. For details ofemission distributions and annual totals we refer to the references listed in this Table. Thisrepresentation of the sources largely corresponds to the a priori scenario used in the inversemodeling study byHouweling et al.[1999a], using an older version of the model (TM2)(referred to as SH99 hereafter). Note that the a-posterioriemission fields derived from theSH99 inversion have not been used here since these emission updates to some extent com-pensate for errors in the transport. For example, part of thea priori to a posteriori emissionchanges compensate for an overestimate by the model of the latitudinal methane gradient,which could partly be explained by an underestimate of the interhemispheric exchange ratein the model. The updated model, used in the present study (TM3), has an improved rep-resentation of interhemispheric transport (see section 4.4), although the simulated methanegradient remains too large. To further improve this, the geographical distribution of naturalwetland emissions has been updated based on process model computations byWalter[1998].This model parameterizes the microbial methane productionand consumption in natural wet-lands, and the transport through plants and soil as a function of soil and vegetation properties,temperature and hydrological conditions [Walter et al., 1996]. The model has been evaluatedand optimized using results of field experiments at a number wetland sites at higher latitudesand a site in the tropics. On the basis of this process model a global emission distributionis obtained with a Northern to Southern Hemisphere emissionratio of 1.38 compared with1.86 for the distribution applied in SH99 based onHein et al.[1997]. This is explained by arelatively large contribution of tropical wetlands in the estimates byWalter [1998].

To improve the representation of microbial methane consumption, sink strengths havebeen used based on a process model byRidgwell et al.[1999], which, unlike the estimates byFung et al.[1991] applied in SH99, also accounts for seasonal changes.In this model, soilmethane oxidation is computed as a function of soil diffusivity and microbial activity. At mid-and high latitudes, oxidation rates in summer are higher than in winter, which is explained bychanges in temperature and soil moisture content. In the tropics, a weak seasonality is foundowing to a much smaller seasonal variation in these climaticfactors. Finally, as proposedby Lelieveld et al.[1998] emissions from industrial waste water have been accounted for,including food and paper industries and oil refineries, estimated at 25±10 Tg(CH4) yr−1

[IPCC, 1992;Environmental Protection Agency (EPA), 1994].

4.3 Measurements

The majority of the measurements used in this study have beenperformed by the NationalOceanic and Atmospheric Administration (NOAA)/ Climate Monitoring and DiagnosticsLaboratory (CMDL) cooperative air sampling network [Dlugokencky et al., 1994b]. At eachstation duplicate flask samples are taken about once a week, and transported to NOAAfor analysis. In addition, at Barrow (Alaska, 71◦19’N, 156◦36’W, 11m above sea level(masl)) and Mauna Loa (Hawaii, 19◦32’N, 155◦35’W, 3397masl) in situ operating Gas Chro-matographs are used, analyzing at a frequency of about 60 samples d−1, which were avail-able to us as hourly averages. The precision of the gas chromatographic analysis of bothin situ and flask samples is estimated at 0.2% [Dlugokencky et al., 1994b, 1995]. Measure-ments “flagged” by NOAA, indicating that the sampled air may have been influenced by localsources, are used in this study. The potential effects of using such samples is discussed in

4.4 Interhemispheric exchange rate 87

section 4.5.In situ measurements at Fraserdale (Canada, 49◦53’N, 81◦34’W, 250masl), southeast of

the Hudson Bay lowlands, have been carried out by the Atmospheric Environment Service(AES) [Worthy et al., 1998]. Measurements were taken from a 40 m tower at a frequencyof 96 d−1. The difference of CH4 calibration scales used by AES and NOAA/CMDL hasbeen accounted for, on the basis of a recent intercalibration presented inGLOBALVIEW-CH4[1999]. Thus, measurements presented in this study are relative to the NOAA/CMDL scale.The precision of the AES gas-chromatographic analyses is also 0.2% [Worthy et al., 1998].

4.4 Interhemispheric exchange rate

As indicated in the previous section, interhemispheric gradients give important informationon the distribution and strength of CH4 emissions. Sulfur hexafluoride (SF6) has been usedto test the interhemispheric tracer exchange rate simulated by our CTM. SF6 is a purely an-thropogenic trace gas, emitted mainly from leakages in electronic insulations and switching.In the atmosphere, SF6 is a virtually inert tracer with an estimated atmospheric residencetime of 800 yr [Morris et al., 1995]. Annually emitted amounts are determined from theatmospheric budget using global transport models. In our model, the observed trend at Neu-mayer (8◦W,71◦S,42masl)) [Maiss and Levin, 1994;Maiss et al., 1996] is well reproducedusing reported emissions byLevin and Hesshaimer[1996], multiplied by 0.937 as proposedby Denning et al.[1999]. This correction factor is introduced to compensatefor differencesbetween the 3D models used inDenning et al.[1999] and the 2D model used byLevin andHesshaimer[1996]. The global emissions were distributed according toelectrical power us-age by country and population density [United Nations, 1994;Tobler, 1995;Denning et al.,1999].

SF6 has been simulated for the period 1989–1993, of which the final year is used toderive annual mean concentrations, again to avoid effects of errors in the assumed initialhemispheric distribution of SF6 on the 1993 fields. A comparison between the observedlatitudinal gradient, as compiled byDenning et al.[1999], and model results is given inFigure 4.1. These results indicate that the interhemispheric transport rate is quite accuratelyreproduced by our model. The computed interhemispheric exchange time, as defined byPrather et al.[1987], amounts to 0.90 yr for 1993. Note that the TM3 resultspresented byDenning et al.[1999] were obtained using ECHAM derived meteorological data, as opposedto ECMWF data that have been used here.

To test the sensitivity of the SF6 test to transport errors, the exchange time in the modelhas been decreased by 15% by: (1) introduction of horizontaldiffusion and (2) reducing thevertical mixing by turbulent diffusion and convection. As shown in Figure 4.1 this leads to adecrease and increase of the simulated SF6 gradient respectively, which approximately cor-responds to the upper and lower limits of simulated SF6 gradients that are still in agreementwith the measurements. A similar uncertainty is associatedwith the simulation of transportfor the long lived trace gases CH3CCl3 and CH4.

4.5 Results and discussion

This section focuses on comparisons between model simulations and measurements at singlelocations. First, however, we show the ability of the model to reproduce the observed global


-90 -60 -30 0 30 60 90Latitude (degree)

2.80

2.90

3.00

3.10

3.20

3.30

3.40

SF

6(pm

ol/m

ol)

Figure 4.1: Measured and simulated SF6 mixing ratios showing the annual mean interhemi-spheric gradients of SF6 for 1993. Circles are measurements; triangles are model simulations;the dashed line represents a 15% decreased exchange time by decreased vertical mixing; thedotted line represents a 15% decreased exchange time by increased horizontal diffusion. Theoffset in the model results has been adjusted to Atlantic Ocean measurements at 40◦S.

scale methane concentration distribution. This provides insight in the extent to which differ-ences between model results and measurements at individualstations are explained by largescale phenomena. In this section, model results at sites arederived by linear interpolation ofthe concentrations simulated for the surrounding model grid boxes.

Measured and simulated annual mean latitudinal concentration gradients of methane arepresented in Figure 4.2 (top panel). It shows that the model overestimates the interhemi-spheric gradient by about 40 nmol/mol (30%), which is a 25% improvement compared toour previous work (SH99). As shown in section 4.4 at most halfof this discrepancy can beexplained by transport errors. The changed natural wetlandemission distribution and moreefficient cross equatorial transport in TM3 resulted in a larger improvement of the simulatedmethane gradient. However, the updated parameterization of soil oxidation tends to increasethe gradient again, owing to an increased contribution of the tropics, as compared with thedistribution byFung et al.[1991] we used previously.

The simulated amplitudes of seasonal cycles show a general overestimation compared to

4.5 Results and discussion 89

a

-90 -45 0 45 90Latitude (degrees)

1.65

1.70

1.75

1.80

1.85

1.90

CH

4(µ

mo

l/m

ol)

MLO

BRW

b

-90 -45 0 45 90Latitude (degree)

0

20

40

60

80

100

CH

4(n

mo

l/m

ol)

MLO

BRW

Figure 4.2: (a) Measured and simulated mixing ratios of CH4 showing the annual mean inter-hemispheric gradients of CH4 for 1993. The offset in the model results has been adjusted tothe South Pole (90S 25W 2810masl). (b) Measured and simulated seasonal cycle amplitudes,estimated from the difference between highest and lowest monthly mean for 1993. Circlesrepresent measurements; and triangles represent model simulations.

observations, particularly in the Northern Hemisphere (Figure 4.2, bottom panel). This biaspoints to errors in the balance between sources with distinct seasonal cycles, such as biomassburning, wetland emissions, and the soil and/or hydroxyl radical sink. In the tropics, wherethe seasonality is partly determined by the movement of the intertropical convergence zone(ITCZ), the overestimated amplitudes may also be related tothe overestimated latitudinalmethane gradient.

For 1993, a tropospheric average trend of 0.2% is computed, in good correspondencewith the observed global averaged increase of 4 nmol mol−1yr−1 and global mean mixingratio of 1715 nmol/mol for this year [Dlugokencky et al., 1998]. Also, we obtain a fairagreement between the observed and simulated annual mean concentration at backgroundstations. As a result, the applied offset adjustments are only small (0.5 and -18.6 nmol/mol


for the Southern and Northern Hemisphere, respectively). The fact that both the observedtrend and the mean concentration are accurately reproducedby the model indicates that thecombination of assumed sources and CH3CCl3-calibrated sinks realistically represents theCH4 budget.

4.5.1 In situ measurements

High frequency in situ measurements at Mauna Loa and Barrow provide useful informationabout methane concentration variations on different temporal and spatial scales. Although arigorous analysis of these scales would require many stations, testing the model at Mauna Loaand Barrow is valuable already. This is partly the case because these stations are located invery different environments, representative of well mixedfree tropospheric air (Mauna Loa),and the continental boundary layer, much closer to (possibly even local) sources of methane(Barrow). Model results and in situ measurements at these stations are presented in Figures4.3 and 4.4. At both locations the model has been “sampled” using the same time schedule asfor the measurements. This synchronized sampling also explains the straight line segmentsin these Figures when there are no measurements performed.

The Mauna Loa station is located at 3.4 kilometer altitude onthe north slope of the MaunaLoa volcano on Hawaii. This geographical location largely determines the daily flow con-ditions at the station. During daytime, radiative heating of the island induces convectiveupslope winds transporting air from lower altitudes to the station. After sunset the land–sea circulation reverses and free-tropospheric air is transported to Mauna Loa by a katabaticdownslope flow [Atlas and Ridley, 1996]. As a result, methane measurements at Mauna Loashow a distinct daily cycle (see Figure 4.3, e.g. days 110–130) with maxima during daytimeand minima during nighttime. Since in the model these up-slope and down-slope winds arenot resolved, the variability of the simulated methane concentration on the time scale of aday is underestimated. In the model the concentration difference between sea level and 3.4km altitude at Hawaii amounts to 22 nmol/mol, which confirms that such local winds maylargely explain the diurnal variation of methane at Mauna Loa.

On the time scale of a week the observed variability is quite well reproduced by themodel (e.g. day 130–150, and 240–270). The amplitude of the seasonal cycle is slightlyoverestimated, as can be seen when comparing the means of December–January to July–August. Partly, this is explained by the overestimated interhemispheric gradient since duringthe July–August period Mauna Loa is relatively frequently influenced by air from the equa-torial tropics [Harris et al., 1992], i.e. from the southern hemispheric “compartment” of theHadley circulation in this part of the year. Also the seasonality of tropical sources and thehydroxyl radical sink may, however, contribute.

At Barrow the picture is quite different. Periods for which the model performs reasonably(e.g. days 250–300) are alternated by periods where model and measurements do not corre-late (e.g. days 60–90). This variable model performance is manifest on different time scales,from reasonable agreement on small scales (e.g. days 200–210) to significant overestimateson the seasonal scale (e.g. days 150–200). The latter is likely related to the representation ofarctic natural wetlands, the largest high latitude source,in particular at relatively high tem-peratures during summer. On smaller scales, model resolution may become a limiting factor.However, a similar model run on 2.5◦x2.5◦ resolution did not lead to significant improve-ments, so that other factors, for example temporal variability of the regional sources, may bemore important. In springtime the amplitude of the simulated daily cycle, which peaks beforesunrise, is strongly overestimated by the model (see, e.g.,the peaks in the model simulations


Mauna Loa (20N 156W 3397masl)

0 10 20 30 40 50 60day of year

1680

1700

1720

1740

1760

1780

CH

4 (n

mol

/mol

) R=0.67


60 70 80 90 100 110 120day of year

16801700

1720

1740

1760

17801800

CH

4 (n

mol

/mol

) R=0.76


120 130 140 150 160 170 180day of year

1680

1700

1720

1740

1760

1780

CH

4 (n

mol

/mol

) R=0.66


180 190 200 210 220 230 240day of year

1660

1680

1700

1720

1740

1760

CH

4 (n

mol

/mol

) R=0.63


240 250 260 270 280 290 300day of year

16601680

1700

1720

1740

17601780

CH

4 (n

mol

/mol

) R=0.84


300 310 320 330 340 350 360day of year

1700

1720

1740

1760

1780

1800

CH

4 (n

mol

/mol

) R=0.59

Figure 4.3: Comparison of in situ measurements at Mauna Loa to model simulations for1993. Squares, measurements; solid lines, simulations. The offset of the model results hasbeen adjusted to match the annual mean at Mauna Loa. “R” denotes the correlation coefficientof residual concentrations of model simulations and measurements. Residuals are defined asthe difference between daily averages and panel (60 d) averages.


Barrow (71N 157W 11masl)

0 10 20 30 40 50 60day of year

1750

1800

1850

1900

1950

CH

4 (n

mol

/mol

) R=0.42


60 70 80 90 100 110 120day of year

1750

1800

1850

1900

1950

CH

4 (n

mol

/mol

) R=0.38


120 130 140 150 160 170 180day of year

1750

1800

1850

1900

1950

CH

4 (n

mol

/mol

) R=0.12


180 190 200 210 220 230 240day of year

1700

1750

1800

1850

1900

1950

CH

4 (n

mol

/mol

) R=-0.0


240 250 260 270 280 290 300day of year

17401760178018001820184018601880

CH

4 (n

mol

/mol

) R=0.44


300 310 320 330 340 350 360day of year

1750

1800

1850

1900

1950

2000

CH

4 (n

mol

/mol

) R=0.01

Figure 4.4: Comparison of in situ measurements at Barrow to model simulations for 1993.Squares, measurements; solid lines, simulations. The offset of the model results has beenadjusted such that the measured and modeled baseline concentrations agree. “R” denotes thecorrelation coefficient of residual concentrations of model simulations and measurements.Residuals are defined as the difference between daily averages and panel (60 d) averages.


for day 150–170). This suggests an overestimate of local sources, or an underestimate ofboundary layer mixing.

4.5.2 Flask measurements

A direct comparison of model results and flask measurements is a difficult test for the model.In fact, whether or not the model reproduces single measurements is of limited value sincethis may largely be accidental. Alternatively, monthly averaged concentrations can be used.As can be deduced from Figures 4.3 and 4.4, weekly flask samples are expected to poorly rep-resent monthly concentration distributions, at least at stations that show similar variabilities.As a consequence, an error is introduced when comparing model-derived means, based onconcentrations computed at each model time step, to averages derived from flask sampling.As pointed out byHaas-Laursen et al.[1997], reducing the frequency of model samplingto the flask sampling frequency does not reduce this error. Rather, a second statistical erroris introduced. Sampling synchronization, however, will reduce such errors since we do notintend to estimate the monthly mean concentration, but an average of a set of measurementsduring a month, of which the individual samples represent the same time.

Figure 4.5 shows the agreement between flask samples (averages of duplicates) and theircorresponding simulations at remote to relatively polluted sites. At background stations suchas Samoa, Ascension Island, and Midway model and flask measurements agree satisfactorily.A substantial part of the (still substantial) variability is reproduced by the model. At CapeGrim significant differences occur in the first half of the year, which, as we will show later,includes a contribution by sampling selection at the observatory. At Mace Head and NiwotRidge the agreement is still fairly good compared with Barrow, indicating that this in situstation is relatively difficult to reproduce by the model. AtTae-Ahn Peninsula there is littleagreement between model and measurements particularly during summer, when the observedvariability is large as a result of changing wind directions, bringing either clean air from thePacific or highly polluted air from the Asian continent, where large methane sources arelocated (see also section 4.5.4).

To illustrate the effect of synchronized sampling on monthly averages, model-derivedhigh frequency (model time step) and low frequency (synchronized) means are comparedwith measurements (Figure 4.6). At remote stations such as the South Pole (89◦59’S, 24◦48’W,2810masl), Samoa (14◦15’S, 170◦34’W, 42masl) and Syowa (69◦00’S, 39◦35’E, 11masl)[not shown], differences between simulated high and low frequency monthly means are neg-ligible. This indicates that a frequency of one sample per week is sufficient to quite accuratelydetermine monthly mean concentrations at these stations. Similar differences are found atsome stations closer to the sources such as Ascension Island, Ocean Station M, and Izana.At stations such as Mace Head, Bermuda, Niwot Ridge, and alsoat Cape Grim (for the lattersee Figure 4.8), however, significantly different model results are obtained using differentsampling frequencies. Here, as expected, weekly samples poorly represent the temporal con-centration distribution. Furthermore, the fact that sampling synchronization improves theagreement between model results and measurements means that the model reproduces ob-served fluctuations of the methane concentration. Effects of sampling frequency on computedmonthly averages are also seen at more polluted sites, e.g. at Baltic Sea, Tae-ahn Peninsulaand Barrow (for the latter see Figure 4.8). At these sites, however, synchronized samplingdoes not significantly improve the agreement with measurements. Here, an important part ofthe variability is simply not reproduced by the model, as wasalready seen in the comparisonswith in situ data at Barrow.


Samoa (14S 171W 42masl)

1992.5 1993.0 1993.5 1994.0 1994.5time (yr)

1620

1640

1660

1680

1700

1720

CH

4 (n

mol

/mol

)

Cape Grim (41S 145E 94masl)

1992.5 1993.0 1993.5 1994.0 1994.5time (yr)

1620

1640

1660

1680

1700

1720

CH

4 (n

mol

/mol

)

Ascension (8S 14W 54masl)

1992.5 1993.0 1993.5 1994.0 1994.5time (yr)

1620

1640

1660

1680

1700

1720

CH

4 (n

mol

/mol

)

Midway (28N 177W 4masl)

1992.5 1993.0 1993.5 1994.0 1994.5time (yr)

1700

1750

1800

CH

4 (n

mol

/mol

)

Mace Head (53N 10W 25masl)

1992.5 1993.0 1993.5 1994.0 1994.5time (yr)

1750

1800

1850

CH

4 (n

mol

/mol

)

Niwot Ridge (40N 106W 3475masl)

1992.5 1993.0 1993.5 1994.0 1994.5time (yr)

1700

1750

1800

CH

4 (n

mol

/mol

)


1992.5 1993.0 1993.5 1994.0 1994.5time (yr)

1800

1850

1900

CH

4 (n

mol

/mol

)

Tae-Ahn Peninsula (37N 126E 20masl)

1992.5 1993.0 1993.5 1994.0 1994.5time (yr)

1800

1850

1900

1950

2000

CH

4 (n

mol

/mol

)

Figure 4.5: Comparison of flask measurements to model simulations for 1993. Circles aremeasurements; squares are simulations; and asterisks are measurements designated as pol-luted by NOAA. Measurements exceeding the scale range are designated as polluted byNOAA. The offset of the model results has been adjusted to theannual mean at Midway(for Northern Hemispheric sites) and the South Pole (for Southern Hemispheric sites).


Izana (28N 16W 2300masl)

J F M A M J J A S O N D month

1700

1720

1740

1760

1780

1800

CH

4 (n

mol

/mol

)

measuredmodel "high frequency" model "synchronized"

Ocean Station "M" (66N 2E 7masl)


1760

1780

1800

1820

1840

1860

1880

CH

4 (n

mol

/mol

)

Ascension (8S 14W 54masl)


1640

1660

1680

1700

1720

CH

4 (n

mol

/mol

)

Mace Head (53N 10W 25masl)


1760

1780

1800

1820

1840

1860

1880

CH

4 (n

mol

/mol

)

Bermuda East (32N 65W 30masl)


1680

1700

1720

1740

1760

1780

1800

1820

CH

4 (n

mol

/mol

)

Niwot Ridge (40N 106W 3475masl)


1700

1720

1740

1760

1780

1800

CH

4 (n

mol

/mol

)

Cape Meares (45N 124W 30masl)


1720

1740

1760

1780

1800

1820

1840

1860

CH

4 (n

mol

/mol

)

Baltic Sea (56N 17E 7masl)


1750

1800

1850

1900

1950

2000

2050

CH

4 (n

mol

/mol

)



1750

1800

1850

1900

1950

2000

CH

4 (n

mol

/mol

)

Figure 4.6: Comparison of monthly means as derived from measurements and model sim-ulations. Squares, measurements with 1σ uncertainties; solid lines, model sampled at eachtime step; dashed lines, model sampled simultaneous to the measurements. The offset of themodel results has been adjusted to the annual mean at Midway (for Northern Hemisphericsites) and the South Pole (for Southern Hemispheric sites).


0.0 0.2 0.4 0.6 0.8 1.0fraction of samples

0.0

0.2

0.4

0.6

0.8

1.0

Cdi

ag. (

rel.

to c

ontin

ents

)

0.00

0.10

0.20

0.30

0.40

0.50

Cape GrimHeimaeyBarrow

Figure 4.7: Distributions of diagnostic tracer concentrations (Cdiag) for windsector selection.Concentrations are expressed relative to an annual averagecomputed for nearby continents.Arrows indicate the scales (left or right y-axis) that correspond to the concentration distribu-tions.

4.5.3 Wind sector selection

At a number of stations air samples are collected only if the wind direction is from a certainwind sector, with the intention to avoid contamination by local sources and, thus, to extendthe scale for which the sampled air is representative. Unless this wind selection is accountedfor in the model, this introduces a representation error when comparing these data to modelresults. For stations such as Cape Grim previous studies on atmospheric CO2 have shownthat these errors cannot be neglected [Ramonet and Monfray, 1996;Law, 1996]. Synchro-nized sampling may compensate for this baseline selection,albeit only to a certain extentsince the observed and modeled wind directions may differ. Furthermore, the clean air sectorin the model can differ from reality since the continents arerepresented at a limited reso-lution. Alternatively, as proposed byRamonet and Monfray[1996], clean air selection canbe accounted for in the model by introducing a222Rn-like diagnostic tracer with an atmo-spheric turnover time of about a week, and a homogeneous source over the continents only.Air parcels of continental origin are identified by elevatedlevels of this diagnostic tracer, andsamples are excluded whenever this tracer exceeds a certainthreshold.

Based on an analysis of ECMWF wind fields,Haas-Laursen et al.[1997] concludedthat at Cape Grim and Heimaey winds are selected from the South–West, and North–East,respectively. Although, this analysis was performed for CO2, the same results apply to CH4

since both gases are analyzed from the same samples. NOAA reports wind sector selectionat Barrow (North to South–East) [Fergusin and Rosson, 1992], although this is not confirmedby Haas-Laursen et al.[1997].

A threshold concentration of the diagnostic tracer is determined from its concentrationdistribution at each station (see Figure 4.7). Ideally, thedistributions would have two modes,representative of continental and marine air. Then we wouldchoose the threshold in be-


tween. In absence of such well defined modes, the change in slope is used as the transitionpoint of continental and marine air. In this way we derive thresholds of 0.1 (Cape Grim), 0.15(Heimaey), and 0.35 (Barrow). Measurements of222Rn [Whittlestone et al., 1996] at CapeGrim and simulated concentrations over the Australian continent [Dentener et al., 1999] con-firm that the Cape Grim threshold is quite realistic for the methane observations at this station.

Figure 4.8 shows the effect of clean air selection at Cape Grim, Heimaey, and Barrow. AtCape Grim, combined sampling synchronization and wind direction selection almost com-pletely explain the overestimation of the simulated monthly means during the first half ofthe year. At Heimaey the simulated concentration maximum inJuly, which is absent in themeasurements, is significantly reduced when wind directionselection is used. Also at Bar-row clean air sector selection improves the agreement between model and measurements,however, model overestimated concentrations remain in June. Samples that are excluded byclean air sector selection are also not used to compute averages of measurements. As a re-sult, corresponding stations in Figure 4.8 show differences in measurements and uncertaintyranges. Drastic changes in the computed 1σ uncertainty intervals indicate that these rangesare caused by the small number of measurements.

4.5.4 Non-local air selection

Measurements at continental stations have the advantage over remote stations that character-istics of the observed concentration distribution may directly point at sources in a particularregion. Their interpretation may, however, be misleading since this region may be smallerthan the smallest scale represented by the model (∼ the grid-scale). To quantify the poten-tial contribution of sub-grid sources, we define a second diagnostic tracer, which is emittedas CH4 in the four grid boxes closest to a station only. A single tracer can be used for allmeasurement sites if it is assigned a sufficiently small lifetime, i.e. such that the “local”tracer emitted for a particular station does not significantly contribute to its concentration atany other station. On the grid-scale, however, its lifetimeshould be long enough to mimicmethane (approximately inert on these scales). A lifetime of one week for the tropics andmid-latitudes and 0.5 week at higher latitudes appeared to satisfactorily meet both require-ments.

From computed methane and diagnostic tracer concentrations a measureR is derived ofthe contribution of “sub-grid” sources to the simulated concentration at a stationi and monthm, defined as

Ri,m =di,m

(CH4i,m−CH4′m), (4.1)

whered denotes the diagnostic tracer concentration,CH4 the methane concentration at thea particular station, andCH4′ the background concentration at a Pacific site at maximumdistance from the continents, at the station’s latitude. Obviously, the contribution of sub-gridsources in the model may differ from the contribution of the corresponding sources in reality.Since it is expected that stations are located at relativelylarge distances from sources, ascompared with their direct surroundings, the real contribution of local sources is not expectedto be much larger than indicated by the model-derivedR. WheneverRi,m exceeds a certainthreshold, the corresponding samples are excluded.

Figure 4.9 shows results of this “local-influence” filter. Inthese comparisons a thresholdof R=25% is chosen. Since at remote stations the methane concentration may almost be equalto the background reference (CH4′), here small local sources of methane may already cause a




1640

1660

1680

1700

1720

CH

4 (n

mol

/mol

)

measuredmodel "high frequency" model, wind sector selection



1640

1660

1680

1700

1720

CH

4 (n

mol

/mol

)


Heimaey (63N 20W 100masl)


1760

1780

1800

1820

1840

1860

CH

4 (n

mol

/mol

)



1760

1780

1800

1820

1840

1860

CH

4 (n

mol

/mol

)



1780

1800

1820

1840

1860

1880

1900

CH

4 (n

mol

/mol

)



1780

1800

1820

1840

1860

1880

1900

CH

4 (n

mol

/mol

)

Figure 4.8: Clean air sector selection. Squares, measurements with 1σ uncertainties; solidlines, model sampled at each time step; dashed lines, model sampled simultaneous to themeasurements, with (left panels) and without (right panel)clean sector selection. Measure-ments at Barrow represent flask samples only. The offset in the model is adjusted as in Figure4.6.




1750

1800

1850

1900

1950

2000

2050

CH

4 (n

mol

/mol

)

measuredmodel "high frequency" model, non-local air selection



1750

1800

1850

1900

1950

2000

2050

CH

4 (n

mol

/mol

)




1780

1800

1820

1840

1860

1880

1900

CH

4 (n

mol

/mol

)



1780

1800

1820

1840

1860

1880

1900

CH

4 (n

mol

/mol

)



1750

1800

1850

1900

1950

2000

CH

4 (n

mol

/mol

)



1750

1800

1850

1900

1950

2000

CH

4 (n

mol

/mol

)

Figure 4.9: Non-local air selection. Squares, measurements with 1σ uncertainties; solidlines, model sampled at each time step; dashed lines, model sampled simultaneous to themeasurements, with (left panels) and without (right panel)non-local air selection. The offsetin the model is adjusted as in Figure 4.6.


R higher than this threshold. To avoid this, we allowed samples to pass the filter irrespectiveof R for diagnostic tracer concentrations smaller than 1 nmol/mol. At the Baltic Sea andBarrow minor effects of filtering locally-influenced data can be seen. The main differencesbetween measurements and model computations, however, remain. At Tae-Ahn Peninsulamost of the measurements during summer are identified as locally affected (data gaps inFigure 4.9 indicate that no sample passed the filter). Averaged over the 1993 samples, theCH4 concentration at Tae Ahn Peninsula is∼100 nmol mol−1 higher than the background(CH4 −CH′

4), with a 37 nmol mol−1 contribution of local sources (d). This indicates thatcomparisons between model and measurements at such stations should be treated with care,since differences between model simulations and measurements may largely be explained bysources that are not resolved by the model.

4.5.5 Influence of wetland emissions

The previous sections showed the importance of consistently comparing model results andmeasurements. However, the question remains what can be learned about atmospheric methanefrom these comparisons. This subsection focuses on two regions that show the most pro-nounced differences between model and measurements; the high latitude stations of theNorthern Hemisphere, and the South China Sea. These sites are all located in regions withimportant sources of methane, in particular, natural and agricultural wetlands.

Northern high latitude wetlandsAt Heimaey and Ocean Station M it is shown that the model overestimates methane con-

centrations in summer (see Figure 4.6 and 4.8). In the previous section it was shown thatthe simulated July maximum at Heimaey may largely be explained by the absence of windsector selection in the model. However, a similar peak is found at Ocean Station M (seeFigure 4.6) and Ny–Alesund (not shown), where, according toHaas-Laursen et al.[1997],no wind sector selection is applied. In contrast, at Barrow such a summertime maximum isabsent (Figure 4.4), although the model significantly overestimates concentrations in June.Figure 4.10 shows comparisons at three additional stations. The results for Alert and MouldBay do not confirm that the model overestimates methane concentrations at higher latitudesin summer. Fraserdale, downwind of the Hudson Bay wetlands,strongly points to a modeloverestimate of wetland emissions. If local contributions>25% are filtered, as explained inthe previous section, it appears that the model is not expected to reliably resolve this sta-tion in summer. To summarize, although the model has difficulties reproducing the seasonalvariation at a number of higher latitude stations of the Northern Hemisphere, this does notunambiguously point to model overestimated emissions.

To determine the potential role of errors in the parameterization of natural wetland emis-sions, two relatively uncertain parameters have been tested (B. P. Walter, personal communi-cation, 1999). Firstly, the parameter that describes the relative change of methane productionrate following a 10 K change of temperature (Q10) has been decreased from 6 to 2. This lowervalue represents the lower limit of previously reported Q10 values, which vary between 1.7and 16 [Dunfield et al., 1993;Valentine et al., 1994;Westermann, 1993].

Secondly, the potential influence of the assumed micro-topography of natural wetlandshas been assessed. Generally, natural wetlands have a typical complex structure of floodedand dry patches, caused by slight surface elevations (hummocks) or depressions (hollows).The standard parameterization prescribes a uniform water table, determined by the balance


Alert (82N 63W 210masl)


1780

1800

1820

1840

1860

CH

4 (n

mol

/mol

)


Mould Bay (76N 119W 58masl)


1760

1780

1800

1820

1840

1860

1880

CH

4 (n

mol

/mol

)

Fraserdale (50N 82W 250masl)


1750

1800

1850

1900

1950

2000

CH

4 (n

mol

/mol

)

Figure 4.10: Comparison of monthly means at locations influenced by natural wetland emis-sions. Squares, measurements with 1σ uncertainties; solid lines, model sampled at each timestep; dashed lines, model sampled simultaneous to the measurements; dotted lines, monthlymeans with a>25% contribution of local sources. The offset in the model isadjusted as inFigure 4.6.


Alert (82N 63W 210masl)


1780

1800

1820

1840

1860

CH

4 (n

mol

/mol

)

measuredstandard modelQ10=2 micro-topography



1780

1800

1820

1840

1860

1880

1900

CH

4 (n

mol

/mol

)



1760

1780

1800

1820

1840

1860

CH

4 (n

mol

/mol

)

Fraserdale (50N 82W 250masl)


1800

1850

1900

1950

2000

CH

4 (n

mol

/mol

)

Figure 4.11: Sensitivity of simulated concentrations to uncertain natural wetland parameters.Squares, measurements with 1σ uncertainties; solid lines, standard simulation; dashed lines,simulation with decreased sensitivity of wetland emissions to temperature changes; dottedlines, simulation accounting for micro-topography (see text).

between precipitation, evaporation, and run-off. The hydrological conditions of hummocksand hollows are, however, relatively insensitive to the water balance, i.e. they remain dryand flooded largely independent of the water supply. As a testit has been assumed that 10%of the northern wetlands (north of 30◦N) are water filled throughout the year, whereas 30%are assumed to be hummocks that are so dry that no significant methane production takesplace. Global and annual totals have been re-scaled to 145 TgCH4 yr−1, in agreement withthe standard simulation.

As illustrated in Figure 4.11, simulated methane concentrations are quite sensitive tothe parameterization of natural wetland emissions. A reduction of Q10 has a particularlystrong influence, significantly improving the agreement between model and measurements atHeimaey and Fraserdale. At Barrow, the model estimate for June improves, but a minimumappears in September which is not seen in the measurements. The importance of Q10 is


Figure 4.12: Definition of source regions affecting South China Sea samples.

confirmed by a comparison for Alert, although it shows that differences between model andmeasurements cannot be explained by a uniform reduction of this parameter only. These testsnevertheless indicate that, instead of errors in the absolute magnitude of wetland emissions,errors in the distribution and timing of wetland emission may largely explain discrepanciesbetween model and measurements.

South East AsiaTo investigate emissions from the Asian continent measurements at the South China Sea

(SCS) in 1992 are used. The samples were taken from cruises from Singapore to Hong–Kongat latitudinal intervals of 3◦ between 3◦ and 24◦N. Ship measurements have the advantageover continental stations, such as Tea-Ahn Peninsula, thatthe contribution of local sourcesis negligible. Figure 4.13 shows simulations and measurements along the South China Seacruise track. Measurements at Guam, a remote Pacific stationat approximately the samelatitude as the cruise measurements, are used to correct forthe model offset. Generally,these results indicate that the SCS methane concentrationsare overestimated by the model.Maximum differences occur in summer, somewhat earlier in the year at the southern (3◦N–6◦N) compared to the northern locations (15◦N–24◦N). Although, differences between highand low frequency sampling of the model are sometimes quite large, this does not alter thegeneral picture. The seasonal pattern is explained by prevailing winds from the continent inwinter, as opposed to the summer monsoon, when air from the tropical Pacific is transportedover Indonesia to the South China Sea. Although continentalsources such as rice paddies arestronger during summer, their influence is masked by a largersupply of background air.

To quantify the contribution of different source regions tothe methane concentrationsas observed over the South China Sea, marked tracer simulations were performed. For this


South China Sea (3N 105E 0masl)


1650

1700

1750

1800

1850

1900

CH

4 (n

mol

/mol

)



1650

1700

1750

1800

1850

CH

4 (n

mol

/mol

)



1680

1700

1720

1740

1760

1780

1800

1820

CH

4 (n

mol

/mol

)



1650

1700

1750

1800

1850

CH

4 (n

mol

/mol

)



1700

1750

1800

1850

CH

4 (n

mol

/mol

)




1700

1750

1800

1850

1900

CH

4 (n

mol

/mol

)

Figure 4.13: Comparison of model and measurements at the South China Seafor 1992.Squares, measurements with 1σ uncertainties; solid lines, model sampled at each time step;dashed lines, model sampled simultaneous to the measurements (no clean air selection). Theoffset in the model is adjusted such that the measured and modeled annual means agree atGuam (13N 145E 2masl).

4.6 Summary and conclusions 105

purpose, four source regions are defined (see Figure 4.12). In the model, methane emissionsfrom each region are represented by a different marked tracer. Except for their sources, allmarked tracers are treated as methane in a standard simulation.

Results of this marked tracer simulation are presented in Figure 4.14. To highlight themost relevant information, only regions with the most important contributions have been in-cluded. At SCS 3◦N, as expected, the Indonesian contribution (regionD) is relatively largesince these measurements are within approximately 300 km north of Singapore. More to thenorth, the contribution of regionD decreases and emissions from regionB, including coun-tries such as Thailand, Laos, and Vietnam, gain importance.Further northwards the contribu-tion of this region again decreases and China (regionA) becomes more important. Thus, theregion contributing strongest to the simulated concentrations changes with the sampling lo-cation, whereas the model overestimates methane over the entire South China Sea. It followsthat the model overestimated concentrations are not explained by emissions from a singleregion, but rather have contributions from regionA, B, andD, with less contribution from thePhilippine regionC. Although the integrated emissions from all four regions largely explainthe observed concentration variations, an important part of the seasonal cycle is explained bysources at larger distance. It follows that, methane simulations pertaining to South-East Asiausing regional models, require that a quite large model domain is accounted for. Therefore,preferably regional models should be nested within global models.

To relate differences between model and measurements to particular source processsesrather than regions, additional simulations have been carried out in which particular sourcesare marked (emitted from all regions). Figure 4.15 shows a large contribution by natural wet-land emissions for SCS 3◦N and 6◦N, most likely related to the natural wetlands of Sumatra.Further north, rice agriculture gains importance, particularly in summer, whereas the sumof other sources, for example, ruminants, fossil fuel, and waste treatment becomes increas-ingly important in winter. Generally, the sum of rice paddy and natural wetland emissionscontributes more than 50% to methane from South East Asian sources, notably when the dif-ferences between simulated concentrations and measurements are largest. This indicates thatthe overestimated concentrations at South China Sea are most likely related to a combinationof these wetland sources.

The advantage of using 1992 for the comparisons at South China Sea is the relativelylarge number of samples that were taken during these particular ship cruises. This year can-not be considered as representative of a larger period, however, since there are indicationsthat methane sources and sinks were influenced by the Mt. Pinatubo eruption of 15 June1991 [Dlugokencky et al., 1994a]. It has been suggested that wetland emissions may havebeen reduced as a result of changed temperature and precipitation patterns caused by the Mt.Pinatubo effluents [Dlugokencky et al., 1996;Hogan and Harris, 1994]. To examine the pos-sible influence of Mt. Pinatubo we repeated our model runs forthe years 1994 and 1996. Asan additional test, we increased the model resolution to 2.5x2.5 degree for 1996. The resultsof these runs (see Figure 4.16) indicate that the differences between model and measurementsare indeed relatively large in 1992, but cannot be explainedby the Mt. Pinatubo effect alone.

4.6 Summary and conclusions

We presented simulations of tropospheric methane using a CTM, and parameterized sourcesand sinks based on the most recently available information.Measurements obtained by high




1650

1700

1750

1800

1850

1900

1950

CH

4 (n

mol

/mol

)

= outside Azian region= . .+ region "D" = . .+_ _+ region "B" = all regions



1650

1700

1750

1800

1850

CH

4 (n

mol

/mol

)

= outside Azian region= . .+ region "A" = . .+_ _+ region "B" = all regions



1650

1700

1750

1800

CH

4 (n

mol

/mol

)



1650

1700

1750

1800

1850

CH

4 (n

mol

/mol

)



1650

1700

1750

1800

CH

4 (n

mol

/mol

)



1650

1700

1750

1800

1850

1900

1950

CH

4 (n

mol

/mol

)

Figure 4.14: Contribution of specified source regions (see Figure 4.12)to the simulatedmethane concentrations at South China Sea. Solid lines, allsources included; dotted lines, allsources except those of the specified source regions; dashedlines, dotted lines plus sourcesfrom the region D (left panels) or region A (right panels); dashed-dotted, dashed lines plussources of region B.




1650

1700

1750

1800

1850

1900

1950

CH

4 (n

mol

/mol

)

= outside Azian region= . .+ swamps = . .+_ _+ rice = all sources



1650

1700

1750

1800

1850

CH

4 (n

mol

/mol

)

= outside Azian region= . .+ swamps = . .+_ _+ rice = all sources



1650

1700

1750

1800

CH

4 (n

mol

/mol

)



1650

1700

1750

1800

1850

CH

4 (n

mol

/mol

)



1650

1700

1750

1800

CH

4 (n

mol

/mol

)



1650

1700

1750

1800

1850

1900

1950

CH

4 (n

mol

/mol

)

Figure 4.15: Contribution of wetland emissions from all specified source regions (see Figure4.12) to the simulated methane concentrations at South China Sea. Solid lines, all sourcesincluded; dotted lines, sources outside the selected domain only; dashed lines, dotted linesplus natural wetland emissions in the selected domain; dashed-dotted, dashed lines plus ricepaddy emissions in the selected domain.


Figure 4.16: Compilation of measurements and model calculations for all South China Seacruises. Circles are based on a simulation at 4x5 degree resolution; and squares are based ona simulation at 2.5x2.5 degree resolution (1996 only).


frequency (60 d−1) in situ and low frequency (weekly) flask sampling have been used toevaluate the model, and identify shortcomings. For background stations, such as AscensionIsland, Guam, and Midway, the agreement is satisfactory. AtMauna Loa, where in situ mea-surements are available, a substantial part of the observedshort-term (∼ week) variabilityis reproduced by the model, indicating that large scale transport is represented well by themodel. Except for a 30% overestimate of the interhemispheric (pole-to-pole) concentrationgradient and a 25% overestimate of the amplitude of seasonalcycles in the Northern Hemi-sphere, the representation of sources and sinks largely explains observed concentrations atbackground stations. Because of the low variability at these stations, monthly mean concen-trations can be derived from weekly samples with reasonableaccuracy. Differences betweensimulated monthly means derived from high and low frequencysampling of the model atremote stations, such as South Pole, and Syowa, are well within the measurement accuracy.

Model simulations indicate that with decreasing distance to the sources, differences be-tween high and low frequency sampling increase, up to 20–30 nmol/mol at Barrow and MaceHead, owing to increased variability on shorter time scales. At stations such as Niwot Ridge,Bermuda East, and Mace Head, the agreement between model andmeasurements is signif-icantly improved when synchronized sampling is applied, i.e. when the model is sampledat the same time and location as the measurements, instead ofsampling at each model timestep. At even shorter distances to the sources, at Tae Ahn Peninsula, Baltic Sea and Barrow,sampling synchronization does not lead to significant improvements since other factors, inparticular the limited model resolution, become dominant.

For coastal observatories that apply clean sector selective sampling, for example, at Bar-row, Cape Grim and Heimaey, the comparability of measured and modeled monthly meanscan further be improved by baseline selection in the model. As suggested byRamonet andMonfray [1996] this can be achieved by simulating a diagnostic tracer that marks continen-tal air, and excluding methane samples whenever this tracerconcentration exceeds a certainthreshold. As previously shown for CO2, also for CH4 discrepancies between simulationsand measurements at Cape Grim can largely be attributed to sampling selection. Similar ef-fects are shown at Heimaey, Iceland. We showed that from the concentration distribution ofthe diagnostic tracer a threshold value can be derived, discriminating between “marine” and“continental” air. This threshold may, however, lead to a different degree of clean air selec-tion in the model as in the measurements. If, in addition, the222Rn concentration is known atthe moment of sampling this can be used to verify this threshold. Generally, we feel that theinterpretation of comparisons between measured and model-derived methane would benefitfrom simultaneous measurements of222Rn, to discriminate marine and continental air, and,for example, SF6 to discriminate air influenced by industrial and biogenic sources.

The usefulness of continental observations is limited by processes on smaller scales thanrepresented by the model. To quantify their potential effects a method is proposed in whichthe contribution of local grid boxes to the simulated concentration is used as a proxy of “sub-grid” sources. Although local effects cannot be filtered from the results, it enables assessmentof the usefulness of a particular station for testing the model, as illustrated for Barrow, BalticSea and Tae Ahn Peninsula. This procedure indicates that forthe present model resolution(3.75◦x5◦) the Korean station is of limited use.

After screening comparisons of model and measurements for representation errors, unex-plained disagreements remain at a number of stations, pointing to shortcomings in the model.The model has difficulties reproducing seasonal variationsat stations at higher latitudes of theNorthern Hemisphere. Comparisons at Heimaey, Ocean Station M, Barrow and Fraserdalesuggest overestimated natural wetland emissions, which isnot confirmed, however, at Alert


and Mould Bay. In addition, representation errors related to wind direction selection and sub-grid sources may largely explain differences between modeland measurements at Heimaeyand Fraserdale, respectively. To examine the potential role of natural wetland emissions, sen-sitivity simulations have been performed in which the temperature dependence of wetlandmethane production and the wetland micro-topography were tested. Model simulations in-dicate that seasonal cycles at Alert, Barrow, Fraserdale, and Heimaey are highly sensitive tothese parameters, which could potentially explain much of the discrepancies between modeland measurements. This high sensitivity suggests that uncertain process model parameterscould be further constraint by inverse modeling.

We further focused on a number of locations in the South-EastAsian region. Three yearsof measurements indicate that emissions over this continent may have been overestimated.Results of marked tracer simulations, in which methane fromdifferent sources and regions isdistinguished, point to natural wetland emissions from Sumatra and Chinese rice paddies aslikely candidates. Recent estimates on the basis of inversemodeling (SH99) and up-scaling[Huang et al., 1997;Denier van der Gon, 2000a] point in the same direction.

We conclude that by applying methods to compensate for representation errors, significantimprovements in the comparison between model results and measurements can be achieved.Differences of these “screened” modeled concentrations and measurements are a better indi-cation for discrepancies in e.g. emission distributions. The same methods can in principal beapplied in inverse modeling, for example, to define a suitable set of measurements or to assessthe uncertainties associated with comparisons of measurements and model simulations. Thiswill be the aim of future work.

Acknowledgments

We thank M. Heimann (MPI for Biogeochemistry) and T. Kaminski (MPI for Meteorology)for useful discussions. Further, helpful comments and suggestions were provided by twoanonymous reviewers. This work has been supported by the Dutch Global Change program,NOP project 951202.

Chapter 5

Simulation of pre-industrialatmospheric methane to constrainthe global source strength ofnatural wetlands

Previous attempts to quantify the global source strength ofCH4 from natural wetlands haveresulted in a range of 90–260 Tg(CH4) yr−1. This relatively uncertain estimate significantlylimits our understanding of atmospheric methane. In this study we reduce this uncertainty bysimulating pre-industrial CH4 with a three dimensional chemistry-transport model. Methanemixing ratios andδ13C-CH4, as deduced from ice cores, and estimates of other pre-industrialsources and sinks are used as constraints. This yields a pre-industrial natural wetland sourcestrength of 163 Tg(CH4) yr−1, with an estimated 2σ uncertainty of 130–194 Tg(CH4) yr−1.The present natural wetland source may be∼10% smaller, owing to drainage and cultivationof wetland area since 1800 A.D. The simulated pole-to-pole concentration difference is foundto be rather insensitive to the assumed relative contributions of important pre-industrialsources and sinks, and, therefore, imposes only a limited constraint on the estimate of naturalwetland emissions. In contrast,δ13C-CH4 could provide robust constraints, but, unfortu-nately, at present reliable measurements are absent. Estimates of the historic development ofanthropogenic CH4 sources, in combination with our model calculations, can largely explainthe increase of methane mixing ratios during the 19th century. Results for the 20th centuryindicate that these historical emission inventories underestimate anthropogenic emissions byat least 10%. Simulations of pre-industrial and present dayisotopic ratios show that thegrowth of anthropogenic sources since 1800 A.D. may have increasedδ13C-CH4 by 3‰.

1Submitted for publication inJournal of Geophysical Research, with F. J. Dentener and J. Lelieveld as co-authors.

111

112 Pre-industrial methane

5.1 Introduction

Methane production by methanotrophic bacteria under anoxic conditions in natural wetlandsconstitutes the most important process by which methane is emitted to the atmosphere. Thequantification of its global source strength remains a scientific challenge after more than adecade of research. Since early estimates byMatthews and Fung[1987], andSeiler andConrad[1987], several studies based on the extrapolation of process related information (up-scaling) and inverse modeling reported global scale sourcestrengths in the range of 90–260Tg(CH4) yr−1, or ∼15–45% of its total source [Aselmann and Crutzen, 1989;Barlett andHarris, 1993;Chappellaz et al., 1993;Cao et al., 1996;Hein et al., 1997;Lelieveld et al.,1998;Walter, 1998]. A more accurate estimate of wetland emissions is of importance sincethis relatively large and uncertain term in the atmosphericmethane budget significantly limitsour understanding of this important chemically reactive greenhouse gas.

Methane releases from wetlands result from a complex interplay of many highly variablefactors, such as organic substrate supply, temperature, hydrological conditions and the com-petition between microbial production and oxidation. The up-scaling approach aims to quan-tify this on the basis of statistics of the geographical distribution of inundated area in combi-nation with a typical emission per unit area [Matthews and Fung, 1987;Seiler and Conrad,1987;Aselmann and Crutzen, 1989;Barlett and Harris, 1993;Chappellaz et al., 1993]. Sincemeasured emission factors represent a limited range of conditions only, this approach is notwell suited for the quantification of wetland emissions. Process-based modeling has the ad-vantage that emissions can be estimated as a function of the process controlling chemical andphysical conditions [Potter, 1997;Cao et al., 1996;Walter, 1998]. Many wetland ecosystems,however, have not yet been studied in sufficient detail to derive reliable parameterizations. Inaddition, the number of field experiments to validate these models are limited, which partic-ularly pertains to tropical wetlands. Therefore, to deriveglobal scale estimates of wetlandemissions by this method large generalizations are needed,which significantly increase theuncertainty of the results. Measurements of concentrations and isotopic ratios of atmosphericmethane can be utilized to constrain wetland emissions, as explored by inverse modeling[Brown, 1993;Kandlikar, 1997;Hein et al., 1997;Houweling et al., 1999a]. The addedvalue of inverse modeling is rather limited by the availability of measurements, as providedby the low-density observational networks such as the National Oceanic and AtmosphericAdministration (NOAA) cooperative air sampling network [Dlugokencky et al., 1994b] andthe Cooperative Atmospheric Data Integration Project [GLOBALVIEW-CH4, 1999]. Particu-larly in the tropics the number of measurements is insufficient and generally too far from thesources to provide significant constraints on wetland emission estimates [Houweling et al.,1999a].

In this study, an alternative method to constrain the natural wetland source is proposedand tested, i.e. by examining the pre-industrial methane budget. A study of the pre-industrialperiod has the advantage that, since anthropogenic sourceswere relatively unimportant, nat-ural wetlands dominated the global methane source. On the other hand, ice core analysesprovide the only observational evidence of the compositionof the pre-industrial atmosphereand, therefore, methane mixing ratios are known over Greenland and Antarctica only. Still,these measurements give a clear indication of the pre-industrial global mean methane con-centration, being∼40% of the present level [Craig and Chou, 1982;Stauffer et al., 1985;Battle, 1993;Chappellaz et al., 1997;Etheridge et al., 1998]. In addition, the differencebetween methane as measured in Arctic and Antarctic ice cores quantifies its pre-industrialnorth-south concentration gradient [Nakazawa et al., 1993;Chappellaz et al., 1997;Etheridge


et al., 1998]. Further, the13C/12C isotopic ratio of methane as measured in Greenland icecores [Craig et al., 1988a] provides a constraint on processes that fractionate these carbonisotopes differently, such as biomass burning (-25‰) and natural wetlands (-60‰) [Levin,1994].

To utilize these measurements, we simulate pre-industrialmethane by means of a three-dimensional (3-D) chemistry transport model (CTM). From ice core measurements it canbe inferred that the anthropogenically induced upward trend of atmospheric methane startedabout 1800 A.D. To minimize the contributions of anthropogenic sources to the overall emis-sions we focus on the preceding period (1500–1800 A.D.). Theglobal wetland source isdetermined by the amount needed to balance the steady state concentration at the observedlevel, given estimates of other pre-industrially active sources and sinks. Upper and lowerlimits of these processes are used to derive a range of wetland emissions that satisfy thepre-industrial methane budget.

This paper is organized as follows: In section 5.2 we introduce our 3-D CTM. Subse-quently, pre-industrial sources (section 5.3) and sinks (section 5.4) are described, includingtheir uncertainties. Section 5.5 presents the simulation of 13C and12C methane isotopes. Insection 5.6 natural wetland emissions are derived from the pre-industrial methane budget,and additional constraints as imposed by the measured interpolar difference andδ13C-CH4

are investigated. Further, in this section we show to what extent the observed exponentialincrease in methane between 1800 A.D. and present is explained by published inventories ofthe historic development of anthropogenic emissions and the natural emissions as derived inthis work. Finally, the impact of anthropogenic emissions on δ13C-CH4 is quantified. Theconclusions are presented in section 5.7.


Model simulations presented in this study have been carriedout using the global three dimen-sional (3-D) Tracer Model 3 (TM3) [Houweling et al., 1998, 1999b;Dentener et al., 1999;Lelieveld and Dentener, 1999]. The geographical resolution applied is 10◦ in the longitudi-nal and 7.5◦ in the latitudinal direction with 19 vertical levels. The vertical levels have beendefined as terrain following coordinates near the surface, pressure levels in the stratosphere,and a hybrid of the two in between. The horizontal and vertical transport of tracers is basedon six hourly mean meteorological fields, including wind, surface pressure, temperature, andhumidity, derived from European Centre for Medium-Range Weather Forecasts (ECMWF)re-analyses for the year 1993. The advective transport is calculated using the “slopes scheme”of Russell and Lerner[1981]. The sub-grid scale convective airmass fluxes are evaluated us-ing the cloud scheme ofTiedke[1989], including entrainment and detrainment in updraftsand downdrafts. Turbulent vertical transport is based on stability dependent vertical diffusion[Louis, 1979].

As a test of boundary layer mixing and regional scale transport, 222Rn simulations atvarious model resolutions have been compared with observations at continental and remotelocations [Dentener et al., 1999]. From this study it follows that measured and simulatedradon concentrations agree quite well; generally, deviations are<50%, and high correlationsof 0.7–0.8 between model and measurements were found. Largescale transport, in particularthe interhemispheric exchange rate, has been validated by acomparison of the simulatedand measured north-south gradient of SF6 [Houweling et al., 1999b]. From this test it wasconcluded that the TM3 simulated interhemispheric exchange time of 0.9 years is accurate


within 15%.Tropospheric chemistry is represented using a modified version of the Carbon Bond

Mechanism 4 [Gery et al., 1989;Houweling et al., 1998], accounting for CH4/CO and non-methane hydrocarbon (NMHC) chemistry, including isoprene. Chemical equations are in-tegrated using a Eulerian Backward Iterative (EBI) scheme,as formulated byHertel et al.[1993]. Emissions of photochemical tracers other than CH4 are based on the GEIA andEDGAR emission inventories [Olivier et al., 1999;Guenther et al., 1995;Yienger and Levy,1995;Benkovitz et al., 1996]. To represent the time evolution of all anthropogenic sourcessince industrialization the historic emission inventory by Aardenne et al.[1999] has beenused. Wet and dry deposition of soluble and reactive tracershas been described byHouwel-ing et al.[1998] andGanzeveld et al.[1997]. Photolysis rates are based on a highly efficientparameterization of the DISSORT radiative transfer code, and a parameterization proposedfor wave-length dependent cross sections and quantum yields accounting for multiple scat-tering by clouds [Krol and Van Weele, 1997;Landgraf and Crutzen, 1998]. O3 transportfrom the stratosphere into the upper level of the TM3 model domain is accounted for by con-straining the ozone concentration in the upper three model layers (<50hPa), based on ozoneconcentrations from the climatology byFortuin and Kelder[1998]. Stratospheric HNO3 istreated similarly, based on UARS-derived O3/HNO3 ratios [Kumer et al., 1997;Bailey et al.,1997].

The chemistry parameterization has been tested by comparing simulated concentrationsof various photochemically active compounds as O3, CO, and NOx with measurements[Houweling et al., 1998;Lelieveld and Dentener, 1999]. Results show that the model satis-factorily reproduces the observed global abundances of these tracers, generally within∼50%.To simulate methane, the distribution of OH radicals is of particular importance. Compar-isons of simulated and measured methyl chloroform (CH3CCl3) concentrations indicate thatOH is quite realistically reproduced by the model, althoughon average 11% lower than ob-served. To simulate CH4 destruction by OH, the applied OH-fields have been adjusted tomatch the observations of CH3CCl3 as further explained in section 5.4.

5.3 Pre-industrial sources

This section describes a pre-industrial scenario of CH4 emissions, including uncertaintyranges. For a summary of the estimated emissions we refer to Table 5.1. Generally, theavailable statistics for this period are rather scarce and incomplete. It should be realized thatthe representations of sources such as rice paddies and ruminants, are based on a few crudeassumptions only. We feel, however, that the limited amountof information available doesnot permit a higher level of detail. Uncertainty ranges of a number of processes can onlybe guessed for lack of global scale estimates of factors thatintroduce uncertainties, such aschanges in wildlife and landcover since pre-industrial times. In some cases such poorly quan-tified factors introduce uncertainties that all point in thesame direction, which explains whysome best guess estimates coincide with upper or lower bounds.

Pre-industrial emissions from domestic ruminants have been conservatively estimated at10 Tg(CH4) yr−1 by Subak[1994]. This is based on the estimated human population den-sity and the level of meat consumption in the 16th century. Asfarming techniques were lessadvanced than at present, the emission per ruminant may havebeen different also. This isrelated, in particular, to the amount and digestability of the food [Crutzen et al., 1986]. Toaccount for this, a global pre-industrial emission factor is applied that represents the present

5.3 Pre-industrial sources 115

Table 5.1.: Pre-industrial Sources of CH4 other than natural wetlands

Process Best Guess Lower Limit Upper Limit

Anthropogenic Sources

Rice agriculture 10 5 15Domestic ruminants 5 5 15Biomass burning 10 5 25Waste treatment 5 0 10

Natural Sources

Termites 20 10 30Wildfires 5 5 5Oceans 15 5 25Volcanoes 3.5 3.5 3.5Wild animals 15 15 15

All fluxes in Tg(CH4) yr−1.

day conditions in developing countries [Subak, 1994;Crutzen et al., 1986]. For wild animalsa global source strength of 15 Tg(CH4) yr−1 has been derived byChappellaz et al.[1993]based on estimated numbers of bisons and buffaloes.Subak[1994] accounts for similar buf-falo and bison emissions, pointing to difficulties in distinguishing pre-industrial domesticfrom wild animals. If we correct for such double countings a total ruminant source of 20Tg(CH4) yr−1 is obtained. It should be realized that this value represents a small numberof relatively important ruminant species only. It may well be that many but less ubiquitousspecies add up to a significant additional source. We tentatively assume that this has con-tributed at most 50%, leading to 30 Tg(CH4) yr−1 as an upper limit of ruminant emissions.

Emissions from pre-industrial rice agriculture have been estimated at 15 Tg(CH4) bySubak[1994]. The approximately ten-fold increase of populationsince industrialization [Du-rand, 1974] and the estimates of current rice paddy emissions of about 60 Tg(CH4) yr−1

[Olivier et al., 1999] suggest that this number is relatively high. However, the rice consump-tion per capita may have been larger since rice made up a larger portion of the Asian diet thantoday [Subak, 1994]. Also, the area needed for the production of rice was relatively largesince cultivation methods were less efficient.Subak[1994] used an emission factor of 0.58 gm−2 day−1 for irrigated rice, and for an average growing season of 136 days. This factor isrelatively high as compared with, for example, 0.35 g m−2 day−1 [Olivier et al., 1999] and 20g m−2 year−1 [IPCC, 1997] (see alsoDenier van der Gon[2000a]). In addition, rain-fed ricewas more commonly practiced than irrigated rice in those days, to which a 40–60% loweremission factor applies [IPCC, 1997]. It is expected that pre-industrial irrigated rice culti-vation still largely resembled the rain-fed conditions. Therefore we consider a pre-industrialrice paddy source of 10 Tg(CH4) yr−1 more realistic. By taking the estimate ofSubak[1994]as an upper limit and using a more conservative emission factor to derive a lower limit, a


range is obtained of 5–15 Tg(CH4) yr−1.Kammen and Marino[1993] estimated that the anthropogenic biomass burning emissions

of CH4 increased from 6.5 to 15 Tg(CH4) yr−1 from 1500 to 1800. These numbers accountfor biofuel combustion, mercantile activities and agricultural land management. Emissionestimates for the mid 19th century byKammen and Marino[1993] andLelieveld et al.[1998]of 9.8 Tg(CH4) yr−1 and 10 Tg(CH4) yr−1, respectively, suggest lower emissions before1800 A.D. On the other hand,Subak[1994] derived a much higher source of 30 Tg(CH4)yr−1 for 1500 A.D., which is explained by a relatively high estimate of emissions from sa-vannah fires. In this study 10 Tg(CH4) yr−1 has been used, well within the range of theseprevious estimates. In addition, emissions from wild fires have been assumed to account for5 Tg(CH4) yr−1 in accord withChappellaz et al.[1993]. By taking the estimate ofSubak[1994] as the upper limit and the estimate for 1500 A.D. byKammen and Marino[1993] pluswildfire emissions as a lower limit, we arrive at an uncertainty range of 10–30 Tg(CH4) yr−1

(anthropogenic plus natural sources).Stern and Kaufmann[1996] estimated that around 1800 A.D. landfills produced∼1% of

the present CH4 release using economic growth as an indicator of waste production. Adopt-ing 40 Tg(CH4) yr−1 [Lelieveld et al., 1998] as the contemporary landfill source strength,this yields an insignificant source of about 0.5 Tg for 1800 and 1.6 Tg for 1860. The latteris in reasonable agreement with the 2.6 Tg(CH4) yr−1 derived by Aardenne et al.[1999] for1860.Lelieveld et al.[1998] estimate the 1850 A.D. waste source at 16 Tg(CH4) yr−1 as thesum of emissions from landfills and waste water. Consideringthe range of these estimates atotal anthropogenic waste source of 5 Tg(CH4) yr−1 seems reasonable. As a lower limit wewill assume that pre-industrial waste emissions can be neglected. This is supported by the ar-gument that methane production from waste may have increased with the number of coveredlandfills, since anaerobic conditions, prevailing in such landfills, are needed for methanogen-esis. As opposed to this, methods for processing solid wasteand waste-water may have beenless effective. To represent the uncertainties introducedby these factors an uncertainty rangeof 0–10 Tg(CH4) yr−1 is applied.

Termite emissions are prescribed based on the work ofSanderson[1996], who usedbiomass densities and emission factors of different species, and identified habitats using highresolution vegetation databases. This yields a global source of 20 Tg(CH4) yr−1. Since mostof these emissions are located in the tropics, emissions areexpected to have remained rel-atively unaffected by anthropogenic activities until the exploitation of tropical rainforests,which became significant several decades ago. According to astudy byMartius et al.[1996]rainforest clearing and the conversion of primary forest topasture land did not significantlychange the amount of methane emitted by termites. Given alternative estimates of the presentday termite source, for example, 12 Tg(CH4) yr−1 by Khalil et al. [1990] and 14 Tg(CH4)yr−1 by Fraser et al.[1990], an uncertainty of 50% is considered, resulting in a range of10–30 Tg(CH4) yr−1.

Emissions from volcanic eruptions and continuously emitting volcanoes are estimatedat 3.5 Tg(CH4) yr−1 [Lacroix, 1993]; these sources have been geographically distributedaccording to volcanic sulphur emissions based on the work ofAndres and Kasgnoc[1998].Since a time scale of a few centuries is short in geological processes, this source is expectedto have remained constant. Considering the low overall amounts we do not account for arange.

A compilation of methane measurements in sea water have beenused to derive methanefluxes from open oceans and coastal zones of 3.6 and 6.1 Tg(CH4) yr−1, respectively [Lam-bert and Schmidt, 1993]. In addition, seepages through the sediments of continental shelfs

5.4 Pre-industrial sinks 117

contribute about 5 Tg(CH4) yr−1 [Hovland et al., 1993]. As a sum of these processes weadopt a total oceanic source of 15 Tg(CH4) yr−1. However,Bates et al.[1996] derived anopen ocean methane flux of only 0.4 Tg(CH4) yr−1 from measurements during ship cruisesover the Pacific, which would significantly reduce the present day estimates. In addition,incidental releases from oceanic reservoirs of methane hydrate and tectonic processes maycontribute, but are poorly quantified. As a result, the present day source is associated with asignificant uncertainty of 5–25 Tg(CH4) yr−1 [Lelieveld et al., 1998]. If the methane contentof the oceans has remained constant since industrialization, the pre-industrial oceanic emis-sion may have been larger than today owing to the increase of atmospheric mixing ratios.Nevertheless, we adopt the present day range for pre-industrial simulations, assuming thatchanges in the ocean-atmosphere exchange are within these uncertainties.

In the next section pre-industrial emissions of natural wetlands are derived from the pre-industrial methane budget. To relate these pre-industrialemissions to present-day emissions,an estimate is required of how pre-industrial and present-day natural wetlands compare. Al-though the documentation on the historic development of wetlands is limited, evidence existsindicating that wetlands have been progressively converted for agricultural, residential orother anthropogenic uses. In addition, peatland has been exploited for fuel. Notably in Eu-rope large fractions of peatland have been reclaimed. For example, in Finland this amountsto 6.4x1010 m2 or 46% of the former peatland area. However, in Canada and Russia, whichtogether account for 80% of the global peatland area, only 1%and 5% has been converted,respectively [Maltby and Immirzi, 1993].Maltby and Immirzi[1993] estimated that globally∼8% of the former peatland area has been drained or otherwise altered. Generally, wetland isclassified as peatland if the input of organic material exceeds the rate of fermentation, whichapplies to approximately 60% of the present wetland area [Lappalainen, 1996]. Chappellazet al. [1993] estimate that the pre-industrial total wetland areawas∼19% larger than today,with the largest wetland losses in the northern mid-latitudes (30◦–60◦N). Since the averagemethane emission per square meter of wetland is larger in thetropics than at temperate andhigh latitudes, the methane emission reduction is smaller than the reduction in wetland area.It should be noted that the 2 estimates of wetland conversion[Maltby and Immirzi, 1993;Chappellaz et al., 1993] do not seem to agree, since this would require that non-peat accu-mulating wetlands are converted preferentially, which is difficult to explain.

Except areal changes, changes in temperature and hydrological conditions as a result ofthe enhanced greenhouse effect may have influenced wetland emissions [Martikainen, 1996].In addition, in relatively polluted regions N and P fertilization may be of importance, sincethe primary production of wetland ecosystems is generally limited by these nutrients [Maltbyand Proctor, 1996]. Temperature increases and increased primary productivity may haveenhanced the production of methane, although to our knowledge these effects have not beenquantified yet. As the combined effect of the factors discussed above we tentatively estimatethat wetland emissions have decreased by∼10% since the beginning of industrialization.

5.4 Pre-industrial sinks

Model simulated present-day OH radical concentrations have been evaluated by comparingsimulated and measured concentrations and trends of methylchloroformHouweling et al.[1999b]. It follows that the model underestimates the annual and global mean hydroxyl rad-ical concentration in the troposphere by 11%. Since there are many possible reasons for thediscrepancy in OH, we simply rescaled the global OH concentrations to match the observed


Table 5.2.: Pre-industrial Sinks of CH4

Process Best Guess Lower Limit Upper Limit

Soil oxidation 12 12 30Stratospheric oxidation 16 12 16Tropospheric oxidation τ=7.2 τ=8 τ=6.5

All fluxes are in Tg(CH4) yr−1, and tropospheric turnover timesτ arein yr.

trend of CH3CCl3 for the period 1980–1992. A simulation for 1860 A.D. indicates a 7.5%higher average hydroxyl radical concentration, as compared with simulations of the contem-porary atmosphere. In this “early industrialization” simulation we apply pre-industrial emis-sions of ozone precursor gases [Aardenne et al., 1999], and assume a pre-ozone hole strato-sphere, while other boundary conditions were taken equal. Photochemical computations havebeen started at 1860 A.D., because from this year on historicemissions were available fromthe database byAardenne et al.[1999]. It is assumed that in 1860 anthropogenic emissionswere still too small to significantly affect global OH concentrations, which is supported bya simulated decrease of OH from 1860–1900 by less than 0.5%. Considering the lack of amethod to evaluate pre-industrial OH concentrations we assume that the methyl chloroformderived scaling factor also applies to pre-industrial simulations.

A 7.5% decrease in OH from pre-industrial times to present compares well with previousestimates, which vary between a 5% increase to a 20% decrease[Martinerie et al., 1995;Crutzen and Bruhl, 1993;Pinto and Khalil, 1991;Thompson et al., 1993;Wang and Jacob,1998]. Most studies point to a reduction of OH due to industrialization, related to relativelystrong increases in CH4 and CO. This is, however, compensated to some extent by increasedNOx emissions and depletion of stratospheric ozone, which haveresulted in a stabilizationor a possible increase of OH in the last two decades [Prinn et al., 1995;Krol et al., 1998].Since OH is largely determined by key compounds such as O3, CO, and NOx, measurementsof their pre-industrial concentrations help to assess how realistic OH is represented. COmeasurements in ice cores samples indicate pre-industriallevels of 91 and 57 nmol mol−1 forGreenland and the South Pole [Haan et al., 1996], respectively, whereas the correspondingannual means derived from the model amount to 67 and 30 nmol mol−1 only. Pre-industrialO3 concentrations have been reconstructed from measurementsby Schonbein’s method inthe second half of the 19th century [Pavelin et al., 1999;Sandroni et al., 1992;Anfossi et al.,1991]. These measurements indicate that the model may overestimate surface O3 by up toa factor 2. Similar discrepancies between model and measurements have been reported forO3 [Wang and Jacob, 1998;Roelofs et al., 1997;Lelieveld and Dentener, 1999] and CO[Wang and Jacob, 1998;Haan et al., 1996]. At present, it is not clear to what extent thesediscrepancies are caused by models errors or experimental uncertainties. Here, uncertaintiesassociated with pre-industrial photochemistry simulations are represented by a±10% errorin global mean OH (see Table 5.2).

To represent atmospheric methane removal by soils we adopt results of a process model

5.5 Isotopic ratios 119

by Ridgwell et al.[1999], who parameterized microbial oxidation in soils as afunction ofmicrobial activity and soil diffusivity. As a first approximation, we assume the magnitude ofthis sink to be linearly dependent on the atmospheric methane mixing ratio, which impliesa reduction by a factor 2.5 of the pre-industrial sink as compared with present. In additionto the atmospheric concentration, however, biological activity is a rate determining factor insoil oxidation, so that a less than linear relationship is probably more realistic. Moreover,in fertilized soils methane oxidation is less efficient. As aconsequence, land use changes,for example, the conversion of forests into cultivated lands, may have reduced soil oxida-tion. Therefore, as an upper limit we assume that the soil sink remained constant sincepre-industrialization, by adopting the estimated presentday sink of 30 Tg(CH4) yr−1.

Turnover times of stratospheric methane oxidation for present-day simulations are basedon 2-D model calculations byBruhl and Crutzen[1993]. As a first approximation, analo-gously to the treatment of the soil sink, it has been assumed that industrialization did notaffect the lifetime of stratospheric methane. Again, this leads to an approximately linear re-lationship between the atmospheric methane abundance and stratospheric oxidation, yieldinga reduction by a factor 2.5 of the pre-industrial stratospheric sink as compared with present.At present, the contribution of chlorine radicals to stratospheric methane oxidation is esti-mated at∼15% (G. M. J. Velders, personal communication, 1999). The stratospheric chlo-rine loading, however, is largely related to anthropogenically produced CFCs. As a result,it can be assumed that, pre-industrially, the contributionof stratospheric chlorine radicalsto methane oxidation was negligible. Further, methane oxidation is the main source of wa-ter vapor in the stratosphere. Therefore, increased methane mixing ratios are expected tohave increased stratospheric water vapor which in turn enhanced the production of OH rad-icals. Since many other chemical feedbacks play a role in thestratosphere, the net changeof stratospheric methane oxidation since pre-industrial times is difficult to estimate. Herewe tentatively assume that, as an upper limit, the stratospheric lifetime of methane was 50%larger than at present.

5.5 Isotopic ratios

Processes that remove or produce methane can be characterized by a process-specific frac-tionation of methane isotopes, owing to isotope dependent rates of the chemical reactionsinvolved. As a result, isotopic ratios of atmospheric methane reflect the relative contribu-tions of these processes. To investigate to what extent measured isotopic ratios constrainpre-industrial sources of methane, its13C/12C ratio has been simulated.

13C/12C ratios are expressed inδ13C-CH4, defined as,

δ13C−CH4 = (13CH4/

12CH4

[13C/12C]PDB−1)×1000, (5.1)

representing the per mil deviation from the Chicago PeeDee Belemnite (PDB) standard([13C/12C]PDB = 0.0112372) [Craig, 1957]. The products of biomass burning essentiallyretain the isotopic signature of the original material (δ13C-CH4 = -20 to -40‰). Anaerobicmicrobial fermentation preferentially produces the lightisotope. However, under oxic con-ditions bacteria also consume this isotope more efficiently. As a result, the net13C/12C ratioof methane produced by sources such as wetlands and landfillsis dependent on the balancebetween methanogenesis and methane oxidation leading toδ13C-CH4 = -50 to -70 ‰. To


Table 5.3.: Isotopic Fractionation

Process δ 13C-CH4 Referencea

Sources

Rice agriculture -64 PB97Domestic ruminants -62 L94Biomass burning -25 L94Waste treatment -55 PB98Natural wetlands -60 L94Termites -57 G96Oceans -40 G96Volcanos -40 G96b

Wild animals -62 L94c

Sinks

Hydroxyl radical 1.0054 C90Stratosphere 1.012 B95Soil oxidation 1.022 T94

All sources in ‰ (relative to the PDB standard), all sinks in KIE(k12/k13).

aPB97,Bergamaschi[1997]; L94,Levin[1994]; PB98,Bergamaschiet al.[1998]; G96,Gupta et al.[1996]; C90,Cantrell et al.[1990]; B95,Brenninkmeijer et al.[1995]; T94,Tyler et al.[1994].

bas fossil gas.cas domestic ruminants.

thermogenically produced fossil CH4 aδ13C-CH4 of about -40‰ applies [Levin, 1994;Quayet al., 1999].

All atmospheric sinks preferentially remove12CH4 and, as a consequence, increaseδ13C-CH4. In particular the oxidation of CH4 by Cl radicals has a relatively large kinetic isotopeeffect (KIE) of 1.066 (k12/k13) at 297 K [Saueressig et al., 1995], which leads to a relatively13C enriched stratosphere as compared with the troposphere. The competing reaction withO(1D) is almost isotope independent (KIE = 1.001) [Davidson et al., 1987]. As the sum ofthese processes a net fractionation of 1.012 in the stratosphere was estimated byBrenninkmei-jer et al.[1995]. In pre-industrial simulations in which methane oxidation by chlorine radicalshas been neglected (scenario WLL in section 5.6), a stratospheric fractionation of 1.0054 isused, since in the absence of Cl radicals the isotopic fractionation is dominated by that of theOH radical reaction. Table 5.3 lists all isotopic fractionations used to simulateδ13C-CH4.


Table 5.4.: Pre-industrial Methane Budget and Inferred Natural Wetland Emissions

Process STa WLL WUL

Natural wetlands 163 83 240All other sources 89 143 53All sinks -252 -226 -293

—- —- —-Trend 0 0 0

All fluxes in Tg(CH4) yr−1.aScenarios: ST is Standard, WLL is lower limit of wetland emissions, and WUL is upper

limit of wetland emissions.

5.6 Results and discussion

5.6.1 Pre-industrial methane

Model simulations have been performed using sources and sinks of methane as defined bythe scenarios ST (Standard), WLL (Wetland Lower Limit), andWUL (Wetland Upper Limit).For the natural wetland source strength we adopt the amount needed to obtain agreementbetween the simulated steady state global mean methane concentration and the global meanconcentration as derived from ice core measurements, i.e. the amount needed to close theglobal CH4 budget. Scenario ST is defined by best guess source and sinks estimates (seeTables 5.1 and 5.2), leading to a best guess estimate of the pre-industrial wetland sourcestrength. WLL and WUL refer to scenarios that lead to lower and upper limits of the naturalwetland source strength, respectively. For example, the WLL scenario shows lower boundsfor all sinks, and upper bounds for sources other than natural wetlands since this allowsclosure of the methane budget with the smallest wetland source (vice versa for WUL). Toreach steady state the model was run for 10 years. Deviationsfrom steady state in the 10thyear are computed using an exponential curve that is fitted toconcentrations as computedfor years 6–10. These functions converge in time to steady state concentrations and isotopicratios.

The observed global mean concentration is obtained by averaging all ice core measure-ments reported byEtheridge et al.[1998] for the period 1500–1800. Measurements weretaken from ice cores drilled at Greenland (Summit, 72◦34’N, 37◦37’W, 3200m) and Antarc-tica (Law Dome, 66◦44’S, 112◦50E, 1390m). The accuracy of the ice core measurements, asdetermined by sample handling and analyses, is claimed to be5 nmol mol−1 (1σ) [Etheridgeet al., 1998].Etheridge et al.[1998] defined the global mean as the concentration measuredat Antarctica plus 37% of the interpolar difference. For a fair comparison between model andmeasurements the same definition of “global mean” is appliedin the model.

Table 5.4 lists the global methane budget and the inferred wetland source strength foreach scenario. It shows that a natural wetland source of 163 Tg(CH4) yr−1 is in optimalagreement with our best guess source and sink scenario (ST).The WLL and WUL scenariosshow that natural wetland source strengths outside the range of 83–240 Tg(CH4) yr−1 are


Table 5.5.: Comparison of Measured and Simulated Pre-industrial Methane Concentrations

Measured ST BB+ WLL WUL

Global averagea(nmol mol−1) 705±5 704 704 705 705Interpolar difference (nmol mol−1) 45±10 46 46 52 43δ13C-CH4 (‰) -49.6±0.7b -50.5 -49.1 -46.9 -51.7

Measurements are taken fromEtheridge et al.[1998] (global mean and interpolar differ-ence) andCraig et al. [1988a] (isotopic ratio). Uncertainties represent 1σ intervals. Allconcentrations are in nmol mol−1.

aThe global average is defined as the concentration at Antarctica plus 37% of the interpolardifference.

bUncertainty as defined byCraig et al.[1988a] (see text).

highly unlikely. These scenarios can be regarded as extremesince the errors of all processeswould have to point in the same direction, which seems quite unrealistic. If it is assumed thatall errors are uncorrelated and Gaussian distributed, thenthe variance of the derived wetlandsource is represented by the sum of the variances of all otherbudget terms. Furthermore, if weassume that the uncertainty ranges, as estimated for the individual processes, represent 95%confidence intervals (∼2σ), then the±2σ uncertainty for natural wetlands amounts to 130–194 TgCH4 yr−1. We emphasize that these figures represent pre-industrial wetland emissions.As discussed in section 5.3, the present-day wetland sourcestrength is expected to be lowerby approximately 10%. This estimate compares well with results reported byLelieveld et al.[1998], who derived a wetland source of 145± 30 Tg(CH4) yr−1 for 1850 A.D. Generally,the source scenario as used in that study is consistent with our ST scenario.

Table 5.5 shows comparisons of the measured and simulated global mean, the interpolardifference, andδ13C-CH4. The scenarios ST, WLL, and WUL all show good correspon-dences between model and measurement derived global means (Table 5.5), which merelyshows that the global methane budget is well balanced by the applied wetland emissions.The interpolar difference has been derived, like the globalmean concentration, from mea-surements presented byEtheridge et al.[1998] averaged over the years 1500–1800 A.D. TheST scenario yields an interpolar difference of 46 nmol mol−1, in good agreement with themeasurements (Table 5.5 and Figure 5.1a). The interpolar differences, as computed for theWLL and WUL scenarios, deviate from the ST scenario by a few nmol mol−1 only. The signsof these differences are explained by the North–South asymmetries of the sources, pointingto larger interpolar differences as the contribution of natural wetlands decreases. Because allthe simulated interpolar differences remain within the measurement uncertainty, they do notprovide further constraints on natural wetland emission estimates.

Isotopic ratios were measured byCraig et al.[1988a] in samples of Greenland ice cores(Dye 3, 65.2◦N, 43.8◦W, 2600m). Theδ13C-CH4 value of -49.6‰ has been corrected fora gravitational separation of 0.3‰Craig et al. [1988b]. Compared with measurements ofcontemporaryδ13C-CH4 [Quay et al., 1991;Lassay et al., 1993;Lowe et al., 1997;Quayet al., 1999] this indicates an increase ofδ13C-CH4 by ∼2‰ since industrialization. After


a

CH4 1800 (nmol/mol)

685 690

690

690700

700

700710

710710720 720 72

0

720

750

750750

750

b

δC13 1800 (per mil)

-50.50

-50.

50

-50.50

-50.

40

-50.

40

-50.

40

-50.

30

-50.30

-50.30

-50.

30

-50.30

-50.

20

-50.

20-50.20

-50.10

-50.10

-50.10

Figure 5.1: (a) Annual mean pre-industrial surface mixing ratios of methane and (b)δ13C-CH4.

the publication ofCraig et al. [1988a] it was recognized that different diffusivities of CH4

isotopes in firn influence the observed isotopic ratios. Thisproblem has not been solvedto date, which explains why noδ13C-CH4 measurements have been reported since. For icesamples that represent pre-industrial times this error maybe limited since, in the absence ofa significant trend of atmospheric methane, the “head-space” air was largely in equilibriumwith the firn. Because of a problem with the dating of the samples published byCraig et al.[1988a], however, only the deepest samples (at 178.5 and 188.0m depth) can be consideredas pre-industrial (J. Chappelaz, personal communication,1999). To comply with this, wehave only used the results of these two measurements. These measurement results shouldbe treated with care, however, since these experimental problems suggest that the relateduncertainties may be significantly larger than the 1σ error of 0.7‰ as reported byCraig et al.[1988a].

Table 5.5 shows that the standard model simulation underestimatesδ13C-CH4 at Green-land by 0.9‰ (see also Figure 5.1b). This suggests that, in the model, the emissions of


“heavy” sources, such as biomass burning, are underestimated, and “light” sources, suchas wetlands and ruminants, are overestimated as compared with the real atmosphere. Anadditional simulation was performed to test the sensitivity of δ13C-CH4 to the relative con-tributions of biomass burning and natural wetland emissions, by increasing and reducingthese sources by 10 Tg(CH4), respectively (scenario BB+ in Table 5.5). Obviously, we couldhave tested other combinations of13C-depleted and enriched processes. The isotopic frac-tionations in Table 5.3 suggest, however, thatδ13C-CH4 is especially sensitive to changesin these particular sources. In addition, their geographical distributions show similar north-south asymmetries since both processes are relatively strongly represented in the tropics.Therefore, the interpolar difference is expected to remainat the realistic value of scenario ST.The results (Table 5.5, scenario BB+) show that an even smaller source adjustment explainsthe discrepancy between the observed and ST-scenario simulatedδ13C-CH4. Indeed, the in-terpolar difference is only weakly affected by the source adjustments. This test shows thatδ13C-CH4 is quite sensitive to the relative contribution of important pre-industrial sources ofmethane. It is emphasized, however, that unless more reliable historicδ13C-CH4 measure-ments become available little can be concluded from these computations. It does show that,potentially, such measurements can make a very useful contribution.

5.6.2 Methane increase during industrialization

The next question is to what extent the exponential growth ofmethane since industrializationis explained by our pre-industrial ST scenario supplemented by estimates of the historicaldevelopment of anthropogenic sources. For this purpose methane simulations have been per-formed spanning a few centuries, using a box model with prescribed emissions and turnovertimes as derived from the global 3-D CTM. Further, a factor isderived from the 3-D CTMto relate the domain integrated methane burden to the globalmean concentration. By using abox model, implicitly the assumption is made that the changeof the global methane distribu-tion over the simulated period can be approximated by scaling the initial distribution with theglobal mean trend. This assumption is verified by comparing “snapshots” of the box modelsimulation with 3-D CTM results for certain years (see Figure 5.2). These comparisons indi-cate that these errors remain relatively small (<15 nmol mol−1).

The box model simulations were carried out for the period 1500–1990 A. D. (see Fig-ure 5.2). Anthropogenic emission scenarios were taken fromStern and Kaufmann[1996]andAardenne et al.[1999]. All natural sources were kept at their pre-industrial (ST) levelthroughout these simulations. As discussed earlier, our pre-industrial scenario accurately re-produces the average of the measurements over 1500–1800 A.D. Since variations over thisperiod are not represented by this scenario methane is underestimated by 35 nmol mol−1 in1800 A.D. From 1800 to 1860 anthropogenic emissions are linearly interpolated between thepre-industrial level and those estimated byAardenne et al.[1999] andStern and Kaufmann[1996], which both start at 1860 A. D. From 1860 to the end of the 19th century model simu-lations and measurements correspond well, indicating thatthe combined estimates of naturaland anthropogenic sources largely explain the observed methane concentrations in the 19thcentury. In the 20th century both theAardenne et al.[1999] andStern and Kaufmann[1996]emission scenarios lead to increasingly underestimated concentrations as compared with theice core measurements. After 1950, also the differences between these simulations increase.The latter is largely explained by differences in estimatesof rice agriculture emissions (100Tg(CH4) yr−1 by Stern and Kaufmann[1996], and 60 Tg(CH4) yr−1 by Aardenne et al.[1999] for 1990) and biomass burning (38 Tg(CH4) yr−1 by Stern and Kaufmann[1996], and


1500 1600 1700 1800 1900 2000year (A.D.)

600

800

1000

1200

1400

1600

1800

CH

4 (p

pbv)

ice coresAardenne et al. 1999Stern & Kaufmann 1996

Figure 5.2: Comparison of measured and model simulated global mean CH4 mixing ratiossince pre-industrial times. The global mean concentrationis defined as the concentration atAntarctica plus 37% of the interpolar difference. Error bars represent differences between thebox model and the global 3-D CTM (in 1700, 1900, and 1950). Thebox model was calibratedto the 3-D CTM using results of 1993.

24 Tg(CH4) yr−1 by Aardenne et al.[1999] for 1990). If all differences between model sim-ulations and measurements are attributed to anthropogenicsources, this comparison pointsto an underestimate of theStern and Kaufmann[1996] sources by∼10% since 1900 A. D.(equivalent to 32 TgCH4 yr−1 in 1980), and∼10% from 1900–1950 and∼15% since 1950for Aardenne et al.[1999] (equivalent to 45 TgCH4 yr−1 in 1980). The uncertainties in theCH3CCl3 test and in the OH change since preindustrial times may, however, largely explainthe emission underestimates as computed forStern and Kaufmann[1996] and forAardenneet al. [1999] until 1980. If a decrease of natural wetland emissions during the 19th and 20thcentury is accounted for (section 5.3) these underestimates become more significant.

5.6.3 δ13C-CH4 during industrialization

To investigate the effect of increased anthropogenic methane sources onδ13C-CH4 a con-temporary 3-D CTM simulation has been carried out using the same emissions as applied byHouweling et al.[1999b]. Computed surface concentrations and isotopic ratios are shownin Figure 5.3. Comparisons of simulated methane mixing ratios and NOAA measurements


a

CH4 1993 (nmol/mol)

1665 16651665

1665

1675 1675

1700

1700

1750

17501800

1800 180018501850

1900

1900

b

δC13 1993 (per mil)

-47.60-47.40

-47.

40

-47.

20

-47.20-47.20

-47.00

-46.

90

-46.90

-46.90

-46.80

-46.80-46.80

Figure 5.3: (a) Annual mean surface mixing ratios of methane and (b)δ13C-CH4 for 1993A. D.

at in-situ and flask sampling stations have been presented inHouweling et al.[1999b]. Ex-cept for a 40 nmol mol−1 or 30% overestimate of the interpolar difference, the results indi-cate that methane mixing ratios are quite realistically reproduced by the model.Quay et al.[1999] derived a global meanδ13C-CH4 of -47.3‰, from the area weighted average of iso-topic ratios as measured at Baring Head (41◦S), Cape Grim (41◦S), Samoa (14◦S), MaunaLoa (20◦N), Olympic Peninsula (48◦N), and Barrow (71◦N) over 1988–1995. By using thesame definition of “global mean”, we simulate a slightly higherδ13C-CH4 of -47.0‰. Whencomparing these isotopic ratios it should be realized, however, that a steady state is assumedin the model, whereas the present atmosphere deviates from steady state by approximately-0.6‰(K. R. Lassay, personal communication, 1999). Therefore the model underestimatesδ13C-CH4 by∼0.3‰. The computed difference between Barrow (-47.2‰) and New Zealand(-46.8‰) of 0.4‰ is in reasonable agreement with 0.5‰ reported byQuay et al.[1999].These results and theδ13C-CH4 of -50.5‰ obtained for the standard pre-industrial scenarioindicate that the increase of anthropogenic sources may have changedδ13C-CH4 by as much

5.7 Conclusions 127

as 3‰.

5.7 Conclusions

This study presents pre-industrial simulations of methane, with the aim to estimate the sourcestrength of natural wetlands and reduce the associated uncertainty. Methane concentrationsandδ13C-CH4 as derived from ice cores, and available information about other sources andsinks of methane are used as constraints. This points to a pre-industrial wetland source of163 Tg(CH4) yr−1. From uncertainty estimates, worse case scenarios have been derived thatprovide lower and upper limits to the emissions from naturalwetlands. This results in a 2σ(95% confidence) uncertainty range of 130–194 TgCH4 yr−1. Cultivation and drainage of thewetlands since industrialization may have reduced the natural wetland emissions by∼10%.

The measured interpolar difference, as deduced from ice cores drilled at Law Dome(Antarctica) and Summit (Greenland), imposes a relativelyweak constraint on the pre-industrial methane budget. This results from the relatively small differences in latitudinaldistribution of important pre-industrial sources. A comparison of measured and simulatedδ13C-CH4 indicates that the model underestimatesδ13C-CH4 by∼0.9‰. This difference canbe compensated for by an assumed shift of wetland emissions to biomass burning by<10Tg(CH4) yr−1, showing the high sensitivity ofδ13C-CH4 to such emission changes. Unfor-tunately, the ice core derivedδ13C-CH4 values, as reported to date, suffer from errors causedby the differential diffusion of isotopes in firn, and are therefore associated with large uncer-tainties. Our model results indicate that future efforts tomeasure methane isotopes in Arcticand Antarctic ice cores will make a highly valuable contribution to our understanding of thepre-industrial methane budget.

Box model simulations show that the observed historical increase of methane mixing ra-tios can be accurately reproduced until the 20th century, using the natural emission estimatesobtained in this study, combined with estimates of the historical increase of anthropogenicsources byStern and Kaufmann[1996] andAardenne et al.[1999]. For the 20th centurytheStern and Kaufmann[1996] inventory leads to the closest agreement between model andmeasurements. Nevertheless, to obtain agreement a 10% increase of anthropogenic sources isneeded (equivalent to 32 Tg(CH4) yr−1 for 1980). If a decrease of natural wetland emissionsis accounted for, an even larger increase of the anthropogenic source estimates is needed. Al-though the uncertainties associated with the representation of OH could also largely explainthese differences.

Simulations of pre-industrial and present day methane indicate an increase ofδ13C-CH4

by ∼3‰ during the industrialization. As a consequence, the observed increase by∼2‰ canbe explained by the growth of anthropogenic sources.

Acknowledgments

We thank J. Chappellaz (Labaratoire de Glaciologie et Geophysique de l’Environnement,CNRS) for kindly providing us an update on the current statusof the isotopic analysis ofice cores. Further, we’d like to acknowledge useful discussions with H. A. C. Denier vander Gon (Department of Soil Science and Geology, WageningenUniversity) and Peter vanBodegom (Department of Theoretical Production Ecology, Wageningen University). Thiswork has been supported by the Dutch Global Change program, NOP project 951202.

Chapter 6

General discussion and futureperspectives

The aim of this thesis is to improve methods and boundary conditions to optimally constrainthe sources and sinks of CH4. Partly, this could be achieved by updating the a priori assump-tions to the current state of understanding of CH4 sources and sinks, for example, by usingthe EDGAR inventory of anthropogenic sources, and by adopting results of process modelsof natural wetland emissions, and soil oxidation. Further,extensions of the NOAA measure-ment network allowed a larger number of stations to be taken into account as compared withprevious studies, including the Pacific and South China Sea ship cruises. This discussion,however, will focus on new insights that result from the methods and parameterizations de-veloped in framework of this thesis. The major conclusions of this thesis are printed in italics.

In Chapter 3 it was concluded thatto realistically simulate CH4 concentrations, exceptfor its surface sources and sinks, accurate representations of tracer transport and atmosphericchemistry are of critical importance. The same is true for the quantification of sources andsinks by inverse modeling, since errors in the representation of transport and chemistry arecompensated by flux adjustments. As shown in Chapter 3, the differences between a poste-riori and a priori emissions could largely be explained by these model errors. To quantifysuch errors, accurate tools to validate transport and chemistry are needed. Particularly, im-proved techniques for validating the global OH distribution would improve the results of CH4inversions.

In Chapter 4 the representation of transport in the model hasbeen improved by the in-crease of horizontal and vertical resolution. Further, theapplication of subsequent years ofreanalyzed meteorological fields in TM3 compared with the repeated use of a single year inTM2 is considered an important step forward. Indeed, the agreement between model simu-lated and measured concentrations of SF6 and CFC-11 improved in TM3 as compared withTM2, also leading to an improved simulated latitudinal concentration gradient of CH4. Stud-ies byDentener et al.[1999];Velthoven and Kelder[1996] also showed that vertical mixingin the model is improved by the increase of model resolution.In Chapter 3, horizontal andvertical diffusion rates have been adjusted to determine the potential impact of a model un-derestimated interhemispheric exchange rate. It should beemphasized that this method isused as a diagnostic tool only, but it was not intended to improve the model performance.

129

130 General discussion and future perspectives

Although the simulated interhemispheric exchange rate canbe improved by the adjustmentof diffusion rates, other aspects of transport that can lesswell be tested may deteriorate.

To improve chemistry, in particular the model simulated OH radical distribution, a NMHCchemistry scheme has been modified for use in global models asdescribed in Chapter 2.Comparisons with other, more extensive, schemes show that this condensed representationof NMHC chemistry performs well for a broad range of chemicalconditions.Simulated O3,CO, NOx, and OH radical concentrations with and without this new chemistry parameteriza-tion clearly indicate the importance of NMHC. Comparisons of simulated and measured O3

concentrations indicate that the representation of tropospheric photochemistry is improved byaccounting for NMHC chemistry.Comparisons of model simulated hydrocarbon concentra-tions, and HNO3 depositions with measurements suggest, however, that the NMHC inducedOH reduction over the continents is overestimated by the model. Notably large discrepanciesare found over tropical rainforests, suggesting that the highly simplified treatment of bio-genic hydrocarbons not yet adequately represents these conditions. In fact, the importanceof biogenic NMHC, as pointed out in Chapter 2, justifies a morerealistic representation ofisoprene at the cost of more CPU. In addition, the present state of knowledge would allow arepresentation of mono-terpenes, the second most important class of biogenic hydrocarbons.It is emphasized, however, that the model overestimated isoprene concentrations over Ama-zonia are probably not explained by chemistry alone, but mayhave important contributionsof subgrid scale processes, such as turbulent mixing. Further, the chemistry of acetone is notrepresented by the scheme, although, according to recent studies, this compound may signif-icantly contribute to HOx production in the free troposphere. The global CBM-4 chemistryscheme can easily be extended, however, to account for acetone sources and sinks, includingits production by NMHC oxidation.

Chapter 3 emphasizes thatthe results of inverse modeling can be improved by better es-timates of a priori uncertainties and spatial and temporal correlations of these uncertainties.The method that is used to derive “local” uncertainties fromglobal uncertainties was mainlychosen for lack of consistent estimates of uncertainties onsmaller than global scales. It wasshown that if sources are assumed to be fully uncorrelated, for lack of a better alternative,this approach leads to unrealistically high 2σ uncertainties (order 600%). Unfortunately, thehigh level of detail of current up-scaling derived emissioninventories is at odds with thedocumentation of the uncertainties. Ideally, each emission inventory would be accompaniedby an uncertainty analysis, comprising estimates of the uncertainty per grid and correlationsamong them. Although this may be feasible only on a very coarse level since in practice dataare lacking, relatively little is needed to substantially improve the treatment of uncertaintiesin our CH4 inversion. The global scale application of process models may provide tools toconsistently quantify uncertainties of emissions and their correlations in space and time.

To improve the inverse modeling method as compared with previous inverse modelingstudies the adjoint technique has been used. An important advantage of this approach overthe “large region” approach as applied, for example, byHein et al.[1997] is that the uncer-tainties in the geographic and temporal distributions of sources can be represented better. As aconsequence, the individual sources are less well constrained, which leads to relatively large,but more realistic, uncertainties as compared with resultsof previous studies. In addition, theadjoint technique helps to reduce biases caused by inhomogeneous sampling. Calculations byKaminski et al.[1999c] for a CO2 inversion indicate that these biases are an important pointof concern. This outcome is, however, dependent on the distribution of sources, sinks, and

General discussion and future perspectives 131

measurement stations, which is different for CO2 and CH4. Therefore, a CH4 specific analy-sis is needed to assess the impact of these biases on the results of a CH4 inversion, which hasnot been performed at present. From such an analysis it couldalso be derived what spatialand temporal resolutions are required to reduce these biases to acceptable levels.

The approach presented in Chapter 3 may appear somewhat inconsistent, because the res-olution of surface fluxes is maximized, while for the OH sink only a single scaling factor isestimated for the global OH distribution. This is justified by the argument that the constraintson the distribution of OH, as imposed by atmospheric chemistry, have poorly quantified un-certainties. In fact, the present treatment of the OH sink isequivalent to assuming that theseuncertainties can be neglected. This approach yields realistic CH4 concentrations if the aposteriori CH4 emissions are applied in photochemistry simulations. Thisis desirable froman atmospheric chemistry point of view, but does not necessarily yield the most accurate CH4emission estimates.

The aim of Chapter 4 is to improve the representation of measurements in CH4 inversions.The pre-treatment of experimental data, as applied in 3, effectively filters out variabilitiesthat cannot be reproduced by the model. This includes variabilities ranging from the timescale of days, related to synoptic events, to interannual variabilities, related to year-to-yearchanges in the general circulation and the sources. Probably, correlations exist between short-term variabilities of transport and measured variabilities of CH4. If such correlations areresolved by the model and utilized by the inversion technique this would result in strongerconstraints on sources and sinks. Further, by studying interannual variabilities of CH4 wemay gain understanding of the variability of sources, whichin turn would help to understandpast fluctuations of the CH4 growth rate. To do this, quasi-stationarity should no longer beassumed.

In the quasi-stationary state set-up, the filtering of short-term and interannual variabilitywas needed for a fair comparison between model and measurements. For a different in-version set-up, a reevaluation is needed of the most efficient way of comparing model andmeasurements. Therefore, we need to know which aspects of the observed variability canbe resolved by the model. As shown in Chapter 4, this largely depends on the geographiclocation of the station that is examined. At the remote station Mauna Loa, the TM3 modelresolves a substantial part of the variability on the time scale of several days (R=0.6–0.8), incontrast to the continental station Point Barrow (R=0–0.4). Further, it was concluded that aweekly sampling interval, as applied at most stations, is insufficient to represent the monthlyconcentration distribution at coastal and continental sites. The agreement between modelsimulations and measurements improved at a number of locations if the model was sampledat the same times as the measurements, and by comparing averages of corresponding sets ofsamples.Also, in Chapter 4 it was shown howrepresentation errors related to wind directionselection in the observations, and unresolved local sources can be reduced.Here, the defi-nition of an appropriate threshold concentration to selectsamples is a crucial factor. Modelresults suggest that measurements of SF6 or 222Rn can be used to establish or verify thesethresholds. As a next step, these methods for reducing representation errors can be applied toinverse modeling, either to select measurements or to increase the observational uncertaintiesto account for representation errors.

As mentioned in section 1.6, a major goal of this thesis is to improve the estimates ofrice field emissions. In this study, inverse modeling has, however, not provided significantconstraints on scales that are small enough to attribute theemissions to rice agriculture. In

132 General discussion and future perspectives

Chapter 3 it was concluded that the relative contributions of wetland and rice paddy emissionsin the latitudinal zone of 10–40◦N are not well constrained by the inversion. A more detailedinvestigation of the southeast Asian stations and South China Sea ship cruises (Chapter 4)indicated that the continental emissions are overestimated by the model, pointing to overes-timated rice paddies emissions. This result is in agreementwith recent up-scaling derivedestimates of rice paddy emissions. It was also concluded, however, that weekly samplingis insufficient to characterize CH4 concentrations distributions at these measurement sites.These results indicate thatto constrain sources of CH4 by inverse modeling in a complexregion such as southeast Asia, it requires an increased model resolution and an increasednumber of stations sampling at relatively high frequencies.

In section 5 it was concluded thatby simulating the pre-industrial atmosphere the uncer-tainty of the present day natural wetland source strength can be reduced.Moreover, model re-sults suggest thatthese estimates may further be improved once pre-industrial isotopic ratios(δ13C-CH4) can reliably be derived from ice cores.If the pre-industrially defined biogenicbackground sources are supplemented by estimates of the historic development of anthro-pogenic sources, the observed exponential increase of CH4 in the 19th and 20th century canlargely be reproduced by the model. For 20th century, these simulations point to an underesti-mate of the anthropogenic emissions by∼10%. However, this remains within the uncertaintyof the OH change over this period. Unfortunately, little constraints on the pre-industrialtrace gas composition are available. The discrepancies of our model results with these mea-surements call for more research. Also, changes of biogenicsources of CH4 over the pastcenturies have not been studied in sufficient detail.

Appendix A: Gas-phase chemistrymechanism

133

134 Appendix A Gas-phase chemistry mechanism

Table 1.: Modified CBM-4 Gas Phase Chemistry Mechanism

Reaction A -E/R n Ref.

Inorganic Reactions

(R1) NO2hν−→ NO+O3 rj.1 BC

(R2) NO+O3 −→ NO2 2.x10−12

-1400 JPL

(R3) NO+HO2 −→ NO2+OH 3.7x10−12

250 JPL

(R4) NO2+OHM−→ HNO3 k0 2.6x10

−303.2 JPL

k∞ 2.4x10−11

1.3

(R5) HNO3hν−→ NO2+OH rj.2 BC

(R6) HNO3+OH −→ NO3 k0a 7.2x10

−15785 JPL

k1a 4.1x10

−161440

k2a 1.9x10

−33725

(R7) NO2+O3 −→ NO3 1.2x10−13

-2450 JPL

(R8) NO+NO3M−→ 2NO2 1.5x10

−11170 JPL

(R9) NO3hν−→ NO rj.3 BC

(R10) NO3hν−→ NO2+O3 rj.4 BC

(R11) NO2+NO3M−→ N2O5 k0 2.2x10

−303.9 JPL

k∞ 1.5x10−12

0.7

(R12) N2O5 −→ NO2+NO3 k(NO2+NO3) JPL

/EQN2O5c

(R13) N2O5hν−→ NO2+NO3 rj.5 BC

(R14) NO2+HO2 −→ HNO4 k0 1.8x10−31

3.2 JPL

k∞ 4.7x10−12

1.4

(R15) HNO4hν−→ NO2+HO2 rj.6 BC

(R16) HNO4+OH −→ NO2 1.3x10−12

380 JPL

(R17) HNO4M−→ NO2+HO2 k(NO2+HO2) JPL

/EQHNO4c

(R18) O3hν−→ 2OH rj.7

dJPL

kO2d

3.2x10−11

70 JPL

kN2d

1.8x10−11

110 JPL

kH2Od

2.2x10−10

JPL

(R19) HO2+O3 −→ OH 1.1x10−14

-500 JPL

(R20) OH+O3 −→ HO2 1.6x10−12

-940 JPL

(R21) H2+OH −→ HO2 5.5x10−12

-2000 JPL

(R22) HO2+OH −→ 4.8x10−11

250 JPL

(R23) HO2+HO2 −→ H2O2b

JPL

(R24) H2O2+OH −→ HO2 2.9x10−12

-160 JPL

(R25) H2O2hν−→ 2OH rj.8 BC


Table 1. (continued)


Methane Chemistry

(R26) CH4+OH −→ CH3O2 2.65x10−12

-1800 JPL

(R27) CH3O2+NO −→ CH2O+HO2+NO2 4.2x10−12

180 JPL

(R28) CH3O2+HO2 −→ CH3O2H 3.8x10−13

800 JPL

(R29) CH3O2+CH3O2 −→ 1.33CH2O+0.67HO2 2.5x10−13

190 JPL

(R30) CH3O2H+OH −→ 0.7CH3O2+0.3CH2O 3.8x10−12

200 JPL

+0.3OH

(R31) CH3O2Hhν−→ CH2O+HO2+OH rj.9 BC

(R32) CH2Ohν−→ 2HO2+CO rj.10 BC

(R33) CH2Ohν−→ CO rj.11 MO

(R34) CH2O+OH −→ HO2+CO 1.0x10−11

JPL

(R35) CH2O+NO3 −→ HNO3+HO2+CO 5.8x10−16

AT

(R36) CO+OH −→ HO2 1.5x10−13

JPL

x(1+0.6 Patm)

NMHC Chemistry

(R37) ALD2+OH −→ C2O3 7.0x10−12

250 G1

(R38) ALD2+NO3 −→ C2O3+HNO3 2.5x10−15

G1

(R39) ALD2hν−→ CH2O+XO2+CO rj.12 LS

+2HO2

(R40) C2O3+NO −→ CH2O+XO2+HO2 3.5x10−11

-180 G2

+NO2

(R41) C2O3+NO2 −→ PAN k0 2.6x10−28

-7.1 ATG

k∞ 1.2x10−11

0.9

(R42) PAN −→ C2O3+NO2 2.0x1016

-13500 G2

(R43) PANhν−→ C2O3+NO2 rj.13 SEN

(R44) C2O3+C2O3 −→ 2CH2O+2XO2+2HO2 2.0x10−12

G1

(R45) C2O3+HO2 −→ CH2O+XO2+HO2 6.5x10−12

G1

+0.79OH+0.21ROOH

(R46) PAR+OH −→ 0.87XO2+0.76ROR 8.1x10−13

G1

+0.11HO2+0.11ALD2

+0.11RXPAR

+0.13XO2N

(R47) ROR −→ 1.1ALD2+0.96XO2 1.x1015

-8000 G1

+0.04XO2N+0.02ROR

+2.1RXPAR+0.94HO2

(R48) ROR −→ HO2 1.6e3 G1

(R49) OLE+OH −→ CH2O+ALD2+XO2 5.2x10−12

504 G1

136 Appendix A Gas-phase chemistry mechanism



+HO2+RXPAR

(R50) OLE+O3 −→ 0.44ALD2+0.64CH2O 4.33x10−15

-1800 STO

+0.25HO2+0.29XO2

+0.37CO+0.9RXPAR

+0.4OH

(R51) OLE+NO3 −→ 0.91XO2+CH2O 7.7x10−15

G1

+0.09XO2N+NO2

+ALD2+RXPAR

(R52) ETH+OHM−→ HO2+1.56CH2O k0 1.x10

−280.8 JPL

+0.22ALD2+XO2 k∞ 8.8x10−12

(R53) ETH+O3 −→ CH2O+0.26HO2 9.1x10−15

-2580 JPL

+0.12OH+0.43CO

(R54) MGLY+OH −→ XO2+C2O3 1.7x10−11

AT2

(R55) MGLYhν−→ C2O3+HO2+CO rj.14 G1

(R56) ISOP+OH −→ 0.85XO2+0.61CH2O 2.54x10−11

410 AT1

+0.58OLE+0.85HO2

+0.15XO2N

+0.03MGLY

+0.63PAR

(R57) ISOP+O3 −→ 0.9CH2O+0.55OLE 12.3x10−15

-2013 AT1

+0.36CO+0.15C2O3

+0.63PAR+0.30HO2

+0.18XO2+0.03MGLY

+0.28OH

(R58) ISOP+NO3 −→ 0.9HO2+0.9ORGNIT 7.8x10−13

WL

+0.45OLE+0.12ALD

+0.08MGLY+0.1NO2

+0.03CH2O

(R59) ROOHhν−→ OH as CH3O2H

(R60) ROOH+OH −→ 0.7XO2+0.3OH 3.x10−12

AT1

(R61) ORGNIT+OH −→ NO2+XO2 1.78x10−12

AT1

(R62) ORGNIThν−→ NO2+HO2 rj.15 RF




Reactions of Operator Species

(R63) XO2+NO −→ NO2 4.2x10−12

180 AT1

(R64) XO2+XO2 −→ 1.7x10−14

1300 AT1

(R65) XO2N+NO −→ ORGNIT 6.8x10−13

G1

(R66) XO2+HO2 −→ ROOH 3.5x10−13

1000 AT1

(R67) XO2N+HO2 −→ ROOH R65xR66

/R63

(R68) RXPAR+PAR −→ 8.x10−11

HT

Rate constants in moleculesp/(cmq*s), (p, q reaction molecularity dependent). Notationof rate coefficients: monomolecular and bimolecular reactions,k = A∗e(−E/RT); three-bodyreactions,k= k0(T)M/(1+k0(T)M/k∞(T))∗F (1/(1+[log(k0(T)M/k∞(T))]2) with k0(T),k∞(T) =A(300/T)n; M, molecular density of air (molecules/cm3) andF = 0.6. Photolysis rates: wave-length and light intensity dependent (see references). CO2, O2, and H2O are not listed inreaction products. BC,Bruhl and Crutzen[1988]; JPL,DeMore et al.[1994]; MO, Moort-gat et al. [1980]; G1,Gery et al.[1989]; G2,Gery et al.[1988]; LS, Leone and Seinfeld[1985]; SEN,Senum et al.[1984]; AT, Atkinson et al.[1992]; AT1, Atkinson[1994]; WL,Wille et al. [1991]; RB, Roberts[1990]; RF,Roberts and Fayer[1989]; HT, Hertel et al.[1993]; STO, W. R. Stockwell (submitted toJournal of Geophysical Research,1997; ATG,based onAtkinson[1994] andGery et al.[1989], note thatF = 0.3 in the three body Troefalloff equation.

aHere k=k0 +k2M/(1+k2M/k1).bHere k=(2.3 ·10−13e(600/T) +1.7 ·10−33Me(1000/T)) · (1+H2O ·1.4e−21e(2200/T)).cEQN2O5,EQHNO4: equilibrium constantsDeMore et al.[1994].dHere k=rj.7·[(kH2O·H2O)/(kH2O·H2O+kN2·N2+kO2·O2)].

Appendix B: NOAA measurementstations

139

140 Appendix B NOAA measurement stations

Table 2.: Measurement Stations

Station Code Location Coordinates

Latitude Longitude Altitude, m

ALT ∗ Alert 82◦27’N 62◦31’W 210ASC∗ Ascension Island 7◦55’S 14◦25’W 54BAL Baltic Sea 55◦30’N 16◦40’E 7BME Bermuda East 32◦22’N 64◦39’W 30BMW Bermuda West 32◦16’N 65◦53’W 30BRW∗ Barrow 71◦19’N 156◦36’W 11CBA Cold Bay 55◦12’N 162◦43’W 25CGO∗ Cape Grim 40◦41’S 144◦41’E 94CMO Cape Meares 45◦29’N 123◦58’W 30GMI∗ Guam 13◦26’N 144◦47’E 2GOZ Gozo 36◦03’N 14◦11’E 30ICE Heimaey 63◦15’N 20◦09’W 100ITN Grifton 35◦21’N 77◦23’W 505IZO Tenerife 28◦18’N 16◦29’W 2300KEY Key Biscayne 25◦40’N 80◦12’W 3KUM ∗ Cape Kumukahi 19◦31’N 154◦49’W 3MBC∗ Mould Bay 76◦15’N 119◦21’W 58MHT Mace Head 53◦20’N 9◦54’W 25MID ∗ Midway 28◦13’N 177◦22’W 4MLO∗ Mauna Loa 19◦32’N 155◦35’W 3397NWR Niwot Ridge 40◦03’N 105◦35’W 3475P01∗ Pacific cruise 10◦00’S 168◦36’W 0P02∗ Pacific cruise 15◦00’S 175◦42’W 0P03∗ Pacific cruise 25◦00’S 178◦00’E 0P04∗ Pacific cruise 30◦00’S 177◦06’E 0PSA∗ Palmer Station 64◦55’S 64◦00’W 10QPC Qinghai Province 36◦16’N 100◦55’E 3810RPB∗ Barbados 13◦10’N 59◦26’W 3

Appendix B NOAA measurement stations 141


Station Code Location Coordinates

Latitude Longitude Altitude, m

SC1 South China Sea cruise 3◦00’N 105◦12’E 0SC2 South China Sea cruise 6◦00’N 107◦18’E 0SC3 South China Sea cruise 9◦00’N 109◦24’E 0SC4 South China Sea cruise 12◦00’N 111◦00’E 0SC5 South China Sea cruise 15◦00’N 112◦18’E 0SC6 South China Sea cruise 18◦00’N 113◦18’E 0SC7 South China Sea cruise 21◦00’N 114◦00’E 0SEY∗ Seychelles 4◦40’S 55◦10’E 3SHM∗ Shemya Island 52◦43’N 174◦06’E 40SMO∗ American Samoa 14◦15’S 170◦34’W 42SPO∗ South Pole 89◦59’S 24◦48’W 2810STM∗ Ocean Station “M” 66◦00’N 2◦00’E 7SYO∗ Syowa 69◦00’S 39◦35’E 11TAP Tae-ahn Peninsula 36◦44’N 126◦08’E 20UTA Utah 39◦54’N 113◦43’W 1320UUM Ulaan Uul 44◦27’N 111◦06’E 914ZEP Ny-Alesund 78◦54’N 11◦53’E 475

∗Station is used to derive the global mean trend.

Bibliography

Aardenne, J. A., F. J. Dentener, C. G. M. Klein Goldewijk, J. Lelieveld, and J. G. J. Olivier,A 1◦x1◦ resolution dataset of historical anthropogenic trace gas emissions for the period1890–1990,Submitted to Global Biogeochem. Cycles, 1999.

Andres, R. J., and A. D. Kasgnoc, A time-averaged inventory of subaerial vulcanic sulfuremissions,J. Geophys. Res., 103, 25,251–25,261, 1998.

Anfossi, D., S. Sandroni, and S. Viarengo, Tropospheric ozone in the nineteenth century: TheMoncalieri series,J. Geophys. Res., 96, 17,349–17,352, 1991.

Apel, E., D. Riemer, W. Brune, I. Faloona, D. Tan, M. A. Carroll, T. Thornberry, andM. Town, Isoprene chemistry in rural forested areas: Significance of nighttime HOx, Posterpresented at IGAC conference, Bologna, Italy, 1999.

Aselmann, I., and P. J. Crutzen, Global distribution of natural freshwater wetlands and ricepaddies, their net primary productivity, seasonality and possible methane emissions,J.Atmos. Chem., 8, 307–358, 1989.

Atkinson, R., Gas-phase tropospheric chemistry of organiccompounds,J. Geophys. Res., 10,1239–1238, 1994.

Atkinson, R., D. L. Baulch, R. A. Cox, R. F. Hampson, J. A. Kerr, and J. Troe, Evaluated ki-netic and photochemical data for atmospheric chemistry: supplement IV,Atmos. Environ.,26, 1187–1230, 1992.

Atlas, E., W. Pollock, J. Greenberg, and L. Heidt, Alkyl nitrates, nonmethane hydrocarbons,and halocarbon gases over the equatorial Pacific Ocean during Saga 3,J. Geophys. Res.,98, 16,933–16,947, 1993.

Atlas, E. L., and B. A. Ridley, The Mauna Loa Observatory Photochemistry Experiment:Introduction,J. Geophys. Res., 101, 14,531–14,541, 1996.

Bailey, P. L., et al., Comparison of cryogenic limb array etalon spectrometer (CLAES) ozoneobservations with correlative measurements,J. Geophys. Res., 101, 9737–9756, 1997.

Barlett, K. B., and R. C. Harris, Review and assessment of methane emissions from wetlands,Chemosphere, 26, 261–320, 1993.

Bates, T. S., K. C. Kelly, J. E. Johnson, and R. H. Gammon, Regional and seasonal variationsin the flux of oceanic carbon monoxide to the atmosphere,J. Geophys. Res., 100, 23,093–23,101, 1995.

143

144 BIBLIOGRAPHY

Bates, T. S., K. C. Kelly, J. E. Johnson, and R. H. Gammon, A reevaluation of the open oceansource of methane to the atmosphere,J. Geophys. Res., 101, 6953–6961, 1996.

Battle, M., Atmospheric gas concentrations over the past century measured in air from firn atthe South Pole,Nature, 383, 231–235, 1993.

Bauer, K., R. Conrad, and W. Seiler, CO production hoherer Pflanzen an naturlichen Stan-dorten,Z. Pflanzenphysiol., 94, 219–230, 1979.

Benkovitz, C. M., M. T. Scholtz, J. Pacyna, L. Tarrason, J. Dignon, E. C. Voldener, P. A.Spiro, J. A. Logan, and T. E. Graedel, Global gridded inventories of anthropogenic emis-sions of sulfur and nitrogen,J. Geophys. Res., 101, 29,239–29,253, 1996.

Bergamaschi, P., Seasonal variations of stable hydrogen and carbon isotopic ratios in methanefrom a Chinese rice paddy,J. Geophys. Res., 102, 25,383–25,393, 1997.

Bergamaschi, P., C. Lubina, R. Konigstedt, H. Fischer, A. C. Veltkamp, and O. Zwaagstra,Stable isotopic signatures (δ 13C, δ d) of methane from European landfill sites,J. Geophys.Res., 103, 8251–8265, 1998.

Bottenheim, J. W., A. Sirios, K. A. Brice, and A. J. Gallant, Five years of continuous PANand ozone at a rural location in eastern Canada,J. Geophys. Res., 99, 5333–5352, 1994.

Bouwman, A. F., D. S. Lee, W. A. H. Asman, F. J. Dentener, K. W. van der Hoek, and J. G. J.Olivier, A global high-resolution emission inventory for ammonia,Global Biogeochem.Cycles, 11, 561–587, 1997.

Brenninkmeijer, C. A. M., D. C. Lowe, M. R. Manning, R. J. Sparks, and P. F. J. vanVelthoven, The13C, 14C, and18O isotopic composition of CO, CH4, and CO2 in the highersouthern latitudes lower stratosphere,J. Geophys. Res., 100, 26,163–26,172, 1995.

Brent, R. P.,Algorithms for Minimization Without Derivatives, chap. 5, Prentice-Hall, Engle-wood Cliffs, N. J., 1973.

Brewer, D. A., T. R. Augustsson, and J. S. Levine, The photochemistry of anthropogenicnonmethane hydrocarbons in the troposphere,J. Geophys. Res., 88, 6683–6695, 1983.

Brown, M., Deduction of fluxes of source gases using an objective inversion algorithm and achemical transport model,J. Geophys. Res., 98, 12,639–12,660, 1993.

Brown, M., The singular value decomposition method appliedto the deduction of the emis-sions and the isotopic composition of atmospheric methane,J. Geophys. Res., 100, 11,425–11,446, 1995.

Bruhl, C., and P. J. Crutzen, Scenarios of possible change in tropospheric temperatures andozone concentrations due to man’s activities, estimated with a one dimensional coupledphotochemical climate model,Clim. Dyn., 2, 173–202, 1988.

Bruhl, C., and P. J. Crutzen, MPIC two-dimensional model,NASA Ref. Publ. 1292, pp. 103–104, 1993.

Butler, J. H., J. W. Elkins, T. M. Thompson, B. D. Hall, T. H. Swanson, and V. Koropalov,Oceanic consumption of CH3CCl3: Implications for tropospheric OH,J. Geophys. Res.,96, 22,347–22,355, 1991.

BIBLIOGRAPHY 145

Cantrell, C. A., R. E. Shetter, A. H. McDaniel, J. G. Calvert,J. A. Davidson, D. C. Lowe, S. C.Tyler, R. J. Cicerone, and J. P. Greenberg, Carbon kinetic isotope effect in the oxidation ofmethane by the hydoxyl radical,J. Geophys. Res., 95, 22,455–22,462, 1990.

Cao, M., S. Marshall, and K. Gregson, Global carbon exchangeand methane emissions fromnatural wetlands: Application of a process-based model,J. Geophys. Res., 101, 14,399–14,414, 1996.

Cao, M. K., J. B. Dent, and O. W. Heal, Methane emissions from China’s paddyland,Agr.Ecosyst. Environ., 55, 129–137, 1995.

Chameides, W. L., and R. J. Cicerone, Effects of nonmethane hydrocarbons in the tropo-sphere,J. Geophys. Res., 83, 947–952, 1978.

Chappellaz, J. A., I. Y. Fung, and A. M. Thompson, The atmospheric CH4 increase since thelast glacial maximum, 1, Source estimates,Tellus, 45B, 228–241, 1993.

Chappellaz, J. A., T. Blunier, S. Kints, A. Dallenbach, J.-M. Barnola, J. Schwander, D. Ray-naud, and B. Stauffer, Changes in the atmospheric CH4 gradient between Greenland andAntarctica during the Holocene,J. Geophys. Res., 102, 15,987–15,997, 1997.

Craig, H., Isotopic standards for carbon and oxygen and correction factors for mass-spectrometric analysis of carbon dioxide,Geochim. Cosmochim. Acta, 12, 133–149, 1957.

Craig, H., and C. C. Chou, Methane: The record in polar ice cores,Geophys. Res. Lett., 9,1221–1224, 1982.

Craig, H., C. C. Chou, J. A. Welhan, C. M. Stevens, and A. Engelkemeir, Isotopic composi-tion of methane in polar ice cores,Science, 242, 1535–1539, 1988a.

Craig, H., Y. Horibe, and T. Sowers, Gravitational separation of gases and isotopes in polarice caps,Science, 242, 1675–1678, 1988b.

Crutzen, P. J., The role of methane in atmospheric chemistryand climate, inRuminant Physi-ology: Digestion, Metabolism, Growth and Reproduction: Proceedings of the Eighth Inter-national Symposium on Ruminant Physiology, edited by W. V. Engelhardt, S. Leonhardt-Marek, G. Breves, and D. Giesecke, pp. 291–315, Ferdinant Enke Verl., Stuttgart, Ger-many, 1995.

Crutzen, P. J., and C. Bruhl, A model study of atmospheric temperatures and concentrationsof ozone,Geophys. Res. Lett., 20, 1047–1050, 1993.

Crutzen, P. J., and P. H. Zimmerman, The changing chemistry of the troposphere,Tellus,43A/B, 136–151, 1991.

Crutzen, P. J., I. Aselmann, and W. Seiler, Methane production by domestic animals, wildruminants, other herbivorous fauna, and humans,Tellus, Ser. B, 271–284, 1986.

Davidson, J. A., C. A. Cantrell, S. C. Tyler, R. E. Shetter, R.J. Cicerone, and J. G. Calvert,Carbon isotopic effect in the reaction of CH4 with OH, J. Geophys. Res., 92, 2195–2199,1987.

146 BIBLIOGRAPHY

DeMore, W. B., S. P. Sander, D. M. Golde, R. F. Hampson, M. J. Kurylo, C. J. Howard,A. R. Ravishankara, C. E. Kolb, and M. J. Molina, Chemical kinetics and photochemicaldata for use in stratospheric modelling,NASA/JPL publ. 94-26 Eval. 11, Jet Propul. Lab.,Pasadena, Calif., 1994.

Denier van der Gon, H. A. C., Changes in CH4 emissions from rice fields from 1960 to the1990s, 1, impacts of modern rice technology,Global Biogeochem. Cycles, in press, 2000a.

Denier van der Gon, H. A. C., Changes in CH4 emissions from rice fields from the 1960s-1990s, 2, the declining use of organic inputs in rice farming, Global Biogeochem. Cycles,in press, 2000b.

Denning, A. S., et al., Three-dimensional transport and concentration ofSF6: A model inter-comparison study (Transcom 2),Tellus, Ser. B, 51, 266–297, 1999.

Dentener, F. J., Heterogeneous chemistry in the troposphere, Ph.D. thesis, Utrecht Univ.,Utrecht, The Netherlands, 1993.

Dentener, F. J., and P. J. Crutzen, Reaction of N2O5 on tropospheric aerosols: Impact on theglobal distribution of NOx, O3, and OH,J. Geophys. Res., 98, 7149–7163, 1993.

Dentener, F. J., J. Feichter, and A. Jeuken, Simulation of the transport of222Rn using on-lineand off-line global models at different horizontal resolutions: A detailed comparison withmeasurements,Tellus, Ser. B, 51, 573–602, 1999.

Derwent, R. G., Evaluation of a number of chemical mechanisms for their application inmodels describing the formation of photochemical ozone in Europe,Atmos. Environ., 24,2615–2624, 1990.

Derwent, R. G., M. E. Jenkin, and S. M. Saunders, Photochemical ozone creation potentialsfor a large number of reactive hydrocarbons under European conditions,Atmos. Environ.,30, 181–199, 1996.

Dlugokencky, E. J., K. A. Masarie, P. M. Lang, P. P. Tans, and E. G. Nisbet, A dramaticdecrease in the growth rate of atmospheric methane in the Northern Hemisphere during1992,Geophys. Res. Lett., 21, 45–48, 1994a.

Dlugokencky, E. J., L. P. Steele, P. M. Lang, and K. A. Masarie, The growth rate and distri-bution of atmospheric methane,J. Geophys. Res., 99, 17,021–17,043, 1994b.

Dlugokencky, E. J., L. P. Steele, P. M. Lang, and K. A. Masarie, Atmospheric methane atMauna Loa and Barrow observatories: Presentation and analysis of in situ measurements,J. Geophys. Res., 100, 23,103–23,113, 1995.

Dlugokencky, E. J., E. G. Dutton, P. C. Novelli, and K. A. Masarie, Changes in CH4 andCO growth rates after the eruption of Mt. Pinatubo and their link with changes in tropicaltropospheric UV flux,Geophys. Res. Lett., 23, 2761–2764, 1996.

Dlugokencky, E. J., K. A. Masarie, P. M. Lang, and P. P. Tans, Continuing decline in thegrowth rate of the atmospheric methane burden,Nature, 393, 447–450, 1998.

Dong, H. M., L. Erda, Y. Li, M. J. Rao, and Q. C. Yang, An estimation of methane emissionsfrom agricultural activities in China,Ambio, 25, 292–296, 1996.

BIBLIOGRAPHY 147

Dunfield, P., R. Knowles, R. Dumont, and T. R. Moore, Methane production and consumptionin temperate and subarctic peat soils: Response to temperature and soil pH,Soil Biol.Biochem., 25, 321–326, 1993.

Durand, J. D., Historical estimates of world population: Anevaluation,Analytical and Tech-nical Reports 10, University of Pennsylvania, Population Center, 1974.

Enting, I. G., C. M. Trudinger, and R. J. Francey, A synthesisinversion of the concentrationandδ13C of atmospheric CO2, Tellus, Ser. B, 47, 35–52, 1995.

Environmental Protection Agency (EPA), International anthropogenic methane emissions:Estimates for 1990,EPA 230-R-94-010, edited by J. Adler, Washington, D. C., 1994.

Erickson, D. J., Ocean to atmosphere carbon monoxide flux: Global inventory and climateapplications,Global Biogeochem. Cycles, 3, 305–314, 1989.

Etheridge, D. M., L. P. Steele, R. J. Francy, and R. L. Langenfelds, Atmospheric methanebetween 1000 A. D. and present: Evidence of anthropogenic emissions and climatic vari-ability, J. Geophys. Res., 103, 15,979–15,993, 1998.

Fergusin, E. E., and R. M. Rosson, Climate monitoring and diagnostics laboratory,Summaryreport 1991, U. S. Dep. of Commer., CMDL rep. 20, Washington, D. C., 1992.

Fortuin, J. P. F., and H. Kelder, An ozone climatology based on ozonesonde and satellitemeasurements,J. Geophys. Res., 103, 31,709–31,734, 1998.

Fraser, P. J., M. A. K. Khalil, R. A. Rasmussen, and A. J. Crowford, Termites and globalmethane: Another assessment,J. Atmos. Chem., 4, 295–310, 1990.

Fung, I., J. John, J. Lerner, E. Matthews, M. Prather, L. P. Steele, and P. J. Fraser, Three-dimensional model synthesis of the global methane cycle,J. Geophys. Res., 96, 13,033–13,065, 1991.

Galliardo-Klenner, L. G., Oxidized nitrogen in the troposphere: The role of lightning, Ph.D.thesis, Stockholm Univ., Stockholm, 1996.

Galliardo-Klenner, L. G., and V. Cooray, Could cloud-to-cloud discharges be as effective ascloud-to-ground discharges in producing NOx?,Tellus, 48B, 641–651, 1996.

Ganzeveld, L., J. Lelieveld, and G. J. Roelofs, A dry deposition parameterization for sulfuroxides in a chemistry and general circulation model,J. Geophys. Res., 103, 5679–5694,1997.

Gery, M. W., G. Z. Whitten, J. P. Killus, and M. C. Dodge, Development and testing of theCBM-4 for urban and regional modelling,Rep. EPA-600/3-88-012, U.S. Environ. Prot.Agency, Research Triangle Park, N. C., 1988.

Gery, M. W., G. Z. Whitten, J. P. Killus, and M. C. Dodge, A photochemical kinetics mech-anism for urban and regional scale computer modeling,J. Geophys. Res., 94, 925–956,1989.

Giering, R., Tangent linear and adjoint model compiler,Users Manual, Max Planck Inst. forMeteorol., Hamburg, Germany, 1996.

148 BIBLIOGRAPHY

GLOBALVIEW-CH4, Cooperative atmospheric data integration project - methane., CD-ROM Also available on internet via anonymous FTP to ftp.cmdl.noaa.gov, path:ccg/ch4/GLOBALVIEW, NOAA CMDL, Boulder, Colorado, 1999.

Goldstein, A. H., S. C. Wofsy, and C. M. Spivakovski, Seasonal variations of nonmethanehydrocarbons in rural New England: Constraints on OH concentrations in the northernmidlatitudes,J. Geophys. Res., 100, 21,023–21,033, 1995.

Goody, R. M., and Y. L. Yung,Atmospheric radation, Oxford University Press, New York,1989.

Guenther, A., et al., A global model of natural volatile organic compound emissions,J. Geo-phys. Res., 100, 8873–8892, 1995.

Gupta, M., S. Tyler, and R. Cicerone, Modeling atmosphericδ13CH4 and the caused of recentchanges in atmospheric CH4 amounts,J. Geophys. Res., 101, 22,923–22,932, 1996.

Haan, D., P. Martinerie, and D. Raynaud, Ice core data of atmospheric carbon monoxide overAntarctica and Greenland during the last 200 years,Geophys. Res. Lett., 23, 2235–2238,1996.

Haas-Laursen, D. E., D. E. Hartley, and T. J. Conway, Consistent sampling methods forcomparing models to CO2 flask data,J. Geophys. Res., 102, 19,059–19,071, 1997.

Hao, W. M., and D. E. Ward, Methane production of global biomass burning,J. Geophys.Res., 98, 20,657–20,661, 1993.

Hao, W. M., M. H. Liu, and P. J. Crutzen, Estimates of the annual and regional releases ofCO2 and other trace gases to the atmosphere from fires in the tropics based on the FAOstatistics for the period 1975-1980, inFire in the Tropical Biota, edited by J. Goldhammer,pp. 440–462, Springer, New York, 1991.

Harris, J. M., P. P. Tans, E. J. Dlugokencky, K. A. Masarie, P.M. Lang, S. Whittlestone,and L. P. Steele, Variations in atmospheric methane at MaunaLoa observatory related tolong-range transport,J. Geophys. Res., 97, 6003–6010, 1992.

Hartley, D., and R. Prinn, Feasability of determining surface emissions of trace gases usingan inverse method in a three-dimensional chemical transport model,J. Geophys. Res., 98,5183–5197, 1993.

Heimann, M., The global atmospheric tracer model TM2,Tech. Rep. 10 ISSN 0940-9327,Dtsch. Klimarechenzentrum, Hamburg, Germany, 1995.

Heimann, M., and C. D. Keeling, A three-dimensional model ofatmospheric CO2 transportbased on observed winds, model description and simulated tracer experiments, inAspectsof Climate Variability in the Pacific and the Western Americas, Geophys. Monogr. Ser.,vol. 55, edited by D. H. Peterson, pp. 237–275, AGU, Washington, D. C., 1989.

Hein, R., Inverse Modellierung des atmospharischen Methan-Kreislaufs unter Verwendungeines drei-dimensionalen Modells des Transports und der Chemie der Troposphare, Ph.D.thesis, Fachbereich Geowiss. der Univ. Hamburg, Hamburg, Germany, 1994.

BIBLIOGRAPHY 149

Hein, R., P. J. Crutzen, and M. Heimann, An inverse modeling approach to investigate theglobal atmospheric methane cycle,Global Biogeochem. Cycles, 11, 43–76, 1997.

Hertel, O., R. Berkowicz, J. Christensen, and O. Hov, Test oftwo numerical schemes for usein atmospheric transport-chemistry models,Atmos. Environ., Part A, 27, 2591–2611, 1993.

Herzberg, G.,Molecular spectra and molecular structure, 2, Infrared andRaman spectra ofpolyatomic molecules, Lancaster Press, New York, 1945.

Hogan, K. B., and R. C. Harris, Comment on ‘A decrease in the growth rate of atmosphericmethane in the Northern Hemisphere during 1992’ by E. J. Dlugokencky et al.,Geophys.Res. Lett., 21, 2445–2446, 1994.

Hough, A. M., Development of a two-dimensional global tropospheric model: Model chem-istry, J. Geophys. Res., 96, 7325–7362, 1991.

Houweling, S., F. J. Dentener, and J. Lelieveld, The impact of non-methane hydrocarboncompounds on tropospheric photochemistry,J. Geophys. Res., 103, 10,673–10,696, 1998.

Houweling, S., T. Kaminski, F. J. Dentener, J. Lelieveld, and M. Heimann, Inverse modellingof methane sources and sinks using the adjoint of a global transport model,J. Geophys.Res., in press, 1999a.

Houweling, S., F. J. Dentener, and J. Lelieveld, The modeling of tropospheric methane; howwell can point measurements be reproduced by a global model?, Submitted to, J. Geophys.Res., 1999b.

Hov, O., N. Schmidbauer, and M. Oehme, C2–C5 hydrocarbons in rural southern Norway,Atmos. Environ., 25A, 1981–1999, 1991.

Hovland, M., A. G. Judd, and R. A. Burke, The global flux of methane from shallow subma-rine sediments,Chemosphere, 26, 559–578, 1993.

Huang, Y., R. L. Sass, and F. M. Fisher, Methane emissions from Texas rice paddy soils,1, Quantative multi-year dependence of CH4 emissions on soil, cultivar and grain yield,Global Change Biol., 3, 479–489, 1997.

IPCC, Climate Change: The IPCC Scientific Assessment, Edited by J. T. Houghton et al.,Cambridge Univ. Press, New York, 1990.

IPCC, Climate change 1992: The supplementary report to the IPCC Scientific Assessment,in Climate change 1994, edited by J. T. Houghton and B. A. C. S. K. Varney, CambridgeUniversity Press, 1992.

IPCC, Radiative forcing of climate change and an evaluationof the IPCC IS92 emissionscenarios, inClimate Change 1994, edited by J. T. Houghton et al., Cambridge Univ.Press, New York, 1994.

IPCC,Guidelines for National Greenhouse Gas Inventories: Reference Manual, chap. 4.3,Agriculture: Methane emission from rice cultivation, Organisation for Economic Cooper-ation and Development (EOCD), 1997.

150 BIBLIOGRAPHY

Isaksen, I. S. A., O. Hov, S. A. Penkett, and A. Semb, Model analysis of the measuredconcentrations of organic gases in the Norwegian Arctic,J. Atmos. Chem., 3, 271–285,1985.

Jacob, J. J., and S. C. Wofsy, Budgets of reactive nitrogen, hydrocarbons, and ozone over theAmazon forest during the wet season,J. Geophys. Res., 95, 16,737–16,754, 1990.

Junge, C. E., and P. J. Gustafson, On the distribution of seasalt over the United States and itsremoval by precipitation,Tellus, 9, 164–173, 1957.

Kames, J., and U. Schurath, Alkyl nitrates and atmospheric nitrates of atmospheric interest:Henry’s law constants and their temperature dependencies,J. Atmos. Chem., 15, 79–95,1992.

Kaminski, T., M. Heimann, and R. Giering, Sensitivity of theseasonal cycle of CO2 at remotemonitoring stations with respect to seasonal surface exchange fluxes determined with theadjoint of an atmospheric transport model,Phys. Chem. Earth, 21, 457–462, 1996.

Kaminski, T., M. Heimann, and R. Giering, A coarse grid three-dimensional global inversemodel of the atmospheric transport, 1, Adjoint model and Jacobian matrix,J. Geophys.Res., 104, 18,535–18,553, 1999a.

Kaminski, T., M. Heimann, and R. Giering, A coarse grid three-dimensional global inversemodel of the atmospheric transport, 2, Inversion of the transport of CO2 in the 1980s,J.Geophys. Res., 104, 18,555–18,581, 1999b.

Kaminski, T., P. Rayner, and M. Heimann, Bias in inversions of atmospheric transport due tocoarse resolution of sources, Poster presented at IGAC conference, Bologna, Italy, 1999c.

Kammen, D. M., and B. D. Marino, On the origin and magnitude ofpre-industrial anthro-pogenic CO2 and CH4 emissions,Chemosphere, 26, 69–86, 1993.

Kanakidou, M., H. B. Singh, K. M. Valentin, and P. J. Crutzen,A two-dimensional studyof ethane and propane oxidation in the troposphere,J. Geophys. Res., 96, 15,395–15,413,1991.

Kanakidou, M., F. J. Dentener, and P. J. Crutzen, A global three-dimensional study of the fateof HCFCs and HFC-134a in the troposphere,J. Geophys. Res., 100, 18,781–18,802, 1995.

Kandlikar, M., Bayesian inversion for reconciling uncertainties in global mass balances,Tel-lus, 49B, 123–135, 1997.

Kastings, J. F., and H. B. Singh, Nonmethane hydrocarbons inthe troposphere: Impact onodd hydrogen and odd nitrogen chemistry,J. Geophys. Res., 91, 13,239–13,256, 1986.

Keeling, C. D., S. C. Piper, and M. Heimann, A three dimensional model of the atmosphericCO2 transport based on observed winds, 4, mean annual gradientsand interannual varia-tions, inAspects of Climate Variability in the Pacific and the WesternAmericas, Geophys.Monogr. Ser.,vol. 55, edited by D. H. Peterson, pp. 305–363, AGU, Washington, D. C.,1989.

BIBLIOGRAPHY 151

Kern, J. S., D. Bachelet, and M. Tolg, Organic matter inputs and methane emissions fromsoils in major rice growing regions of China, inSoils and Global Change: advances in SoilScience, edited by R. Lal, E. Levine, J. Kimble, and B. A. Stewart, CRCLewis Publishers,New York, 1995.

Kern, J. S., Z. Gong, G. Zhang, H. Zhuo, and G. Luo, Spatial analysis of methane emissionsfrom paddy soils in China and the potential for emissions reduction,Nutrient Cycl. Agro-Ecosyst., 49, 181–195, 1997.

Khalil, M. A. K., Atmospheric methane: Sources, sinks, and role in global change, NATOASI Series 1: Global Environmental Change, vol. 13, Springer Verlag, New York, 1993.

Khalil, M. A. K., R. A. Rasmussen, J. R. J. French, and J. A. Holt, The influence of termiteson atmospheric trace gases: CH4, CO2, CHCl3, N2O, CO, H2, and light hydrocarbons,J.Geophys. Res., 95, 3619–3643, 1990.

Komhyr, W. D., S. J. Oltmans, P. R. Franchois, W. F. J. Evans, and W. A. Matthews, Thelatitudinal distribution of ozone to 35 km altitude from ECCozonesonde observations,1985-1987, inOzone in the Atmosphere, R. D. Bojkov and P. Fabian, A. Deepak, Hampton,Va., 1989.

Kreileman, G. J. J., Documentation of a geographically explicit dynamic carbon cycle model,RIVM report, Natl. Inst. for Public Health and the Environ. (RIVM), Bilthoven, TheNetherlands, 1996.

Krol, M., and M. Van Weele, Implications of variation of photodissociation rates for globalatmospheric chemistry,Atmos. Env., 31, 1257–1273, 1997.

Krol, M., P. J. van Leeuwen, and J. Lelieveld, Global OH trendinferred from methylchloro-form measurements,J. Geophys. Res., 103, 10,697–10,711, 1998.

Kumer, J. B., et al., Comparison of correlative data with HNO3 version 7 from the CLAESinstrument deployed on the NASA Upper Atmosphere Research Sattelite,J. Geophys. Res.,101, 9621–9656, 1997.

Lacroix, A. V., Unaccounted-for sources of fossil and isotopically-enriched methane and theircontribution to the emissions inventory: A review and synthesis,Chemosphere, 26, 507–557, 1993.

Lambert, G., and S. Schmidt, Reevaluation of the oceanic fluxof methane: Uncertainties andlong term variations,Chemosphere, 26, 579–589, 1993.

Landgraf, I., and P. J. Crutzen, An efficient method for online calculations of photolysis andheating rates,J. Atmos. Sci., 55, 863–878, 1998.

Lappalainen, E., General review on world peatland and peat resources, inGlobal peat re-sources, edited by E. Lappalainen, pp. 53–64, International Peat Society, 1996.

Lassay, K. R., D. C. Lowe, C. A. M. Brenninkmeijer, and A. J. Gomez, Atmospheric methaneand its carbon isotopes in the Southern Hemisphere: Their time series and an instructivemodel,Chemosphere, 26, 95–109, 1993.

152 BIBLIOGRAPHY

Law, R., The selection of model-generated CO2 data: A case study with seasonal biosphericsources,Tellus, 48B, 474–486, 1996.

Law, R., and I. Simmonds, The sensitivity of deduced CO2 sources and sinks to variations intransport and imposed surface concentrations,Tellus, Ser. B, 48, 613–625, 1996.

Ledbury, W., and E. W. Blair, The partial formaldehyde vaporpressure of aqueous solutionsof formaldehyde, II,J. Chem. Soc., 127, 2823–2839, 1925.

Lelieveld, J., and F. J. Dentener, What controls tropospheric ozone?,Submitted to J. Geophys.Res., 1999.

Lelieveld, J., P. J. Crutzen, and F. J. Dentener, Changing concentration, lifetime and climateforcing of atmospheric methane,Tellus, Ser. B, 50, 128–150, 1998.

Leone, J. A., and J. H. Seinfeld, Comparative analysis of chemical reaction mechanisms forphotochemical smog,Atmos. Environ., 19, 437–464, 1985.

Levin, I., The recent state of carbon cycling through the atmosphere, inCarbon Cycling inthe Glacial Ocean: Constraints on the Ocean’s Role in GlobalChange, edited by R. Zahn,pp. 3–13, Springer Verlag, New York, 1994.

Levin, I., and V. Hesshaimer, Refining of atmospheric transport model entries by the glob-ally observed passive tracer distributions of85krypton and sulfur hexafluoride (SF6), J.Geophys. Res., 101, 16,745–16,755, 1996.

Levy, H., A global three-dimensional time dependent lightning source of tropospheric NOx,J. Geophys. Res., 101, 22,911–22,922, 1996.

Lin, X., O. T. Melo, D. R. Hastie, P. B. Shepson, H. Niki, and J.W. Bottenheim, A case studyof ozone production in a rural area of central Ontario,Atmos. Environ., 26, 311–324, 1992.

Lind, J. A., and G. L. Kok, Henry’s law determinations for aqueous of hydrogen peroxide,methylperoxide, and peroxyacetic acid,J. Geophys. Res., 1991, 7889–7895, 1986.

Liu, S., M. Trainer, F. C. Fehsenfeld, D. D. Parrish, E. J. Williams, D. W. Fahey, G. Hubler,and P. C. Murphey, Ozone production in the rural troposphereand implications for regionaland global ozone distributions,J. Geophys. Res., 92, 4191–4207, 1987.

Louis, J. F., A parametric model of vertical eddy fluxes in theatmosphere,Boundary LayerMeteorol., 17, 187–202, 1979.

Lowe, D. C., M. R. Manning, G. W. Brailsford, and A. M. Bromley, The 1992–1992 atmo-spheric methane anomaly: Southern Hemisphere13C decrease and growth rate fluctua-tions,Geophys. Res. Lett., 24, 857–860, 1997.

Maier-Reimer, E., Geochemical cycles in an ocean general circulation model: Preindustrialtracer distributions,Global Biogeochem. Cycles, 7, 645–677, 1993.

Maiss, M., L. P. Steele, F. R. Francey, P. J. Fraser, R. L. Langenfelds, N. B. A. Trivett, andI. Levin, Sulfur hexafluoride–A powerful new atmospheric tracer,Atmosph. Environ., 30,1621–1629, 1996.

BIBLIOGRAPHY 153

Maiss, M. L., and I. Levin, Global increase of SF6 observed in the atmosphere,Geophys. Res.Lett., 21, 569–572, 1994.

Maltby, E., and P. Immirzi, Carbon dynamics in peatland and other wetland soils, regionaland global perspectives,Chemosphere, 27, 999–1023, 1993.

Maltby, E., and M. C. F. Proctor, Peatlands: Their nature androle in the biosphere, inGlobalpeat resources, edited by E. Lappalainen, pp. 11–20, International Peat Society, 1996.

Martikainen, P. J., The fluxes of greenhouse gases CO2, CH4, and N2O in northern peatlands,in Global peat resources, edited by E. Lappalainen, pp. 29–36, International Peat Society,1996.

Martin, R. S., H. Westberg, E. Allwine, L. Ashman, E. C. Farmer, and B. Lamb, Measure-ments of isoprene and its atmospheric oxidation products ina central Pennsylvania decid-uous forest,J. Atmos. Chem., 13, 1–32, 1991.

Martinerie, P., G. Brasseur, and C. Granier, The chemical composition of ancient atmo-spheres,J. Geophys. Res., 100, 14,291–14,304, 1995.

Martius, C., P. M. Fearside, A. G. Bandeira, and R. Wassmann,Deforestation and methanerelease from termites in Amazonia,Chemosphere, 33, 517–536, 1996.

Matthews, E., Global vegetation and land use: New high-resolution data bases for climatestudies,J. Clim. Appl. Meteorol., 22, 474–487, 1983.

Matthews, E., and I. Fung, Methane emissions from natural wetlands: Global distribution,area, and environmental characteristics of sources,Global Biogeochem. Cycles, 1, 61–86,1987.

Matthews, E., I. Fung, and J. Lerner, Methane emission from rice cultivation: Geographicand seasonal distribution of cultivated areas and emissions,Global Biogeochem. Cycles, 5,3–24, 1991.

Menke, W.,Geophysical data analysis: Discrete inverse theory, Int. Geophys. Ser., vol. 45,Acad. Press, San Diego, CA, 1989.

Midgley, P. M., The production and release to the atmosphereof 1,1,1-trichloroethane(Methyl Chloroform),Atmos. Environ., 23, 2663–2665, 1989.

Midgley, P. M., and A. McCulloch, The production and global distribution of emissions tothe atmosphere of 1,1,1-trichloroethane (methyl chloroform), Atmos. Environ., 29, 1601–1608, 1995.

Minnen, J. G., K. Klein Goldewijk, and R. Leemans, The importance of feedback processesand vegetation transition in the terrestrial carbon cycle,J. Biogeogr., 22, 805–814, 1996.

Moortgat, G. K., W. Klippel, K. H. Mobius, W. Seiler, and P. Warneck, Laboratory measure-ments of photolytic parameters for formaldehyde,Tech. Rep. Rep. FAA-EE-80-47, U.S.Dep. of Transp., Washington, D. C., 1980.

Morris, R. A., T. M. Miller, A. A. Viggiano, J. F. Paulson, S. Salomon, and G. Reid, Effectsof electron and ion reactions on atmospheric lifetimes of fully fluorinated compounds,J.Geophys. Res., 100, 1287–1294, 1995.

154 BIBLIOGRAPHY

Moxim, W. J., H. L. II, and P. S. Kasibhatla, Simulated globaltropospheric PAN: Its transportand impact on NOx, J. Geophys. Res., 101, 12,621–12,638, 1996.

Muller, J. F., Images: A three-dimensional chemical transport model of the global tropo-sphere,J. Geophys. Res., 100, 16,445–16,490, 1995.

Mulquiney, J. E., and J. A. Norton, A new inverse method for trace gas flux estimation:State-space model identification and constraints,J. Geophys. Res., 103, 1417–1427, 1998.

Mulquiney, J. E., J. A. Taylor, A. J. Jakeman, J. P. Norton, and R. G. Prinn, A new inversemethod for trace gas flux estimation: Application to tropospheric CFCl3 fluxes,J. Geophys.Res., 103, 1429–1442, 1998.

Nakazawa, T., T. Machida, M. Tanaka, Y. Fujii, S. Aoki, and O.Watanabe, Differences ofthe atmospheric CH4 concentration between the Arctic and Antarctic regions in preindus-trial/agricultural era,Geophys. Res. Lett., 20, 943–946, 1993.

Novelli, P. C., L. P. Steele, and P. P. Tans, Mixing ratios of carbon monoxide in the tropo-sphere,J. Geophys. Res., 97, 20,731–20,750, 1992.

Olivier, J. G. J., A. F. Bouwman, J. J. M. Berdowski, C. Veldt,J. P. J. Bloos, A. J. H. Viss-chedijk, C. W. M. van der Maas, and P. Y. J. Zandveld, Sectoralemission inventories ofgreenhouse gases for 1990 on a per country basis as well as on 1◦x1◦, Environmental Sci-ence and Policy, 2, 241–263, 1999.

Olivier, J. G. J., et al., Description of EDGAR version 2.0: Aset of global emission inven-tories of greenhouse gases and ozone depleting substances for all anthropogenic and mostnatural sources on a per country basis on 1◦ x 1◦ grid, RIVM Rep. 771060 002/TNO MEPRep. R96/119, Natl. Inst. for Public Health and the Environ., Bilthoven,The Netherlands,1996.

Olsen, J., et al., Results from the Intergovernmental Panelof Climatic Change photochemicalmodel intercomparison (Photocomp),J. Geophys. Res., 102, 5979–5991, 1997.

Olson, J. S., J. A. Watts, and L. J. Allison, Carbon in live vegetation of major world ecosys-tems,ORNL 5862, Oak Ridge Lab., Oak Ridge, Tenn., 1983.

Oltmans, S. J., and H. Levy II, Ozone measurements from a global network of surface sites,ozone in the troposphere and stratosphere, part 2, inProceedings of the quadrennial ozonesymposium, pp. 19–24, NASA Conf. publ. 3266, 1992.

Oltmans, S. J., and H. Levy II, Surface ozone measurements from a global network,Atmos.Environ., 28, 9–24, 1994.

Oltmans, S. J., W. D. Komhyr, P. R. Franchois, and W. A. Matthews, Tropospheric ozone:Variations from surface and ECC ozonesonde observations, in Ozone in the Atmosphere,edited by R. D. Bojkov and P. Fabian, A. Deepak, Hampton, Va.,1989.

Paulson, S. E., and J. H. Seinfeld, Development and evaluation of a photooxidation mecha-nism for isoprene,J. Geophys. Res., 97, 20,703–20,715, 1992.

BIBLIOGRAPHY 155

Pavelin, E. G., C. E. Johnson, S. Rughooputh, and R. Toumi, Evaluation of pre-industrialsurface ozone measurements made using Schonbein’s method, Atmospheric Environment,33, 919–929, 1999.

Penkett, S. A., The seasonal variation of nonmethane hydrocarbons in the free troposphereover the North Atlantic Ocean: Possible evidence for extensive reaction of hydrocarbonswith the nitrate radical,J. Geophys. Res., 98, 2865–2886, 1993.

Petit, J. R., et al., Climate and atmospheric history of the past 420,000 years from the Vostocice core, Antarctica,Nature, 399, 429–436, 1999.

Pinto, J. P., and M. A. K. Khalil, The stability of tropospheric OH during ice ages,Tellus, Ser.B, 43, 347–352, 1991.

Plumb, R. A., and X. Zheng, Source determination from trace gas observations: An orthog-onal function approach and results for long-lived gases with surface sources,J. Geophys.Res., 101, 18,569–18,585, 1996.

Poppe, D., et al., Gas-phase reactions in atmospheric chemistry and transport models:A model intercomparison,Tech. rep., EUROTRAC, Garmisch-Partenkirchen, Germany,1996.

Potter, C. S., An ecosystem simulation model for methane production and emission fromwetlands,Global Biogeochem. Cycles, 11, 495–506, 1997.

Prather, M., M. McElroy, S. Wofsy, G. Russell, and D. Rind, Chemistry of the global tropo-sphere: Fluorocarbons as tracers of air motion,J. Geophys. Res., 92, 6579–6613, 1987.

Prinn, R. G., R. F. Weiss, B. R. Miller, F. N. Alyea, D. M. Cunnold, D. E. Hartley, P. B. Fraser,and P. G. Simmonds,Global weighted-average concentration and trend of OH based on15 years of ALE/GAGE CH3CCl3 data, Int. Assoc. of Meteorol. and Atmos. Phys., Fuji-Yoshida, Jpn., 1994, paper presented at Joint Meeting on Global Atmospheric Chemistry.

Prinn, R. G., R. F. Weiss, B. R. Miller, J. Huang, F. N. Alyea, D. M. Cunnold, P. B. Fraser,D. E. Hartley, and P. G. Simmonds, Atmospheric trends and lifetime of CH3CCl3 andglobal average hydroxyl radical concentrations based on 1978–1994 ALE/GAGE mea-surements,Science, 269, 187–192, 1995.

Prinn, R. G., et al., Global average concentration and trendfor hydroxyl radical deductionfrom ALE/GAGE trichloroethane (methyl chloroform) data for 1978-1990,J. Geophys.Res., 97, 2445–2461, 1992.

Quay, P., J. Stutsman, D. Wilbur, A. Snover, E. J. Dlugokencky, and T. Brown, The isotopiccomposition of atmospheric methane,Global Biogeochem. Cycles, 13, 445–461, 1999.

Quay, P. D., et al., Carbon isotopic composition of atmospheric CH4: Fossil and biomassburning source strengths,Global Biogeochem. Cycles, 5, 25–47, 1991.

Ramonet, M., and P. Monfray, CO2 baseline concept in 3-D atmospheric transport models,Tellus, Ser. B, 48, 502–520, 1996.

Rayner, P. J., I. G. Enting, and C. M. Trudinger, Optimizing the CO2 observing network forconstraining sources and sinks,Tellus, Ser. B, 48, 433–444, 1996.

156 BIBLIOGRAPHY

Reichle, H. G., V. S. Conners, J. A. Holland, R. T. Sherrill, H. A. Wallio, J. C. Casas, E. P.Condon, B. B. Gormsen, and W. Seiler, The distribution of middle tropospheric carbonmonoxide during early October 1984,J. Geophys. Res., 95, 9845–9856, 1990.

Ridgwell, A. J., S. J. Marshall, and K. Gregson, Consumptionof atmospheric methane bysoils: A process-based model,Global Biogeochem. Cycl., 13, 59–70, 1999.

Roberts, J. M., Review article: Atmospheric chemistry of organic nitrates,Atmos. Environ.,24, 243–287, 1990.

Roberts, J. M., and R. W. Fayer, UV absorption cross section of organic nitrates of potentialatmospheric importance and estimation of atmospheric lifetimes,Environ. Sci. Technol.,23, 945–951, 1989.

Roelofs, G. J., and J. Lelieveld, Distribution and budget ofO3 in the troposphere calculatedwith a chemistry general circulation model,J. Geophys. Res., 100, 20,983–20,998, 1995.

Roelofs, G.-J., J. Lelieveld, and R. van Dorland, A three-dimensional chemistry/general cir-culation model simulation of anthropogenically derived ozone in the troposphere and itsradiative climate forcing,J. Geophys. Res., 102, 23,389–23,401, 1997.

Roemer, M., Trends of tropospheric ozone over Europe, Ph.D.thesis, Utrecht Univ., Utrecht,The Netherlands, 1996.

Roselle, S. J., The sensitivity of regional ozone modeling to biogenic hydrocarbons,J. Geo-phys. Res., 96, 7371–7394, 1991.

Russell, G., and J. Lerner, A new finite-differencing schemefor the tracer transport equation,J. Appl. Meteorol., 20, 1483–1498, 1981.

Saeki, T., T. Nakazawa, M. Tanaka, and K. Higuehi, Methane emissions deduced from a two-dimensional atmospheric transport model and surface measurements,J. Met. Soc. Jap., 76,307–324, 1998.

Sanderson, M. G., Biomass of termites and their emissions ofmethane and carbon dioxide:A global database,Global Biogeochem. Cycles, 10, 543–557, 1996.

Sandroni, S., D. Anfossi, and S. Viarengo, Surface ozone levels at the end of the nineteenthcentury in South America,J. Geophys. Res., 97, 2535–2539, 1992.

Saueressig, G., P. Bergamaschi, F. N. Crowley, H. Fischer, and G. W. Harris, Carbon kineticisotope effect in the reaction of CH4 with Cl atoms,Geophys. Res. Lett., 22, 1225–1228,1995.

Schwartz, S. E., and W. H. White, Solubility equilibria of the nitrogen oxides and oxiacids indilute aqueous solution,Adv. Environ. Sci. Eng., 4, 1–45, 1981.

Seiler, W., and R. Conrad, Contribution of tropical ecosystems to global budgets of tracegases, inThe geophysiology of Amazonia, vegetation and climate interactions, edited byR. E. Dickinson, pp. 133–140, John Wiley, New York, 1987.

Senum, G. I., Y. N. Lee, and J. S. Gaffney, Ultraviolet absorption spectrum of peroxyacetylnitrate and peroxypropionyl nitrate,J. Phys. Chem., 88, 1269–1270, 1984.

BIBLIOGRAPHY 157

Shepson, P. B., D. R. Hastie, K. W. So, H. I. Schiff, and P. Wong, Relationship between PAN,PPN and O3 at urban and rural sites in Ontario,Atmos. Environ., 26A, 1259–1270, 1992.

Shepson, P. B., K. G. Anlauf, J. W. Bottenheim, H. A. Wiebe, N.Gao, K. Muthuramu, andG. I. Mackay, Alkyl nitrates and their contribution to reactive nitrogen at a rural site inOntario,Atmos. Environ., 27A, 749–757, 1993.

Singh, H. B., L. J. Salas, and W. Viezee, Global distributionof peroxyacetyl nitrate,Nature,321, 588–591, 1986.

Singh, H. B., et al., Atmospheric measurements of peroxyacetyl nitrate and other organicnitrates at high latitudes: Possible sources and sinks,J. Geophys. Res., 97, 16,511–16,522,1992.

Singh, H. B., et al., Summertime distribution of PAN and other reactive nitrogen species inthe northern high-latitude atmosphere of eastern Canada,J. Geophys. Res., 99, 1821–1853,1994.

Spivakovski, C. M., et al., Three dimensional climatological distribution of tropospheric OH:Update and evaluation,J. Geophys. Res., in press, 1999.

Stauffer, B., G. Fischer, A. Neftel, and H. Oeschger, Increase of atmospheric methanerecorded in an Antarctic ice core,Science, 229, 1386–1388, 1985.

Stern, D. I., and R. K. Kaufmann, Estimates of global anthropogenic methane emissions1860–1993,Chemosphere, 33, 159–176, 1996.

Stockwell, W. R., R. P. Middleton, J. S. Chang, and X. Tang, The second generation regionalacid deposition model chemical mechanism for regional air quality modeling,J. Geophys.Res., 95, 16,343–16,367, 1990.

Stockwell, W. R., F. Kirchner, M. Kuhn, and S. Seefeld, A new mechanism for regionalatmospheric chemistry modeling,J. Geophys. Res., 102, 25,847–25,879, 1997.

Strand, A., and O. Hov, A two-dimensional global study of tropospheric ozone production,J.Geophys. Res., 99, 22,877–22,895, 1994.

Subak, S., Methane from the house of Tudor and the Ming dynasty: Anthropogenic emissionsin the sixteenth century,Chemosphere, 29, 843–854, 1994.

Tans, P. P., A note on isotopic ratios and the global atmospheric methane budget,GlobalBiogeochem. Cycles, 11, 77–81, 1997.

Tarantola, A.,Inverse Problem Theory, Methods for Data Fitting and Model Parameter Esti-mation, Elsevier, New York, 1987.

Thompson, A. M., J. A. Chappellaz, I. Y. Fung, and T. L. Kucsera, The atmospheric CH4increase since the last glacial maximum, 2, Interaction with oxidants,Tellus, Ser. B, 45,242–257, 1993.

Tiedke, M., A comprehensive mass flux scheme for cumulus parameterization in large-scalemodels,Mon. Weather Rev., 117, 1779–1800, 1989.

158 BIBLIOGRAPHY

Tissot, B. P., and D. H. Welte,Petroleum formation and occurrence, Springer-Verlag, NewYork, 1984.

Tobler, W., Population database of the consortium of international Earth science informationnetwork (CIESIN) and the environmental system research institute, inc. (ESRI), throughthe national center for geographic information and analysis.,NCGIA Technical Report TR-95-6, Dept. of Geography, University of Santa Barbara, Sante Barbara, California, 1995.

Tonnesen, S., and H. E. Jeffries, Inhibition of the odd oxygen production in the carbon bondfour and generic reaction set mechanisms,Atmos. Environ., 28, 1339–1349, 1994.

Trainer, M., E. Y. Hsie, S. A. McKeen, R. Tallamraju, D. D. Parrish, F. C. Fehsenfeld, andS. C. Liu, Impact of natural hydrocarbons on hydroxyl and peroxy radicals at a remote site,J. Geophys. Res., 92, 11,879–11,894, 1987.

Trampert, J., and R. Snieder, Model estimations biased by truncated expansions: Possibleartifacts in seismic tomography,Science, 271, 1257–1260, 1996.

Tyler, S. C., P. M. Crill, and G. W. Brailsford,13C/12C fractionation of methane during oxi-dation in a temperate forested soil,Geochim. Cosmochim. Acta., 58, 1625–1633, 1994.

United Nations, 1992 energy statistics yearbook,United Nations Publication Sales No. E/F.94XVII.9, Dept. for Economic and Social Information and Policy Analysis, Statistical di-vision, New York, 1994.

Valentine, D. W., E. A. Holland, and D. S. Schimel, Ecosystemand physiological controlsover methane production in northern wetlands,J. Geophys. Res., 99, 1563–1571, 1994.

Velthoven, P., and H. Kelder, Estimates of stratospheric exchange: Sensitivity to model for-mulation and horizontal resolution,J. Geophys. Res., 101, 1429–1434, 1996.

Walter, B., Development of a process-based model to derive methane emissions from naturalwetlands for climate studies, Ph.D. thesis, Fachbereich Geowissenschaften der UniversitatHamburg, Hamburg, Germany, 1998.

Walter, B. P., M. Heimann, R. D. Shannon, and J. R. White, A process-based model to derivemethane emissions from natural wetlands,Geophys. Res. Lett., 23, 3137–3734, 1996.

Wang, Y., and D. J. Jacob, Anthropogenic forcing on the tropospheric ozone and OH sincepreindustrial times,J. Geophys. Res., 103, 31,123–31,135, 1998.

Warneck, P., Chemistry of the Natural Atmosphere,Int. Geophys. Ser., vol. 41, p. 757, Acad.Press, San Diego, Calif., 1988.

Wesely, M. L., Parameterization of surface resistance to gaseous dry deposition in regional-scale numerical models,Atmos. Environ., 23, 1293–1304, 1989.

Westermann, P., Temperature regulation of methanogenesisin wetlands,Chemophere, 26,321–328, 1993.

Whittlestone, S., S. D. Schery, and Y. Li, Pb-212 as a tracer for local influence on air samplesat Mauna Loa observatory,J. Geophys. Res., 101, 14,777–14,785, 1996.

BIBLIOGRAPHY 159

Wille, U., E. Becker, R. N. Schindler, I. T. Lancer, G. Poulet, and G. LeBras, A dischargeflow mass-spectrometric study of the reaction between the NO3 radical and isoprene,J.Atmos. Chem., 13, 183–193, 1991.

WMO, GAW data, Greenhouse gases and other atmospheric gases, WMO WDCGG data cat-alogue WDCGG No. 13, World Meteorological Organization World data centre for green-house gases, Japan Meteorological Agency, Tokyo, 1997.

World Meteorological Organization,Scientific assessment of ozone depletion 1994: Globalozone research and monitoring project, Geneva, Switzerland, rep. 37 ed., 1995.

Worthy, D. E. J., I. Levin, N. B. A. Trivett, A. J. Kuhlmann, J.F. Hopper, and M. K. Ernst,Seven years of continuous methane observations at a remote boreal site in Ontario, Canada,J. Geophys. Res., 103, 15,995–16,007, 1998.

Yao, H., Y. B. Zhuang, and Z. L. Chen, Estimation of methane emission from rice paddies inmainland China,Global Biogechem. Cycles, 10, 641–649, 1996.

Yienger, J. J., and H. Levy, Empirical model of global soil-biogenic NOx emissions,J. Geo-phys. Res., 100, 11,447–11,464, 1995.

Yokouchi, Y., Seasonal and diurnal variation of isoprene and its oxidation products in a semi-rural area,Atmos. Environ., 28, 2651–2658, 1994.

Zimmerman, P. R., et al., Measurements of atmospheric hydrocarbons and biogenic emissionfluxes in the Amazon boundary layer,J. Geophys. Res., 93, 1407–1416, 1988.

Summary

The research presented in this thesis focuses on the quantification of sources and sinks of at-mospheric methane and the reduction of the associated uncertainties. For this purpose exist-ing information is used of the involved processes and measurements of atmospheric methaneconcentrations. Rice paddies of Southeast Asia receive special attention since these are par-ticularly important and highly uncertain sources of methane. Like CO2 and N2O, methane isan important greenhouse gas in the Earths atmosphere. Nowadays, an important fraction ofthis methane is emitted by anthropogenic sources. Therefore it is to be expected that futureconcentration changes will be strongly related to changes in human activities. To predict thesechanges and exert an influence if needed, a solid understanding of the involved mechanismsis essential.

The inverse modeling technique is well suited for source quantification and uncertaintyreduction. By this method it is computed what size sources and sinks should have in order toobtain agreement between simulated and observed concentrations. In addition, the computedsources strengths should be in agreement with the current understanding of the underlyingprocesses (“a priori” information). To improve the a prioriinformation by means of mea-sured concentrations, relationships are needed between changes of sources and the resultingchanges of atmospheric concentrations. To compute these relationships we make use of anatmospheric transport model. The successfulness of this method depends on various factorsas:

• The availability of measurements and a priori information.

• The accuracy of the atmospheric transport model.

• The applied inverse modeling technique.

In addition, estimates of uncertainties are of crucial importance since they determine theweights that are received by each information element (measurement of a priori information)in this analysis. This Ph.D. thesis aims at improving all these aspects.

In the first chapter a new parameterization of photo-chemistry is described and tested. Arealistic representation of photo-chemistry is of importance since the chemical reaction ofmethane with OH constitutes the most important sink of atmospheric methane. As an ex-tension to current parameterizations, a scheme has been developed that represents chemicalinteractions of nonmethane hydrocarbon compounds (NMHC).In this scheme, as an im-portant improvement compared with other schemes, NMHC are described with a minimumnumber of compounds and reactions. This is relevant since the computational requirementsof these schemes are relatively large.

Test show that the new reduced scheme still yields reliable results, with deviations frommore extensive schemes within 50%. The differences betweencomputed OH levels are less,

161

162 Summary

on average smaller than 10%. Our computations indicate thatthe chemistry of NMHC hasimportant effects on the global distribution of the OH radical, leading to a reduction overthe continents and an increase over the oceans. Comparisonsbetween measured and simu-lated O3 concentrations indicate that tropospheric chemistry is simulated more realistically ifNMHC chemistry is accounted for. However, comparisons between measured in computedNMHC concentrations indicate that the reduction of OH over the continents is overestimated,likely related to the chemistry of isoprene.

The second chapter presents source and sink estimates of CH4 derived from inverse mod-eling, by using the adjoint technique. This technique allows to distinguish a relatively largenumber of different sources and sinks. In previous studies these fluxes were defined as aver-ages over relatively large regions only (scale of continents), for lack of this possibility. Thedisadvantage is that it must then be assumed that the distributions of the fluxes over theseregions are perfectly well known. With the adjoint technique the size of regions can be re-duced to the spatial resolution of the model (grid size). Although for the applied model theseregions remain quite large (∼800 x 1000km), still this leads to a more realistic representationof methane sources and sinks.

An important conclusion of this study is that the source estimates derived from inversemodeling are sensitive to model errors, such as errors in theassumed OH distribution, and theexchange of CH4 between the Northern and the Southern Hemisphere. Differences betweenthe a priori and computed (a posteriori) CH4 source strengths can even largely be explainedby such model errors. This applies particularly to the globally integrated source strengthand the north-south distribution. To reduce these errors, except for more realistic processparameterizations, it requires more accurate validation techniques.

Further, it is concluded that sources can be described more realistically if the uncertaintyof a priori information is quantified better. This pertains particularly to correlations of uncer-tainties in space and time. For example, if it is known that particular sources have no seasonalvariation, it can be required that such sources, a priori emitting at a constant rate throughoutthe year, can only be adjusted such that this is also true for the a posteriori estimate. Thisis equivalent to assuming that the uncertainties of these sources are positively correlated intime (R=1 in this case). At present, the information about such correlations is scarce, but, ifavailable, it would make an important contribution to the quantification of sources and sinks.

Chapter 3 focuses on methods of comparing measured and modelcalculated CH4 concen-trations. The idea behind this is that information on sources, as present in the measurements,should be utilized as efficient as possible. By “efficient as possible” it is meant that thevariability of the measurements should be utilized as long as it can be represented by themodel. Since the model has a limited spatial and temporal resolution, model results cannotbe compared with variations of measured concentrations that are caused by local or “subgrid” sources (sources that are not explicitly described bythe model). If this requirement isnot satisfied, i.e. differences between model and measurements can partly be explained bythe fact that they do not represent the same, this is regardedas a representation error. Exceptfor effects of local sources, representation errors are also caused by wind-sector selection.By this we mean that at some stations samples are taken only ifthe wind is from a relativelyclean direction, with the aim to measure background conditions only. Methods are presentedto quantify, and possibly minimize, the effects of wind-sector selection and local sources.

Comparisons of measurements and model calculations indicate that the usual weekly sam-pling interval is insufficient to represent the monthly concentration distribution at many sta-tions. If this is the case, model and measurements can betterbe compared if the model is

Summary 163

“sampled” at the same times as the measurements are taken. Usually, monthly mean modelconcentrations are obtained by averaging the computed concentrations after each model timestep.

The quantification of rice paddy emissions turns out to be more complicated than antici-pated. In southeast Asia, except rice paddies, other important sources of methane are activeas coal mining and cattle-breeding. To estimate the emissions of rice paddies, these differentsources should be distinguishable. As expected, this requires a high model resolution. It wasnot expected, however, that the available number of measurements would be insufficient. Thisis true for the number of measurement sites as well as the applied sampling frequency. Com-parisons between model calculated and measured concentrations over the South China Seaindicate, however, that emissions from the Asian continentare overestimated by the model.Results of marked tracer simulations (each source is assigned a different color of CH4) showthat these overestimates or most likely related to rice paddy emissions.

In Chapter 4 pre-industrial simulations of methane are presented with the aim to re-duce uncertainties of natural sources, particularly, natural wetlands. Because anthropogenicsources were relatively unimportant a few centuries ago, the pre-industrial level of methaneis mainly explained by natural processes. Further, it is assumed that the source strength ofnatural sources has remained rather constant since the beginning of industrialization. Pre-industrial methane concentrations can be derived from ice cores from Greenland and Antarc-tica. Based on these measurements and estimates of other sources and sinks, the pre-industrialnatural wetland source strength is estimated at 163±30 Tg(CH4) yr−1. Current emissionsare possibly∼10% smaller as a result of inundation and cultivation of natural wetlands.

Further, information about the pole-to-pole concentration difference and the13C/12C iso-topic ratio can be used to constrain natural wetland emissions. Particularly the latter couldmake a potentially important contribution to the reductionof uncertainties of pre-industrialsources. “Potentially” because reliable methods to determine pre-industrial isotopic ratiosare lacking at present. Finally, we show that the model is able to largely reproduce the ob-served exponential growth of CH4 in the period 1750–1990 A.D. These computations indicatethat historic emission inventories underestimate anthropogenic sources in the 20th century by∼10%, although this is within the uncertainties of possible changes of OH during this century.

Samenvatting

Het onderzoek in dit proefschrift richt zich op het schattenvan de bronnen en verwijdering-processen (sinks) van atmosferisch methaan (CH4) en het reduceren van de hieraan gerela-teerde onzekerheden. Hierbij wordt gebruik gemaakt van beschikbare kennis van de betrok-ken processen en metingen van methaanconcentraties in de atmosfeer. Bijzondere aandachtgaat uit naar emissies van rijstvelden in de regio Zuidoost Azie, die een belangrijke en onze-kere bron van methaan vormen. Methaan is, naast bijvoorbeeld CO2 en N2O, een belangrijkbroeikasgas in de atmosfeer. Tegenwoordig is dit methaan voor een belangrijk deel afkom-stig van antropogene bronnen. Het is dan ook te verwachten dat mogelijk toekomstige con-centratieveranderingen grotendeels bepaald zullen worden door veranderingen in menselijkeactiviteiten. Om deze veranderingen te kunnen voorspellenen hierop, indien nodig, invloeduit te kunnen oefenen, is kennis van de betrokken processen van essentieel belang.

Een geschikte techniek om bronsterktes te schatten en onzekerheden te reduceren is in-verse modellering. Daarbij wordt berekend hoe groot de bronnen moeten zijn om de gemetenmethaanconcentraties in de atmosfeer te kunnen verklaren.Daarnaast moeten de berekendebronsterktes in overeenstemming zijn met de huidige kennisvan de betrokken processen (“a-priori” informatie). Om met behulp van concentratiemetingen de a-priori informatie te ver-beteren moeten relaties bekend zijn tussen veranderingen in bronsterktes en veranderingen inatmosferische concentraties als gevolg hiervan. Om deze relaties te berekenen wordt gebruikgemaakt van een model dat transportprocessen in de atmosfeer simuleert. Het succes vandeze methode hangt af van verschillende factoren zoals:

• De beschikbaarheid van metingen en a-priori kennis.

• De betrouwbaarheid van het atmosferisch transport model.

• De toegepaste inverse modelleringstechniek.

Daarnaast zijn schattingen van onzekerheden van essentieel belang, omdat die bepalen hoe-veel gewicht elk afzonderlijk informatie-element (metingof a-priori kennis) krijgt in dezeanalyse. Dit proefschrift richt zich op al deze aspecten.

In hoofdstuk 2 wordt een nieuwe numerieke berekeningsmethode van troposferische fo-tochemie beschreven en getest. Een realistische beschrijving van fotochemie is belangrijkomdat de chemische reactie van methaan met het OH radicaal inde atmosfeer het belangrijk-ste verwijderingmechanisme vormt voor atmosferisch methaan. Als uitbreiding op gangbaremethoden in mondiale modellen is een schema ontwikkeld dat de chemische interacties vanhogere koolwaterstoffen (>C1) beschrijft. Het nieuwe schema is afgeleid van een schema datspecifiek bedoeld is om fotochemie in verstedelijkte gebieden te simuleren. Dit schema is

165

166 Samenvatting

zodanig aangepast dat ook achtergrondomstandigheden realistisch worden beschreven. Eenbelangrijke verbetering ten opzichte van andere schema’s is dat hogere koolwaterstoffen meteen gering aantal componenten en reacties worden beschreven. Dit is van belang omdat be-rekeningen met hogere koolwaterstoffen relatief veel rekentijd kosten.

Het nieuwe gereduceerde schema levert nog steeds betrouwbare resultaten, met afwij-kingen binnen de 50% ten opzichte van uitgebreidere schema’s. De verschillen voor OHzijn nog kleiner, gemiddeld minder dan 10%. De chemie van hogere koolwaterstoffen heefteen belangrijke invloed op de atmosferische verdeling van het OH radicaal, wat leidt tot eenverlaging boven de continenten en een verhoging boven de oceanen. Vergelijkingen tussengemeten en model berekende O3 concentraties wijzen erop dat troposferische chemie rea-listischer gesimuleerd wordt als interacties met hogere koolwaterstoffen worden verrekend.Daarentegen duiden vergelijkingen van gemeten en berekende koolwaterstof concentratiesop te sterke OH reducties over de continenten, waarschijnlijk gerelateerd aan de oxidatie vanisopreen.

Hoofdstuk 3 presenteert bronschattingen van CH4 op basis van inverse modellering, ge-bruikmakend van de zogenaamde adjoint-techniek. Met deze techniek is het mogelijk omeen relatief groot aantal verschillende bronnen en sinks teonderscheiden. Voorheen werden,bij gebrek aan deze mogelijkheid, slechts bronnen beschreven die representatief waren voorgrote gebieden (schaal van continenten). Hierbij werd aangenomen dat de verdeling van debronnen over elk gebied exact bekend is. Bij de adjoint-methode kan de afmeting van bron-nen gereduceerd worden tot het ruimtelijk oplossend vermogen van het model. Hoewel ditbij het toegepaste model nog steeds resulteert in vrij groteregio’s (∼800 x 1000km), wordenmethaan bronnen en sinks hierdoor toch aanmerkelijk realistischer beschreven.

Een belangrijke conclusie van hoofdstuk 3 is dat de resultaten van inverse modelleringgevoelig zijn voor modelfouten, zoals fouten in de aangenomen verdeling van OH, en de uit-wisseling van CH4 tussen het noordelijk en zuidelijk halfrond. Verschillen tussen de a-priorien berekende (a-posteriori) CH4 bronsterkte kunnen zelfs in belangrijke mate verklaard wor-den door zulke modelfouten. Dit geldt in het bijzonder voor de mondiaal geintegreerde bron-sterkte en de noord-zuid verdeling. Om deze fouten te reduceren zijn, behalve realistischereprocesbeschrijvingen, ook nauwkeurigere validatietechnieken nodig.

Een tweede belangrijke conclusie is dat bronnen realistischer kunnen worden geschat alsonzekerheden met betrekking tot a-priori kennis beter gekwantificeerd zijn. Dit geldt in hetbijzonder voor de correlaties van deze onzekerheden in ruimte en tijd. Als bijvoorbeeld be-kend is dat bepaalde bronnen nagenoeg geen seizoensvariatie vertonen kan worden vereistdat zulke bronnen, die a-priori constant emitteren, alleenzodanig aangepast worden dat dita-posteriori nog steeds geldt. Dit is equivalent met de aanname dat de onzekerheden van dezebronnen positief gecorreleerd zijn in de tijd (R=1 in dit geval). Kennis van zulke correla-ties is in het algemeen schaars, maar zou, indien beschikbaar, het schatten van emissies inbelangrijke mate kunnen verbeteren.

Hoofdstuk 4 gaat dieper in op het vergelijken van gemeten en met het model berekendeCH4-concentraties. Het achterliggende idee is dat informatieover bronnen, zoals aanwezigin metingen, zo efficient mogelijk moet worden benut. “Zo efficient mogelijk” betekent hierdat de variabiliteit in de metingen moet worden gebruikt zolang het model in staat is die tereproduceren. Aangezien het model een zekere minimale ruimtelijke en temporele schaalrepresenteert, kunnen modelresultaten bijvoorbeeld nietvergeleken worden met variaties ingemeten concentraties die veroorzaakt worden door lokale of wel “subgrid” bronnen (bron-nen die niet expliciet beschreven worden door het model). Als aan deze voorwaarde niet

Samenvatting 167

voldaan is en het verschil tussen modelresultaten en metingen deels veroorzaakt wordt door-dat beiden niet hetzelfde representeren, dan wordt dit beschouwd als een representatiefout.Representatieouten ontstaan, behalve door effecten van locale bronnen, ook door windsectorselectie. Hiermee wordt bedoeld dat op sommige stations alleen monsters worden genomenals de wind uit een richting waait waar geen belangrijke vervuilingsbronnen voorkomen, methet doel zoveel mogelijk de achtergondconcentratie te meten. In hoofdstuk 4 worden metho-den gepresenteerd om de effecten van windrichtingselectieen locale bronnen te kwantificerenen, zo mogelijk, te minimaliseren.

Uit vergelijkingen tussen model berekeningen en metingen blijkt dat de gangbare weke-lijkse monstername op veel stations niet voldoende is om de maandelijkse concentratiever-deling weer te geven. In dit geval kunnen model en metingen beter vergeleken worden doorhet model op dezelfde tijdstippen te “bemonsteren” als de echte atmosfeer. Voorheen washet gebruikelijk om maandgemiddelde concentraties te berekenen op basis van de berekendeconcentratie na elke modeltijdstap.

Het kwantificeren van emissies uit rijstvelden door middel van inverse modellering blijktgecompliceerder te zijn dan aanvankelijk werd aangenomen.De regio Zuidoost Azie kent,naast rijstvelden, nog andere belangrijke bronnen van methaan, zoals kolenmijnen en veeteelt.Om de emissies van rijstvelden te kunnen schatten moeten deze verschillende bronnen vanelkaar onderscheiden kunnen worden. Dat hiervoor een hoge model resolutie nodig is wasvoorzien, maar niet dat het aantal beschikbare metingen hiervoor niet toereikend is. Dit geldtzowel voor het aantal meetstations in de regio Zuidoost Azi¨e als de toegepaste meetfrequen-tie. Vergelijkingen tussen model berekende en gemeten concentraties over de ZuidchineseZee geven wel aan dat de emissies van het Aziatische continent overschat worden. Door mid-del van modelberekeningen waarbij gebruik gemaakt wordt van gemarkeerde tracers (elkebron emitteert een eigen “kleur” methaan) wordt afgeleid dat deze overschattingen hoogst-waarschijnlijk gerelateerd zijn aan rijstveldemissies.

In hoofdstuk 5 wordt de pre-industriele atmosfeer gesimuleerd met als doel de onze-kerheid van natuurlijke bronnen, in het bijzonder natuurlijke wetlands (zoals moerassen), teverkleinen. Aangezien antropogene bronnen enkele eeuwen geleden relatief onbelangrijkwaren, wordt de pre-industriele methaanconcentratie voornamelijk verklaard door natuur-lijke processen. De aanname die hierbij gemaakt wordt is datde bronsterkte van natuurlijkewetlands sinds het begin van de industriele ontwikkeling nagenoeg constant is gebleven. Pre-industriele CH4-concentraties kunnen afgeleid worden uit de methaanconcentratie in gas datopgesloten is in het ijs van Groenland en Antarctica. Op basis van deze metingen en schattin-gen van andere bronnen en sinks wordt een pre-industriele natuurlijke wetland bron berekendvan 163±30 Tg(CH4) j−1. Huidige emissies zijn mogelijk∼10% lager als gevolg van hetdroogleggen en in cultuur brengen van natuurlijke wetlands.

Verder kan kennis worden ontleend aan het concentratieverschil tussen Groenland enAntarctica en de13C/12C isotoopverhouding van CH4. Vooral dit laatste kan een potenti-eel belangrijke bijdrage leveren aan het reduceren van de onzekerheden van pre-industrielebronnen. “Potentieel” omdat op dit moment geen betrouwbaremethoden bestaan om pre-industriele isotoopverhoudingen te meten. Tot slot blijkt dat het model grotendeels in staatis om de waargenomen exponentiele groei van CH4 over de periode 1750–1990 A.D. te re-produceren. Verder volgt uit deze berekeningen dat historische emissie-inventarisaties deantropogene bronnen gedurende de 20e eeuw met∼10% onderschatten, hoewel dit binnen deonzekerheid valt van mogelijke OH verandering gedurende deze eeuw.

Nawoord

Het proefschrift is af! Een jaar geleden zag de planning er nog tamelijk onrealistisch uit.Zeker afgaande op de produktiviteit van de jaren ervoor. Zo zie je maar weer hoe onzekerextrapolaties zijn (sorry, vanaf nu geen wetenschap meer).Waarschijnlijk heeft die planninger juist voor gezorgd dat het nu klaar is, ik ken mezelf, een beetje pressie kan geen kwaad.Niet dat ik in de jaren ervoor mijn tijd heb lopen verkwanselen, maar op het laatst kwamen depuzzelstukjes aardig bij elkaar, ofwel konden investeringen van voorafgaande jaren verzilverdworden. Ik kijk daarom met tevredenheid terug op de achterliggende periode en realiseer metegelijkertijd dat ik het voor een belangrijk deel te dankenheb aan anderen. Ik maak daaromgraag van deze gelegenheid gebruik om een aantal mensen te bedanken, die dat meer dantoekomt.

Allereerst Jos, je bent erin geslaagd je op te stellen als ideale promotor. Enerzijds gundeje me veel vertrouwen en vrijheid om mijn eigen ideeen uit teproberen. Anderzijds, op mo-menten dat belangrijke beslissingen genomen moesten worden of het onderzoek vast dreigdete lopen was je standpunt heel duidelijk. Hoewel ik me daar opsommige van die momentenmoeilijk in kon vinden, bleken die achteraf inderdaad verreweg de beste opties te zijn. Het isinspirerend om te zien met hoeveel energie je je op een heel breed vakgebied stort, en hoeveeldat in een korte tijd als hoogleraar al opgeleverd heeft. Datis iets wat aanstekelijk werkt enook in belangrijke mate bijgedragen heeft aan mijn motivatie.

Frank, ik kan me herinneren dat ik jou ooit in Wageningen voorstudent heb uitgemaakt.Toch was ik in die tijd vast niet de enige die je voor een erg slimme student aanzag. De bij-voeglijke naamwoorden kloppen trouwens wel, niet alleen als het gaat om wetenschappelijkeideeen maar vooral ook om een nogal ad-remme opmerkingsgave. Ik was nogal eens op-zoek naar bevestigingen of wat ik produceerde wel wetenschappelijk te verantwoorden viel.Daarin heb jij erg nuttige bijdragen geleverd, verrassend genoeg ook wanneer het onderwerpniet bepaald op je eigen terrein lag. Ik heb het erg getroffenmet jou als kamergenoot enco-promotor.

Richard, het volgende citaat drukt jouw bijdrage aan het proefschrift duidelijk uit “Thebest source of information, however, is often other experienced users” (A Guide to Latex2ε).Zo iemand heb je dus nodig en in mijn geval vervulde jij de rol van Latex Goeroe. Ik weeteigenlijk amper hoe ik je passend kan bedanken, want jouw hulp heeft me heel erg veelellende bespaard. Misschien kan ik het goed maken door de layout van dit proefschrift aan jeop te dragen.

Geertjan, Maarten en Bram keken iets meer vanaf de zijlijn mee maar stonden altijd klaarom gegevens aan te leveren en mee te denken. De laatste twee ook in het bijzonder bedanktvoor het op peil houden van mijn fysieke conditie, een beetjepressie woensdags om 12 uurkon geen kwaad. De eerste twee bedankt voor een onvergetelijke uitvoering van Slade’s“Merry X-mas Everybody” in het Onderonsje. Verder alle medeAIO’s bedankt voor het

169

170 Nawoord

prettige werkklimaat en geslaagde chemie etentjes. In het bijzonder Bert en Laurens voor hetmeeliften naar Wageningen, al dan niet met gierende V-snaar.

Hugo Denier van der Gon en Nico van Breemen, voor een verfrissende kijk vanuit eenandere discipline. Ik heb bewondering voor de inspanning die jullie leverden om over degrenzen van je eigen vakgebied te kijken en iets gezamelijksop te zetten. Hetzelfde geldtook voor Peter van Bodegom en Peter Verburg. Het dwong mij om geen denkstappen overte slaan, wat uiteindelijk hielp om m’n eigen ideeen scherper voor ogen te krijgen. Hopelijkheeft het jullie iets gebracht, mij in ieder geval wel.

Ed Dlugokencky, one can rightly question what the results ofnumerical models reallymean. One thing is sure, without measurements the answer would be easy. Even if measure-ments are available, the question remains whether comparisons with model results make anysense. With your help I was able to avoid a number of pittfalls. Further, your openness inplacing data to everyones disposal should serve as a good example too others.

Thomas Kaminksi en Martin Heimann, without your help I wouldn’t have made muchprogress in inverse modelling compared with my predecessors. Particularly, I want to thankThomas for his patient explanations of mathematical principles. If you ever decide to quitscience, you’ll still be able to make it as a good teacher.

Verder wil ik Marcel Portanger en Piet Jonker bedanken voor de computer- en netwerk-ondersteuning. Alles in dit proefschrift staat of valt met de beschikbaarheid van computers.Een gedachte om nachtmerries van te krijgen, maar door jullie inspanningen kwamen diegelukkig nooit uit.

Last but not least Sandra, je wilt dit wel niet maar ik doe het toch. Als er iemand heeftmoeten zuchten onder mijn krappe planningen dan ben jij het.Zeker in de wetenschap datje eigen stuwmeer aan vrije dagen als maar erger aanzwelde. Zo tegen deadlines kon ik vrijmeedogenloos zijn in mijn prioriteiten. Dat je, ondanks datalles, achter me bent blijven staansiert je. Dat is me veel waard (meer dan een proefschrift).

Curriculum Vitae

Sander Houweling is geboren op 18 maart 1970 te Hilversum. In1988 behaalde hij zijnVWO diploma aan het Christelijk College Stad & Lande te Huizen. Vanaf 1988 tot 1994volgde hij de studie Moleculaire Wetenschappen aan de Landbouw Universiteit Wageningen(tegenwoordig Wageningen University & Research Centre). Afstudeervakken werden uitge-voerd bij achtereenvolgens Fysische en Kolloıd-chemie onder leiding van Dr. Wonders enDr. van Leeuwen aan de karakterisering van lood op aerosol, Luchtkwaliteit onder leidingvan Prof. Dr. Adema aan de oxydatie van SO2 in wolkendruppeltjes en Organische Chemieonder leiding van Dr. F. Griepink en Dr. T. van Beek aan de gas-chromatografische analysevan biologische signaalstoffen in luchtmonsters. Vanaf januari tot juni 1994 volgde hij eenstage aan het Geofysisch Instituut te Bergen, Noorwegen, onder leiding van Prof. Dr. Ø. Hovmet als onderwerp het modelleren van de fotochemie van biologische koolwaterstoffen en heteffect hiervan op ozon concentraties over het Europese continent.

Van september 1994 tot september 1995 was hij werkzaam als tijdelijk onderzoeker aande vakgroep Luchtkwaliteit van de Landbouw Universiteit Wageningen en werd een beknopteparameterisatie ontwikkeld van koolwaterstof-chemie voor toepassing in globale 3D chemie-transportmodellen. Van september 1995 tot september 1999 werkte hij als assistent in op-leiding aan dit proefschrift, aanvankelijk aan de vakgroepLuchtkwaliteit, maar vanaf Januari1996 aan het Instituut voor Marine and Atmosferisch onderzoek Utrecht (IMAU), UniversiteitUtrecht. Gedurende deze periode zijn enkele werkbezoeken doorgebracht in het buitenland,namelijk aan CNRS, Gif Sur Yvette, Frankrijk, bij Dr. N. Poisson en Dr. M. Kanakidou,the Nationaal Oceanic and Atmospheric Administration (NOAA) te Boulder, VS, bij Dr. P.Tans en Dr. E.J. Dlugokencky, en Max Planck Institute (MPI) for Meteorology, Hamburg,Duitsland, bij Dr. T. Kaminski en Dr. M. Heimann.

Vanaf december 1999 is hij werkzaam als tijdelijk onderzoeker aan het IMAU, Universi-teit Utrecht.

171

global modeling of atmospheric methane sources and sinksinez/msri-ncar_carbonda/...methane is the...

Documents