Factors controlling variability in the oxidative capacity of the troposphere since the Last Glacial Maximum

Abstract. The oxidative capacity of past atmospheres is highly uncertain. We present here a new climate–biosphere–chemistry modeling framework to determine oxidant levels in the present and past troposphere. We use the GEOS-Chem chemical transport model driven by meteorological fields from the NASA Goddard Institute of Space Studies (GISS) ModelE, with land cover and fire emissions from dynamic global vegetation models. We present time-slice simulations for the present day, late preindustrial era (AD 1770), and the Last Glacial Maximum (LGM, 19–23 ka), and we test the sensitivity of model results to uncertainty in lightning and fire emissions. We find that most preindustrial and paleo climate simulations yield reduced oxidant levels relative to the present day. Contrary to prior studies, tropospheric mean OH in our ensemble shows little change at the LGM relative to the preindustrial era (0.5 ± 12 %), despite large reductions in methane concentrations. We find a simple linear relationship between tropospheric mean ozone photolysis rates, water vapor, and total emissions of NOx and reactive carbon that explains 72 % of the variability in global mean OH in 11 different simulations across the last glacial–interglacial time interval and the industrial era. Key parameters controlling the tropospheric oxidative capacity over glacial–interglacial periods include overhead stratospheric ozone, tropospheric water vapor, and lightning NOx emissions. Variability in global mean OH since the LGM is insensitive to fire emissions. Our simulations are broadly consistent with ice-core records of Δ17O in sulfate and nitrate at the LGM, and CO, HCHO, and H2O2 in the preindustrial era. Our results imply that the glacial–interglacial changes in atmospheric methane observed in ice cores are predominantly driven by changes in its sources as opposed to its sink with OH.


Introduction
The oxidative capacity of the troposphere -primarily char-35 acterized by the burden of the four most abundant and reactive oxidants (OH, ozone, H 2 O 2 , and NO 3 ) -determines the lifetime of many trace gases of importance to climate and human health, including air pollutants and the greenhouse gas methane (Fiore et al., 2012).Ozone itself ab-40 sorbs both longwave and shortwave radiation, and its tropospheric enhancement since the Industrial Revolution is estimated to have contributed the third largest radiative forcing effect of any gas over this period, after CO 2 and methane (IPCC, 2007).Oxidants also impact the evolution, lifetime, 45 and physical properties of inorganic, and primary and secondary organic aerosol particles, modulating their direct and indirect radiative effects (Liao et al., 2003;Tie et al., 2005;Leibensperger et al., 2011).The tropospheric oxidative capacity may perturb terrestrial and marine ecosystems via ox-50 idative stress and by altering the deposition of oxidized nutrients such as nitrate (Shepon et al., 2007;Sitch et al., 2007;Collins et al., 2010).Understanding how tropospheric oxidative capacity has changed over the past is critical for understanding its chemical, climatic and ecological consequences.

55
The oxidation capacity of past atmospheres, however, is uncertain.In this work, we introduce a new model framework to improve estimates of the oxidative capacity on glacialinterglacial time scales, and we examine the implications of our results for current understanding of the observed ice-core 60 record of methane.
Tropospheric oxidant levels respond to meteorological conditions, changes in emissions of key chemical species, and changes in surface and stratospheric boundary conditions (Spivakovsky et al., 2000;Lelieveld et al., 2002; Holmes sulting changes in tropospheric oxidants and identifies their controlling factors, with particular focus on OH.Section 6 evaluates and discusses the model results in the context of the ice-core record.Section 7 discusses the implications of our results for the methane record.We conclude with a summary section. 2 Methods: Project and Model Description Fig. 1 illustrates the stepwise offline coupling of the ICECAP project framework.Archived meteorology for multiple timeslices from the GISS ModelE GCM (Schmidt et al., 2006) are applied to the global vegetation models BIOME4-TG (Kaplan et al., 2006) to determine land cover characteristics and LPJ-LMfire (Pfeiffer et al., 2013) to calculate dry matter consumed by fires.The archived meteorology and land cover products are then used to drive the GEOS-Chem CTM of tropospheric composition (Bey et al., 2001), which includes online linearized stratospheric chemistry (McLinden et al., 2000).

The GISS ModelE general circulation model
The ModelE version of the GISS GCM of global climate is described in detail by Schmidt et al. (2006).The GISS GCM has a long history in investigations of past atmospheres (e.g., Webb et al., 1997;Joussaume et al., 1999;Delaygue et al., 2000;Shindell et al., 2003;Schmidt and Shindell, 2003;LeGrande et al., 2006;Rind et al., 2001Rind et al., , 2009)).We use here ModelE with 4 • latitude by 5 • longitude horizontal resolution and 23 vertical layers extending from the surface to 0.002 hPa.We archive mean meteorological fields from ModelE for input to GEOS-Chem at temporal resolution of 6 h for three-dimensional fields (e.g., winds, convective mass fluxes) and 3 h for mixing depths and surface fields (e.g., surface temperature, solar radiation, albedo).Climate in ModelE is forced by different greenhouse gas levels from ice-core measurements, orbital parameters, topography, and fixed sea ice and sea surface temperatures (SSTs) as summarized in Table 2 and discussed in Section 3.For each scenario the GCM is intialized for five years, with four subsequent years of meteorology then archived for the terrestrial vegetation and chemical transport models.We discuss the forcings applied to the GISS climate model for each of the scenarios and the resulting climate in Section 3.1.

The BIOME4-TG global vegetation model
BIOME4-TG (Kaplan et al., 2006) is an equilibrium terrestrial vegetation model that calculates the global distribution of 28 biomes and a suite of associated biogeochemical state variables and fluxes.The model is driven by climatological mean monthly climate, soil texture in two layers, and atmospheric CO 2 concentrations.BIOME4-TG has been used extensively for simulating past vegetation distributions and for evaluating paleoclimate simulations from GCMs against 225 paleoecological archives (e.g., Kaplan, 2002;Kaplan et al., 2003Kaplan et al., , 2006;;Harrison and Prentice, 2003).For this study, we drive BIOME4-TG with a standard baseline climatology (Kaplan et al., 2006) to which we add monthly mean anomalies of surface temperature, precipitation, and solar radiation 230 from the GISS ModelE paleo and preindustrial climate scenarios.The BIOME4-TG outputs passed to GEOS-Chem are biome classification and each biome's constituent plant functional types (PFTs), and aggregate leaf area index (LAI).The PFTs and LAI are used for determining terrestrial plant 235 emissions of VOCs (Section 4.2); LAI and biome classes are used in calculating dry deposition velocities in a resistancein-series model in GEOS-Chem (Wesely, 1989).For anthropogenic land use at present, we superimpose a present-day crop fraction mask from the Model of Emissions of Gases 240 and Aerosols from Nature (MEGAN) version 2.1 (Guenther et al., 2012).

2.3
The LPJ-LMfire global biomass burning model LPJ-LMfire (Pfeiffer et al., 2013) is a dynamic global vegetation model designed to simulate biomass burning from 245 natural and human-cased wildfires in preindustrial time.The model further accounts for intentional burning of agricultural land and pastures in preindustrial societies.Similarly to BIOME4-TG, we drive LPJ-LMfire with a standard baseline climatology (Pfeiffer et al., 2013) to which we add monthly 250 mean anomalies of mean temperature, diurnal temperature range, precipitation, total cloud cover fraction, surface and wind speed from the GISS ModelE scenarios.Monthly anomalies of lightning ground-strike rate determined by GEOS-Chem (Section 4.5) are also used to drive LPJ-LMfire 255 in the paleoclimate experiments.We archive monthly total dry mass of biomas burned for use in GEOS-Chem.

The GEOS-Chem chemical transport model
The GEOS-Chem global CTM (version 9-01-03, with modifications as described; http://www.geos-chem.org)260 simulates tropospheric ozone-NO x -CO-VOC-BrO x -aerosol chemistry.Emissions and their coupling with the GCM and vegetation model outputs are the subject of Section 4. The original description of the chemical mechanism is by Bey et al. (2001).The coupled sulfur-nitrate-ammonium 265 aerosol simulation is described by Park et al. (2004), with aerosol thermodynamics computed via the ISORROPIA II model (Fountoukis and Nenes, 2007) as implemented by Pye et al. (2009).Sea salt aerosol is described by Jaeglé et al. (2011), primary and secondary organic aerosol by 270 Heald et al. (2011), and black carbon by Wang et al. (2011).Recent updates of relevance for this study include those to oceanic sources and sinks of acetone (Fischer et al., 2012), optical properties of dust (Ridley et al., 2012), and snow scavenging and washout of aerosols (Wang et al., 2011).
We include online bromine chemistry (Parrella et al., 2012), whose catalytic destruction of ozone may be an important mechanism for limiting preindustrial ozone levels.
In their study of effect of climate change on surface air quality in the near future, Wu et al. (2007) coupled GEOS-Chem with the GISS III version of the GISS GCM (Rind et al., 2007).Here, we build upon the Wu et al. (2007) framework, using meteorology from GISS ModelE to drive GEOS-Chem chemistry.Primary differences between the Wu et al. (2007) implementation and this work include the prescription of land cover from an offline global terrestrial vegetation model (Sections 2.2-2.3 and 4.2), a linearized stratospheric ozone scheme, adjusted convective transport and rainout schemes to reflect the updated moist convection in ModelE (Kim et al., 2012), and implementation of the GISS quadratic upstream tracer advection scheme (Prather, 1986) for improved consistency with the GCM.We discuss these updates in detail below.
For this work, we prescribe atmospheric methane concentrations, imposing a meridional gradient as summarized in Table 3.The preindustrial and LGM simulations use observed methane values from Antarctic, Andean, and Greenland ice cores in the NOAA National Geophysical Data Center (NGDC) Paleoclimatology Program (http://www.ncdc.noaa.gov/paleo/paleo.html).Our present-day values are from flask observations of the NOAA Global Monitoring Division database (http://www.esrl.noaa.gov/gmd/).We assume methane is uniformly mixed up to the tropopause in each latitudinal band.In the present day and preindustrial, methane increases with increasing latitude in the Northern Hemisphere, reflecting its terrestrial source.The record of tropical methane exists from a single ice core and implies that tropical methane at the LGM maintained preindustrial levels.This ice core was likely contaminated by in situ methanogenic production (Campen et al., 2003), and it is unlikely that a gas with a chemical lifetime of 10 years could maintain such a large tropical-extratropical gradient.We therefore assume tropical methane to be 4% higher than recorded at Antarctica.This is similar to the present-day gradient, and consistent with earlier model studies of LGM methane (Kaplan et al., 2006).
We calculate stratospheric ozone online in GEOS-Chem following the Linoz linearized chemical scheme (McLinden et al., 2000), with coefficients appropriate for stratospheric chemistry in the LGM (Rind et al., 2009) and the preindustrial (C. McLinden, personal communication, 2009).GEOS-Chem typically uses satellite observations of total ozone columns to determine the attenuation of shortwave radiation in its calculation of tropospheric photolysis rates.Our use of Linoz allows for calculation of photolysis rates more consistent with changing climate and chemical conditions.
In addition to ozone, we also calculate an array of 20+ other species in the stratosphere to account for chemical fluxes across the tropopause.For example, the stratosphere is a net source of NO x into the troposphere and a net sink of tropospheric CO.Stratospheric concentrations of these species are calculated from monthly climatological 3D production rates and loss frequencies archived from the Global Modeling Initiative (GMI) CTM driven by MERRA assimilated meteorology for 2004-2010 (Allen et al., 2010;Duncan 335 et al., 2007;Considine et al., 2008) and re-gridded to the ModelE grid.We approximate the paleo-stratosphere by scaling the production rates of NO x and HNO 3 by the ratio of the primary stratospheric nitrogen precursor (N 2 O) observed in NOAA/NGDC ice cores (preindustrial/present = 0.83;

340
LGM/present = 0.77) (Fluckiger et al., 2002), and similarly for CO by the ratio of ice-core methane (preindustrial/present = 0.41; LGM/present = 0.32).Bromine species are the exception to the linearized treatment, and stratospheric concentrations are instead prescribed as described by Parrella et al.

345
(2012) and do not change between climates.
To reach equilibrium with respect to stratospheretroposphere exchange, each scenario was initialized over 10 years, repeatedly using the first year of archived meteorology.Simulations for analysis use the three remaining years 350 in each climate scenario.

Evaluation of present-day simulation
We evaluate the ICECAP present-day simulation against a suite of sonde, aircraft, satellite, and surface measurements of trace gases, aerosols, and radionuclides, and we also com-355 pare ICECAP with the standard GEOS-Chem driven by assimilated meteorology (GEOS4 and GEOS5).We refer the reader to the supplemental materials for the details, and summarize here.Tropospheric vertical mixing performs better than, or on par with, the GEOS simulations as inferred by 360 comparisons to observed vertical profiles of 222 Rn and surface ratios of 7 Be/ 210 Pb.Stratospheric age-of-air reaches 4-5 years by 35 km in our simulations, 1-2 years longer than either GEOS simulation, but in better agreement with observations (Waugh and Hall, 2002).We find a tropospheric resi-365 dence time for 210 Pb-containing aerosols against deposition of 7.2 d, within the range of previous model studies: 6.5-12.5 d (Turekian et al., 1977;Lambert et al., 1982;Balkanski et al., 1993;Koch et al., 1996;Guelle et al., 1998a,b;Liu et al., 2001).
The simulated present-day stratospheric columns of ozone

380
(SCO) in our model have a small mean bias of +1.5% versus a 2004-2010 climatology from satellite (Ziemke et al., 2011).However, we overestimate SCO over the poles by 5-50% due to an overly vigorous Brewer-Dobson circulation, particularly over Antarctica during austral spring (>100%) as our implementation of Linoz does not include the heterogeneous reactions of the ozone hole.In the troposphere, we underestimate tropical ozone columns by 20% and overestimate extratropical columns by 45% versus satellite climatologies (Ziemke et al., 2011).Comparison with ozonesondes indicates that the tropical bias is in the free and upper troposphere and not at the surface, and that the extratropical bias is a springtime feature in both hemispheres, consistent with overestimated transport from the stratosphere.Overly vigorous mass fluxes from the stratosphere to the troposphere have historically been a problem with CTMs (e.g., Liu et al., 2001), and some models have prescribed cross tropopause fluxes of ozone (e.g., Wu et al., 2007).Here we elect to accept the bias in the cross-tropopause flux of ozone in the present day in order to examine the role of the tropical overhead ozone columns on photolysis in different climates.

Climate forcings
Table 2 summarizes the different greenhouse gas levels, orbital parameters, topography, and fixed sea ice and sea surface temperatures (SSTs) used to force each of our four simulations.In this section and throughout the paper, we report most changes relative to the state of the preindustrial.Simulations of LGM climate are sensitive to assumptions about ice sheet topography and tropical SSTs (Rind, 2009), and these assumptions have implications for the chemical composition of the troposphere, as they affect transport, surface emissions, and the surface UV albedo.We use here the reconstruction of ice sheet thickness and extent from Licciardi et al. (1998).At the LGM, large permanent land ice sheets 3-4 km thick existed over northern North America (Laurentide and Cordilleran), Greenland, and northwestern Eurasia (Fennoscandian).The accumulation of water in terrestrial ice sheets caused sea levels to drop 120 m, exposing an additional 3% of the Earth's surface as land, particularly in Southeast Asia and Oceania, the Bering Strait, and the southeastern continental shelf of South America (Clark and Mix, 2002).
Fig. 2 shows the annual mean SST distribution used in each past simulation.SSTs in the preindustrial and presentday simulations are from the Hadley Centre sea ice and SST version 1 (HadISST1) data set (Rayner et al., 2003).For the LGM, large uncertainty exists over the extent of cooling of tropical SSTs, which strongly affects LGM dynamics because of its influence on meridional temperature gradients and low latitude circulation (Rind, 2009).To bound the range of likely conditions, we simulate two potential realizations of the LGM climate driven by different SSTs.For our "warm-LGM" simulation, we use the SST reconstruc-tion from Climate: Long range Investigation, Mapping, and 435 Prediction (CLIMAP, 1976), with an average change in SST within 15 • of the equator (∆SST 15 • S−15 • N ) of −1.2 • C. The CLIMAP reconstruction indicates that much of the subtropical Pacific was warmer in the LGM than in the preindustrial, a controversial result (e.g., Rind and Peteet, 1985).For 440 our "cold-LGM" simulation, we use SSTs from Webb et al. (1997), who achieved ∆SST 15 • S−15 • N of −6.1 • C by imposing an ocean heat transport flux in the GISS GCM.The recent Multiproxy Approach for the Reconstruction of the Glacial Ocean surface (MARGO Project Members, 2009) concluded 445 that the SST change was likely colder than that implied by the CLIMAP reconstruction (∆SST 15 • S−15 • N = −1.7 • C) with a more subdued warming of the subtropical Pacific gyres, but not as cold as in Webb et al. (1997).

450
Table 4 shows mean values for the globe and large-zonal bands of meteorological variables in each simulated climate scenario.Relative to the preindustrial, surface air temperatures are 1.4 • C and 6.8 • C colder in the warm-LGM and cold-LGM scenarios respectively, driven by the decreases in 455 greenhouse gases and SSTs and the increase in topography (Fig. 3, left column).The changes in surface temperature to the LGM are much greater in magnitude than the simulated 0.5 • C increase from the preindustrial to the present day (not shown).Tropospheric water vapor is highly sensitive to 460 temperature (e.g., Held and Soden, 2006;Sherwood et al., 2010), and tropical specific humidity in our warm-LGM and cold-LGM scenarios decrease by 10% and 38% relative to the preindustrial.The residual circulation of the stratosphere also weakens with decreasing levels of greenhouse gases in our simula-475 tions.Stratospheric circulation is characterized by net ascent in the tropics and descent in the extratropics, driven by stratospheric breaking of upward-propagating planetary waves of tropospheric midlatitude origin (Holton et al., 1995).The vertical component of the residual circulation (w * ) weakens in magnitude relative to the preindustrial by an average of 16% and 32% respectively in the warm-LGM and cold-LGM scenarios at 50 hPa in the tropics (12 • S-12 • N), and increases by 5% in the present day.There are corresponding changes in the extratropics.The mass-weighted mean stratospheric 485 age of air (Waugh and Hall, 2002) deduced from an SF 6 -like tracer in GEOS-Chem is 9-15% greater in the LGM than the preindustrial, consistent with the weakening residual circulation; the present-day age of air is 8% lower than the preindustrial.
That the strength of the residual circulation is correlated with greenhouse gas levels is a robust result of GCMs (Garcia and Randel, 2008), including at the LGM (Rind et al., 2001(Rind et al., , 2009)), although not clearly supported by observations (Engel et al., 2008).The decreased thermal wind shear at the LGM reduces the strength of the subtropical jet, which in turn weakens wave penetration into the stratosphere and leads to deceleration of the residual circulation relative to the preindustrial; the converse holds for the preindustrial to present (Shepherd and McLandress, 2011).Weakened tropical deep convection at the LGM and increased static stability about the tropopause from stratospheric radiative warming also impedes mixing across the tropopause.In addition, the tropopause height decreases in each colder climate; in the cold-LGM scenario it is 2 km lower than in the preindustrial.
The slowing of the residual circulation of the stratosphere together with decreases in tropopause height during the LGM contribute toward enhancement of the SCO and have implications for photolysis rates in the tropical troposphere, as we describe in Section 5.1.

Evaluation of simulated LGM climate
Fig. 4 demonstrates that our GISS ModelE climate simulations for the LGM are in broad agreement with other LGM model studies and an independent reconstruction of regional LGM climate from pollen records.The left panels compare the zonal mean temperature and precipitation changes of our two LGM simulations to the mean and range of those values in the models participating in the Paleoclimate Modelling Intercomparison Project Phase II (PMIP2) (Braconnot et al., 2007a,b).Each of the seven models that simulated the LGM for PMIP2 had a fully-coupled ocean-atmosphere, and did not prescribe SSTs as we do here.Our LGM scenarios bound the PMIP2 ensemble mean change of −2.3 • C in the tropics relative to the preindustrial.Our simulations find decreases in zonal mean precipitation of 10-50% within 10 • of the equator, and of 10-90% within 40 • of the poles, and little change in the midlatitudes as in the mean of the PMIP2 ensemble.Precipitation increases up to 25-33% in the tropics poleward of the equator in our simulations, where the PMIP2 ensemble mean is zero.Our zonal mean LGM precipitation changes are of greater magnitude than the range of PMIP2.
Fig. 4 also compares our modeled results and those of PMIP2 to an inverse reconstruction of regional climate using paleoecological archives by Bartlein et al. (2011).Temperature changes in the warm-LGM simulation relative to the preindustrial better match values inferred from the pollen data than do those in the cold-LGM simulation, with mean absolute biases of 4.4 • C and 5.1 • C, respectively (the PMIP2 bias is 3.5 • C).Both PMIP2 and our simulations have difficulty in matching the inferred changes in precipitation, except over Africa, where the cold-LGM performs well.The pollen record has a small sample size (n = 98 at the LGM) and a limited geographic distribution but does provide some confidence in our choice of LGM scenarios as brackets of a range of probable climates.545 4 Tropospheric emissions Fig. 5 summarizes the gas-phase emissions and Fig. 6 the direct and precursor aerosol emissions in our different simulations.Wherever possible, emissions respond to changing meteorology from the GCM, as described below.We test the 550 sensitivity of our results to three different emission scenarios in the preindustrial and LGM atmospheres: a low fire and a high fire scenario, which together span the range of estimated past fire magnitudes (Section 4.6), and a scenario in which we fix the total lightning source in all climates to compare to 555 our other simulations in which total lightning changes with climate (Section 4.5).

Anthropogenic
Present-day anthropogenic emissions from fuel combustion, agriculture, and industry are prescribed from the Emission 560 Database for Global Atmospheric Research (EDGAR) base inventory for 2000 (Olivier, 2001), overwritten with regional inventories for the United States (EPA NEI99), Canada (CAC), Mexico (BRAVO, Kuhns et al., 2003), Europe (EMEP, Auvray and Bey, 2005) and South and East Asia 565 (Streets et al., 2003(Streets et al., , 2006)).These inventories all have monthly variability, some with additional weekly cycles.Biofuel emissions (Yevich and Logan, 2003) and aircraft emissions are constant.All emissions are scaled as described by van Donkelaar et al. (2008) to 1995 values for our present-570 day simulation.Our preindustrial and LGM simulations assume no anthropogenic emissions, aside from a small biofuel burning source in the preindustrial that is 45% of the present-day value, estimated by scaling and distributing the 1850 country-level inventory from Fernandes et al. (2007) 575 by the gridcell-level of population between 1770 and 1850 from the History Database of Global Environment (HYDE v3.1 Klein Goldewijk et al., 2011).Anthropogenic emissions since the preindustrial yield large perturbations to the tropospheric budgets of reactive nitrogen, carbon, and sulfur 580 (Fig. 5 and Fig. 6), predominantly in the northern midlatitudes.

Biogenic -Land
Terrestrial plant emissions of VOCs in GEOS-Chem follow the MEGAN scheme, in which each model grid cell is as-585 signed a base emission rate based on its local vegetation.MEGAN adjusts these local base emission rates by applying scaling factors that are time-varying functions of photosynthetically available radiation (PAR), leaf area index (LAI), and a 15-day moving average of recent surface air temperature (Guenther et al., 1995(Guenther et al., , 1999(Guenther et al., , 2000(Guenther et al., , 2006(Guenther et al., , 2012)).Emissions respond positively to increases in each driving variable.For all scenarios in this study, we calculate the base biogenic VOC emissions using the PFT distributions from BIOME4-TG.To derive these emissions, we first perform a multiple regression, mapping the default distribution of MEGAN version 2.1 emission factors onto the present-day PFT distribution in BIOME4-TG.We then apply the resulting emission factors, shown in Table 5, onto the distribution of PFTs calculated by BIOME4-TG for the preindustrial and LGM climates.
Fig. 5c shows the resulting biogenic emissions from our approach, and Fig. 7a the spatial distribution of changes in isoprene emissions between the different climate simulations.Global terrestrial plant emissions decrease by 7% in the present day relative to the preindustrial because of large-scale conversion of natural vegetation types to modern croplands (which have lower VOC yields), particularly over southern Brazil and northern India.At the LGM, terrestrial plant emissions decrease by 17% and 51% relative to the preindustrial in our warm-LGM and cold-LGM scenarios respectively, predominantly driven by decreases over South America and Africa.These decreases are caused in part by a simulated reduction in tropical broadleaf forests and LAI in the colder and drier tropics (especially in the Amazon and Congo in the cold-LGM scenario), but the driving factor is predominantly the leaf-temperature effect.Hence, we simulate severe reduction in plant emissions of VOCs in the cold-LGM scenario with its > 6 • C of cooling in tropical surface temperatures.However, additional land area exposed on the Sunda and Gulf of Carpentaria continental shelves causes a large regional increase in plant emissions that is stronger in the warm-LGM scenario due to warmer temperatures and increased PAR from decreased cloud cover than in the cold-LGM scenario.Our LGM results span the range of previous Emissions of NO x from soil microbial activity follow the parameterization of Yienger and Levy (1995) as implemented by Wang et al. (1998) and are dependent on land cover type, precipitation, and temperature.Present-day emissions are 16% higher than the preindustrial, mostly reflecting the use of fertilizers (Fig. 5a).Soil emissions decrease in each successively colder past climate simulation, driven by the associated decreases in precipitation and surface temperature.
In the warm-LGM, decreases in precipitation and temperature are offset by additional tropical land mass, and global soil emissions decrease by only 2% relative to the preindustrial.In the cold-LGM, large decreases in temperature and precipitation (especially over southern Brazil) to a 40% decrease in global soil emissions relative to either the preindustrial or warm-LGM.

645
Phytoplankton and bacteria activity in the mixed-layer of the ocean may represent a local source or sink of acetone to the atmosphere.We assume constant ocean acetone concentrations in all climates (15 nM), but allow the sign and magnitude of the ocean-atmosphere flux to vary locally with sur-650 face temperature and atmospheric concentration as described by Fischer et al. (2012).In all scenarios, the ocean acts as a net sink for acetone.Biogenic production of dimethylsulfide (DMS) is represented in the model as the product of seawater DMS concen-655 trations and sea-to-air transfer velocities (Park et al., 2004).We use distributions of surface ocean concentrations of DMS from Kettle et al. (1999) in all climates, but transfer velocities are parameterized as a function of SST and surface wind speeds (Nightingale et al., 2000a,b).At the LGM, cooler 660 SSTs together with our assumption that no production occurs beneath sea ice yield an 11% reduction in DMS emissions relative to either the preindustrial or present-day (Fig. 6a).
Bromocarbon emissions from oceanic macroalgae and phytoplankton are the dominant bromocarbon precursors for 665 tropospheric Br y (Warwick et al., 2006;Law et al., 2007).Emissions are prescribed from Liang et al. (2010) as implemented by Parrella et al. (2012), and remain constant between climate scenarios.

Volcanic 670
Volcanic emissions of SO 2 are from the AEROCOM inventory (Diehl, 2009) as implemented by Fisher et al. (2011) and vary spatially and daily.The emissions are released vertically into the troposphere as in Chin et al. (2000).We apply emissions from a typical recent year (1995) to all scenarios, with 675 2.5 Tg S a −1 of eruptive emissions and 8.9 Tg S a −1 of passive outgassing.Volcanic activity may vary over glacial periods following redistribution of surface mass stresses from ice sheets (Kutterolf et al., 2013), but we do not examine that possibility here.

Lightning
Lightning is a major natural source of NO x (Schumann and Huntrieser, 2007).The standard lightning NO x source in GEOS-Chem is described by Murray et al. (2012).Lightning flash rates in the model follow the parameterization of Price 685 and Rind (1993Rind ( , 1992Rind ( , 1994) ) and reflect changes in convective cloud top heights.We determine for the present day a uniform scaling factor (=7.0) such that the simulated global mean flash rate matches the 46 flashes s −1 observed by satellites (Christian et al., 2003).We use the same scaling factor 690 for all climates, but allow global mean lightning activity to change with changes in deep convection.We assume a uniform production rate of 310 mol N per all flash types (corre-sponding to 6.3 Tg N a −1 in the present day), and distribute emissions vertically following Ott et al. (2010).Fig. 7b shows changes in the lightning NO x distribution.The Price and Rind (1993, 1992, 1994) formulation strongly responds to decreases in the vertical extent of deep convection, and lightning activity therefore decreases in each successively colder climate.Relative to the preindustrial, lightning is 5% more abundant at the present day and 20% to 45% less abundant in the warm-LGM and cold-LGM scenarios respectively (Fig. 5a).The primary reductions in lightning abundance at the LGM are simulated over Africa and South America, with only very small reductions in temperate and boreal latitudes, and local increases in lightning over the exposed continental margins in the tropics.
Although lightning is known to be strongly correlated with surface temperatures on diurnal through decadal time scales (Williams et al., 2005), how it may respond to changing meteorology on longer scales is uncertain (Williams, 2005).Therefore we also perform a suite of "fixed lightning" simulations in which we uniformly scale the average NO x yield per flash in each climate scenario such that we maintain a constant total lightning NO x yield of 6.3 Tg N a −1 .

Fires
Both natural and human-ignited fires have likely played a large role in trace gas emissions since the LGM (Pyne, 2001).For the present day, we prescribe monthly mean emissions of pyrogenic trace gases and aerosols from Yevich and Logan (2003).However, there is large uncertainty in the amount of fire activity in the past.Many previous model studies assumed that preindustrial fire emissions were considerably lower than those estimated for the present; Crutzen and Zimmermann (1991) suggested a 90% reduction.More recently, a global synthesis of charcoal accumulation rates suggested that preindustrial fire emissions were similar to or slightly higher than those in the present day (Power et al., 2008), while the record of black carbon deposition at Greenland and Antarctica also implies that burning was mostly constant from the preindustrial through to the present (McConnell et al., 2007;Bisiaux et al., 2012).Measurements of δ 13 C and δ 18 O in ice-core CO at Antarctica appear to indicate that Southern Hemispheric fire emissions were substantially higher in the preindustrial than at present (Wang et al., 2010), but this view is controversial (van der Werf et al., 2013).The extremely limited charcoal data that exist for LGM suggest that fire activity was at a historical minimum (Power et al., 2008), but geographic coverage is very poor, and only a small fraction of the continents are represented.Model representations of fire activity show large disagreements in the trends over glacial-interglacial and preindustrial to present-day time horizons (Thonicke et al., 2005;Fischer et al., 2008;Pfeiffer and Kaplan, 2013).Given the wide range of estimates for fire activity since the LGM, we test a range of emissions for each climate, as we describe below.
For all simulations, we use distributions of dry matter consumed per PFT from the LPJ-LMfire model (Pfeiffer et al., 2013) coupled with emission factors from the Global Fire 750 Emissions Database version 3 (GFED3) (van der Werf et al., 2010) for NO x , CO, SO 2 , black carbon, OC, and 17 additional gas-phase species.LPJ-LMfire yields fire emissions in the preindustrial and LGM scenarios that are 3-4 times higher than those for the present day, primarily reflecting the 755 absence of active fire suppression in the preindustrial world and decreases in passive fire suppression due to less fragmentation in the landscape and reduced livestock grazing.These results represent an upper limit of fire activity in the past.For a lower limit, we estimate fire emissions by scaling the 760 LPJ-LMfire total dry matter consumed to match emissions implied by the charcoal accumulation rates from the Global Charcoal Database (preindustrial = 100% and LGM = 10% of the present-day 4.2 Pg a −1 dry matter consumed) (Power et al., 2008(Power et al., , 2010;;van der Werf et al., 2010).We apply both 765 estimates to GEOS-Chem, and refer to these simulations as the "high fire" and "low fire" scenarios in each climate.Totals for these scenarios are shown in Fig. 5 and Fig. 6.
Fig. 7c shows the changing distribution of CO emissions in the low fire preindustrial emissions scenario.While the total 770 dry matter consumed is the same for the low fire and presentday simulations, there are changes in the geographic distribution of emissions.Fire activity decreases in the present-day extratropics due to active and passive fire suppression, and increases in the tropics due to deforestation and agricultural 775 practices.As tropical fires have higher emission factors per dry matter consumed (van der Werf et al., 2010), total emissions are higher in the present day by ∼25% relative to the preindustrial, despite dry matter consumed remaining constant.In the high fire scenario, emissions are lower almost 780 everywhere relative to preindustrial in both the present-day and LGM scenarios (not shown).

Wind-driven
Sea salt aerosol evasion from the ocean is described by Jaeglé et al. (2011), and responds to SSTs and surface wind speeds.

785
Emissions at the present day are 3% higher than the preindustrial.Emissions at the LGM are 4% and 25% lower than the preindustrial, largely reflecting SSTs and extent of sea ice.Debromination of sea salt is an important source of bromine radicals to the troposphere, and responds to sea salt aerosol as 790 described by Parrella et al. (2012) and has been updated with size-dependent bromine depletion factors from Yang et al. (2008).The debromination source increases by 3% at the present day and decreases by 7-9% at the LGM relative to the preindustrial.We do not consider here the possible enhancement due to increased particle acidity from present-day anthropogenic emissions, or the reduction with increased alkalinity at the LGM (Sander et al., 2003).
We prescribe monthly mean dust concentrations in GEOS-Chem, as the coarsely resolved surface winds from the GCM are too weak to mobilize a realistic dust burden.We use 3D mineral dust fields from Mahowald et al. (2006), with LGM burdens 125% greater than in the preindustrial and present.The drier conditions of the LGM promote large increases in dust mobilized from the expanded Sahara desert and Tibetan plateau relative to the preindustrial and present, as well as from the exposed continental shelf of South America.
5 Results: Changes in the tropospheric oxidants Fig. 8 shows the tropospheric mean airmass-weighted concentrations for each of the four major oxidants simulated in each of the four climates (present day, preindustrial, warm-LGM, and cold-LGM), and three sensitivity studies for the past climates (high fire, low fire, and fixed lightning).For comparison, we also show a present-day simulation driven by assimilated GEOS4 meteorology for 1994-1996 instead of ModelE meteorology.Anthropogenic emissions in the two present-day simulations are identical.We find reduced oxidative capacity relative to the present day in all climates, except for OH in the fixed lightning LGM simulations, and H 2 O 2 in the preindustrial high fire scenario.In general, oxidant burdens decrease with decreasing surface temperature, and changes in OH are well-buffered compared to the other oxidants.We discuss these changes in oxidants in more detail below.Unless otherwise indicated, all changes are reported as the mean percent change ± standard deviation (1σ) in the 18 pairwise combinations of the 6 LGM relative to the 3 preindustrial simulations, and the 3 pairwise combinations of the ModelE present day vs. 3 preindustrial.Comparison with the PMIP2 ensemble and with reconstructions of LGM climate from pollen records (Section 3) and ice-core records (Section 6) suggest that the low-fire scenarios and the warm-LGM climate are the most likely representations of the past atmospheres of our ensemble members; we refer to their combination as our "best estimate."

Hydroxyl radical
Tropospheric mean OH for each simulation is shown in the top left panel of Fig. 8. OH burdens in our ModelE-driven simulations increase by +7.0 ± 4.3% in the present day relative to the preindustrial (+5.3% in our best estimate).The ensemble mean shows little change at the LGM relative to the preindustrial: +0.5 ± 12% (+1.7% best estimate).The spread in our estimated OH change is large, but much less than found by most prior studies (cf.Table 1).OH is largely insensitive to variations in fire emissions.We diagnose these changes below.
Fig. 9 shows the global chemical budget for OH in each simulation.The non-linear reactions controlling OH are complicated, but well-known (e.g., Logan et al., 1981;Spivakovsky et al., 2000;Lelieveld et al., 2002).Primary pro-850 duction is by photolysis of tropospheric ozone in the presence of water vapor, with secondary production from rapid chemical cycling of the HO x family (HO x ≡ OH + RO 2 ; RO 2 ≡ HO 2 + organic peroxy radicals).Loss is primarily via reaction with CO in all climates, but reaction with 855 methane, other VOCs, and their degradation products (which include CO) are also important.Absolute production and loss rates consistently decrease in each successively colder climate, and do so mostly independently of our different emission scenarios, implying the changes are primarily driven by 860 meteorology.However, the response of mean OH as seen in Fig. 8 is more buffered than the overall rates of its production and loss reactions.
Present-day mean OH increases +7.0 ± 4.3% relative to preindustrial levels, which is much less than the perturbations 865 to emissions -e.g., the 35-52% increase in NO x .Our results are consistent with the recent multi-model ACCMIP study, which reported a mean ∆OH of −0.6 ± 8.8% for 1850-2000(Naik et al., 2013)).To understand the relatively small impact of anthropogenic emissions on OH, we follow the approach 870 of Wang and Jacob (1998), who derived a linear relationship between OH and emissions using the steady state equations of the NO x -HO x -CO-ozone system.They found that mean OH varies with the ratio S N /(S 3/2 C ), where S N represents total NO x emissions (mol a −1 ) and S C represents the source 875 of reactive carbon (CO and hydrocarbons; mol a −1 ).We also find a linear relationship between OH and S N /(S 3/2 C ) in our preindustrial and present-day simulations (R 2 = 0.81; n = 2 present day + 3 preindustrial simulations).∆OH is small because S N /(S 3/2 C ) has changed little from the preindustrial 880 to the present, as human activity has increased both S N and S C (Thompson et al., 1993).
On glacial-interglacial timescales, the dependence of OH on S N /(S 3/2 C ) alone cannot explain the simulated variability (R 2 = 0.06; n = 11 simulations).Instead, we find lin-885 ear dependence by including two variables that Wang and Jacob (1998) had assumed remained constant in their derivation: the mean tropospheric ozone photolysis frequency, J O3 (s −1 ), and water vapor concentration, which we represent by specific humidity, q (g H 2 O/kg air), i.e., 890 This linear relationship explains 72% of the simulated variability in tropospheric mean OH across our 11 simulations (Fig. 10).Our simulations indicate that J O3 and q are sufficiently variable on glacial-interglacial time scales to effect 895 changes greater than those from emissions.To first order, J O3 q may be thought of as controlling the variability in primary production of HO x , and S N /(S 3/2 C ) in regulating HO x equilibrium partitioning between OH and RO 2 , which taken together determine mean OH (although, the ratio S N /(S 3/2 C ) 900 also influences ozone production and therefore HO x production).
Mean OH independently varies most in our 11 simulations with the photolysis component (J O3 ) of Equation 1 (R 2 = 0.26), although its weak correlation emphasizes the necessity of considering all three factors together.Specific humidity q is a strong function of temperature (Sherwood et al., 2010), decreasing by 23% in the warm-LGM scenario and 50% in the cold-LGM scenario, and is more weakly correlated with mean OH (R 2 = 0.18).
Changes in J O3 reflect changes in shortwave radiation attenuation through overhead ozone columns, aerosol concentrations, and clouds; overhead ozone accounts for most of the change (J O3 vs. SCO: R 2 = 0.85).Total cloud cover does not change sufficiently in our simulations to have a significant impact on global photolysis rates.Over high latitudes in summer the LGM, increased snow and ice cover over both land and ocean enhance UV surface albedos and thus J O3 .
Fig. 11 shows the absolute changes in stratospheric columns of ozone in our simulations.Relative to the preindustrial simulation, tropical stratospheric ozone columns have decreased 4-6% (10)(11)(12) in the present day, and have increased 3-6% (10-12 DU) in the warm-LGM and 10-13% (26-30 DU) in the cold-LGM simulations.The increases in tropical overhead ozone at the LGM are driven in the model by: (a) reduced catalytic destruction of ozone from reduced stratospheric halogen and NO x burdens, (b) the deceleration of the residual stratospheric circulation (cf.Section 3.2) that slows the transport of ozone away from the region of net photochemical production in the tropics to the polar stratosphere, which also decreases extratropical SCO, and (c) decreases in the tropical tropopause height, which increases stratospheric mass, especially in the cold-LGM scenario.These increases in stratospheric ozone are slightly offset by the influence of warmer stratospheric temperatures at the LGM.The converse holds true for explaining the preindustrial to presentday changes (increased catalyzed ozone loss, a strengthened residual circulation, and a slight increase in tropopause height).Our results are consistent with Rind et al. (2009), who also simulated increases in stratospheric ozone at the LGM using Linoz online within the GISS III GCM, and with climate-driven changes in stratospheric ozone seen in satellite observations (Neu et al., submitted) and model studies (Hsu and Prather, 2009;Hegglin and Shepherd, 2009) for the present day and near future.
The dependence of OH on S N /(S 3/2 C ) explains why the simulated OH burden is insensitive to fire emissions but highly sensitive to lightning.Fire emissions of NO, CO, and VOCs in the model assume fixed emission factors per dry matter consumed (van der Werf et al., 2010), so large absolute differences in burning yield relatively small changes in S N /(S 3/2 C ).In contrast, lightning and soil microbial activity are large sources of NO x with potentially large changes on glacial-interglacial time scales and so can strongly alter S N /(S 3/2 C ).Large changes in emissions of biogenic VOCs 955 could also potentially perturb the S N /(S 3/2 C ) ratio.However, we find interglacial changes in biogenic VOCs are strongly correlated to soil NO x emissions (R 2 = 0.92), muting the impact of either on OH.
That we simulate small and even negative ∆OH at the

960
LGM relative to the preindustrial is contrary to most earlier studies (Table 1).LGM-preindustrial pairs), or if biogenic VOC emissions decrease sufficiently such that the increase in S N /(S 3/2 C ) can overcome the counter-effects of reduced water vapor and increased overhead ozone.curring over the land regions where lightning NO x decreases the most (Fig. 7b).Decreased production rates in the tropics outweigh the reduction in loss frequencies from reduced CO, methane, and NMVOCs, driving net negative anomalies in most regions in all simulations.One exception is over 1000 Sundaland, where increased lightning NO x over the exposed continental shelves locally enhances OH.Another exception is over the Indian subcontinent in the cold-LGM scenario, where severe reductions in biogenic emissions also increase OH.In the polar regions, where OH production is relatively 1005 slow, OH increases in response to reductions in CO transported from lower latitudes and in polar methane concentrations (Table 3).The increase is greatest in LGM summer over regions with enhanced snow and ice cover (e.g., the Southern Ocean and the Laurentide ice sheet) and thus higher UV surface albedos relative to the preindustrial or present day.The polar changes are inconsequential for global OH, but are of potential interest for the interpretation of measurements of trace gases from ice cores.

Tropospheric ozone
The tropospheric ozone burden decreases in each progressively colder scenario (Fig. 8).Relative to the preindustrial, present-day ozone increases +28 ± 11% (+35% best estimate), somewhat less than the mean change of +41 ± 9% from 1850 to 2000 in the ACCMIP ensemble (Young et al., 2013) or the +40-48% changes simulated by changing emissions in GEOS-Chem by Sofen et al. (2011) and Parrella et al. (2012).Our more tempered response likely reflects the relatively high cross-tropopause flux of ozone in our model, and in the case of Sofen et al. (2011) and Parrella et al. (2012), our much higher fire emissions in the preindustrial.Our simulated changes of −20 ± 9% (−20% best estimate) in the LGM tropospheric ozone burden relative to the preindustrial are consistent with previous estimates (Table 1).
To understand these changes, we examine the ozone budget in each simulation (Fig. 13).Photochemical production increases in the present day, relative to the preindustrial, but decreases in the two LGM scenarios in response to reductions in photolysis rates and in emissions of NO x , CO and VOCs.Ozone production efficiencies (OPE) per molecule NO x emitted (Liu et al., 1987), as shown in the middle panel of Fig. 13, have decreased in the present day by 15% to 37%, and range from 30% decreases to 45% increases in the LGM relative to the preindustrial, although changes from the LGM are small within each of the three emission scenarios.
The cross-tropopause ozone flux decreases in each colder climate in response to the deceleration of the stratospheric circulation (Section 3.2) together with decreases in extratropical lower stratospheric ozone (Section 5.1 and Fig. 11) where most stratosphere-to-troposphere transport occurs.
Nonetheless, transport of ozone from the stratosphere becomes an increasingly important fraction of total ozone sources in each colder climate, increasing from 17% in the present day to 33% in the cold-LGM scenario.
The bottom panel of Fig. 13 gives the lifetime of tropospheric ozone in each simulation.Loss is primarily by photolysis in the presence of water vapor (i.e., primary production of OH), but also by reaction with HO x radicals; both decrease in each cold climate as described in Section 5.1 and Section 5. Bromine represents an increasingly important component (16 to 34%) of the total loss rates in each successively colder climate, as the net loss of ozone from the cycling of BrO x radicals (shown as hatching in the top panel of Fig. 13) re-mains relatively constant between simulations.This reflects the nearly constant emissions of Br y .The depositional loss frequencies decrease in each successively colder climate scenario, reflecting decreases in LAI and total vegetated area 1065 (despite the exposed continental margins), and to lesser extent, decreases in surface friction velocities.
The multi-model comparisons by Stevenson et al. (2006) and Young et al. (2013) show a tight linear relationship between ozone burdens and NO x emissions.Our simulations 1070 extend the relationship of the tropospheric ozone burden with NO x into the different climate and emission scenarios of the Last Glacial Maximum, with an almost identical slope to Young et al. (2013).In the 10 simulations driven by ModelE meteorology, 94% of the variability in the tropospheric ozone 1075 burden, X O3 , is explained by differences in NO x emissions alone (X O3 = [160 Tg] + [52 Tg / Tmol a −1 ] S N ), likely reflecting that the majority of the atmosphere is NO x -limited (Sillman et al., 1990).However, ozone is also strongly correlated with Fig. 14 shows the changes in the temporal and spatial distribution of ozone relative to the preindustrial in our low fire simulations, which have similar patterns to the other emissions scenarios.In the preindustrial simulations, tropo-1085 spheric ozone columns peak at the poles during their respective spring months due to transport from the stratosphere, consistent with Mickley et al. (2001).In the present day, ozone increases everywhere due to anthropogenic precursor emissions, particularly in the temperate latitudes of the 1090 northern hemisphere.Extratropical ozone also increases in both hemispheres because of enhanced transport from the stratosphere, especially at the poles.Most of the LGM troposphere has reduced ozone relative to the preindustrial driven by the changes in climate and emissions.Greenland expe-1095 riences local enhancements in stratospheric downwelling at the LGM, probably driven by the presence of the Laurentide and Fennoscandian ice sheets.
Despite inclusion of bromine chemistry, our simulated reduction in global ozone at the preindustrial is not sufficient 1100 to explain the low concentrations derived from surface air measurements taken in the late 19th century (e.g., Pavelin et al., 1999).Reproduction of these low values remains a long-standing issue in the modeling of past changes in tropospheric ozone (Wang and Jacob, 1998;Mickley et al., 2001; 1105 Shindell et al., 2003;Lamarque et al., 2005;Horowitz, 2006;Parrella et al., 2012;Young et al., 2013).

Hydrogen peroxide
H 2 O 2 responds sensitively to different climates and emission scenarios (Fig. 8), changing by +18 ± 22% in the present 1110 day and by −36 ± 21% at the LGM relative to the preindustrial (+31% and −23% best estimate).Its tropospheric budget is given in Table 6.H 2 O 2 is both a reservoir species for HO x and the dominant HO x sink.The main loss mechanism for H 2 O 2 , and thus HO x , is via uptake by cloud droplets and subsequent deposition.Despite reductions in precipitation in each successively colder climate that increase the H 2 O 2 lifetime against deposition, we find that the decline in HO x production in the colder climates (Section 5.1) governs the changes in H 2 O 2 .Our simulations indicate that H 2 O 2 , unlike OH, is very sensitive to changing fire emissions, with H 2 O 2 levels in the high fire scenarios 40-70% greater than in the low fire scenarios.Fire emissions generate peroxy radicals (RO 2 ) and therefore can enhance total HO x (although not OH, for reasons described in Section 5.1).The sensitivity of H 2 O 2 to fire emissions but not to lightning make ice-core measurements of this molecule a potentially useful proxy for paleo fire abundance.H 2 O 2 could also reflect changes in biogenic VOC emissions, as they also generate RO 2 .
Fig. 15 shows the spatiotemporal distribution of H 2 O 2 and its changes in our low fire simulations.Production of H 2 O 2 is greatest in the tropics, coincident with maximum production of OH, and therefore HO x .However, the solubility of H 2 O 2 causes high loss in the intertropical convergence zone, leading to the maxima in H 2 O 2 concentrations in the subtropical lower free troposphere.At the LGM, precipitation increases in the subtropics and decreases in the tropics relative to the preindustrial in our simulations (Fig. 3), but these changes are not enough to offset reductions in primary production, and H 2 O 2 decreases almost everywhere.The exception is over Southeast Asia in the warm-LGM, where enhanced reactive carbon emissions from biogenics and fires over the exposed continental shelves (Fig. 7) promote RO 2 production.
In the present day relative to the low fire preindustrial scenario, H 2 O 2 increases everywhere.
The overall decrease in H 2 O 2 levels has implications for the oxidation pathways of SO 2 , which is primarily oxidized to sulfate in the aqueous phase by H 2 O 2 within cloud droplets.If a greater fraction of SO 2 is instead oxidized by OH in the gas phase, it would lead to additional particle formation with associated direct and indirect climate effects.

Nitrate radical
The chemistry of the nighttime troposphere is dominated by the nitrate radical (NO 3 ; Wayne et al., 1991).In the absence of solar radiation, almost all NO x is converted to NO 2 , which may react with ozone to produce NO 3 .The tropospheric burden changes by +130 ± 70 % at the present day and −46 ± 26 % at the LGM, relative to the preindustrial (+180% and −48% best estimate).The tropospheric burden of NO 3 (X NO3 ) in each of the 11 simulations shown in the bottom right panel of Fig. 8 almost exactly follows the total source of reactive nitrogen (X NO3 = 0.014 ppt / Tg N a −1 × S N ; R 2 = 0.97).How NO 3 changes in different climates is therefore sensitive to assumptions made about lightning, fire and soil emissions.

Comparison with ice-core record
Observations of reactive species in ice cores are limited for the preindustrial, and even more so for the LGM (Wolff, 2012), but offer our only constraints for the chemical composition of past atmospheres.Comparison of our model simula-1170 tions with the ice-core record is important for both evaluation of the model, and for interpretation of the observations, particularly for placing polar observations in the context of the entire troposphere.However, many reactive species are suspected or known to undergo post-depositional changes (e.g., 1175 Hutterli et al., 2002Hutterli et al., , 2003;;Haan et al., 1996), which complicates interpretation.
Formaldehyde (HCHO) undergoes substantial postdepositional processing via evaporation followed by photolysis.The degree of post-depositional processing is strongly 1180 influenced by local temperature and the snow accumulation rate (Hutterli et al., 2002(Hutterli et al., , 2003)).The available HCHO data have not been corrected since that realization, though our simulations qualitatively match measurements from Greenland (Staffelbach et al., 1991) and Antarctica (Gillett 1185(Gillett et al., 2000)).Our results indicate that HCHO was relatively stable through the past 1000 years and then increased in the 20th century following increases in VOC precursors (70% simulated vs. 100% observed increases at Greenland; +40% vs. +40-100% at Antarctica).At the LGM, however, the 1190 Antarctic ice core indicates a 100% increase relative to the preindustrial, while we simulate 27-36% decreases in surface concentrations.This discrepancy may be due to the temperature dependence of HCHO preservation in ice cores, with increased HCHO preservation in the very cold temperatures 1195 of the LGM (Hutterli et al., 2003).
The ratio of HCHO to methane in ice cores has been identified as a potential proxy for polar OH abundance, as production of HCHO in the remote atmosphere results primarily from methane oxidation (Staffelbach et al., 1991).
Given the large uncertainty in ice-core HCHO (Hutterli et al., 2002(Hutterli et al., , 2003)), it is unclear what to make of uncorrected HCHO/methane ratios.Our simulations, like those of Levine et al. (2011a) show that this ratio at most offers local information, and only in the polar regions where methane is the dom-1205 inant source of HCHO.Tropospheric mean OH (weighted by the methane reaction rate constant) is strongly correlated with tropospheric mean ratio of HCHO/methane in the preindustrial and present-day simulations (R 2 = 0.82; n = 5 simulations) but uncorrelated in the LGM scenarios.The ero-1210 sion of the correlation between tropospheric mean OH and HCHO/methane over glacial-interglacial timescales may reflect variability in NMVOC precursors to HCHO, stronger sensitivity of the CH 4 + OH reaction to temperature relative to other NMVOCs, changes in the meridional distribution of 1215 methane, and/or changes in photolysis loss frequencies of HCHO.
H 2 O 2 is the only oxidant directly preserved in ice cores, but like HCHO it undergoes post-depositional processing that should be considered (Hutterli et al., 2003).Ice cores from the West Antarctic Ice Sheet (WAIS) considered representative of surface air concentrations indicate H 2 O 2 was relatively constant over the past 500 years, but has at least doubled since 1950 (Frey et al., 2006); uncorrected data from Greenland cores also show a large 20th-century increase (Fuhrer et al., 1993;Anklin and Bales, 1997).Compared to our preindustrial low fire scenario, deposition of H 2 O 2 increases by 31% over Greenland in the present day, consistent with the ice-core record.However, over the WAIS, surface concentrations increase by only 5% and deposition shows no significant change due to decreases in precipitation in our model here.In their simulations, Lamarque et al. (2011) attribute the late-20th century increase in the WAIS record to the ozone hole, whose heterogeneous chemistry we do not simulate here.The preindustrial high fire simulation is inconsistent with the ice-core data, as it implies lower H 2 O 2 surface concentrations and deposition everywhere in the present day relative to the preindustrial.For the LGM, uncorrected measurements from Law Dome in East Antarctica imply decreases in H 2 O 2 relative to the preindustrial (vanOmmen and Morgan, 1996), generally matched by reduced deposition fluxes in our simulations despite enhanced surface concentrations.Surface concentrations over East Antarctica generally increase relative to the preindustrial in our various LGM simulations because of enhanced polar OH production (Section 5.1) and decreased precipitation (Section 3.2), with large variability reflecting the wide spread in the LGM fire emissions as HO x precursors.
Our present-day simulation matches the mean CO surface concentration of 49 ppbv measured at the South Pole by the NOAA Global Monitoring Division for the 1990s, as well as the Southern Hemispheric gradient in CO.Carbon monoxide has been recorded in ice-core air bubbles over the past two millennia.Wang et al. (2010) measured ice-core concentrations of CO at the South Pole of 47 ± 4 ppbv for 1770.Our low fire preindustrial scenario yields 35 ppbv surface CO at the South Pole, consistent with Wang et al. (2010), given that as much as 10 ppbv of CO in Antarctic ice cores may derive from in situ production (Haan and Raynaud, 1998).Our high fire simulations yield much higher surface concentrations of 63 ppbv at the preindustrial, a value that is greater than any over the entire CO ice-core record at Antarctica.
Black carbon deposited in Greenland ice cores increased from 1850-1950 due to human industrial activity, but then declined to the 1800 values by the late 1950s (McConnell et al., 2007).Records of black carbon deposition to Antarctica for 1850-2000 imply mostly constant deposition rates (Bisiaux et al., 2012).Squaring these records of black carbon deposition with our model results is challenging.Our high fire scenario for the preindustrial implies 40-50% decreases in black carbon deposition to both Greenland and Antarctica from the 1770s to the present day, not consistent with observations.Our low fire preindustrial scenario yields a better match with the observations over Antarctica, with similar rates of depo-sition in the 1770s and the present day.But this same sce-1275 nario indicates large, 40-500% increases in black carbon deposited to Greenland in the present day relative to the 1770s.Reconciling both the black carbon and CO ice-core records with current understanding of trends in biomass burning is a direction for future research.
1280 Antarctic ice cores show increased sea salt aerosol deposition during cold periods, perhaps as a result of enhanced sea salt emission from frost flowers across the greater expanse of sea ice (Wolff, 2003;Wolff et al., 2006;Fischer et al., 2007;Rothlisberger et al., 2008;Röthlisberger et al., 2010;Wolff 1285Wolff et al., 2010)).GEOS-Chem assumes no source of sea salt from sea ice (Fig. 6e; Jaeglé et al., 2011), and therefore predicts 70-90% decreases in sea salt deposition to Antarctica in the LGM climate relative to the preindustrial.
The ∆ 17 O of ice-core sulfate measures the departure 1290 from mass-dependent fractionation in its oxygen isotopes (Thiemens, 2006).It provides a constraint for assessing the relative importance of different sulfate chemical production pathways, i.e., gas-phase oxidation by OH versus aqueousphase production via H 2 O 2 or ozone, as the latter imparts a 1295 large ∆ 17 O signature (Savarino et al., 2000).Measurements from the Vostok ice core in Antarctica imply that gas-phase oxidation by OH contributed up to 40% more to sulfate production during the LGM relative to the interglacial periods that came before or after (Alexander et al., 2002). 1300 Our results appear consistent with these measurements of ∆ 17 O of sulfate.In our LGM scenarios, surface concentrations of OH at the poles increase driven primarily by the reduction in methane, CO, and overhead ozone, and increases in surface UV albedo.In addition, the fraction of SO 2 oxi-1305 dized in the gas phase by OH increases throughout the LGM troposphere, reflecting a global decrease in aqueous-phase chemistry in the drier climate.In a companion paper (E.Sofen, manuscript in prep), we offer an extensive interpretation of ∆ 17 O measurements in sulfate and examine the utility 1310 of ∆ 17 O as a proxy for oxidant variability, a practice that has been recently questioned (Levine et al., 2011a).
Unlike sulfate, nitrate aerosol undergoes substantial postdepositional photochemical processing via evaporation of HNO 3 (Rothlisberger et al., 2002) and photolysis to produce 1315 NO x (Honrath et al., 1999(Honrath et al., , 2000)).Photolysis of snow nitrate to produce NO x it thought to be dominant (Davis et al., 2008;Erbland et al., 2012).The degree of post-depositional processing depends mainly on snow accumulation rate.In East Antarctica where the snow accumulation rate is low, it 1320 is thought that all of the nitrate originally deposited to the snow is photolyzed to NO x during austral spring and summer (Erbland et al., 2012).This NO x is released to the boundary layer where it is re-oxidized locally and redeposited to the snow surface as nitrate.The ∆ 17 O signature in East 1325 Antarctic ice-core nitrate thus reflects the relative abundance of local atmospheric oxidants ozone and RO 2 .At locations with higher snow accumulation rates such as Greenland and West Antarctica, the ∆ 17 O of nitrate may reflect a combina-tion of the local summertime ozone/RO 2 ratio and that ratio in the NO x source regions.
For the present day, our low fire simulations yield a 4% increase in the ozone/RO 2 ratio over Greenland and a 7% decrease over the WAIS, qualitatively consistent with the expected sign changes from ice-core measurements of ∆ 17 O of nitrate (Alexander et al., 2004;Sofen et al., in review).For the LGM, we calculate 20-40% increases in the ozone/RO 2 ratio over East Antarctica relative to the preindustrial, and up to a 50% increase over Greenland.Over Greenland, the increases result from increases in both tropospheric ozone from local enhanced stratospheric downwelling (Fig. 14) and reductions in RO 2 .Over Antarctica, ozone decreases in both LGM scenarios, so decreases in RO 2 largely dominate the response.Observations of nitrate ∆ 17 O in the LGM are not yet published, but are currently being collected.
7 Implications for the methane budget Table 7 gives the global methane lifetime against OH oxidation in the troposphere (τ CH4,OH ) for each simulation, which largely determines the removal efficiency of methane from the atmosphere.We find τ CH4,OH in our present-day simulations decreases by 7 ± 3% relative to the preindustrial.In the warm-LGM scenario, τ CH4,OH changes little from the preindustrial (0.2 ± 6%), but increases by 35 ± 16% in the cold-LGM scenario.In our most likely representation of change across the glacial-interglacial period, with low fire and warm- LGM, τ CH4,OH is 4.5% higher at the LGM than in the preindustrial.We diagnose these changes below.
The value of τ CH4,OH is given by the global burden (Tg) divided by the total loss rate in the troposphere (Tg a −1 ), i.e., It therefore responds to convolved changes in the tropospheric OH and methane distributions and the strong temperature kinetic effect (2.4% K −1 at 273K) on the reaction rate constant, k CH4+OH (Lawrence et al., 2001;Sander et al., 2011;John et al., 2012).The temperature changes alone cause τ CH4,OH to increase by 4.2% in the warm-LGM scenario and 20% in the cold-LGM relative to the preindustrial, and decrease by 1.4% at the present day.At the LGM, this effect is partially countered by the shift of the methane maximum from the northern high latitudes to the tropics, which by itself leads to a 1.9% decrease in τ CH4,OH .Changes in the methane meridional distribution are negligible from the preindustrial to present, despite the large mean change.The kinetic temperature and methane distribution effects are then offset or enhanced by changes in OH responding to the factors outlined in Section 5.1.In the present day, the kinetic effect is augmented by the increase in tropospheric mean OH, and the lifetime decreases relative to the preindustrial.In the warm-LGM scenario, the net effect of the changes 1380 in temperature and methane distribution is small, and therefore changes in OH dominate τ CH4,OH .In the cold-LGM scenario, the temperature kinetic effect guarantees that τ CH4,OH increases relative to the preindustrial.
That τ CH4,OH remains nearly constant or increases at the

1385
LGM in our simulations puts a higher burden on sources in explaining the glacial-interglacial variability of atmospheric methane concentrations recorded in ice cores.In their model study, which did not include a paleo-stratosphere, Levine et al. (2011b) concluded that the effects of temperature and 1390 biogenic VOCs cancel out each other such that observed changes in ice-core methane must be driven primarily by sources.Our findings confirm those of Levine et al. (2011b), and extend them to include changes in overhead ozone and NO x , which together may even drive decreased OH at the

1395
LGM relative to the preindustrial.Assuming no large changes occurred in the minor loss mechanisms for methane (soils and stratospheric OH;IPCC, 2007), the source of methane emissions scales as its loss by OH in the troposphere (Table 7 and Fig. 9).We estimate that 1400 the methane loss frequency in the stratosphere changes less than 1% since the LGM in our simulations; the effect of reduced stratospheric water vapor at the LGM is largely offset by temperature increases.Our results imply that relative to the preindustrial, total methane emissions are 150% higher in 1405 the present, and are reduced by 42-65% at the LGM, with a 50% reduction our best estimate.These estimated reductions at the LGM are greater than the 29-42% decrease in wetland emissions simulated by the PMIP2 ensemble members (Weber et al., 2010), and much larger than the 16% and 23% 1410 decreases in natural methane emissions simulated by Kaplan et al. (2006) and Valdes et al. (2005) over the same period.However, we note that if the tropical fraction of total methane in LGM were to increase by 33% as implied by a Bolivian ice core -an extremely unlikely scenario (Section 2.4) -the 1415 implied methane source would in fact be consistent with the previous bottom-up estimates, as the integrated loss rate of methane (denominator of Eqn. 2) would increase despite the reductions in OH and temperature.

Discussion and Conclusions 1420
We present a stepwise, offline-coupled modeling framework for simulating paleo-oxidant levels since the Last Glacial Maximum.Our goal is to test within a 3D model framework the sensitivity of tropospheric oxidants over a range of likely forcings from climate-driven and anthropogenic 1425 changes.Archived meteorology from the GISS ModelE GCM is used to drive the BIOME4-TG global equilibrium terrestrial vegetation model and the LPJ-LMfire fire model.We also apply the ModelE meteorology to the GEOS-Chem CTM, together with the BIOME4-TG vegetation structure and LPJ-LMfire emissions.We then perform detailed simulations of tropospheric composition with an online-coupled linearized stratospheric chemistry scheme.
We simulate four different time slices with the GCM: the present day (ca.1990s), preindustrial (ca.1770s), and two possible realizations of the Last Glacial Maximum (LGM; 21ka), one significantly colder than the other reflecting uncertainty in the extent of tropical cooling at the LGM.The climate simulations are forced by prescribing greenhouse gas levels, orbital parameters, SSTs, sea ice extent, and ice sheet topography.Reductions in greenhouse gases drive decreases in tropospheric temperatures and water vapor, and decelerate the stratospheric residual circulation, consistent with earlier work (Braconnot et al., 2007a;Bartlein et al., 2011;Rind et al., 2009).Relative to preindustrial, tropical surface air temperatures are 1.4 • C colder in the warm-LGM scenario and 6.8 • C colder in the cold-LGM scenario; the preindustrial to present-day change is +0.5 • C. Reduced stratospheric circulation leads to an increase in tropical stratospheric ozone columns, which strongly affect tropospheric photolysis rates.
Tropospheric trace gas and aerosol emissions respond to climate, and we simulate reductions in most trace gas emissions in cooler climates.Terrestrial plant VOC emissions are influenced by both land cover and meteorology.A decrease in the present day relative to the preindustrial results from expansions in cropland.In the LGM, the cooler temperatures ultimately drive decreases in plant VOC emissions relative to the preindustrial.Soil microbial emissions of NO x also decrease with reduced temperatures and precipitation fluxes at the LGM.Lightning responds to the vertical extent of deep convection, and therefore decreases at the LGM, especially in the cold-LGM scenario.We also test a scenario in which we fix total lightning emissions.Estimates of biomass burning emissions in past atmospheres are highly uncertain; we test the extremes in the range of estimates at the LGM (10% and 200% of present) and the preindustrial (100% and 320% of present), using distributions of dry matter burned from the LPJ-LMfire model.
We find net reductions in all oxidants in past atmospheres relative to the present day, except for H 2 O 2 in the high fire preindustrial simulation, and OH in the fixed lightning LGM scenarios.OH is relatively well-buffered in comparison with the other oxidants.It is relatively insensitive to changes in anthropogenic and fire emissions, but sensitive to lightning.In our simulations, OH increases at the present day by +7.0 ± 4.3%, and shows little change at the LGM (0.5 ± 12%), both relative to the preindustrial.A linear relationship with J O3 qS N /(S 3/2 C ) explains 72% of the total variability seen in the mean OH burdens of our 11 simulations.J O3 and q largely regulate primary production of HO x , whereas S N /(S 3/2 C ) largely influences HO x partitioning between OH and RO 2 (but also primary production).
Most previous studies lacked realistic paleo-stratospheres and simulated larger increases in OH at the LGM (cf.Table 1).Our work suggests that OH changes were more 1485 tempered, and possibly even reduced at the LGM.Of the various emission sources, lightning is most capable of perturbing S N /(S 3/2 C ), as it makes up a relatively large fraction of S N (37±16% of S N in past atmospheres; 14% in the present day), and S N ≪ S C .Unlike anthropogenic or fire emissions, 1490 NO x from lightning also lacks any co-emissions of reactive carbon, and it is relatively uncorrelated with a competing emission source, as soil NO x emissions are with those of biogenic VOCs.Therefore, we conclude that the three factors OH is most sensitive to on glacial-interglacial time scales 1495 are overhead ozone, tropospheric water vapor, and lightning emissions.These factors should be given particular attention in future studies.
Ozone in our simulations increases from the preindustrial to the present by 24 ± 11%, and decreases by 12 ± 10% to 1500 the LGM.Our changes are consistent with prior studies but more tempered, likely reflecting the high bias in the source of ozone transported from the stratosphere.That source decreases in magnitude in the colder climates following the deceleration of the stratospheric circulation, but nevertheless 1505 becomes an increasingly large fraction of the total source.We extend to LGM conditions the finding that global mean ozone from 1850 through 2100 depends strongly on S N (Stevenson et al., 2006;Young et al., 2013).Ozone production efficiencies per unit NO x emitted increase in each successively 1510 colder atmosphere.
H 2 O 2 responds to changes in both climate and emissions, varying more strongly in its production pathways across the different scenarios than in its loss rates via wet deposition.We find that H 2 O 2 abundance is very sensitive to fire emis-1515 sions, as these emissions generate HO x .Ice-core H 2 O 2 may therefore act as a useful proxy for paleo fire emissions as well as for biogenic emissions, another HO x source.Nitrate radical is directly related to the total source of reactive nitrogen, S N (R 2 = 0.97).the LGM relative to the preindustrial, implying that the large glacial-interglacial variability seen in ice-core methane is predominantly driven by changing methane sources.Our results imply that methane emissions decreased by 50% at the LGM relative to the preindustrial, a greater change than reported in previous bottom-up estimates.Webb et al., 1997) scenarios.Both difference plots share a common colorbar for each variable.The green dots indicate regions with significant changes, as determined by a Student's t-test with degrees of freedom adjusted for autocorrelation (Wilks, 2011).The changes relative to the preindustrial in present-day surface temperatures and precipitation fluxes are small, and significant only at the poles and over small areas of the tropics, respectively (not shown).That study used the present day as its baseline, and we follow that convention here.We sample our simulations and the PMIP2 ensemble at the locations of the sediment core measurements.Fig. 5. Gas phase emissions used for each climate scenario.We test two different fire emission scenarios, a high fire scenario with past fire emissions from the LPJ-LMfire model (Pfeiffer et al., 2013), and a low fire scenario in which we scale the model output by sediment records of charcoal accumulation rates (Power et al., 2008).We also perform a fixed lightning suite of simulations in which we scale the total lightning NOx source to be constant in different climates.Fig. 8. Tropospheric mean mass-weighted oxidant burdens, calculated with the tropopause determined from the thermal lapse rate.For comparison with ACCMIP (Young et al., 2013), the horizontal yellow lines for ozone show the tropospheric burden calculated using a 150 ppb chemical tropopause.C ), where JO 3 is the mean tropospheric ozone photolysis frequency (s −1 ), q is the tropospheric mean specific humidity (g/kg), and SN and SC are the tropospheric sources of reactive nitrogen (mol N a −1 ) and reactive carbon (mol C a −1 ).The line shows a reduced major axis regression fit.SC is calculated as the sum of emissions of CO and NMVOCs, and an implied source of methane equal to its loss rate.Each molecule of isoprene yields an average 2.5 carbons that react within the gas-phase mechanism; the other NMVOCs have negligible global impact.b Mass-weighted mean stratospheric age of air (Waugh and Hall, 2002), calculated with an SF6-like inert tracer.

Fig. 3
also shows the changing distribution of precipita-465 tion in each past scenario (right column).The decreases in greenhouse gases and water vapor increase the tropical tropospheric lapse rate at the LGM, reducing the depth and intensity of tropical deep convection (e.g.,Sherwood et al., 2010) and weakening the Hadley circulation.With reduced 470 upwelling near the equator, precipitation decreases in the tropics and increases in the subtropics, especially in the warm pools in the CLIMAP reconstruction.
estimates: Valdes et al. (2005) simulated a 50% reduction in biogenic VOC emissions relative to the preindustrial, Kaplan et al. (2006) a 33% reduction, and Adams et al. (2001) a 29% reduction. 680 Fig. 12 shows the modeled spatiotemporal distribution of 985 OH concentrations in the preindustrial scenario and the absolute changes in the present-day and LGM scenarios relative to the preindustrial.Only results from the low fire scenarios are shown, as the other emission scenarios have similar spatiotemporal patterns.Most of the change from the prein-990 dustrial to present occurs in the boundary layer over the temperate latitudes of the northern hemisphere in summer, reflecting the rise of anthropogenic emissions.At the LGM, the changes in overhead ozone and water vapor decrease OH throughout most of the tropics, with maximum decreases oc-995 3. The lifetimes of the present-day simulation driven by ModelE (23 d) is comparable to that of the multimodel ensemble of Stevenson et al. (22 ± 2 d) 2006).

Fig. 2 .
Fig.2.Top panel shows preindustrial sea surface temperature (SST), as reconstructed by proxies(Rayner et al., 2003).Bottom two panels are the absolute changes in SST during the warm-LGM (bottom left; CLIMAP, 1976) and cold-LGM (bottom right;Webb et al., 1997) scenarios, relative to the preindustrial, with the LGM coastlines and ice sheets.The difference plots share a common colorbar.The change in SST from the preindustrial to the present day is relatively small (not shown).

Fig. 3 .
Fig. 3. Simulated surface air temperatures (left column; • C) and precipitation fluxes (right column; mm d −1 ).Top panels: Preindustrial conditions.Bottom panels, absolute change from preindustrial to warm-LGM (middle row; CLIMAP, 1976) and cold-LGM (bottom row;Webb et al., 1997) scenarios.Both difference plots share a common colorbar for each variable.The green dots indicate regions with significant changes, as determined by a Student's t-test with degrees of freedom adjusted for autocorrelation(Wilks, 2011).The changes relative to the preindustrial in present-day surface temperatures and precipitation fluxes are small, and significant only at the poles and over small areas of the tropics, respectively (not shown).

Fig. 4 .
Fig. 4. Evaluation of climate changes for (a) mean surface temperature and (b) zonal mean precipitation flux.The left panels compare our ModelE simulations to the range of zonal mean changes from the preindustrial in the Paleoclimate Multimodelling Intercomparison Project Phase II (PMIP2;Braconnot et al., 2007a).The latitudinal axis is weighted by relative zonal area.The right panels compare our simulation to the pollen-based climate reconstruction fromBartlein et al. (2011) using box-and-whisker plots; dots indicate outliers.That study used the present day as its baseline, and we follow that convention here.We sample our simulations and the PMIP2 ensemble at the locations of the sediment core measurements.

Fig. 7 .
Fig. 7. Spatial distribution of annual mean preindustrial emissions, and the present-day and LGM absolute changes in these emissions relative to the preindustrial for (a) isoprene from terrestrial plants, (b) NOx from lightning, and (c) CO from fires in the low fire scenario.The difference plots share the common color bars at right.

Fig. 9 .Fig. 10 .
Fig. 9. Global annually averaged chemical production and loss rates of OH in each simulation, by contributing reaction.The numbers along the bottom are tropospheric airmass-weighted mean OH.

Fig. 13 .
Fig. 13.Top panel: Budget of tropospheric ozone in each simulation.Production and loss rates are for the odd oxygen family (Ox ≡ O3 + O + NO2 + 2NO3 + organic nitrates + HNO4 + 3N2O5 + HNO3 + BrO + HOBr + BrNO2 + 2BrNO3).Transport from the stratosphere is inferred by mass balance in the troposphere with chemical production, loss and deposition.Tropopause calculated with thermal lapse rate definition.Diagonal hatching indicates net magnitude of ozone destroyed by BrO + hν → O3 + Br and Br + O3 → BrO + O2.Middle panel: Ox production efficiency per unit NOx emitted.Bottom panel: Lifetime of tropospheric ozone against chemical loss and deposition.

1520
Our simulations are broadly qualitatively consistent with ice-core records of ∆ 17 O in sulfate and nitrate at the LGM, and CO, HCHO, black carbon, and H 2 O 2 in the preindustrial.Comparison with black carbon, CO and H 2 O 2 preserved in ice cores indicate that fire emissions during the 1525 preindustrial and LGM likely fall closer to the lower end of the estimated range.Our simulations indicate that the HCHO/methane ratio cannot serve as a proxy for global OH variability, even from future tropical ice cores.Finally, our work provides greater certainty in the break- 1530down of the atmospheric methane budget at the LGM.Prior studies disagree as to the relative role of sources versus sinks in explaining the large glacial-interglacial variability in atmospheric methane concentrations recorded in ice cores.We simulate nearly constant or increased methane lifetimes at 1535